The NEXT LAST R600 Rumours & Speculation Thread [Archive] - Page 18

Jawed

02-May-2007, 23:31

And would you be so kind to compare vs G80 texturing power?
Too many variables and not enough known about the vertex buffer fetch, texture buffer fetch and filtered fetch of both architectures, I'm afraid.

For purely bilinear texture fetches, R600 has half the rate of G80 per clock (16 versus 32). In theory G80 should never fall below this rate.

For trilinear filtering, it's not possible to say what R600's rate is. There's every likelihood G80 is twice as fast. Same applies to anisotropic filtering.

Because these are unified architectures, the shaders can request vertex fetches, interleaved with filtered texture fetches. Additionally, non-filtered fetches may also be interleaved. I don't remember seeing any information comparing how G80 handles these 3 types of fetches (rusty memory alert). Consequently I don't know if they are all performed by the single group of TA and TF pipelines.

Whether they are or not, trying to theorise about their performance is bloody hard, because there are so many momentary factors at play (the interleaving I mentioned is just one).

If G80 does indeed send vertex and texture buffer requests through its TF pipelines, then it will indeed suffer in terms of absolute throughput. But the workload of VFs (at least in DX9 games) shouldn't be very dramatic, while non-filtered fetches have historically always clogged up the filtering units...

Jawed

fellix

02-May-2007, 23:33

Well, that said, a cache of any reasonable size is helpful, of course, considering the physical restrictions of the design. The other thing is how the cache array is implemented and organized.
Take a look at the Cell/B.E. (a.k.a. raytracing monster), although not everything there is in the category of caching. ;)

Mat3

02-May-2007, 23:38

However, considering AMD is really focused on GPGPU is 1 meg L2 really wasted in that arena?

Is it possible that R600 is trying to be too many things in too many market segments?

After all it appears that AMD/ATI is trying to position it as a top graphics performer, a top physics processor, and a top GPGPU unit.

So is it possible they expended a lot of transistors on things that would benefit GPGPU greatly but have low to minimal impact on 3D rendering?

Regards,
SB

A lot of what I've read of the R600 reminds me of what I've read about the physx PPU.

Jawed

02-May-2007, 23:54

You mean 40/80 cycles for propagation across the die and for all the MC related scheduling? According to Bob, the 40 would be too low, right? Where does the 200 come from?
40 is for a crossbar. 200 is for DDR memory. So the baseline in our little "ring bus is doomed" scenario is 240 clocks, apparently.

Anyway, if your numbers are reasonable, it's a 16% buffering penalty just for transport.
But the penalty for DDR is 5x bigger than that for a ring-bus compared with a crossbar (200 compared with 80-40=40). Sure, it's an extra mile, that 16%, but look how many miles have had to be travelled to build the GPU in the first place. With the architecture in place, surely sizing buffers is, relatively, trivial. The metrics for all these buffers still have to be in place, scaling for worst-case scenarios is still needed, etc.

Jawed

Galduta

03-May-2007, 00:01

when you provdie a screenshoot of R600XT with fps on it, someone here say: "hey, tester, one pic means meaningless. we want to see avg fps."

ok, when you provide a avg fps using fraps with FP16 HDR on, then the guy said:" What a magnificent dx8 speed."

Does DX8 support FP16 HDR? fine, I will not provide any 2900XT realgame test score anymore. I don't deserve this attitude.

1280*1024 HIGH SETTINGS HDR on

from the first step out of cellar, until reach the village

2007-05-02 12:04:07 - XR_3DA
Frames: 7027 - Time: 35704ms - Avg: 196.812 - Min: 93 - Max: 333

Mao , this score not is in DX 9, not is with HDR .... in one 8800 GTX the average is 30- 40 fps in this point - in DX 9 - . And sorry ;)

PSU-failure

03-May-2007, 00:10

I seem to remember that it has 1x6pin; 1x8pin connectors and that if you don't use 6&8 then you can't overclock.

You can use 6&6 but cannot overclock.

As far as I know, this could be complete BS...

If 8pin PEG power connector had something else than 2 more grounds (+12v/sense + ground for example, allowing a significantly lower voltage drop) it would sound ok, but as is I strongly doubt the card's driver could know what you're using.

Unknown Soldier

03-May-2007, 00:19

Maybe so, but until the Official release I guess we are in the dark.

US

Unknown Soldier

03-May-2007, 00:59

As far as I know, this could be complete BS...

If 8pin PEG power connector had something else than 2 more grounds (+12v/sense + ground for example, allowing a significantly lower voltage drop) it would sound ok, but as is I strongly doubt the card's driver could know what you're using.

For optimal performance and Overdrive overclocking you need to use one 2x3 pin PCIe power connector and one 2x4 pin PCIe power connector.

Of course this could be Fud
http://www.fudzilla.com/index.php?option=com_content&task=view&id=791&Itemid=1

US

Dalton Sleeper

03-May-2007, 01:20

I've got OCZ GameXStream 850w with 70A on 12v rail, you think it'll work with x-fire?

What PSU did the guys who benched use?

memberSince97

03-May-2007, 01:27

I've got OCZ GameXStream 850w with 70A on 12v rail, you think it'll work with x-fire?

What PSU did the guys who benched use?

I'm Sure...

Rangers

03-May-2007, 02:08

So it does seem like R600 is going to be heavily texture limited..sigh when is ATI(AMD) gonna wake up..

I hope they at least put 24 texture units into the R600 refresh.

One thing you knew Nvidia would do is make sure they have tons of texture throughput with G80..and sure enough they did, with 32 units.

Unknown Soldier

03-May-2007, 02:09

I've got OCZ GameXStream 850w with 70A on 12v rail, you think it'll work with x-fire?

What PSU did the guys who benched use?

Does it have a 8pin connector?

I was gonna get the 700W version then realised that I might be making a mistake and cancelled.
OCZ GameXStream 700W Power Supply

I do see that the 850W has two PCI-e 2x6 pin which is good, and I also see a 2x4/1x8pin(although they say it's for the CPU/Mobo?), so it looks good although people might argue differently.

http://www.ocia.net/reviews/ocz850/page4.shtml
http://www.cluboverclocker.com/reviews/power/ocz/gamexstream850/index.htm

I wonder if there is any chance for X-Fire (Dual) R600's working on the 700W version?

US

Kaotik

03-May-2007, 02:15

Unknown Soldier

03-May-2007, 02:21

Ye, was wondering .. so it won't do hey?

US

Dalton Sleeper

03-May-2007, 02:37

Well, my PSU isn't very old, hope they send a few cables in the x-fire package, if not where can you get such cables?

Unknown Soldier

03-May-2007, 02:46

http://www.tweaknews.net/reviews/thermaltake_toughpower_1000w_cable_management_powe r_supply/index3.php

Thermaltake's ToughPower 1000w supports 3x8pin and 3x6pin

US

INKster

03-May-2007, 03:06

A person i trust received a R600 XT sample for testing, and found out that he couldn't overclock the core more than 2 to 3MHz, even with improved air cooling (he's an experienced and unbiased source, so i tend to believe him).
It's not yet known why are the power and pin requirements are so restrictive, but he did test with two high-end PSU's (an OCZ GameXtream 850W and a PC Power & Cooling 850W).

Is the 8pin plug absolutely needed for even basic overclocking or is it an intentional driver lock out ?
If that's so, then the price advantage of the R600 XT might be negated by the need to purchase a new -and certainly expensive- power supply.
Where's the 8800 GTS 320MB competitor ? A R600 XL would sound terrific, if true.

Silent_Buddha

03-May-2007, 03:08

Mao , this score not is in DX 9, not is with HDR .... in one 8800 GTX the average is 30- 40 fps in this point - in DX 9 - . And sorry ;)

Actually with HDR ON but Dynamic Light OFF the 8800 GTX averages much higher than that. Techreport has the 8800 GTX at stock averaging 137.4 with HDR ON at 2560x1600 res.

http://techreport.com/reviews/2007q2/geforce-8800ultra/index.x?pg=4

So, the numbers that Mao posted seem extremely reasonable with HDR ON and Dynamic Lighting OFF.

Regards,
SB

Silent_Buddha

03-May-2007, 03:17

Does it have a 8pin connector?

I was gonna get the 700W version then realised that I might be making a mistake and cancelled.
OCZ GameXStream 700W Power Supply

I do see that the 850W has two PCI-e 2x6 pin which is good, and I also see a 2x4/1x8pin(although they say it's for the CPU/Mobo?), so it looks good although people might argue differently.

http://www.ocia.net/reviews/ocz850/page4.shtml
http://www.cluboverclocker.com/reviews/power/ocz/gamexstream850/index.htm

I wonder if there is any chance for X-Fire (Dual) R600's working on the 700W version?

US

You may not be able to overclock but it should run it just fine. 450 watts max for 2xR600's. Leaves roughly 250 for the rest of the system. Assuming you aren't running 20x15k RPM SCSI drives in your computer you should be more than fine. That is, of course, assuming you have a quality 700 watt powersupply which should be the case with OCZ, as FSP group manufactures most of their power supplies.

Or with a 2x6pin -> 1x8pin converter you should be able to overclock with the same powersupply as it has 4x6pin connectors. Of course, then you wouldn't be able to do Crossfire without using 1x4 pin -> 1x6pin converters. Boy would that end up being mess. Not to mention Overclocking a Crossfire setup on a 700 watt PSU might be pushing it. :D

Regards,
SB

Unknown Soldier

03-May-2007, 03:18

Is the 8pin plug absolutely needed for even basic overclocking or is it an intentional driver lock out ?

That's what mao5 said and it's seems intentional as he said drivers older than 8.361 worked with 2x6pins and the 8.361 drivers only worked with 1x6 & 1x8 .. so seems pretty intentional.

Hopefully ATI will sort it out with 2x6pins again in future driver releases.

keep on disclosering:

If you want to oc XT with 8.361 drv, your psu must have 6pin and 8pin line, or a 6pin-8pin converter line, otherwise, you can not oc 1MHz with 8.361 drv.

6-6 works on the drv below 8.361, but 6-6 lost its oc function on 8.361. I think it will also lost oc function above 8.361

US

LLB

03-May-2007, 03:27

Actually with HDR ON but Dynamic Light OFF the 8800 GTX averages much higher than that. Techreport has the 8800 GTX at stock averaging 137.4 with HDR ON at 2560x1600 res.

http://techreport.com/reviews/2007q2/geforce-8800ultra/index.x?pg=4

So, the numbers that Mao posted seem extremely reasonable with HDR ON and Dynamic Lighting OFF.

Regards,
SB

From the above link:

"For this test, we set the game to its "maximum" quality settings at 2560x1600 resolution. Unfortunately, the game crashed on both GeForce and Radeon cards when we set it to use dynamic lighting, so we had to stick with its static lighting option."

LLB

Silent_Buddha

03-May-2007, 03:54

From the above link:

"For this test, we set the game to its "maximum" quality settings at 2560x1600 resolution. Unfortunately, the game crashed on both GeForce and Radeon cards when we set it to use dynamic lighting, so we had to stick with its static lighting option."

LLB

Which is basically just repeating what I said. :)

Regards,
SB

silent_guy

03-May-2007, 05:38

40 is for a crossbar. 200 is for DDR memory. So the baseline in our little "ring bus is doomed" scenario is 240 clocks, apparently.
Well, Bob was hinting at 100 cycles to cross the die. If that's the case, the difference is 100 cycles. Sure it's trivial to increase buffers and a ring is not the harbinger of Doom. Just more expensive, especially taking into account the overdimensioning, and reducing perf/W, which hasn't exactly been a forte of ATI lately.

Anyway, I guess I made my point. :wink:

Silent_Buddha

03-May-2007, 06:46

_xxx_

03-May-2007, 07:38

I've got OCZ GameXStream 850w with 70A on 12v rail, you think it'll work with x-fire?

What PSU did the guys who benched use?

That 12V rail will never, ever deliver 70A, not even in their dreams.

