The NEXT LAST R600 Rumours & Speculation Thread

[ninelven]

I propose a new naming scheme:

Name: Viseon... oh come on.

Number: (represents generation) abbreviated G1

Status: (place in the food chain) first letter represents segment, second letter represents place in that segment. First letters: L,M,X (low, medium, Xtreme!!!)(note higher letter = better). Second letters: R,S,T. (higher letter = better).


Sample card lineup:

AMD Viseon G1.0 XT
AMD Viseon G1.0 XS
AMD Viseon G1.0 XR
...
AMD Viseon G1.0 MT
etc...
...
AMD Viseon G1.0 LT
etc...


Refreshes denoted with decimal:
AMD Viseon G1.1 XT

Generational leaps get a new number:
AMD Viseon G2.0 XT

Hurray for a short, consistent, logical naming system.

[Cuthalu]

That's very good naming scheme. However, it's way too logical, good and simple to be used by AMD (or nVidia). :smile:

[Sobek]

http://forum.catchride.com/style_emoticons/default/puke.gif

Just Radeon 2 GT/XT/XTX thanks. Poor/Average/Highend users :wink:

[R300King!]

I say call it the Radeon-and-on :D

[_xxx_]

That's very good naming scheme. However, it's way too logical, good and simple to be used by AMD (or nVidia). :smile:

Yup, not confusing enough for the less knowlegable customers. Otherwise, how would you sell stuff like the 9200 512MB or such?

[hoom]

Naming is decided by marketing & marketing likes short sensible names like AMD ATI Radeon HD X39050 XTX Limited Edition Extreme Model Deluxe :roll:

[SugarCoat]

Naming is decided by marketing & marketing likes short sensible names like AMD ATI Radeon HD X39050 XTX Limited Edition Extreme Model Deluxe :roll:

i'm confused i thought the XTX was the OEM only version, i just bought an XTL, i thought that was the right one because the XTV has half the shader engines and the XTG has twice the clock rate with half the ram. If the XTX isnt OEM only then whats the XTL? If i bought the wrong card im going to be so angry!!11!


personally i have a simple way to solve the naming confusion, i just go to the store and say to the guy "give me the one that costs way wayyy wayyyyyy too much money"

actually i just lurk here and learnded it all as the new stuff pops up

[PatrickL]

Could be worse. Just imagine if Leadtek was making AMD based cards :razz:

[neliz]

personally i have a simple way to solve the naming confusion, i just go to the store and say to the guy "give me the one that costs way wayyy wayyyyyy too much money"

You still end up with the XFX Geforce 8800 Ultra Extreme XXX OC model.

[BrynS]

Well, at least the Tunis Press Briefing appears (http://www.theinquirer.net/default.aspx?article=39100) to be underway.

[nicolasb]

Well, at least the Tunis Press Briefing appears (http://www.theinquirer.net/default.aspx?article=39100) to be underway.Surely it must now be time for the LAST LAST rumours and speculation thread...?

[Techno+]

instead of G1, G2 etc. make it 1G, 2G, 3G just like mobile phones:cool:

[Arty]

"Preview" or should I say all_rumors_so_far_compiled :sly:

http://it-review.net/index.php?option=com_content&task=view&id=1288&Itemid=1

[nutball]

"Preview" or should I say all_rumors_so_far_compiled :sly:

http://it-review.net/index.php?option=com_content&task=view&id=1288&Itemid=1

24-bit anti-aliasing? Sounds ground-breaking.

[Sound_Card]

"Preview" or should I say all_rumors_so_far_compiled :sly:

http://it-review.net/index.php?option=com_content&task=view&id=1288&Itemid=1

Some how they complied the info that G80 is 80nm.

[Sound_Card]

This is a intresting bit from that artical...



R600 has a built-in sound chip. As simple as that. That makes quite a lot of interesting things possible.


Let's discuss this sound-chip thingy a bit further and use our imagination. One of the things ATi-engineers have been thinking about is something we could call sound physics. The "usual" physics is supported with R600, but having a sound chip onboard - that's a very interesting thing. But let's start with an example.

You're playing your favorite game (pick your name here), and you're firing like crazy. You move, and you hit a table that has a glass on. Glass falls down from the table and crushes into a thowsand pieces. You go on as if nothing has happened.
Next day, you come to a same place, do the very same thing and - you'll get the same sound, because that's the way things are mostly done today. But what if you could always have some set of parameters that could "morph" the sound (change it) with respect to different speed you bump into the table, different angle the glass falls on the floor etc? The sound would change, thus making the game more realistic and interesting. It does sound interesting, right? We absolutely want this feature and give it a standing ovation, especially since this is done directly on the graphic card (less delay, less overhead, faster response time, easier to manipulate etc). We heard about this back in January and we've been thinking about this for months. Being a musician and a sound engineer on a part-time basis I understand the complexity of this solution and the need for it on a completely another level. Now I wonder if this could somehow be used in audio applications. Gee, wasn't this the very same idea we've been having for months now? Hmmm...

nice!

[3dilettante]

Do they have R600 down as a 65nm chip?

[Robin B]

So its 65 nm now ??

Not bad for a 12-layer 65nm part, huh? With 700 milion transistors, working at 750MHz and higher clock, with 512MB GDDR3 and price somewhere in the $450 region?

[Sound_Card]

Do they have R600 down as a 65nm chip?

First and foremost - the designs for some parts of the new series are 65nm, according to our sources. This is the first surprise we stumbled upon when we were discussing some info we got from different partners. There were some reports about going with 80nm and then 65nm while products are already in shops, but scrap this. 65nm only. High-end parts have 320 stream processing units and 512-bit 8-channel memory interface, if information that partners have been kindly providing us are true.

So they say...

[IbaneZ]

http://www.fudzilla.com/index.php?option=com_content&task=view&id=656&Itemid=1

ATI Radeon HD 2900 XT TURBO SILENT Heatpipe 512MB GDDR3 VIVO TV-Out/Dual DVI (PCI-Express) - Retail

overclockers.co.uk is a little early. :)

[Geeforcer]

You know, I am looking at the pictures of R600 heatsink... that thing is impressive. I wonder how much it costs.

[3dilettante]

I'm wondering if that site got their numbers mixed up.

They have R600 consuming upwards of 270 W. If everything's supposedly 65nm, then one would think R600 wouldn't be burning the rumored wattage of the 80nm part.

[neliz]

This is a intresting bit from that artical...

bla blaSound Physicsbla bla

nice!


Ouch! "Sound Physics" comes from an Inq/Fud article where, as usual the lousy grasp of english turned "Sound and Physics" into "Sound Physics".. which got it-review wetting it's panties.. .nothing to worry about...

[neliz]

http://www.fudzilla.com/index.php?option=com_content&task=view&id=657&Itemid=34

copper

http://www.fudzilla.com/images/stories/2007/April/graphics/r600_pic_2.jpg

Fudo says the 8pin is only for overclocking/overdrive...