(70 x 12 = 840W, so that would leave 10W for the rest of the machine, which is of course bonkers)

Arnold Beckenbauer

03-May-2007, 07:42

Which is basically just repeating what I said. :)

Regards,
SB

No.:smile:
Static lighting: better than HL2, nice shadows.
Dynamic lighting: OMG, fp16 hdrr, parallax mapping, great shadows...
You cannot turn off dynamic lighting without turning off hdrr.
http://www.tweakguides.com/STALKER_5.html

silent_guy

03-May-2007, 08:06

So, then all of ATI's engineers were blowing up our arses by claiming the Ringbus scales better with high speed and a wider bus?
No, they weren't. Because it's true that it makes it easier to layout and scales well. So if you think up front that a crossbar become too much of a hassle during the chip backend phase, it's the right thing to do. It just costs area that could have been used from something that increases performance or save on power.

Galduta

03-May-2007, 08:10

Actually with HDR ON but Dynamic Light OFF the 8800 GTX averages much higher than that. Techreport has the 8800 GTX at stock averaging 137.4 with HDR ON at 2560x1600 res.

http://techreport.com/reviews/2007q2/geforce-8800ultra/index.x?pg=4

So, the numbers that Mao posted seem extremely reasonable with HDR ON and Dynamic Lighting OFF.

Regards,
SB

Silent_Buddha, the techreport (http://techreport.com/reviews/2007q2/geforce-8800ultra/index.x?pg=4) bench is wrong . Not is posible - with DX 9 render - . You play Stalker ? Please you can capture a screen at 98 fps - the minimun in techreport (http://techreport.com/reviews/2007q2/geforce-8800ultra/index.x?pg=4) .Somebody can do it?

Silent_Buddha

03-May-2007, 08:42

Yeah, probably time I got off my lazy butt and installed Stalker. Bought almost a month ago and still haven't gotten around to playing it. Although it sounds like it doesn't support AA very good so may not be playing for long.

If TR did do that DX8 path, it's absolutely dishonest that they used a DX9 screenshot on that page.

Regards,
SB

03-May-2007, 08:43

Just got word that the HIS HD2900XT will run at 750Mhz core and 850Mhz mem.

Bob

03-May-2007, 08:47

So, then all of ATI's engineers were blowing up our arses by claiming the Ringbus scales better with high speed and a wider bus?
It's also not being scaled all that well. If it the ring bus had 21 stops (16 memory partitions of 32-bits, 4 texture/shader clients, PCI-E), you'd have a point. But I count just five stops. Might as well have used point-to-point in that case. Wiring layout is just as complicated, and you get to save on scheduling logic and buffering.

Ah well, maybe in R650.

Domell

03-May-2007, 08:56

Rys

03-May-2007, 09:03

It's also not being scaled all that well. If it the ring bus had 21 stops (16 memory partitions of 32-bits, 4 texture/shader clients, PCI-E), you'd have a point. But I count just five stops. Might as well have used point-to-point in that case. Wiring layout is just as complicated, and you get to save on scheduling logic and buffering.

Ah well, maybe in R650.
When you see a diagram of the controller, you only see the DRAM stops and possibly PCIe. There are actually dozens of active clients after that.

EDIT: I say that because you mix in clients to the controller in your description, just in case it's not clear why I said so :smile:

Pete

03-May-2007, 09:09

There are results in STALKER

Radeon HD2900xt - 42fps (http://filexoom.com/files/2006/10/15/25426/My%20Documents/Pictures/20070502_9e34f00a8b1f5577bb65WUzbsDRJTWEw.jpg)
GF8800GTS 320MB - 59fps (http://filexoom.com/files/2006/10/15/25426/My%20Documents/Pictures/XR_3DA%202007-05-02%2021-45-36-51.jpg)

Here is the link (http://66.249.91.104/translate_c?hl=en&ie=UTF-8&oe=UTF-8&langpair=zh-CN%7Cen&u=http://we.pcinlife.com/thread-758765-1-1.html&prev=/language_tools)

If it is truth then is no good for ATI

And an 8800GTX at 41fps. (http://forum.beyond3d.com/showpost.php?p=980309&postcount=4144) I can't see square one, I'm so dizzy. And still the weird washed-out screen for R600.

Galduta

03-May-2007, 09:14

Yeahhh but in my 8800 GTX the score is 41 ;) . What is the CPU ? The settings, the resolution of the 8800 GTS. This not is the correct method . Stalker have a bench . This is not serious .

Rys

03-May-2007, 09:20

A whole massive chunk of the backend of this thread hasn't been serious, when it comes to possible R600 performance. The semi-random Stalker shots have to stop, because they're not being collected into anything useful and they're serving no honestly good purpose at this point. A moratorium on that kind of post will hopefully reduce signal:noise, so it'd be appreciated if that was observed.

And while I love snakes as much as the next guy, please......

silent_guy

03-May-2007, 09:21

When you see a diagram of the controller, you only see the DRAM stops and possibly PCIe. There are actually dozens of active clients after that.

EDIT: I say that because you mix in clients to the controller in your description, just in case it's not clear why I said so :smile:

Are you saying that there are much more ring stops that are reserved to clients? (Even more increased latency, traffic entropy, and scheduling difficulty, BTW)

That would go somewhat against the info on the slides, where the ring stop block diagram shows clearly that it is responsible to select and arbitrate which client is going to inject onto the ring.

I have high hopes for your article on the 14th! :wink:

Rys

03-May-2007, 09:26

Are you saying that there are much more ring stops that are reserved to clients? (Even more increased latency, traffic entropy, and scheduling difficulty, BTW)

That would go somewhat against the info on the slides, where the ring stop block diagram shows clearly that it is responsible to select and arbitrate which client is going to inject onto the ring.

I have high hopes for your article on the 14th! :wink:
I was just checking what Bob meant by 'stop', effectively, since he mixed in controller clients when talking about a scaled version of the controller might be like, in terms of stop count.

And here's hoping the article goes well :smile:

Martin Eddy

03-May-2007, 09:34

How many amps does it have on 12V? That's all that matters there.
And yes, those on the first link, where there's 2x4 pin stuff, they are for mobo, not video cards.
I'm not sure if there's any PSU's out there yet with 8pin PCIe plug

My 6 month old Antec True Power 480 has 2 of each of 6 pin and 8 pin plugs. When I bought it I was wondering what the hell the 8 pin plugs were for. :lol:

_xxx_

03-May-2007, 09:55

My 6 month old Antec True Power 480 has 2 of each of 6 pin and 8 pin plugs. When I bought it I was wondering what the hell the 8 pin plugs were for. :lol:

Well in a 480W PSU, I'd say for nothing useful at all.

Dalton Sleeper

03-May-2007, 09:58

Hope that the card doesn't get 100 degrees like fudzilla wrote, alot of heat and noise for crossfire users.

nicolasb

03-May-2007, 10:17

Indeed, Fudo is busy dissing the R600 heat output and cooling system today. Some linkage for you....

Hot:

http://www.fudzilla.com/index.php?option=com_content&task=view&id=803&Itemid=1

...this card has great potential to burn your fingers.

A chap informed us that after the card was powered down for two minutes he still couldn't touch the back of the cooler. A separate source reported that the card runs at about 100 degrees Celsius which is ridiculously hot.

So an R600 can at least heat your apartment and don't even think about putting one in a plastic case as, it might melt it....

Noisy:

http://www.fudzilla.com/index.php?option=com_content&task=view&id=800&Itemid=1

...The trouble is that in 3D the card is noisier than the Geforce 8800 GTX. The Radeon HD 2900XT cooler has only two heat pipes and it is not as massive as the G80 GTX ones and this probably makes the fan spin at higher speed and thus making the unnecessary noise....

Gelanin

03-May-2007, 10:50

Pressure

03-May-2007, 10:53

Indeed, Fudo is busy dissing the R600 heat output and cooling system today. Some linkage for you....

Hot:

http://www.fudzilla.com/index.php?option=com_content&task=view&id=803&Itemid=1

Noisy:

http://www.fudzilla.com/index.php?option=com_content&task=view&id=800&Itemid=1

Yeah, must be a slow news day...

neliz

03-May-2007, 10:54

Indeed, Fudo is busy dissing the R600 heat output and cooling system today. Some linkage for you....

Hot:

http://www.fudzilla.com/index.php?option=com_content&task=view&id=803&Itemid=1

Again.. fud.. there is a plastic cover to tunnel the air, did the guy run the card without the plastic cover (to touch it) and inherently.. without fan?

bludragon

03-May-2007, 10:57

To me, it looks like the best way to figure out the performance of the 2900xt is to compare the benches here:

http://forums.vr-zone.com/showthread.php?t=148492

to x1950xtx vs 8800 gtx benches, e.g. here:

http://www.anandtech.com/video/showdoc.aspx?i=2979&p=4

From that, I would say that the 2900xt is almost at the performance of 8800 gtx in hdr / shader intesive titles, but a little slower in other titles. Overall it is probably safe to say it is slightly slower than the gtx, but faster than the gts.

If it's priced similarly to the gts, then it will be the better buy, of course, the gtx will come down in price over time, but it may take a little while for this to happen - 768 vs 512 meg of memory might mean worse margins for nVidia, although you have to believe that each gpu will cost them less than it does for amd, as they've had time to improve the yields.

The interesting part (to me) will be how the midrange cards compare. nVidia now has the mindshare of its top-end cards being the fastest, and it looks very unlikely that AMD can disprove that, so AMD marketers better hope their midrange is faster, as even then it will be a tough sell.

vertex_shader

03-May-2007, 11:39

Indeed, Fudo is busy dissing the R600 heat output and cooling system today. Some linkage for you....

Hot:

http://www.fudzilla.com/index.php?option=com_content&task=view&id=803&Itemid=1

Noisy:

http://www.fudzilla.com/index.php?option=com_content&task=view&id=800&Itemid=1

Now everyone know why the flame design is on the card, the card will burn :lol:

Chalnoth

03-May-2007, 11:41

Well, if this is an issue, it'll be immediately obvious with the initial reviews. I wouldn't worry about it now myself. I'm a bit more interested in how the card renders.

Unknown Soldier

03-May-2007, 12:01

http://www.tweaknews.net/reviews/thermaltake_toughpower_1000w_cable_management_powe r_supply/index3.php

Thermaltake's ToughPower 1000w supports 3x8pin and 3x6pin

US

Found another PSU - Cooler Master Real Power Pro RS-A00-EMBA (1000W)
http://www.hexus.net/content/item.php?item=7639
Cooler Master Real Power Pro RS-A00-EMBA (1000W)
http://www.xbitlabs.com/articles/other/display/1000w-psu-roundup_3.html

4x6pin PCI-e
2x8pin PCI-e

and another - Ultra X3 1000W
http://enthusiast.hardocp.com/article.html?art=MTMxNywyLCxoZW50aHVzaWFzdA==

4x6pin PCI-e
2x8pin PCI-e

PC Power & Cooling Turbo-Cool 1KW-SR (1000W)
http://www.xbitlabs.com/articles/other/display/1000w-psu-roundup_7.html

4x6pin PCI-e
2x8pin PCI-e

SilverStone Olympia OP1000 (1000W)
http://www.xbitlabs.com/articles/other/display/1000w-psu-roundup_9.html

4x6pin PCI-e
2x8pin PCI-e

US

flopper

03-May-2007, 12:07

at 1600x1200 one card is more than enough.

Its all what is gonna happen with dx10, if crysis, alanwake and other games pushes physics then I guess a board with 3 pci-exp slots might be sumtin to go for, but as far, its a tech that is a bit off yet.

300w overclocked is just hilarious.
lucky I use watercooling ;)

vertex_shader

03-May-2007, 12:24

HIS english webpage back online, hd2900xt removed :smile:

compres

03-May-2007, 12:44

So, then all of ATI's engineers were blowing up our arses by claiming the Ringbus scales better with high speed and a wider bus?

I'm a bit puzzled as to why they implemented it then if there is absolutely no benefit considering they already had a crossbar architecture that presumably worked well.

Regards,
SB

The people here commenting on the ring bus are really oversimplifying the situation. Both topologies have their advantage, and to know what is best for r600 for example, we simply don't have the details and probably never will.

Most likely than not, the whole r600 chip is designed around the ring bus, making good use of it. I just checked the max latency estimates and they seem quite reasonable for a worst case scenario of going around the whole chip, this situation I would spect to be avoided most of the time. Even if they did nothing to avoid the worst case we could assume 33.333% average of this kind of xfer to happen, so it would be more realistic to assume a 53nS delay(.33*80+.66*40)nS. And as I said, they likely have means to reducing the worst case scenario even more with a whole network designed to work with this.

Since this is a ring bus and not a token bus all channels can be used at the same time given enough buffering, and I bet they made sure of this, so the utilization should be hight.

Take into consideration it is a 1024 in both directions, 2048 bit wide bus. Assuming it is close to what those diagrams show in the slides and a 750mhz clock, that is close to 200GB/s internal BW per pair of point to point connections between neighbor ring stops. In the best case, if every stop is "talking" only to its neighbor, the total bandwidth would be 4 times of that, or 800GB/s(overestimate).

Yeah well, hopefully you get my point. If not basically ATI engineers are likely to be smarter than some people might imply...

PS: By scaling you have to decide if in performance or hardware complexity. For example the crossbar scales with performance well, but on hardware complexity it is not. So it is not a one size fits all.

Edit:
We have contradicting internal widths:
http://www.forum-3dcenter.org/vbulletin/showpost.php?p=5450882&postcount=510
http://forums.vr-zone.com/showthread.php?t=148492

nicolasb

03-May-2007, 13:15

Take into consideration it is a 1024 in both directions, 2048 bit wide bus. Are you sure about that? I thought it was 1024-bit total - 512 in each direction.

Unknown Soldier

03-May-2007, 13:17

Are you sure about that? I thought it was 1024-bit total - 512 in each direction.

:D yes, that's how I understand it.

US

Gelanin

03-May-2007, 13:18

Are you sure about that? I thought it was 1024-bit total - 512 in each direction.

Seems like so from the presentation slides on vrzone.

compres

03-May-2007, 13:44

Arrrrg!!!! Not fun to search on such a long thread...

I think the VR Zone slides look more legit, but this is why I got confused on the internal width of the ring bus:

http://www.forum-3dcenter.org/vbulletin/showpost.php?p=5450882&postcount=510

Edit:
http://forums.vr-zone.com/showthread.php?t=148492
Perhaps the chineese interpreted each magenta colored bus line in the vr-zone/amd slide as a 512bit channel? There is room for confusion there...

EasyRaider

03-May-2007, 15:13

So, you're discussing >700 W PSUs...

A typical C2D + 8800GTX system tops out below 250 W in games. With C2Q + 8800GTX SLI, a quick estimate gives 400W. Add a big stack of drives and some heavy overclocking, and it'd still be hard to get much higher than 500 W.

I say get a Corsair HX620, best PSU money can buy as far as I'm concerned. If an XT Crossfire setup turns out to be too demanding, the blame really should be directed at ATI.

IbaneZ

03-May-2007, 15:17

As some of our readers actually didn't believe that their bellowed R600XT might need a 750W or more here is a proof. AMD clearly states you need a 750W or better power for this hot beast.

http://www.fudzilla.com/index.php?option=com_content&task=view&id=810&Itemid=1

He has posted a pic now.

I still think it's fake or BS. :grin:

Skinner

03-May-2007, 15:23

http://www.fudzilla.com/index.php?option=com_content&task=view&id=810&Itemid=1

He has posted a pic now.

I still think it's fake or BS. :grin:

And I wanted to go 2900CF.. :D

I don't believe this counts for quality psu's

EasyRaider

03-May-2007, 15:27

That's ridiculous. Many 8800GTX systems run perfectly stable with less than half that wattage.

Suflex

03-May-2007, 15:29

http://www.fudzilla.com/index.php?option=com_content&task=view&id=810&Itemid=1

He has posted a pic now.

I still think it's fake or BS. :grin:

Unfortunately the document looks genuine to me, unless he had intentionally tampered with it.

Sound_Card

03-May-2007, 15:35

http://www.theinquirer.net/default.aspx?article=39370

HDMI dongle revealed.

PSU-failure

03-May-2007, 15:36

http://www.fudzilla.com/index.php?option=com_content&task=view&id=810&Itemid=1

He has posted a pic now.

I still think it's fake or BS. :grin:
I remember sucessfull operation of R580 boards on 300-350watt PSUs though, while it's stated to require a 450watt one...

As ever, stating that it requires a 750watt PSU simply means they don't want you to stick your shiny HD2900XT with a 550watt LC Power and such ridiculously overestimated PSUs.

By the way, the GF8800 Ultra/680i SLI/x6800 tops out at roughly 335watt drain at the wall.

nicolasb

03-May-2007, 15:40

There is a massive difference between "ATI recommends a 750W power supply" and "a system using R600 will consume 750W". Those two statements are not remotely similar in meaning.

The actual PSU requirement will be a certain number of amps drawn from the 12V rail. But the average idiot consumer doesn't know what an amp or a 12V rail is, all he knows about a PSU is "how many Watts is it?" So ATI has no choice but to recommend a PSU in those terms.

ATI is not saying "you need 750W". What they're saying is "if your power supply is officially 750W, then your system will be fine, even if it's the worst, crappiest, least stable and failing-to-come-anywhere-its-paper-specifications so-called 750W PSU in the world." So absolutely any PSU officially rated at 750W will easily be able to supply enough amps on the 12V rail to avoid any problems.

What that means in practice is that a high-quality power supply will be more than adequate, even if it's far less than 750W. I wouldn't anticipate problems with a good quality 450W PSU (unless you want to go Crossfire).

EasyRaider

03-May-2007, 15:43

There is a massive difference between "ATI recommends a 750W power supply" and "a system using R600 will consume 750W". Those two statements are not remotely similar in meaning.

The actual PSU requirement will be a certain number of amps drawn from the 12V rail. But the average idiot consumer doesn't know what an amp or a 12V rail is, all he knows about a PSU is "how many Watts is it?" So ATI has no choice but to recommend a PSU in those terms.

ATI is not saying "you need 750W". What they're saying is "if your power supply is officially 750W, then your system will be fine, even if it's the worst, crappiest, least stable and failing-to-come-anywhere-its-paper-specifications so-called 750W PSU in the world." So absolutely any PSU officially rated at 750W will easily be able to supply enough amps on the 12V rail to avoid any problems.

What that means in practice is that a high-quality 450W power supply will be more than adequate (for a single card).
Yes, of course. But it's still a huge disadvantage.

If true, a lot of people will stay away just to be able to use a cheaper PSU. And for the silence crowd, NVidia will once again be the brand of choice.

PSU-failure

03-May-2007, 15:43

Unfortunately the document looks genuine to me, unless he had intentionally tampered with it.

Genuine or not, it's just a way to avoid PMPO watts marketed PSUs to be used with their cards.

I've recently seen they now even sell those highly overestimated PSUs for up to 680watt...

Unknown Soldier

03-May-2007, 15:46

Hmm.. could've sworn the Dongles were revealed before with the Pics showing the whole R600/RV6xx range. ;)

US

vertex_shader

03-May-2007, 15:47

http://www.fudzilla.com/index.php?option=com_content&task=view&id=810&Itemid=1

He has posted a pic now.

I still think it's fake or BS. :grin:

Official recommendation always high, AMD and NV not want kill the PSU makers :wink:

fellix

03-May-2007, 15:50

AMD and NV not want kill the PSU makers :wink:
Certainly not the maker, but the PSU itself. :lol:

Suflex

03-May-2007, 15:56

Genuine or not, it's just a way to avoid PMPO watts marketed PSUs to be used with their cards.

I've recently seen they now even sell those highly overestimated PSUs for up to 680watt...

Yes you're correct. It just doesn't "look" good.

EasyRaider

03-May-2007, 15:59

I think it would have been better to just state the wattage for the card alone, and point to a guide for how to calculate total needed. Let people figure it out by themselves, or get someone knowledgeable to help them. A 750 W recommendation will needlessly scare away some potential buyers.

(edit)
Well, I guess it doesn't matter. It's a disaster no matter what.

Dalton Sleeper

03-May-2007, 16:15

I just got 2*6pin (PCI-e) and a few molex left. Now I need 2*molex-to-8pin adapters, will this work?

http://www.frozencpu.com/products/3200/psu-147/4-pin_Molex_to_8-pin_12V_Power_Connector.html?tl=g11c28s91&id=ibQrxw2R

Kaotik

03-May-2007, 16:16

http://www.theinquirer.net/default.aspx?article=39370

HDMI dongle revealed.

Unless you ignore the ones showed earlier attached to cards, of course

nicolasb

03-May-2007, 16:16

I've recently seen they now even sell those highly overestimated PSUs for up to 680watt...That in itself might be a reason for recommending 750W, of course. Most of the truly crap PSUs are rated at <750W, so insisting on 750W means you avoid the worst. :)

nicolasb

03-May-2007, 16:17

I just got 2*6pin (PCI-e) and a few molex left. Now I need 2*molex-to-8pin adapters, will this work?That sounds like a Very Bad Idea (tm). Just use your two 6-pin connectors and you'll be fine. (Unless you have a pathological need to overclock).

Dalton Sleeper

03-May-2007, 16:20

That sounds like a Very Bad Idea (tm). Just use your two 6-pin connectors and you'll be fine. (Unless you have a pathological need to overclock).

May I ask why it's a Very bad Idea? If it's recommended with both why not do so?

Pressure

03-May-2007, 16:32

Hmm.. could've sworn the Dongles were revealed before with the Pics showing the whole R600/RV6xx range. ;)

US

Indeed, didn't come as a surprise to me.

Arnold Beckenbauer

03-May-2007, 16:38

http://www.fudzilla.com/index.php?option=com_content&task=view&id=810&Itemid=1

He has posted a pic now.

I still think it's fake or BS. :grin:

Operating System: ... Windows Vista ready
Where can I get an operating system, that is Vista ready? :wink:
Next one:
For optimal performance, we recommend requires ...
:roll:

Fornowagain

03-May-2007, 16:44

fellix

03-May-2007, 16:45

Where can I get an operating system, that is Vista ready? :wink:
Er, WinXP with DX10 (somehow). :wink:

silent_guy

03-May-2007, 17:01

Take into consideration it is a 1024 in both directions, 2048 bit wide bus. Assuming it is close to what those diagrams show in the slides and a 750mhz clock, that is close to 200GB/s internal BW per pair of point to point connections between neighbor ring stops. In the best case, if every stop is "talking" only to its neighbor, the total bandwidth would be 4 times of that, or 800GB/s(overestimate).
What you're basically suggesting is that the ring is not only used for memory traffic, but also for other intra-chip communication. It's possible. But the closer you get to the theoretical maximum BW of a ring, the harder it becomes to avoid freak behavior. (Including the possibility of a deadlock.)
If you have a few hours of free time, you should try once to model a ring with a scripting language and see how it behaves as you increase its utilization. The spread in latency goes through the roof.

Yeah well, hopefully you get my point. If not basically ATI engineers are likely to be smarter than some people might imply...
If they're using it for other stuff than just MC traffic, they need all the smarts they can get!

Bob

03-May-2007, 17:02

When you see a diagram of the controller, you only see the DRAM stops and possibly PCIe. There are actually dozens of active clients after that.
I don't know about you, but I see crossbars between those clients.

compres

03-May-2007, 17:25

What you're basically suggesting is that the ring is not only used for memory traffic, but also for other intra-chip communication. It's possible. But the closer you get to the theoretical maximum BW of a ring, the harder it becomes to avoid freak behavior. (Including the possibility of a deadlock.)
If you have a few hours of free time, you should try once to model a ring with a scripting language and see how it behaves as you increase its utilization. The spread in latency goes through the roof.

I did say in that post that there are advantages and disadvantages in either topology. I said nothing about what goes through the ring, beats me.