[Geeforcer]

I'm wondering if that site got their numbers mixed up.

They have R600 consuming upwards of 270 W. If everything's supposedly 65nm, then one would think R600 wouldn't be burning the rumored wattage of the 80nm part.

Reading the article, the whole thing is full of "we don't know", "we think", "it seems" and "it's rumored".

BTW, I must be the only person in the world who thinks that spending GPU transistors on sound processing is a dubious idea.

[Cuthalu]

Wow, if it's truly silent, meaning passive, it may have somewhat low energy consumption after all. That would be awesome. :)

[Kaotik]

I'm wondering if that site got their numbers mixed up.

They have R600 consuming upwards of 270 W. If everything's supposedly 65nm, then one would think R600 wouldn't be burning the rumored wattage of the 80nm part.

those 270W and whatever numbers are just plain and simple bollocks, a while back the 270W rumor got up from AMD Q&A Session video, where some reporter/journalist/whatever says "I saw a benchmark of R600 with 270W blahblah" - AMD never mentioned anything about it being true, and I think we can safely say that those benches people "saw" were nothing but BS back then?

[trinibwoy]

This is a intresting bit from that article...

Good luck getting API support for that mate :) Assuming that any of it is true in the least of course.

[Geeforcer]

Wow, if it's truly silent, meaning passive, it may have somewhat low energy consumption after all. That would be awesome. :)

Passive? This is how "HIS ATI Radeon X1950 XT ICEQ 3 TURBO SILENT Heatpipe 256MB GDDR3 VIVO TV-Out/Dual DVI" (at least the name is not too long) looks like:


http://www.overclockers.co.uk/pimg/GX-067-HT_400.jpg

[Geeforcer]

those 270W and whatever numbers are just plain and simple bollocks, a while back the 270W rumor got up from AMD Q&A Session video, where some reporter/journalist/whatever says "I saw a benchmark of R600 with 270W blahblah" - AMD never mentioned anything about it being true, and I think we can safely say that those benches people "saw" were nothing but BS back then?

Or what people saw was work in progress never ment for public consumption.

[Kaotik]

Yes, there is "sound chip" in R600 and RV630 & RV610 - but it's meant for nothing but to get the sound out via HDMI with no extra cabling required, it won't be any damn EAX XXX monster, and probably won't be doing DDLive (though I'm not sure would it be even needed on HDMI either)

[Frank]

I'm wondering if that site got their numbers mixed up.

They have R600 consuming upwards of 270 W. If everything's supposedly 65nm, then one would think R600 wouldn't be burning the rumored wattage of the 80nm part.
R600 is almost surely 80nm. And 270 W is very unlikely, as it exceeds the specs of the combined power sources.

But I am pretty sure they have their top model burn the maximum allowed (225 W), simply because they need the performance increase that gives them.

[Kaotik]

Or what people saw was work in progress never ment for public consumption.

mm, that's a possibility, too.
In the same Q&A video, before that question in fact, they were asked how many watts for that teraflop box, they didn't give straight answer, but said "you're looking at about 200w per card"

[Frank]

Yes, there is "sound chip" in R600 and RV630 & RV610 - but it's meant for nothing but to get the sound out via HDMI with no extra cabling required, it won't be any damn EAX XXX monster, and probably won't be doing DDLive (though I'm not sure would it be even needed on HDMI either)
As far as I get it, there is no real sound chip as such, but mostly something that does some post processing to make it easier to add sound to the HDMI output.

[Kaotik]

As far as I get it, there is no real sound chip as such, but mostly something that does some post processing to make it easier to add sound to the HDMI output.

Nah, it would then still need the cables to get the sound from the actual soundcard / integrated sound chip to the card, so it has to have "full capabilities" if for nothing else, but for straight pass through sound (that's what the "real soundchips/cards" do with digital sound output anyway)

[Skinner]

Where are the leaks, just a little Stalker/Obivion sneak peak would be nice.
There should be some perf. info to the press right now?

[Valzic]

It-review.net has a fallup writeup on the R600 on its site.


(http://it-review.net/index.php?option=com_content&task=view&id=1289&Itemid=105)

More details.

"The 2600 series should be up and running up to 800MHz clock speed, using unique TSMC 65nm+ process. The 2400 series also has 320 SP's, and we're talking about superscalar ALU as well (dedicated branch execution units and texture address processors)."

Interesting. Just have to figure out if they adding speculation or getting feedback from Africa.

[NocturnDragon]

Yes, there is "sound chip" in R600 and RV630 & RV610 - but it's meant for nothing but to get the sound out via HDMI with no extra cabling required, it won't be any damn EAX XXX monster

You are talking like you know it... are you so sure? ;)

[CJ]

Ouch! "Sound Physics" comes from an Inq/Fud article where, as usual the lousy grasp of english turned "Sound and Physics" into "Sound Physics".. which got it-review wetting it's panties.. .nothing to worry about...

Urhm... no... it doesn't have to come from the usual Inq/Fud article... In February - way before The Inq/Fud came up with it - a trustworthy source already mentioned a rumour of AMD doing "Sound Physics" next to "normal physics". At the time he didn't know for sure if it was true, but seeing as this word has been popping up it's head more frequently in the last couple of weeks... I wouldn't be surprised if R600 actually was capable of Sound Physics...

[Kaotik]

NocturnDragon, of course everything is possible, but it would eat quite a bit more transistors at that point, I mean, X-Fi for example has 51million transistors (which of ~20 million is logic, the rest cache), and even Audigy has over 4.6 million transistors.
So yes, I'd throw the possibility of being EAX monster etc out of the window, I can't see them spending even just that 4 million transistors just for the sound, especially on the lower end chips.

Valzic, 65nm was known for both RV610 and RV630, the 800MHz clockspeed rumor floated up few days ago I think, but wtf is that "320 SP's" on 2400 thing? :???:

[Frank]

Nah, it would then still need the cables to get the sound from the actual soundcard / integrated sound chip to the card, so it has to have "full capabilities" if for nothing else, but for straight pass through sound (that's what the "real soundchips/cards" do with digital sound output anyway)
Nah, drivers can do those things fine.

[Kaotik]

Nah, drivers can do those things fine.

Then why do the current cards with HDMI include SPDIF-in to get the audio for HDMI ? :???:

[Frank]

Then why do the current cards with HDMI include SPDIF-in to get the audio for HDMI ? :???:
Because they don't have the possibility to do the post processing?

[Kaotik]

Because they don't have the possibility to do the post processing?

What other "post processing" do you need except to make sure that the video & audio go out "in sync", which the current cards do with the digital audio signal they get via SPDIF-In?