Maybe I should put it in other words:

This chip is sufficiently complex and can have features that help it overcome the issues of a ring topology.

If they're using it for other stuff than just MC traffic, they need all the smarts they can get!

There is no way with the information you know right now(which amount to close to nothing) to just call this design decision as plain bad...

Maybe I misinterpret you, if so sorry and would you care to clarify your point?

Fornowagain

03-May-2007, 17:32

About the PSU, here (http://forums.overclockers.co.uk/showpost.php?p=9247744&postcount=18)

Hi there

Our system had an Enermax 1000W PSU.

I have the ATI Tech/Fact sheet here so upto 200W is requirement at stock clocks with 2x6pin connectors.

If overclocking then you need 1x8pin and 1x6pin connected and then it consumes upto 225W.

I hope ATI don't bash me for stating this but I don't want misleading information such as that going around.

Fact is a decent quality 400-500W PSU will run of these cards fine if your running a Core 2 DUO system or Athlon X2 based system. :)

Arty

03-May-2007, 17:49

In other news, AMD released (http://www.techpowerup.com/?30555) a AMD GPU Clock Tool to reviewers. It works with RV630, RV610 and all previous AMD products too.

And a couple of posts from Shamino..

Confirming (http://forums.vr-zone.com/showpost.php?p=3653059&postcount=10) the MSRP
The price tag should surprise :)

On Fuad's story (http://forums.vr-zone.com/showpost.php?p=3652213&postcount=39) of R600 cooler reaching 100C
if everyone belives whatever they read without checking the links, then they dserve to be mislead since they are alrdy misled to begin with.

Well so you say they're dumb , then so be it, you dont lose anything.

Geeforcer

03-May-2007, 17:50

anaqer

03-May-2007, 17:57

Remember how everybody was freaking out big time about the dual power connectors of the 6800U and the OMGLOLHUEG 480W official PSU recommendation? This is basically round two... sure, the numbers are different but what it boils down to seems the same to me - if you're even *thinking* about dropping one of these in your box, chances are, you're current PSU is powerful enough already.

So... what he said.

PSU-failure

03-May-2007, 18:16

The PSUgate is a non-story. I think nicolasb nailed it - AMD is just trying to cover all the no-name, 2-lb PSUs out there. I seem to remember a 100W+ gap between PSU Nvidia recommended for 6800 and the one most people could run it with. NBD.
In fact, NV40 is quite hungry...

G70 is specified to require a bigger PSU but draws about the same (GTX vs Ultra), the same goes for the G71.

All this BS about PSU requirement wouldn't have been a problem if some fools didn't marketed the "German engineered" and such overestimated and even underprotected PSUs. (some don't even meet european standards, but given no one complains about that they don't worry about their self-claimed compliance)

Fornowagain

03-May-2007, 18:18

Link (http://forums.overclockers.co.uk/showpost.php?p=9247924&postcount=27)

Hi there

Also regarding this 100c nonsence.

This is making my blood boil now.

AMD/ATI let me tell these guys the real facts about this card because the utter crap on the internet is literally that, CRAP!

The card is quiet, in fact its more or less silent even after a good gaming session. It also stays in the 70c-80c under full load whilst overclocked and still remains quiet.

At stock speeds its generally around the 60c-70c mark and the heatsink does not burn to the touch like an 8800GTX or X1950 XTX, so the card can be handled even after a mad benchmarking session.

AnarchX

03-May-2007, 18:18

Real HD 2900XT temperatures?
- 28°C room
- 58°C idle
- 60°C 3sec after benchmark
- 52°C case or card-ambient after hours of oc-tests
http://we.pcinlife.com/viewthread.php?tid=759069&page=2&fromuid=167461#pid14366626

bdotobdot2

03-May-2007, 18:38

Dalton Sleeper

03-May-2007, 18:46

According to this page (a few days old) AMD will skip the 80nm, quite confusing when everyone write different things:shock: .

http://www.digitimes.com/mobos/a20070423PD213.html

IbaneZ

03-May-2007, 19:29

http://www.xtremesystems.org/forums/showthread.php?t=142955

I'll keep an eye on this thread over at xs.

The "OC masters" have their cards now, and hopefully one of them will get drunk and talk to much. :lol:

Edit: Wee! My 600th post. I'm such a spammer.

EasyRaider

03-May-2007, 19:42

The PSUgate is a non-story.
Depends on your POV. For noise-conscious people like me, energy efficiency is a big deal. And it looks like NVidia wins big time in that department.

Silent_Buddha

03-May-2007, 19:53

What's there to complain about, the only news from people that actually have the card and are under NDA state that it's pretty much as quiet as a 8800 GTX idle and under load.

Myself, I'm planning on running one in my shuttle case with a 450 watt PSU.

I seriously wouldn't be surprised if this could run on a modest system with a core 2 duo, 1 HD, and a high quality 350 watt PSU, assuming they even sell PSU's that low with 2x6 pin connectors.

Unless of course the Vista 64 drivers suck in which case it might still be a toss up between this and a 8800 GTX. Still a 399 USD price point is mighty attractive.

Regards,
SB

vertex_shader

03-May-2007, 19:56

EasyRaider

03-May-2007, 19:56

if you're even *thinking* about dropping one of these in your box, chances are, you're current PSU is powerful enough already.
Still, if the card can really draw 200 W or more, there will be a very significant number of potential buyers with PSUs on the weak side. The question is not if AMD will lose sales, it's how many.

cadaveca

03-May-2007, 20:02

It's also part of the reason the card is so cheap. You can buy new psu AND HD2900XT, for the cost of the competitor's product.:lol: I understand that part of the reason 8800U is so expensive is the high-grade ram, and the AMT of it, but it then also strikes me as good reason for there to NOT be an XTX card w/ 1gb of ram, or AMD would not be able to release the card for so cheap. The GDDR used on these AMD cards is not so expensive as the memory on nV's cards, but 1GB of GDDR4 is a whole different story!

Geo

03-May-2007, 20:04

The question is not if AMD will lose sales, it's how many.

Sure, that's not that unusual with high-end parts. The length of 8800 GTX no doubt lost some sales (or at least moved them to GTS) as well.

EasyRaider

03-May-2007, 20:06

What's there to complain about, the only news from people that actually have the card and are under NDA state that it's pretty much as quiet as a 8800 GTX idle and under load.
It's not just about the cooler. How much heat it dumps in the case and how much heat is generated in the PSU due to load are both very important considerations for a quiet system.
Myself, I'm planning on running one in my shuttle case with a 450 watt PSU.

I seriously wouldn't be surprised if this could run on a modest system with a core 2 duo, 1 HD, and a high quality 350 watt PSU, assuming they even sell PSU's that low with 2x6 pin connectors.
Total wattage is only a small piece of the picture. 12V rails are the first spec to look at these days. But I agree, 350 W should be fine. You'll just need a 500W PSU anyway for the connectors. (edit)More importantly, it's hard to find a 350 W PSU with sufficient 12V capacity.

Silent_Buddha

03-May-2007, 20:31

It's not just about the cooler. How much heat it dumps in the case and how much heat is generated in the PSU due to load are both very important considerations for a quiet system.

True, but according to Gibbo (taken with a large dosage of salt) AMD told him the R600 draws about 200 watts in normal operation and about 225ish when overclocked. Again taken with a large grain of salt. He could just be saying that to drum up demand for R600 in preparation of being able to sell them.

So, is another 20-30 watts over and above an 8800 GTX going to suddenly cause a powersupply to heat up much more with a R600 vs. a G80?

And as far as I can tell, most of the heat is vented out the back of the case. Although the memory chips on the back of the card would dump some into the case.

And anyone looking for a truly silent HTPC isn't really going to be using an HD 2900 XT or a 8800 GTX anyways. Myself I'm looking at either a passively cooled 8500 from Nvidia or a passively cooled RV610 from ATI for my HTPC. Depending on which one handles HD video decoding better (H.264 AND VC-1).

Regards,
SB

chavvdarrr

03-May-2007, 20:46

I just wonder if/when R600 comes and it really needs these rumoured 225W ...
will all the people who wrote again and again that "theinq" and "Fudo" were writing FUD about how power hungry the card is ("220W? no way!"), will these people write that they were wrong?

Sheeesh, now the fact that XT with 512MB can need 225W is almost officially confirmed... I don't wanna touch the XTX ground and imagine how much it may need.

No doubt R600 is a beast in a kind. Hopefully in no way it will be associated with P4 the "Presshot" :P

EasyRaider

03-May-2007, 20:52

So, is another 20-30 watts over and above an 8800 GTX going to suddenly cause a powersupply to heat up much more with a R600 vs. a G80?
I was thinking compared to the GTS. But then again, we don't know much about performance yet. It could be a worthwhile tradeoff for most.

And anyone looking for a truly silent HTPC isn't really going to be using an HD 2900 XT or a 8800 GTX anyways.
I'm not looking for a dedicated HTPC at all, I use my gaming system for movies. Since ATI uses a smaller process, I was hoping they would improve on NVidia's already decent energy efficiency. It looks like I'll be waiting quite a while for a card to play Crysis. But that's OK, no hurry.

Silent_Buddha

03-May-2007, 21:22

True, I probably should have compared it to a GTS 640 since it appears to be closest to that in DX9 games. Although with all the FUD flying around the web at the moment, I'm not pinning it down at anything until proper reviews come out.

I'm wondering if it'll be possible to truly compare apples to apples anymore as it appears Nvidia and ATI now have different AA seetings. I'm sure the standard 4xMSAA will still be around, but that's rather boring when you have these other hybrid AA settings.

Regards,
SB

Kaotik

03-May-2007, 21:28

Without any real proof or anything, the max capacity of XT's default cooler is 210W (aka what it can still cool), so it means that even with overdrive OC, it won't go over that.

Razor1

03-May-2007, 21:37

Without any real proof or anything, the max capacity of XT's default cooler is 210W (aka what it can still cool), so it means that even with overdrive OC, it won't go over that.

Not all the power supplied to the card goes to heat, some of it has to used to some work :wink:

Silent_Buddha

03-May-2007, 21:44

Which fits in quite well with the 200 watts under full load at default clocks that Gibbo is saying.

I'd imagine hardcore o/c'rs will will probably replace the default cooler anyways.

I'll still hold judgement until some actual reviews show up. At this point it's anyone's guess what is BS and what is correct out there.

Regards,
SB

PS - I bet noone can guess I'm sitting at home bored. ;)

PSU-failure

03-May-2007, 21:57

Not all the power supplied to the card goes to heat, some of it has to used to some work :wink:
Which work?

The only "work" is the fan rotation which isn't much compared to the total power a GPU eats up.

A GPU isn't a mechanical part and the efficiency formula you refer to applies only to this. The power has to be distributed between the GPU itself, the memory, the VRMs and the remaining components of the board, just like an R580 itself never eats more than something like 60watt according to ATITool/RivaTuner, although this doesn't include memory controllers' supply voltage.

Sound_Card

03-May-2007, 21:57

wow, I can't believe people skipped right over what Vertex shader posted.

Wizzard @ techpowerup.com has a 1ghz R600 card. The qustion is, is it a OC, or stock? Because if that is a OC, then that is one danm fine OC. 250w is most impressive.

http://i21.photobucket.com/albums/b299/Genocide737/tool2.gif
http://i21.photobucket.com/albums/b299/Genocide737/tool1.gif

http://www.techpowerup.com/?30555

Geeforcer

03-May-2007, 22:06

Sound_Card

03-May-2007, 22:10

CJ: "Just got word that the HIS HD2900XT will run at 750Mhz core and 850Mhz mem (http://forum.beyond3d.com/showpost.php?p=980795&postcount=4281)."

Then that is a one danm fine OC.

Sound_Card

03-May-2007, 22:10

What if it's the XTX?

RobertR1

03-May-2007, 22:14

What if it's the XTX?

:runaway:

Natoma

03-May-2007, 22:16

What if it's the XTX?