[neliz]

Urhm... no... it doesn't have to come from the usual Inq/Fud article... In February - way before The Inq/Fud came up with it - a trustworthy source already mentioned a rumour of AMD doing "Sound Physics" next to "normal physics". At the time he didn't know for sure if it was true, but seeing as this word has been popping up it's head more frequently in the last couple of weeks... I wouldn't be surprised if R600 actually was capable of Sound Physics...

Then Inq knows the term, but not how to apply it.. typical klok / klepel situation.

http://www.theinquirer.net/default.aspx?article=39042
We hear that Sound Physics has been buzzed for quite some time, and we are massaging our contacts to bring you as much detail as possible. It seems that R600 GPU will be less CPU-bound than it has been in the past, since a lot more parts of the scene will be calculated and rendered on the GPU itself.

R600 combines graphics with audio and physics process in a whole another way, and those 64 vect5D shader pipelines yield in 320 scalar units total, directly comparable to Nvidia's 128 scalar ones. Now, 128 scalar units at 1.35 GHz is still a tad more efficient that ATi's own 320 at 740 MHz, but ATi's pipeline is more complex than it was previously thought.

[Geeforcer]

It-review.net has a fallup writeup on the R600 on its site.


(http://it-review.net/index.php?option=com_content&task=view&id=1289&Itemid=105)

More details.

"The 2600 series should be up and running up to 800MHz clock speed, using unique TSMC 65nm+ process. The 2400 series also has 320 SP's, and we're talking about superscalar ALU as well (dedicated branch execution units and texture address processors)."

Interesting. Just have to figure out if they adding speculation or getting feedback from Africa.

To me this sounds like nonsense. "65nm+ process"? 320 SPs for 2400-series?

[neliz]

To me this sounds like nonsense. "65nm+ process"? 320 SPs for 2400-series?

That you didn't know, all Radeon HD's will be the same, just the price will be different.. AMD really wanted the performance crown in all segments this time.. That's why they are selling millions to OEM's

[Kaotik]

That you didn't know, all Radeon HD's will be the same, just the price will be different.. AMD really wanted the performance crown in all segments this time.. That's why they are selling millions to OEM's

Yeh, and in reality, the huge HSFs and extra powerplugs on 2900's are just for the imago there, i mean, it would look silly if most powerful card in the town would be looking like some lowend with short PCB and small 1-slot cooler, no extra powerplugs etc :grin:

[Frank]

What other "post processing" do you need except to make sure that the video & audio go out "in sync", which the current cards do with the digital audio signal they get via SPDIF-In?
Well, whatever you would like and GPGPU can do, I think.

[Geeforcer]

That you didn't know, all Radeon HD's will be the same, just the price will be different.. AMD really wanted the performance crown in all segments this time.. That's why they are selling millions to OEM's

LOL....:grin:... In related news, it was just revealed that every Radeon HD 2xxx, from 2100 to 2985.75 will have TWO R600, one 80nm and one 65nm one. AMD is shooting for 100% marketshare.

[Kaotik]

Well, whatever you would like and GPGPU can do, I think.

But that would eat resources from the 3D department, which shouldn't be an option here, so it pretty much has to have it's own audio department too, just like UVD is separate from the 3D department.
Potentially with any card capable of GPGPU functions, you could make it act as a sound card, but this is not the case with current cards even though X1600 surely could do GPGPU calculations just like it's bigger brothers, but could they manage to make it function as a video card and a sound card at once? I doubt it.

If the drivers could handle getting the audio signal without extra cabling, it would have been done before, but it hasn't because they can't, so the sound department of R6xx/RV6xx handles that part as fully functional if not for anything else, but for pass through audio.

[Sound_Card]

AMD claims R600 is 50x more powerful than G80 in geometry operations.

http://it-review.net/index.php?option=com_content&task=view&id=1289&Itemid=1


The part that I find disturbing and somehow must be a typo, is that 2400 is claimed to have 320 SP's. I mean what the hell?

[Razor1]

50 times that really sounds like BS, the rest of that site/article also smells funny too.

[neliz]

Well.. 50x should be 50% methinks, we already heard that R600 will be a lot better than G80 at Geometry.

[Razor1]

yeah was thinking the same thing but they might be talking about amount of scalar operations again, and not real world use. I don't see how if the g80 is able to keep up with its scalar units in pixel and vertex shaders, why would they be weaker in geometry shading?

[Recall]

Guys, what is 24 bit anti aliasing?

http://it-review.net/index.php?option=com_content&task=view&id=1288&Itemid=1&limit=1&limitstart=6

[Kaotik]

Guys, what is 24 bit anti aliasing?

http://it-review.net/index.php?option=com_content&task=view&id=1288&Itemid=1&limit=1&limitstart=6

It's someone reading something wrong and then writing up what he read as it was true.

[Razor1]

way too many mistakes in that article, sounds like the guy just grabbed a bunch of stuff of the net and put numbers by them or something.

[nutball]

way too many mistakes in that article, sounds like the guy just grabbed a bunch of stuff of the net and put numbers by them or something.

Probably angling for a job at T'Inq.

[INKster]

Probably angling for a job at T'Inq.

Or at Fudzilla, judging by the names of the writers and their similar bad english and dubious news sources...

[Arnold Beckenbauer]

AMD claims R600 is 50x more powerful than G80 in geometry operations.

http://it-review.net/index.php?option=com_content&task=view&id=1289&Itemid=1


Not AMD! It was a FSX developer, who said, R600's is more powerfull working on GS.
"The 2600 series should be up and running up to 800MHz clock speed, using unique TSMC 65nm+ process. The 2400 series also has 320 SP's, and we're talking about superscalar ALU as well (dedicated branch execution units and texture address processors)."
Wow, superscalar. G80's architecture is scalar, R600's architecture is superscalar. Wow!
Me is a man. Clark Kent is Superman! ::wink:
But wait, G80's architecture is superscalar, too. Superscalar means, a processing unit can do more than one instruction in a single pipeline.

[INKster]

Superscalar means, a processing unit can do more than one instruction in a single pipeline.

And in a single clock cycle. ;)

[nutball]

That you didn't know, all Radeon HD's will be the same, just the price will be different.. AMD really wanted the performance crown in all segments this time.. That's why they are selling millions to OEM's

And losing millions of dollars. It all fits into place now. :yep2:

[Graham]

TMU's have 64-bit HDR textures bilinear filtering at true speed, as well as 128-bit FP textures working at half speed. There's also support for 32-bit shared exponent texture format (RGBE 9:9:9:5)


Do we know if full speed 64bit extends to backbuffer formats/blending too? (is this the case for G80 too?)

Afaik RGBE9995 is a standard DX10 texture format, right? Still awesome though :razz:

[Kaotik]

Do we know if full speed 64bit extends to backbuffer formats/blending too? (is this the case for G80 too?)

Afaik RGBE9995 is a standard DX10 texture format, right? Still awesome though :razz:

I think you should just forget all that IT-review site says :yep2:

[compres]

OMFG still no real news, no leaks, no nothing...!

:cryes:

[Arty]

OMFG still no real news, no leaks, no nothing...!

:cryes:
An NDA that actually works! :razz:

[trinibwoy]

If we don't have a full set of slides and performance numbers by wednesday i'll be very surprised. People WILL talk...

[Natoma]

Either people are so stunned by the performance that they can't get a word out, or they're so horrified by the performance that it's not worth wasting the bandwidth to say anything. :twisted:

[INKster]

An NDA that actually works! :razz:

Intense heat, nice beaches and free drinks at AMD's discretion go a long way... :D

[Ailuros]

50 times that really sounds like BS, the rest of that site/article also smells funny too.


I didn't even reach that far reading it to see that the author doesn't have a single clue:

First news - R600 has Programmable Tesselation Unit, featured in the XBOX chip, as well.

After that I closed the link; thanks to the author for wasting 30 seconds of my time :D

[Ailuros]

Guys, what is 24 bit anti aliasing?

http://it-review.net/index.php?option=com_content&task=view&id=1288&Itemid=1&limit=1&limitstart=6

Multisampling without transparency antialiasing ROFL :D (yes it's a joke)

[Ailuros]

Either people are so stunned by the performance that they can't get a word out, or they're so horrified by the performance that it's not worth wasting the bandwidth to say anything. :twisted:

*sticks finger in the air* after all this time reading the signs, following traces and sniffing dog poo here and there my blurry crystal ball says that R600 will have it's advantages yet not end up to be a "G80-killer". Ironically something like that has been posted/predicted so many times in recent times that it's not even funny anymore.

AMD will most likely go for a price war, which in essence means that NV will also cut down immediately prices and all potential D3D10 GPU buyers have all the reason in the world to be exited. The consumer wins again and we all will live happily ever after *awwwww....* ;)

[seahawk]

Another price war for AMD ?

And given that they are late to the market, while G80 most likely has paid itself of already ...... - was the last conforence call not bad enough ?

[Cuthalu]

And another 65nm rumour: http://www.digitimes.com/mobos/a20070423PD213.html
Someone has too much fun creating misinformation, or we'll have unexpected surprise. :eek:

[Geeforcer]

You know, Digitimes are not always right but if I were a betting men I'd start putting more money on 65nm then before. I still think 80nm R600 is more likely than 65nm though.

[Kaotik]

It's 80nm, and will never be anything else.
The dieshots alone tell enough, as we know about the transistor counts.

When the 65nm cores will come, which might of course be really soon, they will be R650's, not R600's.

[dizietsma]

The Digitimes article is interesting in that it says that the delay is due to GDDR4 shortage and not because AMD wanted top to bottom synchronous launch, in fact they say the mid range parts will be released later ! If that is true it shows that AMD were ecconomical with the truth.

[vertex_shader]

It's 80nm, and will never be anything else.
The dieshots alone tell enough, as we know about the transistor counts.

When the 65nm cores will come, which might of course be really soon, they will be R650's, not R600's.

Who said AMD not have 80nm R600 GPU's for the sample cards? :wink:

I think no one except 65nm r600 in may, not even NV, so when its true will be a huge surprise.

[Kaotik]

Who said AMD not have 80nm R600 GPU's for the sample cards? :wink:

I think no one except 65nm r600 in may, not even NV, so when its true will be a huge surprise.

It's been on every single shot on every single card we've seen, do you honestly think it even as a possibility that the retail cards would actually have 65nm chip?
And even if they did, it would be R650, not R600 :wink:

[vertex_shader]

It's been on every single shot on every single card we've seen, do you honestly think it even as a possibility that the retail cards would actually have 65nm chip?
And even if they did, it would be R650, not R600 :wink:

Possibility is always there, maybe its only 0.1% but there :wink:

[vertex_shader]

And another 65nm rumour: http://www.digitimes.com/mobos/a20070423PD213.html
Someone has too much fun creating misinformation, or we'll have unexpected surprise. :eek:

R610/R630 :smile:

[Gouhan]

@ a rumored $399 I'm really liking the R600.
Sure it may not wpe the floor with 8800GTX, but who cares its cheap and faster than 8800GTS.
can't argue with 64x5-way SIMD units :D

[vertex_shader]

http://img243.imageshack.us/img243/4866/atihd2900keyfeaturesky0mm1.jpg

[_xxx_]

@ a rumored $399 I'm really liking the R600.

When oh when will this stupid shite of a rumour die? :roll:

[leoneazzurro]

http://www.dailytech.com/article.aspx?newsid=7043

no word yet on AA performance, and it would be interesting to see where it is CPU limited and where it is not.

[Arnold Beckenbauer]

http://img243.imageshack.us/img243/4866/atihd2900keyfeaturesky0mm1.jpg

Free Black Box Game from VALVe? What? Return of the famous HL2 voucher? :lol:

[trinibwoy]

no word yet on AA performance, and it would be interesting to see where it is CPU limited and where it is not.

They have the specs listed right in the article ;)

It's 2x faster than the GTS in HL2:EP1 :shock: :lol:

[vertex_shader]

http://www.dailytech.com/article.aspx?newsid=7043
Expect AMD to pull the wraps off its DirectX 10 product line up in mid-May, with value, midrange and high end models to boot. AMD’s flagship ATI Radeon HD 2900-series will have two models at launch – the ATI Radeon HD 2900 XTX and the HD 2900 XT. The ATI Radeon HD 2900 XTX models feature 1GB of GDDR4 memory while the lower HD 2900 XT features 512 MB.

Every source say different things :smile:

Maybe i'm wrong, but in this (http://images.dailytech.com/nimage/4535_large_R600_2.jpg) picture looks like the card have 2x6pin psu connector.

Edit: In this (http://images.dailytech.com/nimage/4534_large_R600_1.jpg) picture looks like 1x6 pin 1x8pin :smile:

[leoneazzurro]

They have the specs listed right in the article ;)

It's 2x faster than the GTS in HL2:EP1 :shock: :lol:

Yes, I saw it. But anyway it's not easy to see which game is CPU limited at that resolution.
I.e. I could dare to say that CoH should be pretty CPU limtied, but i.e. I could not say that so easily about FEAR, even if that could be the reason the results are almost equal.
And anyway, I would have a look at the AA/AF scores (and IQ, of course), as they are more important for an high-end card.

EDIT: I also expect that XTX model should have not only different amount and speed of memory, but higher clock speed, too. Otherwise in 99% of the tests, I expect almost no performance difference, and this does not justify the price

[Gouhan]

When oh when will this stupid shite of a rumour die? :roll:
Calm down. Nothing's dying it is actually $399
I'm willing to bet on it are you?

[Kaotik]

Every source say different things :smile:

Maybe i'm wrong, but in this (http://images.dailytech.com/nimage/4535_large_R600_2.jpg) picture looks like the card have 2x6pin psu connector.