Well it's one of 3 possibilities:

1) Stock XTX
2) OC'd XTX
3) OC'd XT

If it were #1 or #3, :-D. Anyway, we'll know in 2 weeks.

jimmyjames123

03-May-2007, 22:18

Well obviously it is an overclocked score. Sampsa says: "Actually there is already news about R600 overclocking".

http://www.xtremesystems.org/forums/showpost.php?p=2168101&postcount=18

Whether this is an XT or XTX, and under what conditions he was able to achieve the overclock, who knows. But clearly that is not stock clock speed.

Natoma

03-May-2007, 22:20

I'm hoping it's an aircooled XT. ;)

Realistically though it's probably LN or some other form of exotic cooling.

PSU-failure

03-May-2007, 22:22

Well it's one of 3 possibilities:

1) Stock XTX
2) OC'd XTX
3) OC'd XT

If the stock XTX is at 1GHz stock, which is more than 600GFlops peak processing power, next step will be the Tflop in a chip war.

Razor1

03-May-2007, 22:24

Which work?

The only "work" is the fan rotation which isn't much compared to the total power a GPU eats up.

A GPU isn't a mechanical part and the efficiency formula you refer to applies only to this. The power has to be distributed between the GPU itself, the memory, the VRMs and the remaining components of the board, just like an R580 itself never eats more than something like 60watt according to ATITool/RivaTuner, although this doesn't include memory controllers' supply voltage.

The GPU and memory arn't mechincal parts, but where is the heat coming from? Electricity converted to heat is coming from somewhere, if the cooler can dissipate up to 210watts of heat, the card if supplied 200 watts, all that wattage isn't going to be turned to heat, the card is doing something when the computer is on right?

edit: And that somewhere is resistance.

Sound_Card

03-May-2007, 22:32

Ok, I asked wizz if that is a OC, and if yes, with stock cooling. And if it was a XT or XTX. I doubt he can answer that though...

:roll:

Sound_Card

03-May-2007, 22:33

If that is a OC'ed XT on on stock cooling, I'm going to go crazy.

PSU-failure

03-May-2007, 22:37

the card is doing something when the computer is on right?
That's electrons displacement, 100% of the power a GPU eats up has to be considered as following Ohm's law as there is no "visible" work.

Data processing itself can't be considered as work as there is no moving part.

Natoma

03-May-2007, 22:37

Ok, I asked wizz if that is a OC, and if yes, with stock cooling. And if it was a XT or XTX. I doubt he can answer that though...

:roll:

Whether it's an OC'd XT or XTX, 1Ghz core clock with stock cooling is :shock:. That represents a 33% increase over stock XT core speeds as far as we know. That's absolutely phenomenal; better than anything we've seen in the land of OC'ing in the last few generations.

Razor1

03-May-2007, 22:40

That's electrons displacement, 100% of the power a GPU eats up has to be considered as following Ohm's law as there is no "visible" work.

Data processing itself can't be considered as work as there is no moving part.

Alright I used the wrong word :wink: what would you like to use other then work?

Silent_Buddha

03-May-2007, 22:48

I believe what he's saying is that all electricity consumed by the video card (other than the fan) is converted to heat in the process of crunching data.

Regards,
SB

PSU-failure

03-May-2007, 22:51

Alright I used the wrong word :wink: what would you like to use other then work?
Data processing only affects the global resistivity of the chip, thus it consumes and heats more or less depending on the load.

The most efficient the manufacturing process is, the least the resistivity which depends on traces length and section, capacitance which lengthens the transistors switching time... That's why a smaller process doesn't always mean lower power consumption.

Razor1

03-May-2007, 22:57

yes I know that, but not all of enegy is always convertered to heat or is it?

PSU-failure

03-May-2007, 23:03

Oh, and I forgot...

From the supposedly real slides, would the R600 be SIMD capable or purely VLIW based?

Given the relatively low instructions and processing units count, VLIW and structured data storage would imply a very good efficiency I think, the caches and registers allowing complex calculations requiring many passes through the ALUs to be efficient too.

Sound_Card

03-May-2007, 23:05

Whether it's an OC'd XT or XTX, 1Ghz core clock with stock cooling is :shock:. That represents a 33% increase over stock XT core speeds as far as we know. That's absolutely phenomenal; better than anything we've seen in the land of OC'ing in the last few generations.

What is even more impressive about it is that you can't change voltage in AMD gpu Tool. So that OC was done on stock voltage. :razz:

And since it was done on stock volatge, that pretty much answers one of my qustions for wizz. That being if it was on the stock cooler or not. Why change the cooler if you can't change the voltage?:razz:

Chalnoth

03-May-2007, 23:05

yes I know that, but not all of enegy is always convertered to heat or is it?
It is in this case. I mean, there might be some amount of electromagnetic radiation due to the oscillating current (accelerating charges release radiation), but that's going to be small compared to the heat.

mao5

03-May-2007, 23:06

Razor1

03-May-2007, 23:11

It is in this case. I mean, there might be some amount of electromagnetic radiation due to the oscillating current (accelerating charges release radiation), but that's going to be small compared to the heat.

Learn something new every day, pretty cool!

Brian118

03-May-2007, 23:12

mao5

03-May-2007, 23:13

The bottom shot is static lighting...

All dynamic, R600XT' quality is better, look at the shadow.

http://bbs.souyo.com/uploadimage/igdbol_200754_5162.jpg

Razor1

03-May-2007, 23:14

Almost the same scene, and I will not reply to any comment.

default R600XT+default E6600:

http://www.chiphell.com/attachments/month_0705/20070504_8103b314d5baf1cb1644j0W06TnO2439.jpg

660/2200MHz 8800GTX+3519.9MHz E6400

http://bbs.souyo.com/uploadimage/igdbol_200754_9461.jpg

PIC Quality compare:

http://www.chiphell.com/attachments/month_0705/20070504_45ff8a1b4c68fa57397bJHWQCSPB5grb.jpg

source:

http://www.chiphell.com/viewthread.php?tid=3407&extra=page%3D1&page=29 (http://www.chiphell.com/viewthread.php?tid=3407&extra=page%3D1&page=29)

Those screenshots are at different angles, look at the ground, are they even using the same settings? It looks like the ground in the r600 is blurred and not just a little bit, alot!

Chalnoth

03-May-2007, 23:17

Learn something new every day, pretty cool!
Yup. That's the basic concept on which transmitter antennas work. They basically send an oscillating current across the length of the antenna, and if the length of the antenna and the wavelength of that frequency of light are close to one another, it'll emit radiation.

Modern-day CPUs/GPUs are still significantly smaller than the wavelength of the light for the frequency they run at, but in a few years the released radiation may be a significant factor. The radiation, of course, would be in the microwave range and completely harmless, but it would be another barrier to increased clock speeds.

Corwin_B

03-May-2007, 23:44

Sooo, If I were able to buy an 8800gtx now. Do you guys think I should wait for the R600? I have been unable to keep up with this thread so I don't know if it is supposed to be a good card or not. Thanks.

I think you should wait a bit. Even if the R600 doesn't interest you once revealed, there's a chance it may trigger a price war and some GTX prices could go down as a result. :grin:

Brian118

03-May-2007, 23:47

I think you should wait a bit. Even if the R600 doesn't interest you once revealed, there's a chance it may trigger a price war and some GTX prices could go down as a result. :grin:

yea, thanks for the advice. I probably am going to wait, it's just that I'M SO SICK OF IT. I have been waiting ever since the 8800 came out.

tEd

03-May-2007, 23:50

Those screenshots are at different angles, look at the ground, are they even using the same settings? It looks like the ground in the r600 is blurred and not just a little bit, alot!

R600 seems to use less AA in this shot , i assume 2x and 4x for the gtx. Just checked on my x1900xtx using 4x. The texture difference comes from the different angle. I get about 21/22 frames on my x1900xtx with all on high and 4xAA,8xAF

edit: corrected fps of x1900xtx as the 16/17fps were with aAA enabled

russo121

03-May-2007, 23:57

Yup. That's the basic concept on which transmitter antennas work. They basically send an oscillating current across the length of the antenna, and if the length of the antenna and the wavelength of that frequency of light are close to one another, it'll emit radiation.

Modern-day CPUs/GPUs are still significantly smaller than the wavelength of the light for the frequency they run at, but in a few years the released radiation may be a significant factor. The radiation, of course, would be in the microwave range and completely harmless, but it would be another barrier to increased clock speeds.

I'm not understanding what you're saying - "that frequency of light" ?
What as light to do with antennas?

PSU-failure

04-May-2007, 00:09

R600 seems to use less AA in this shot , i assume 2x and 4x for the gtx. Just checked on my x1900xtx using 4x. The texture difference comes from the different angle. I get about 16/17 frames on my x1900xtx with all on high and 4xAA,8xAF

If it uses less AA, how do you explain the fact that the pillar shows aliasing on the G80 screen only?

But there seems to be a LoD difference at least, since the G80 loses lots of details...

tEd

04-May-2007, 00:18

If it uses less AA, how do you explain the fact that the pillar shows aliasing on the G80 screen only?

But there seems to be a LoD difference at least, since the G80 loses lots of details...

huh , r600 shows more aliasing on the pillar imo. Also looking at picture , G80 shows better AA quality overall.

Comparing the 4xAA on my x1900xtx , i'm pretty sure the r600 uses <4xAA

mao5

04-May-2007, 00:23

There are results in STALKER

Radeon HD2900xt - 42fps

http://filexoom.com/files/2006/10/15/25426/My%20Documents/Pictures/20070502_9e34f00a8b1f5577bb65WUzbsDRJTWEw.jpg

GF8800GTS 320MB - 59fps

http://filexoom.com/files/2006/10/15/25426/My%20Documents/Pictures/XR_3DA%202007-05-02%2021-45-36-51.jpg

Here is the link

http://66.249.91.104/translate_c?hl=en&ie=UTF-8&oe=UTF-8&langpair=zh-CN%7Cen&u=http://we.pcinlife.com/thread-758765-1-1.html&prev=/language_tools

If it is truth then is no good for ATI

I don't like some one cheating on 320m, pay attention to the shadow of ladder on the wall, and the density of grass on ground, need I say more on the 320m's cheating?

X1950Pro:
http://bbs.ccidnet.com/attachment/Mon_0705/41_104272_ee7da89147ba5af.jpg

8800 320m:
http://bbs.ccidnet.com/attachment/Mon_0705/41_104272_7ea351f50cf46a0.jpg

mao5

04-May-2007, 00:26

And an 8800GTX at 41fps. (http://forum.beyond3d.com/showpost.php?p=980309&postcount=4144) I can't see square one, I'm so dizzy. And still the weird washed-out screen for R600.

so you should play by yourself in the game, at that spot, the game engine will generate cyclone randomly, so when cyclone come, the pic will show washed-out screen.

Bob

04-May-2007, 00:31

Given the relatively low instructions and processing units count, VLIW and structured data storage would imply a very good efficiency I think
By that logic, Itanium and NV30 should have been runaway successes.

Fornowagain

04-May-2007, 01:04

I'm not understanding what you're saying - "that frequency of light" ?
What as light to do with antennas?
Light and radio waves belong to the electromagnetic spectrum, radio waves are just much longer wavelengths.

Speccy

04-May-2007, 01:05

By that logic, Itanium and NV30 should have been runaway successes.
Would you class R300 a success?

russo121

04-May-2007, 01:30

Light and radio waves belong to the electromagnetic spectrum, radio waves are just much longer wavelengths.

I know that, but this doesn't make any sense:

"Modern-day CPUs/GPUs are still significantly smaller than the wavelength of the light for the frequency they run at, ..."

Is he mixing wavelenght of the light with wavelength of the frequency the chip is operating?!

Chalnoth

04-May-2007, 01:36

I know that, but this doesn't make any sense:

"Modern-day CPUs/GPUs are still significantly smaller than the wavelength of the light for the frequency they run at, ..."

Is he mixing wavelenght of the light with wavelength of the frequency the chip is operating?!
Remember that the frequency within the chip is a frequency at which electrons oscillate back and forth. If they have to do this over a long enough distance, approaching the wavelength of light at the same frequency, then the chip will start to radiate like mad.