Edit: In this (http://images.dailytech.com/nimage/4534_large_R600_1.jpg) picture looks like 1x6 pin 1x8pin :smile:

There has been sample cards with both 2x6pin and 1x8 1x6pin configs, the max wattage can't be more than 225W (and since you need that small "safety zone", i'd say 220W max)

[aeryon]

http://www.fudzilla.com/index.php?option=com_content&task=view&id=661&Itemid=1

XTX in Q3 because of GGDDR4 issues, well not surprising...

[Kaotik]

nevermind, posted already it seems

[Arnold Beckenbauer]

They have the specs listed right in the article ;)

It's 2x faster than the GTS in HL2:EP1 :shock: :lol:

R600's launch driver versus G80's launch driver? Who knows...
* Our benchmarks for Half Life 2: Episode 1 showed an abnormal framerate for the NVIDIA 8800 GTS card that scaled with lower resolutions -- we are still investigating what occurred there.

[trinibwoy]

Yeah those HL2 scores are definitely borked. Dailytech must be using some pretty custom benchmarks too cause those GTS scores look low across the board compared to other reviews.

[Rangers]

Nvmind

[nicolasb]

Some odd things lurking in the DailyTech posting.

Although the R600 GPU supports HDMI audio and video output, the our reference design only features dual dual-link DVI.Not unexpected, but intensely annoying to me if all of the R600 cards turn out like that. :evil:

AMD’s flagship ATI Radeon HD 2900-series will have two models at launch – the ATI Radeon HD 2900 XTX and the HD 2900 XT. The ATI Radeon HD 2900 XTX models feature 1GB of GDDR4 memory while the lower HD 2900 XT features 512 MB.So, XTX available at launch (they claim), not delayed into June.

The ATI Radeon HD 2900 XT is poised to have a street price approximately the same as the GeForce 8800 GTS, which currently has a suggested retailer price of $499.So, $499 rather than $399? :(

[Pressure]

Some odd things lurking in the DailyTech posting.

Not unexpected, but intensely annoying to me if all of the R600 cards turn out like that. :evil:


As far as I understood they will come as dual dual-link DVI but feature cables that converts it to HDMI.

[neliz]

R600's launch driver versus G80's launch driver? Who knows...

8.361 vs 158.19

8.361 were reported the SLOW driver, remember inq's article about 8.461 saves the day

[neliz]

As far as I understood they will come as dual dual-link DVI but feature cables that converts it to HDMI.

It would seem to mee they'd use a DMS-59 connector instead of a normal DVI connector since that one is able to transmit a lot more data than a normal DVI or dual-link DVI connector.

http://en.wikipedia.org/wiki/DMS-59

[vertex_shader]

8.361 vs 158.19

8.361 were reported the SLOW driver, remember inq's article about 8.461 saves the day

The 8.361 save the day (http://www.fudzilla.com/index.php?option=com_content&task=view&id=525&Itemid=34), 8.35 the slow driver, this only come from Fu(a)dZilla :smile:

[_xxx_]

Calm down. Nothing's dying it is actually $399
I'm willing to bet on it are you?

Yes. $50, via PayPal?

And we're talking about the XT and the average price in e-tail on launch day.

[Gouhan]

Yeah! $50...
I'm only reporting what I am seeing. I couldn't believe $399 myself but it seems true at least as the MSRP

[vertex_shader]

Yes. $50, via PayPal?

And we're talking about the XT and the average price in e-tail on launch day.

You not want loose :smile:
When MSRP 399$ than launch day in store the card can cost easily 450$.

[neliz]

The 8.361 save the day (http://www.fudzilla.com/index.php?option=com_content&task=view&id=525&Itemid=34), 8.35 the slow driver, this only come from Fu(a)dZilla :smile:

ah yes... thanks 8.36 is Cat 7.4

[jamis]

Yes. $50, via PayPal?

And we're talking about the XT and the average price in e-tail on launch day.
LOL, getting cold feet? It's MSRP, not the price the e-tailers set on launch

[_xxx_]

Ok, it's a deal :cool:

[_xxx_]

LOL, getting cold feet? It's MSRP, not the price the e-tailers set on launch

MSRP is usually higher than online street prices, that would be even easier.

But ok, if you wish let's say MSRP. You choose.

[jamis]

MSRP is usually higher than online street prices, that would be even easier.

But ok, if you wish let's say MSRP. You choose.
Not at launch. And I'm not betting anything here..:lol:

[vertex_shader]

MSRP is usually higher than online street prices, that would be even easier.

Not in the first few days, some store sell the 8600gts for 250$.
Avabialility is a key when the prices going closer to msrp or lower than msrp.

[_xxx_]

jamis, I meant Gouhan should choose.

[_xxx_]

Not in the first few days, some store sell the 8600gts for 250$.
Avabialility is a key when the prices going closer to msrp or lower than msrp.

Well, they boldly claim they'll be having a hardy-hard-harder-than-ever launch, so... ;)

[Sound_Card]

MSRP is usually higher than online street prices, that would be even easier.

But ok, if you wish let's say MSRP. You choose.

On launch day, MSRP is lower than what retailers set.

[vertex_shader]

Well, they boldly claim they'll be having a hardy-hard-harder-than-ever launch, so... ;)

Than we see some days after launch hd2900xt for 399$ :smile:
Good avabialility is not the only key, when the product sell like hot cake with higher price than msrp the stores has no reason to sell it cheaper :smile:

[Geo]

Ok, it's a deal :cool:

If you guys are going to use our forums to make wagers, then we get to ban . . . heaven forbid. . . welchers, right? :smile:

[_xxx_]

Sorry, that wasn't meant to become "B3D going online-casino" or such :oops:

Feel free to remove it all and we'll keep this private.

[erick]

I think Geo meant that he would have to start banning those who don't pay up for the lost wager ;)

[Jawed]

I've got my fingers crossed that the MSAA comes in 8x per loop (instead of 2x with prior GPUs): 8x, 16x, 24x. That would be groovy.

Also, what if "superscalar" in this case means "multiple scalar instructions per pixel in parallel". That would be interesting...

Jawed

[_xxx_]

Oh, I do always keep my promises and pay my dues.

[Evildeus]

I think more probablye it will be 6x, 12x, 18x and 24x like before with a big boost in 24x with 2 cards i suppose :twisted:

[Geo]

http://www.dailytech.com/ATI+Radeon+HD+2900+XT+Performance+Benchmarks/article7043.htm

DT claims to have some test numbers they ran themselves. Tho price-wise it seems to me it would have been nice to compare to both a GTS and a GTX, as it looks like XT will split them at street price initially. And the res and aa/af options are a bit mid-rangy in my opinion.

[nelg]

If you guys are going to use our forums to make wagers, then we get to ban . . . heaven forbid. . . welchers, right? :smile:

Bans should be handed out when the house does not gets its cut. :twisted:

[neliz]

Damn Geo, you're slow! (and use confusing URL's)

http://www.dailytech.com/article.aspx?newsid=7043

no word yet on AA performance, and it would be interesting to see where it is CPU limited and where it is not.

[trinibwoy]

Damn Geo, you're slow!

For real :lol:

[Mariner]

Reading some of the discussion comments after the Dailytech article, Kris Kubicki admits that there seems to be something funky up with the HL2:EP1 benchmarks and these should be ignored. He also says that the relatively low resolution was selected because this was the max supported by the monitor used when they had their brief time hands-on with the card.

Finally:

We'll get more cards and benchmarks later this week.

[nicolasb]

Latest from the Inq: http://www.theinquirer.net/default.aspx?article=39145

Highlights:

this second gen unified shader marchitecture part has 64 Vect5D units and seeing the 320 scalar unit number, we can now say this is official. Please bear in mind that the rumour that the chip has 64 vect4+scalar was wrong from day one.

Vect4+Scalar is not directly comparable to Nvidia, Vect5D is. So, we have a superscalar marchitecture with 320 scalar processsing units designed in-line with built-in Tessellation hardware, relatively similar to the Xenos chip from Xbox 360. But performance is far beyond we have imagined, and hearing that this is not limited to a certain version of DX10 only speaks of better days for PC gaming.

Geometry performance is vastly improved, as is expected from vect5D based marchitecture, and gamedevs can now use Geometry Shaders, unlike certain green products.

Anisotropic filtering and AntiAliasing were vastly improved as well, and R600 marchitecture marks off the start of "NextGen AA". A new AA mode is called Custom Filter AA (CFAA), and it peaks at 24x. 8x MultiSample AA mode was completely changed, so expect some fireworks and comparisons between G80 and R600 hardware.

ATi AVIVO also makes a return in a form of Universal Video Decoder (UVD), scoring 128 on HQV test, and HD2900XT will of course, support high-resolution HDMI (1.2 spec), featuring AC3 5.1 output.HDMI 1.2?!?! Not 1.3! :evil::evil::evil: I hope that's wrong! :(

Clocks are set at 742 MHz for the GPU and 1.65 GHz for 512MB of GDDR-3 memory.

[trinibwoy]

Vect4+Scalar is not directly comparable to Nvidia, Vect5D is. So, we have a superscalar marchitecture with 320 scalar processsing units....

Geometry performance is vastly improved, as is expected from vect5D based marchitecture, and gamedevs can now use Geometry Shaders, unlike certain green products.

http://www.mysmiley.net/imgs/smile/confused/confused0046.gif

Aaaaaargh!!!! I never knew ignorance could cause physical pain.

[_xxx_]

Vect4+Scalar is not directly comparable to Nvidia, Vect5D is

Huh, what? *boggle*

EDIT: hehe

[Razor1]

think I should coin the term and phrase "inqinizite it!" :razz:

[Sound_Card]

ATi introduces a new AA Mode called Custom filtering AA.(CFAA) going to 24xCFAA.

http://www.theinquirer.net/default.aspx?article=39145

Anisotropic filtering and AntiAliasing were vastly improved as well, and R600 marchitecture marks off the start of "NextGen AA". A new AA mode is called Custom Filter AA (CFAA), and it peaks at 24x. 8x MultiSample AA mode was completely changed, so expect some fireworks and comparisons between G80 and R600 hardware. Of course, all existing AA tech is supported.

[Andrew Lauritzen]

Huh, what? *boggle*
Seriously... I have no idea what half of that article is trying to say. I think I'm gonna wait for B3D's excellent architecture articles for R600 :) In the mean time, I will feel guilty for even clicking on an Inquirer link...

[Valzic]

A number of links to other sites will be posted here as more and more information will be linked. Some will be fact and some will be Speculation.

I have been a lurker on this site for many years. I use this site as a means to further understand graphics. But probably the key thing about this site is the excellent discussion and debunking of information.

It is a really positive for B3D and I am sure I am not the only one reading forums and enjoying the discussions and getting some real analysis of data shown on the internet on Graphics.

So I hope to have an enjoyable learning experience reading the comments here. Please be patient with us less knowledgeable readers and let us know what is true and what is not. Ignorance does not necessarily cause physical pain.

:smile:

[Jawed]

Custom filtering AA
Aren't future versions of D3D going to allow programmer-defined AA patterns?

Also there's EATM modes and screendoor transparency anti-aliasing (erm, are they the same thing?).

Sample-level screen-door transparency using programmable transparency sample masks (http://v3.espacenet.com/textdoc?DB=EPODOC&IDX=US2007070082&F=0)

Described are a graphics processing unit (GPU) and a sample-level screen-door transparency technique for rendering transparent objects. The GPU includes a scan converter and a shader. The scan converter identifies pixels to be processed for rendering a transparent object and divides each pixel into a plurality of samples. The shader generates, for one of the identified pixels, an application developer-specified transparency sample mask indicating which samples of the pixel are to be suppressed when determining a color of the pixel. Execution of an application developer-specified sample mask command produces a pattern of bits that map to samples of the pixel. The values of the bits determine which samples of the pixel may be used and which samples are to be suppressed when determining a color of the pixel.

[0017] The degree of realism attained by graphics applications when using screen-door transparency to render overlapping transparent objects or objects of graduated transparency can depend upon the dithering and randomness of the sample masks used to approximate the transparency of the objects. Each transparency sample mask determines which samples of a pixel may contribute to the displayed color of the pixel. The transparency sample mask can identify these samples positively (by identifying each sample that may contribute to the color), negatively (by identifying each sample that does not contribute to the color, i.e., suppressed samples), or by a combination of positive and negative identification of the samples. In general, the number of samples contributing to the pixel's color determines the transparency of the object depicted by those samples: the fewer the number of contributing samples, the more transparent the object appears. The present invention enables developers of graphics application programs to specify one or more functions to generate transparency sample masks that achieve a desired level of dithering, randomness, or both. The ability to specify a mask-generating function gives creative license to graphics application developers when implementing screen-door transparency at the sub-pixel or sample level. In addition, this flexibility extends to the use of currently unknown randomizing and dithering algorithms or functions that may serve to produce a transparency sample mask for achieving a desired transparency effect.

Jawed

[nicolasb]

I can't be *rsed with posting actual links, but the latest from Fudzilla is:

- R600 won't overclock unless you plug in an 8-pin power connector.

- RV670 will probably be here in Q3.

- R650 will be here in Q4 and compete against G90.

[eSa]

Well, the 64 Vec5 units is always interesting. Obviously it is 4+1, but main thing is of course how flexible those vector units are. How fine grain data they can process and wether you can "divide" them down to perform simple MADD-operations. Which would be cool.

[erick]

- R650 will be here in Q4 and compete against G90.