PSU-failure

04-May-2007, 02:01

Would you class R300 a success?
R300 was a successfull tradeoff, nothing more.

GPUs are far more complex this time, we'll see how the SIMD penalty the G80 has to face compares to the R600's supposed bandwidth penalty.

Rangers

04-May-2007, 03:21

wow, I can't believe people skipped right over what Vertex shader posted.

Wizzard @ techpowerup.com has a 1ghz R600 card. The qustion is, is it a OC, or stock? Because if that is a OC, then that is one danm fine OC. 250w is most impressive.

http://www.techpowerup.com/?30555

I think this is the only place R600 can really make up ground. It's so TMU limited the only place it can achieve enough texture throughput to really shine is simply with high clocks.

As of now it has ~30% clock advantage over a stock 8800GTX (575mhz), which does give it the equivalent of 21 TMU's comparitivly rather than 16 just by the clock difference. Of course, one problem is the vendor OC'd 8800's around that clock up to 650, but ignoring them.

Anyway, by one gigahertz the TMU deficit in R600 versus a stock GTX is virtually erased. Maybe if AMD could at least push XTX stock clocks to 850mhz or so, it could make some difference.

Sound_Card

04-May-2007, 03:27

I think this is the only place R600 can really make up ground. It's so TMU limited the only place it can achieve enough texture throughput to really shine is simply with high clocks.

As of now it has ~30% clock advantage over a stock 8800GTX (575mhz), which does give it the equivalent of 21 TMU's comparitivly rather than 16 just by the clock difference. Of course, one problem is the vendor OC'd 8800's around that clock up to 650, but ignoring them.

Anyway, by one gigahertz the TMU deficit in R600 versus a stock GTX is virtually erased. Maybe if AMD could at least push XTX stock clocks to 850mhz or so, it could make some difference.

I really don't see how R600 is TMU limited. Can we even be correct by even saying TMU anymore?:shock:

Topman

04-May-2007, 04:41

No tips about Crossfire SuperAA modes?

I expect new modes:

Super AA 16X = 8X MSAA + 8X MSAA
Super AA 18X = 8X MSAA + 8X MSAA + 2X SSAA

Combined modes with CFAA? (Possible?)
Super AA 18X (alt) = 8X MSAA + 8X MSAA + CFAA 2X narrow filter
Super AA 20X = 8X MSAA + 8X MSAA + CFAA 4X wide filter
(CF SSAA + CFAA makes no sense)
And the Ultimate AA mode!
Super AA 32X = 8X MSAA + 8X MSAA + CFAA edge detection filter

I expect that SuperAA works with non-stardard framebuffer formats, since X1K SuperAA only works in stardard FB format (no FP16 or 10 bit support).
And I want 16X MSAA in D3D caps too.

bye

(sorry for my bad english) :cry:

silent_guy

04-May-2007, 04:59

I did say in that post that there are advantages and disadvantages in either topology. I said nothing about what goes through the ring, beats me.
The MC's have a combined bandwidth of 150Gbps. As you correctly point out, the ring has a combined bandwidth of almost 800Gbps. Your suggestion that this excess BW will somehow be useful, automatically implies that there will also be non-MC traffic on the ring, since the MC's definitely are not going to use it.
If this is not what you were thinking, what do you suggest it will be used for?

This chip is sufficiently complex and can have features that help it overcome the issues of a ring topology.
The ring will do an excellent job transporting 150Gbps over an 800Gbps bus. That's what I meant before by 'overdimensioning'.

There is no way with the information you know right now(which amount to close to nothing) to just call this design decision as plain bad...
A block diagram that shows everything down to the small crossbars, the location of arbiters, schedulers etc.? There are industrial espio...intelligence companies who make a business selling this kind of info to competitors. I don't recall ever seeing this kind of bus detail in a presentation about GPU. If ATI is not BS'ing all of us with their slides, the information we have now is as good as it gets.

Anon Lamer

04-May-2007, 05:30

I'm not understanding what you're saying - "that frequency of light" ?
What as light to do with antennas?

Radio waves are invisible light. So are microwaves and x-rays, two other common forms of invisible light you must have encountered.

DemoCoder

04-May-2007, 05:56

I believe what he's saying is that all electricity consumed by the video card (other than the fan) is converted to heat in the process of crunching data.

Because of conservation laws, if the energy didn't go into heat, or radiation, it would have to be stored somehow in the chip, either via chemical or nuclear processes (reactions that consume energy), or by doing physical work inside the chip (moving matter around, such as the case with rod-logic gates)

Thermodynamics and information theory tell us the ultimate penalty of data processing is erasure. Everything else can in theory by made to cost zero energy, but erasing a bit can't be made cheaper than k*t*log(2)

_xxx_

04-May-2007, 07:11

yes I know that, but not all of enegy is always convertered to heat or is it?

In this case, it pretty much is. The converted electrical energy used for crunching of data gets converted to heat as well. And since the GFX card doesn't output any energy, it's all "lost".

_xxx_

04-May-2007, 07:17

Remember that the frequency within the chip is a frequency at which electrons oscillate back and forth. If they have to do this over a long enough distance, approaching the wavelength of light at the same frequency, then the chip will start to radiate like mad.

Not the oscillations of electrons are the problem, the bigger problem is the signal propagation without reflections. Refections are "teh evil" in HF-technology.

neliz

04-May-2007, 08:31

w00h00, the black box voucher...

http://www.fudzilla.com/index.php?option=com_content&task=view&id=821&Itemid=1

So besides Ep2, TF2 and portal you get DoD:Source for free to cover the distribution time.

chavvdarrr

04-May-2007, 10:38

w00h00, 82 degrees outside of the case
http://www.fudzilla.com/index.php?option=com_content&task=view&id=829&Itemid=1

neliz

04-May-2007, 10:40

w00h00, 82 degrees outside of the case
http://www.fudzilla.com/index.php?option=com_content&task=view&id=829&Itemid=1

I have some feeling that w1zzard is helping fudo with the benchmark scores of late...

82 degrees does sound a lot like the temperatures on the 19xx's

I have the ATI X1900 XTX that I recently purchased from Newegg. Installed the Arctic Cooling device on it as well that I had read good things about. The card idle in what I believe to be a good temp for idle (51 C). But in peak gaming I am seeing it climb as high as 85-88C.

http://www.tech2.com/india/reviews/pc-addon-cards/asus-radeon-1950xtx/2923/1
Idling the card runs at a core temperature of 58-59 degrees

Also checked some temperatures of GTX/GTS's and the temperatures are all in that reported range

http://www.rage3d.com/board/showthread.php?t=33873843
Mine seems rather hot to me. I mean right now at idle its 66c And under load it goes to 78-79c. And thats with the side of my case off. I have emailed BFG asking them about but have not gotten a reply yet.
Mines idling at 64C and pushed to the max ive seen 81C but shutoff temp seems to be 127C so there still headroom. I used nvtray to get the old cp back the new cp + ntune doesnt give me any temp readings. I played FEAR, Oblivion and FarCry for a few hours each and no hickups.

http://www.tones.be/forum/viewtopic.php?p=510566&sid=9b90c72f88031f887dc8b4b08791ebc8

Adam and Jamie have another R600 urban legend/myth busted!

(and when you say the temperature is at the edge of the card.. that's where heat spreaders are for. we just have to believe the overclocking util for the real values..

Plano

04-May-2007, 10:43

Well in that pic person is measuring temp of power component(s) not core.

leoneazzurro

04-May-2007, 10:47

w00h00, 82 degrees outside of the case
http://www.fudzilla.com/index.php?option=com_content&task=view&id=829&Itemid=1

That is the temperature of power MOS or components for voltage regulation, not the chip core. The MOS or passive components of the supply circuit go often at that temperature or more

Ooops, already posted :)

neliz

04-May-2007, 10:48

Well in that pic person is measuring temp of power component(s) not core.

right-o .. if the chip itself ran at 100 degrees, the unprotected back of the chip (also visible in that shot) would be able to fry eggs..

vertex_shader

04-May-2007, 11:22

w00h00, 82 degrees outside of the case
http://www.fudzilla.com/index.php?option=com_content&task=view&id=829&Itemid=1

This is a example how Fruitzilla trying to save himself after he write pure BS yesterday :wink:

XMAN26

04-May-2007, 12:15

I don't like some one cheating on 320m, pay attention to the shadow of ladder on the wall, and the density of grass on ground, need I say more on the 320m's cheating?

X1950Pro:
http://bbs.ccidnet.com/attachment/Mon_0705/41_104272_ee7da89147ba5af.jpg

8800 320m:
http://bbs.ccidnet.com/attachment/Mon_0705/41_104272_7ea351f50cf46a0.jpg

Hey mao5, hate to be a ball buster, but the screen shots are so different, its like night and day(pun intended). On the 1950 link you posted, there are fewer clouds in the sky=more sun light=better better shadow casting. In the 320M screen, there are more clouds=less light=dampened/reduced shadows. You could have 2 people play that game at the same exact time and neither would start the game or be at that point and have the exact same lighting/time of day sequence.

compres

04-May-2007, 12:24

The MC's have a combined bandwidth of 150Gbps. As you correctly point out, the ring has a combined bandwidth of almost 800Gbps. Your suggestion that this excess BW will somehow be useful, automatically implies that there will also be non-MC traffic on the ring, since the MC's definitely are not going to use it.
If this is not what you were thinking, what do you suggest it will be used for?

I don't know what they will use it for, but I think given the limitations you have rightly stated that it might be mostly for memory, but we don't know yet.

The ring will do an excellent job transporting 150Gbps over an 800Gbps bus. That's what I meant before by 'overdimensioning'.

I agree.

A block diagram that shows everything down to the small crossbars, the location of arbiters, schedulers etc.? There are industrial espio...intelligence companies who make a business selling this kind of info to competitors. I don't recall ever seeing this kind of bus detail in a presentation about GPU. If ATI is not BS'ing all of us with their slides, the information we have now is as good as it gets.

That information is not enough to judge the decision of implementing the ring, IMHO of course.

Nebula

04-May-2007, 12:26

right-o .. if the chip itself ran at 100 degrees, the unprotected back of the chip (also visible in that shot) would be able to fry eggs..

Like the Nvidia Riva cards back in the ol' days! :lol:

Galduta

04-May-2007, 12:26

It seems that AMD gives to reviewers of the 2900 XT a Enlight 500W with active PFCâ€¦.

Nuestro confidente, llamÃ©mosle SeÃ±or X, se ha puesto en contacto conmigo para contarme algÃºn dato mÃ¡s.

Sobre la fuente de 750W os podÃ©is quedar tranquilos, porque el equipo que envÃ*a AMD junto a la grÃ¡fica tiene una Enlight de 500W con PFC activo, asÃ* que por si quedaba alguna duda...

http://www.gpumania.com/foro/viewtopic.php?t=8419&postdays=0&postorder=asc&start=15

Fudo Fudo.. bad boy

But

Y bueno, me siguen confirmando que la grÃ¡fica produce mucho calor cuando trabaja a tope, mucho mucho

And good, to me they continue confirming that the graph produces much heat much much when it works in full

Nebula

04-May-2007, 12:28

Hey mao5, hate to be a ball buster, but the screen shots are so different, its like night and day(pun intended). On the 1950 link you posted, there are fewer clouds in the sky=more sun light=better better shadow casting. In the 320M screen, there are more clouds=less light=dampened/reduced shadows. You could have 2 people play that game at the same exact time and neither would start the game or be at that point and have the exact same lighting/time of day sequence.

The shadow quality is same on both. One image is darker but changing the brightness with an image editor reveals that there are no difference except for different character postures and color of the sky (lightning).