That just sounds so wrong on so many levels :(

[fellix]

I think R600 will target a thread size of 32 fragments, which should mean 4 cycles switching time in the thread dispatcher, for the pool of 64 vec[5] units (8*8 !?).

[SirPauly]

Aren't future versions of D3D going to allow programmer-defined AA patterns?

Also there's EATM modes and screendoor transparency anti-aliasing (erm, are they the same thing?).

Sample-level screen-door transparency using programmable transparency sample masks (http://v3.espacenet.com/textdoc?DB=EPODOC&IDX=US2007070082&F=0)





Jawed


Thanks, :)

[IbaneZ]

The quality settings for the games were as follows:

Call of Duty 2 - Anisotropic filtering, 4xAA (in game), V-Sync off, Shadows enabled, a high number of dynamic lights, soften all smoke edges and an insane amount of corpses.

Company of Heroes - High shader quality, High model quality, Anti-aliasing enabled (in game), Ultra texture quality, high quality shadows, high quality reflections, Post processing On, High building detail, High physics, high tree quality, High terrain detail, Ultra effects fidelity, Ultra effects density, Object scarring enabled and the model detail slider all the way to the right.

F.E.A.R. - 4x FSAA (in game), maximum light details, shadows enabled, maximum shadow details, soft shadows enabled, 16x anisotropic filtering, maximum texture resolution, maximum videos, maximum shader quality.

Half Life 2: Episode 1 - High model detail, high texture detail, high shader detail, reflect all water details, high shadow detail, 4x multi-sample AA (in-game), 16x anisotropic filtering, v-sync disabled, full high-dynamic range.

Pretty much as high as you can crank it.


I guess that's dailytechs test dude who wrote that.

[Silent_Buddha]

F.E.A.R. - 4x FSAA (in game), maximum light details, shadows enabled, maximum shadow details, soft shadows enabled, 16x anisotropic filtering, maximum texture resolution, maximum videos, maximum shader quality.

If FEAR isn't CPU limited in the test Dailytech did, then R600 must have improved shadow performance quite a bit. I haven't look at benchies in a long time and my memory might be going out on me, but didn't ATI used to take a much larger hit for soft shadows in FEAR than NV hardware?

Regards,
SB

[IbaneZ]

Unless something changed -- Sven got an XTX about 2 hours after Anh got the XT. Sven is benchmarking now and we should have that up this week.

http://www.dailytech.com/article.aspx?newsid=7043&commentid=131535&threshhold=-1&red=493#comments

Come on Sven! :wink:

No delay for the XTX then I guess.

[Razor1]

If FEAR isn't CPU limited in the test Dailytech did, then R600 must have improved shadow performance quite a bit. I haven't look at benchies in a long time and my memory might be going out on me, but didn't ATI used to take a much larger hit for soft shadows in FEAR than NV hardware?

Regards,
SB

actually concidering the gf7's and x1x00 ATi had a less of a hit I thought? Dynamic branching performance comes into play there.

Anyways, a member here told me about the problems with FEAR with MSAA and softshadows on after my mini review of the g80 scaling with different clocks for mem and core, even with the latest patch softshadows don't work with MSAA activated so the FEAR numbers really don't account for soft shadows.

[Silent_Buddha]

Thanks for clearing that up, it's been a long time since I played FEAR and never used soft shadows since they looked ugly in that game. More like double shadows than actual soft shadows IIRC.

Regards,
SB

[Jawed]

This patent application:

Vertex data processing with multiple threads of execution (http://v3.espacenet.com/textdoc?DB=EPODOC&IDX=US2006033757&F=0)

A method for processing video image data including a plurality of different image data types begins by providing tasks to be performed on each different image data type. The image data is divided into a plurality of groups based on the image data type. A set of arithmetic operations required to accomplish the tasks provided for the corresponding image data type is determined. Each arithmetic operation is assigned to one of a plurality of commonly used arithmetic units which performs the arithmetic operation, whereby each image data type is transformed in accordance with the corresponding provided tasks. The transformed image data of each group is combined, completing the processing.

http://www.cupidity.f9.co.uk/b3d93.gif

is interesting because it uses two scalar ALUs (170 and 175) and feeds them with out of order code to process vertex data. Each ALU has a dedicated scheduler, and a set of semaphores are used to track per-thread data dependency.

The result is that 3 sub-programs (vertex position, vertex colour and texture) are executed for each vertex. Each program can have arbitrary complexity, and each vertex can have an arbitrary number of colours/textures, each of which is effectively an independent thread.

The basic idea of the patent is that instead of processing each sub-program for each vertex sequentially, "one instruction at a time", potentially leaving either the multiplier or adder idle, the schedulers issue dependency-freed instructions from the pool of available threads. It's very much like the out of order, decoupled, ALU and TMU scheduling we know from R5xx.

Next, it's possible to imagine this scaled-up. Instead of a single multiplier and a single adder, you could have a SIMD array for each. Additionally other types of instruction (most obviously, special function) can have their own ALU array.

So I'm wondering if this kind of architecture is the basis for R6xx's ALU pipes? Is this the origin of the "superscalar" buzzword.

Jawed

[trinibwoy]

I guess that's dailytechs test dude who wrote that.

Yeah, guess I'll wait on proper drivers then because the XT should be mopping the floor with the GTS with AA enabled even at 12x10.

[fellix]

So I'm wondering if this kind of architecture is the basis for R6xx's ALU pipes? Is this the origin of the "superscalar" buzzword.
16 groups of 4 vec[5] (lovely quads ;)), chained by four in a cluster (filling a single thread "burst") and attached texture address & sampling hardware -- that would make for 16/32 fragment batch size and sits perfectly with the [rumored] 16 ROP orthogonal config.

[ChrisRay]

Vec4 + scalar not comparable to Nvidia? Huh? My head hurts.

[vertex_shader]

http://www.dailytech.com/article.aspx?newsid=7043&commentid=131535&threshhold=-1&red=493#comments

Come on Sven! :wink:

No delay for the XTX then I guess.

I hope Sven have bigger resolution monitor, and not testing the xtx in 1280x1024, what is the waste of the time :lol:

[jimmyjames123]

Here is a breakdown of 8800 GTX 768MB vs HD 2900 XT 512MB vs 6800 GTS 640MB. Seems like 8800 GTX has a significant advantage in almost all cases (note that the 8800 GTX data is from xbit labs using a very old Forceware 97.02 driver and is only using an Athlon FX-60 CPU--and the 8800 GTX gets big performance gains with new drivers and with highest end Core 2 Duo CPU--see review here: http://www.xbitlabs.com/articles/video/display/gf8800-games_26.html)



Gaming: Maximum Quality, 1280x1024

NVIDIA GeForce 8800 GTX 768MB
AMD ATI Radeon HD 2900 XT 512MB
NVIDIA GeForce 8800 GTS 640MB


Call of Duty 2
92.4 FPS
73.5 FPS
56.7 FPS

Company of Heroes
102.5 FPS
92.1 FPS
90.1 FPS

F.E.A.R.
91.0 FPS
84.0 FPS
83.3 FPS

Half Life 2: Episode 1
108.2 FPS
112.0 FPS
57.4 FPS * (<-- this number looks way off, since 8800 GTS is almost the same performance as 8800 GTX at this setting in an Xbit labs test)

Oblivion
57.2 FPS
47.9 FPS
39.5 FPS



Now we know why it would be difficult for AMD to place the HD 1900 XT at the $599-649 price point.