_xxx_

04-May-2007, 12:43

Like the Nvidia Riva cards back in the ol' days! :lol:

Oh, I remember that! I had a minor shock when I realized that the STB sticker on the chip (no coolers/fans back then...) was all but fried, it was black like burned paper after 2-3 weeks and fell off. I RMA'd it and got another Riva card from ELSA with a passive cooler on it (I think that was the first one with a cooler back then) :lol:

nicolasb

04-May-2007, 15:23

Fudo back-pedals furiously on the 750W issue:

http://www.fudzilla.com/index.php?option=com_content&task=view&id=830&Itemid=1

â‚¬80 PSU enough for 2900XT

Our sources confirmed that the Radeon HD 2900 XT works great with a Silverstone element 500W. This PSU has two power rails and each can provide 18 Amps. So this should be enough and the card works well under these conditions.

This PSU has two separate rails for the graphic card and it is doing the job well and it can power the Core 2 Duo system at the same time.

The good news is that at least in Germany this PSU costs under â‚¬80. You can buy it here. (http://geizhals.at/eu/?a=228631&t=alle&plz=&va=nachnahme&vl=de&v=l#filterform)

But for Crossfire you need 750W+.

PatrickL

04-May-2007, 15:29

There is no forum at fudzilla that people need to link here every crap he posts ? What about judt delete the bookmark and ignore his site instead of giving him more hits each day ?

IbaneZ

04-May-2007, 15:32

So there's still hope for my 2 years old Hiper 480W?

Specs: http://www.nordichardware.se/skrivelser_img/328/Type-R-label.jpg

Pretty cheap upgrade if I can use my old PSU. And a bunch of free Valve games to add to my steam account. :smile:

KTE

04-May-2007, 16:13

OK. The fella who mentions that it's very easily possible to draw more than 225W with the 1x 8-pin and 1x 6-pin PCIe connectors that Kombatants' post showed is perfectly correct.

What the card actually draws at stock is around the 200W mark AFAI can tell. Temps are very good under full load and its quiet compared to a X1900XTX. OC'd, watercooled+OC'd by AIB or customer, it certainly has a lot of potential for more, so AMD has been very wise with this choice and only done the right thing with providing the 8-pin connectors. It's becoming the future anyway.

PCP&C are making a new revision for their Quad Silencer 750 to include the new 8-pin connectors. Corsair will ship adapters out if you email them. SilverStone has included them in a newer revision of their Olympia/Decathlon lineup too.

And before someone mentions the obvious, yes I know PCISIG specs for the upcoming universally accepted PCIe 2.0. They are not released to anyone unless you're a member.

1x 6-pin PCIe connector can draw +12V of 6.25A i.e. 75W
1x 8-pin PCIe connector can draw +12V of 12.5A i.e 150W
PCIe slot through motherboard can draw +12V of 5.5A & +3.3V of 3A i.e. ~75W

So yeah, they have expanded to incorporate peak draw of 300W for a certain reason. Power consumption is only set to rise.

And BTW, all this mess and people still believe or even pay attention to Inq/DT/Fud ?

Seriously guys. :no:

Yes, it's FUD in every sense of the word, to the point that it now looks like a "get my back" attempt from what I know. Just wait for the reviews and you'll see.

Kinc aka Marcus aka a good known overclocker also with the cards and the DFI RD600 will test and OC them to the limits very soon over at XS and NH. AFA NDA guys go, let me tell you something: they already believe that it's the biggest thing in, to make DreamHack Summer 2007 one of the best ever, if not the best. :wink:

So just hold away from to conjectural sayings out there and no digging at Kombatant young fella's. <3

mao5

04-May-2007, 16:31

Hey mao5, hate to be a ball buster, but the screen shots are so different, its like night and day(pun intended). On the 1950 link you posted, there are fewer clouds in the sky=more sun light=better better shadow casting. In the 320M screen, there are more clouds=less light=dampened/reduced shadows. You could have 2 people play that game at the same exact time and neither would start the game or be at that point and have the exact same lighting/time of day sequence.

need I say more? look at the shadow and grass desity!

X1950Pro:
http://bbs.ccidnet.com/attachment/Mon_0705/41_104272_ee7da89147ba5af.jpg

8800 320m:
http://bbs.ccidnet.com/attachment/Mon_0705/41_104272_7ea351f50cf46a0.jpg

R600XT
http://www.chiphell.com/attachments/month_0705/20070502_4d1abfef69420e8d9c7cNvwPThwHVaVn.jpg

Dalton Sleeper

04-May-2007, 16:35

What the card actually draws at stock is around the 200W mark AFAI can tell.
...
1x 6-pin PCIe connector can draw +12V of 6.25A i.e. 75W
1x 8-pin PCIe connector can draw +12V of 12.5A i.e 150W
PCIe slot through motherboard can draw +12V of 5.5A & +3.3V of 3A i.e. ~75W
...

Well, if not using the 8-pin then it's only 75w+75w, and it's way less than 200w:shock: is it enough?

nAo

04-May-2007, 16:43

need I say more? look at the shadow and grass desity!
why don't you cut this crap? it's getting ridiculous now

nicolasb

04-May-2007, 16:44

Well, if not using the 8-pin then it's only 75w+75w, and it's way less than 200w:shock: is it enough?With R600 you have to use either two 6-pin connectors, or one 6-pin and one 8-pin.

The two possible configurations are:

stock speed only: 6-pin + 6-pin + slot = theoretical maximum of 225W (probably 200W draw in reality).
overclocking: 6-pin + 8-pin + slot = theoretical maximum of 300W (probably <225W draw in reality).

Dalton Sleeper

04-May-2007, 16:54

With R600 you have to use either two 6-pin connectors, or one 6-pin and one 8-pin.

The two possible configurations are:

stock speed only: 6-pin + 6-pin + slot = theoretical maximum of 225W (probably 200W draw in reality).
overclocking: 6-pin + 8-pin + slot = theoretical maximum of 300W (probably <225W draw in reality).

Ok, thanx for clearing that out.

I thought I could run X-Fire with two 6-pin, now i must get two adapters anyway, think I'm going for the 8-pin...

nicolasb

04-May-2007, 17:12

Ok, thanx for clearing that out.

I thought I could run X-Fire with two 6-pin, now i must get two adapters anyway, think I'm going for the 8-pin...If you're using adaptors then you will need two spare 6-pin connectors to make one 8-pin connector.

To run Crossfire without overclocking you therefore need FOUR 6-pin connectors.

To run overclocked Crossfire you need one of the following:

- two 6-pin connectors and two 8-pin connectors
- one 8-pin connector, four 6-pin connectors and one adaptor
- six 6-pin connectors and two adaptors

Dalton Sleeper

04-May-2007, 17:46

Found a molex to 8pin in another post, but someone said it wouldn't work. Isn't there any molex to 8-pin (or 6-pin) adapters? Otherwise I can't run X-Fire without changing PSU.

flippin_waffles

04-May-2007, 17:48

why don't you cut this crap? it's getting ridiculous now

I think what he's refering to, is that there are no horizontal shadows for the rungs, which does seem a little odd.

IbaneZ

04-May-2007, 18:03

Ok, it's friday, and I'm a little drunk. But ain't it weird that people (me included) are so excited about a card that's probably as fast as a 7 months old 8800 GTS? Hype is a weird thing. :smile:

Cheers!

I'm gonna get really tanked tonight. :lol:

_xxx_

04-May-2007, 18:03

Well, if not using the 8-pin then it's only 75w+75w, and it's way less than 200w:shock: is it enough?

You'd use the other 6-pin instead of the 8-pin, though. So add another 75W to that.

If it should indeed draw that much, a 750W PSU won't cut it for crossfire.

EasyRaider

04-May-2007, 18:07

If it should indeed draw that much, a 750W PSU won't cut it for crossfire.
:?:
And how did you reach that conclusion?

Dalton Sleeper

04-May-2007, 18:10

You'd use the other 6-pin instead of the 8-pin, though. So add another 75W to that.

If it should indeed draw that much, a 750W PSU won't cut it for crossfire.

750w? I'm using OCZ GameXStream 850w, but maybe it won't do.
And I realized now that I can run one XT with two 6-pin, but I want two XT.

EasyRaider

04-May-2007, 18:17

Don't be silly. Why the hell would it not do?

russo121

04-May-2007, 18:23

Excuse me this is OT, but, thanks Chalnoth, Fornowagain, and Anon Lamer because I didn't know that "light" is also considered an electromagnetic radiation of any wavelength.

Fornowagain

04-May-2007, 18:32

Excuse me this is OT, but, thanks Chalnoth, Fornowagain, and Anon Lamer because I didn't know that "light" is also considered an electromagnetic radiation of any wavelength.
Its all photons to me mate.

Dalton Sleeper

04-May-2007, 18:41

What about this adapter, will it work? Will it deliver the same as the original 6-pin?

http://www.scan.co.uk/Products/ProductInfo.asp?WebProductID=154684&source=Froogle

Chalnoth

04-May-2007, 18:50

What about this adapter, will it work? Will it deliver the same as the original 6-pin?

http://www.scan.co.uk/Products/ProductInfo.asp?WebProductID=154684&source=Froogle
Provided your power supply is up to it, yes. But you should use power connectors from different cables, not two from the same cable, and nothing else that is power hungry should be on either wire.

Fornowagain

04-May-2007, 18:54

What about this adapter, will it work? Will it deliver the same as the original 6-pin?

Will it work? Sure.

It'll have two 12v conductors, a normal 6 pin PCIe has three. So it doesn't have the same load capacity, but I wouldn't worry about it. Just make sure to feed from two different leads from the PSU to spread the load.

Anarchist4000

04-May-2007, 18:56

The cards have no way of sensing how much power is actually behind a given connector. You could use a single molex to 6 pin if you really wanted to but you'd probably want some serious cabling to do it.

If you have one of the single rail PSUs(PCP&C for example), it's just a matter of having thick enough wires to support the current they will be carrying. And of course a large enough PSU to begin with. You could probably even make the adapters yourself if you had the right tools and parts.

Dalton Sleeper

04-May-2007, 19:06

I got 4 rails with 12V, 18-20A on every one. But the 6-Pin connectors, didn't ATI include it, I mean, they have to send atleast two with a single card to get it running. I got two, and if they include two then it's done.

neliz

04-May-2007, 19:23

Just for information.
Salland, who had the Sapphire 2900XT for sale at â‚¬380 (and who were subsequently kicked in the nads by ATI) now have the Asus 2900XT on pre-order at â‚¬396.
Their core speed says unknown, memory at 1.65Ghz.

AnarchX

04-May-2007, 19:53

Specviewperf v9
http://img405.imageshack.us/img405/2517/x2900xt102401319944ik7.pnghttp://img405.imageshack.us/img405/5362/x2900xt10240231a133qw8.png
http://www.xtremesystems.org/forums/showthread.php?t=143104

3DMark-protection?

Wirmish

04-May-2007, 19:56

Hum...

An Asus 2900XT on pre-order at 396 euros. :sad:
A Sapphire 2900XT for sale at 380 euros. :smile:

But why not a Sapphire HD 2900XT for 365 euros ? :grin:

http://i114.photobucket.com/albums/n273/Wirmish/RadeonHD2900XT-s3.jpg

Seems much more interesting to me... :wink:

Silent_Buddha

04-May-2007, 19:57

There is no forum at fudzilla that people need to link here every crap he posts ? What about judt delete the bookmark and ignore his site instead of giving him more hits each day ?

Because just like people like reading the Enquirer or the Star for gossip, they like reading Fudo and the Inq for technical gossip. Both have about the same grounding in reality I'd imagine. :wink:

Although why post it here continuously, I haven't a clue.

Regards,
SB

Frank

04-May-2007, 20:00

I'm pretty sure I could simply power all the extra power needs for an X600 from a single 4-pin connector with my 580 W power supply. My main concern would be the contact resistance.

Not all PSUs are created equal.

Silent_Buddha

04-May-2007, 20:11

Specviewperf v9

** image cut **

http://www.xtremesystems.org/forums/showthread.php?t=143104

3DMark-protection?

Is it true what that poster said on that forum? That those scores are better than a Quadroplex? I usually don't pay much attention to industry OGL benches.

Regards,
SB

Lux_

04-May-2007, 20:13

Specviewperf v9What resolution?

For comparison, 8800 GTX is half as fast ( 2560x1600 - E6300 2.80GHz): http://www.mvktech.net/content/view/3552/39/1/9/

AnarchX

04-May-2007, 20:23

What resolution?

I do not know, but if you looking at the results @ dailytech (http://www.dailytech.com/ATI+Radeon+HD+2900+XT+Performance+Benchmarks/article7043.htm), it should be higher than 1280x1024.

satein

04-May-2007, 20:29

What resolution?

For comparison, 8800 GTX is half as fast ( 2560x1600 - E6300 2.80GHz): http://www.mvktech.net/content/view/3552/39/1/9/

If you look into the picture posting score, it stated that the monitor is 17" CMC something.... So, I would suggest this is a score on testing at 1280x1024 resolution.

Sound_Card

04-May-2007, 20:31

Originally Posted by Sampsa
I will stick with NDA unlike some other people who attended Tunis event and already posted pretty much all the details of the card. There are also quite alot false information floating around the internet regarding Radeon HD 2900 XT.. main source is one particular site which is not so hard to guess.

Anyway, there is actually quite alot overclocking headroom in R600 GPU and AMD GPU Clock Tool seems to be working just fine. I had a problem with Powerstrip beta-version which was limited to driver's limits. Now we just need to wait for new version of ATITool which brings voltage adjustments for R600 and I'll run tests under dry ice cooling. Hopefully that'll be next week.

PS. 1000/1000 MHz clock speeds at techPowerUp!'s screenshot are pretty modest :fact:

Kombatant, nice to see you again

That put a good smile on my face.:razz:

Razor1

04-May-2007, 20:32

Is it true what that poster said on that forum? That those scores are better than a Quadroplex? I usually don't pay much attention to industry OGL benches.

Regards,
SB

not a quardoplex lol, quadro fx, which yes better then last gen gf7 quadro fx, not the g8's.

Cuthalu

04-May-2007, 20:36

If 1000/1000MHz is "pretty modest", this might fare well against G80 when oc'd. :grin:

Natoma

04-May-2007, 20:40

If 1000/1000 is pretty modest, one wonders why AMD didn't clock the R600 higher. Maybe we've been mislead on the true default clocks of the XT and XTX. :lol:

nutball

04-May-2007, 20:41

If 1000/1000 is pretty modest, one wonders why AMD didn't clock the R600 higher. Maybe we've been mislead on the true default clocks of the XT and XTX. :lol:

or mislead on the overclocks.

neliz

04-May-2007, 20:55

Well, if we go back a couple of years we heard about the 1ghz mark first for the r600.

If we like to compare.. the 2900XT will be for R600 what the 1800XL was for the R520. So Getting up with a Ghz XTX sounds much better for a fighting chance.. It would also make the pricepoint of the XT logical.. it's yesteryears XL.
And with no overclock utils available for DT, they were indeed misled by whoever thought of it an XTX at XT clocks with no way to push it...

Sound_Card

04-May-2007, 21:03

Wizard hit 1ghz on stock volts which would most definitely imply hitting that clock is somewhat moddest.:grin:

3dilettante

04-May-2007, 21:40

A few boards that may or may not have hit 1 GHz isn't enough to say R600 can reliably hit that clock speed.

Considering the limited number of examples, it's possible those high clocking chips happen to be of a small number of chips with good leakage and power characteristics.

R600 as a design may be artificially capped by an AMD-imposed TDP limit, or the bin splits above 750 MHz are not acceptable within the power envelope.

vertex_shader

04-May-2007, 22:13

Specviewperf v9
IMG]http://img405.imageshack.us/img405/2517/x2900xt102401319944ik7.png[/IMG] IMG]http://img405.imageshack.us/img405/5362/x2900xt10240231a133qw8.png[/IMG]
http://www.xtremesystems.org/forums/showthread.php?t=143104

3DMark-protection?

HD2900XT 1024mb gddr4? :wink:

flippin_waffles

04-May-2007, 22:16

HD2900XT 1024mb gddr4 :wink:

I was just going to say the same. That's not the first screenshot i've seen with 1024mb ddr4 specs. Curious.

[edit]

Maybe crossfire???

XMAN26

04-May-2007, 22:29

need I say more? look at the shadow and grass desity!

X1950Pro:
http://bbs.ccidnet.com/attachment/Mon_0705/41_104272_ee7da89147ba5af.jpg

8800 320m:
http://bbs.ccidnet.com/attachment/Mon_0705/41_104272_7ea351f50cf46a0.jpg

R600XT
http://www.chiphell.com/attachments/month_0705/20070502_4d1abfef69420e8d9c7cNvwPThwHVaVn.jpg

Your are seeing something that isn't there other than the fact that both ATI shots have much more sunshine in them than the Nvidia one does. Go outside ona nice sunny day and you will get shadows like in the ATI pics, go out on cloudy days and you'll get shadows more in line with the Nvidia screenie. And the grass looks the same to me in all three shots. and that is after staring at them each for the last 5 minutes.

wishiknew

04-May-2007, 22:29

... so the r600 has p4 syndrome.

Silent_Buddha

04-May-2007, 22:57

... so the r600 has p4 syndrome.

Only in the case that you also assume that the G80 has Athlon XP (not 64) syndrome. :wink: After all it appears they both perform similar just that one is Designed for higher clocks at that performance.

Regards,
SB

PSU-failure

04-May-2007, 23:02

Does anyone have seen anything susceptible to be an NDA break on XS forums?

It has been down for a moment...

no-X

04-May-2007, 23:04

... so the r600 has p4 syndrome.
Really? G80s ALUs are clocked at 1500MHz. R600s at 750MHz...

nelg

04-May-2007, 23:04

Wizard hit 1ghz on stock volts which would most definitely imply hitting that clock is somewhat moddest.:grin:

Taken with the likelihood that someone who's claim to fame is that of an overclocker, receiving a cheery picked card for that various purpose would not be to much of a stretch.

Anarchist4000

04-May-2007, 23:11

If the cards are hitting those clocks can't we safely assume that the 2900XT is using clocks that keep it in the 225W power envelope? Not that it was clocked to the max to keep pace with the 8800GTX.

flippin_waffles

04-May-2007, 23:14

Does anyone have seen anything susceptible to be an NDA break on XS forums?

It has been down for a moment...

Yeah, regarding that, i've had nothing but problems with that site since they switched servers. So it's really nothing new as far as I can tell.

PSU-failure

04-May-2007, 23:16

Wizard hit 1ghz on stock volts which would most definitely imply hitting that clock is somewhat moddest.:grin:

Take note of the smilie used, what does this tiny creature points out?

Maybe it's the NDA... :idea:

Twinkie

04-May-2007, 23:28

1Ghz sure does sound nice. (possible XT version that comes with 1024mb GDDR4 ram?)

Things are getting more interesting by the day!

w0mbat

04-May-2007, 23:34

Does anyone have seen anything susceptible to be an NDA break on XS forums?

It has been down for a moment...

Yes, there was a Thread with benches 8800TGTX OC vs. HD2900 XT OC which broke the NDA (8800GTX was faster all the time).

Twinkie

04-May-2007, 23:40

Yes, there was a Thread with benches 8800TGTX OC vs. HD2900 XT OC which broke the NDA (8800GTX was faster all the time).

Did anyone save the results? :twisted:

flippin_waffles

04-May-2007, 23:57

Yes, there was a Thread with benches 8800TGTX OC vs. HD2900 XT OC which broke the NDA (8800GTX was faster all the time).

But then Perkam said it was removed because the results were on pre-release hardware, and didn't reflect the results that users under NDA were seeing. Dunno but that's what he said.

BlizzardOne

05-May-2007, 00:14

I was just going to say the same. That's not the first screenshot i've seen with 1024mb ddr4 specs. Curious.

[edit]

Maybe crossfire???

Nope, I have CF'd X1950 XTX's, and it shows 512 Mb in my CCC, even though the two cards have 1024 Mb between them:

http://img520.imageshack.us/img520/3248/image1ui2.jpg

Galduta

05-May-2007, 00:31

R600 preview (http://it-review.net/index.php?option=com_content&task=view&id=1310&Itemid=91)

After a week of time "off" to travel around Europe to acquire as much data as possible, we're back on our R(V)6xx track. A lot of interesting things happened in the last week and we decided to share them with you. After the launch event in Tunis, everything calmed down a little bit and we were able to acquire more info about the actual products we already covered. Ready to go back in the R(V)6xx world?

tEd

05-May-2007, 00:40

I'm wondering if they gonna expose hardware PCF in d3d9.

ow and amd/valve (http://www.steampowered.com/activateATI/index.html)

Rangers

05-May-2007, 01:30

R600 preview (http://it-review.net/index.php?option=com_content&task=view&id=1310&Itemid=91)

Expect more news very soon. Like - next week, full R600 reviews and, of course, some CrossFire testing, as well. We wouldn't want to leave you in the dark...

So they're going to break NDA (or not under one)?

PSU-failure

05-May-2007, 01:46

So they're going to break NDA (or not under one)?
I think the real NDA end won't be the launch day as it's never been the case...

Now, I could just be wrong but I see it somewhere in the next week.

Sc4freak

05-May-2007, 02:29

Nope, I have CF'd X1950 XTX's, and it shows 512 Mb in my CCC, even though the two cards have 1024 Mb between them:

http://img520.imageshack.us/img520/3248/image1ui2.jpg

I never knew the X1950XTX came with DDR4. Learn something new every day.

hoom

05-May-2007, 03:00

Um, that was the whole point of X1950XTX :wink:

INKster

05-May-2007, 04:57

Um, that was the whole point of X1950XTX :wink:

I think he was being ironic. ;)
DDR4 is (or will be) one thing. GDDR4 is another. Much like DDR3 is very different from GDDR3, both technically and in concrete hardware applications.

KTE

05-May-2007, 05:15

Specviewperf v9
[IMG]http://img405.imageshack.us/img405/2517/x2900xt102401319944ik7.png
http://www.xtremesystems.org/forums/showthread.php?t=143104

3DMark-protection?
1: He needs to run it on Vista and 3DMark will run.
2: Its not the XT retail version, its the OEM workstation card with 1GB GDDR4.
3: Kinc and Sampras from XS have the retail XT with 512MB GDDR3, so look to their threads for results. :wink:

The retail XT is very good, just hold on and see the reviews from guys who know what they're doing. Most guys online that I've seen with the cards producing benches already published are still using the older drivers that had issues and were showing underpeformance.

Any guy with the GPU will confirm that, like kinc, gibbo, wizzard and so forth.

Topman

05-May-2007, 05:24

from xtremesystems (link in previous post)

http://www.iamxtreme.net/video/r600/2900XT1024_04.PNG
Not bad for a bad card(XT OEM 1GB) :twisted: with older drivers...

Default clocks (VGA)...

Imagine > 1000/1000... :D

bye

Sobek

05-May-2007, 05:29

Out of curiosity, does anyone know what an x1950xtx gets on a similar setup? (Core2Q overclocked and so on).

Topman

05-May-2007, 05:41

http://www.free4up.com/view.axd?fn=255139044010069.jpg&Thumbnail=1 (http://www.free4up.com/ShowImage.aspx?fn=255139044010069.jpg)
lol

http://hdtvsg.blogspot.com/

PS: a external PSU conector? WTF?
PS2: fixed

Cuthalu

05-May-2007, 05:50

"Forbidden
Your client does not have permission to get URL / from this server."
:sad:

Galduta

05-May-2007, 05:55

http://service.futuremark.com/compare?3dm05=2820199

QX6700@3.70 8800GTX 650/1450/2100 3mark 2005

3DMark Score20228 3DMarks

Game Tests

GT1 - Return To Proxycon------80.0 fps http://service.futuremark.com/images/1x1pix.gif

GT2 - Firefly Forest--------------59.5 fps http://service.futuremark.com/images/1x1pix.gif

GT3 - Canyon Flight-------------111.3 fps http://service.futuremark.com/images/1x1pix.gif http://service.futuremark.com/images/1x1pix.gif