[vertex_shader]

Now we know why it would be difficult for AMD to place the HD 1900 XT at the $599-649 price point.

We know nothing, using 2 different test for comparsion not a good idea, i can say in other word too, its stupid thing (sorry) :wink:

[jimmyjames123]

Of course it is two different tests, but use some logic too. The Xbit labs test of the 8800 GTX was using very old Forceware drivers, and an Athlon FX CPU, both which would be not nearly as flattering as the newer Forceware drivers and Core 2 Duo processor that were used in the Daily Tech testing. All of the Xbit data on the prior page was at 4xAA and 16xAF though.

Surely you don't believe that those old linked Xbit results overestimate the 8800 GTX performance. :D

In fact, at the resolution and settings that Daily Tech tested at, the 8800 GTX would be largely CPU limited. It would perform much better using a Core 2 Duo 6600 or higher, while the 8800 GTS 640MB would gain very little. See these CPU scaling tests: http://www.firingsquad.com/hardware/geforce_8800_gtx_gts_core_2_performance/page5.asp

The Daily Tech benchmarks are not the be-all end-all of performance indicator either. Several of the tests did not even stress the GPU's very much.

[vertex_shader]

Of course it is two different tests, but use some logic too. The Xbit labs test was using very old Forceware drivers, and an Athlon FX CPU, both which would be not nearly as flattering as the newer Forceware drivers and Core 2 Duo processor that were used in the Daily Tech testing.

Surely you don't believe that those old linked Xbit results overestimate the 8800 GTX performance. :D

In fact, at the resolution and settings that Daily Tech tested at, the 8800 GTX would be largely CPU limited. It would perform much better using a Core 2 Duo 6600 or higher, while the 8800 GTS 640MB would gain very little. See these CPU scaling tests: http://www.firingsquad.com/hardware/geforce_8800_gtx_gts_core_2_performance/page5.asp

The Daily Tech benchmarks are not the be-all end-all of performance indicator either. Several of the tests did not even stress the GPU's very much.

May 14th you get your answers :wink:

No one asked the question yet, why AMD give a chance for dailytech benchmark the cards 3 weeks before launch?

[IbaneZ]

Sven is having fun. :lol:

http://www.dailytech.com/Overclocking+the+R600/article7044.htm

It's the XT though, but it overclocks ok.

[jimmyjames123]

May 14th you get your answers :wink:

No one asked the question yet, why AMD give a chance for dailytech benchmark the cards 3 weeks before launch?


We've been waiting six or seven months, what's another three weeks huh? ;)

Why do you think AMD "gave them a chance" (assuming that is the case) to benchmark the cards 3 weeks early? The benchmarks certainly don't look bad in any way compared to the 8800 GTS, so at least it could generate more buzz and provide some hard data where data has been sorely lacking.

As I've said all along, from reading in between the lines, most likely the AMD cards will not have a very pronounced performance advantage in the majority of games currently tested today. Rather, they will offer a nice package of good performance and good features at a good price. No one can argue against that combination :)

[Arty]

Now we know why it would be difficult for AMD to place the HD 1900 XT at the $599-649 price point.
Definitely! These cards were intended for only 12x10 only. In fact, reviewing higher resolutions or higher IQ settings have been banned by the International Hardware Reviewing Forum. ;) :lol:

[jimmyjames123]

Definitely! These cards were intended for only 12x10 only. In fact, reviewing higher resolutions or higher IQ settings have been banned by the International Hardware Reviewing Forum. ;) :lol:

Note that the tests were done with 4xAA and 16xAF, although in several cases the cards were not very stressed, no doubt. That said, if performance was as good as the 8800 GTX in general, then don't you think the 512MB XT model would sell for higher than $449-499 MSRP? The 8800 GTX has an advantage in terms of pure performance on these games at these settings tested, and there is little doubt that would translate to higher resolutions and higher IQ settings as well. That's not a knock on the XT, just saying you get what you pay for at a very minimum. Clearly, the 8800 GTS either needs to be overclocked or moved into a lower price point to offer a similar price-to-performance ratio as the XT, based on these prelim numbers.

[Kaotik]

http://www.dailytech.com/Overclocking+the+R600/article7044.htm

Overclocking the XTX apparently, at stock(?) clocks of 745/800 it scores 13547 in 3DMark06

[jimmyjames123]

http://www.dailytech.com/Overclocking+the+R600/article7044.htm

Overclocking the XTX apparently, at stock(?) clocks of 745/800 it scores 13547 in 3DMark06

With memory running stock at 1.6Ghz on the XTX, he is able to get 13547. Interesting that the other guy got 11447 for the XT which has memory at 1.65Ghz. So what is cut down in the XT to give such a big difference in score?

It's interesting that with a core overclock of 100Mhz (745Mhz to 845Mhz), and memory overclock of 200Mhz (1.6Ghz to 2.0 Ghz), the 3dmark score only increase by 450 points. So even with a 15% increase in core clock and 25% increase in memory clock, the 3dmark score only goes up by 3 or 4%? The card must be very CPU limited at that resolution.

[leoneazzurro]

http://www.dailytech.com/Overclocking+the+R600/article7044.htm

Overclocking the XTX apparently, at stock(?) clocks of 745/800 it scores 13547 in 3DMark06

This is very strange: 800 MHz is XT memory speed (2x800=1600), not XTX
DDR4 at 1600 MHz? Really strange :???:

[neliz]

XTX at 745? fudo says the XT runs at 742. interesting..

[neliz]

RD790 just got PCI Sig certified..

http://www.pcisig.com/developers/compliance_program/integrators_list/pcie/

It looks more and more like a completely new family testbed

[SirPauly]

This is very strange: 800 MHz is XT memory speed (2x800=1600), not XTX
DDR4 at 1600 MHz? Really strange :???:

I believe those were over-clocked XT numbers and he is currently testing the XTX is what I did get from that blog.

[Cuthalu]

I believe those were over-clocked XT numbers and he is currently testing the XTX is what I did get from that blog.

How do you explain the difference of a bit more than 2000 points? They used QX6800 instead of QX6700, but that doesn't explain the point difference

[vertex_shader]

http://www.dailytech.com/Overclocking+the+R600/article7044.htm

Overclocking the XTX apparently, at stock(?) clocks of 745/800 it scores 13547 in 3DMark06

XT have 825mhz (http://dailytech.com/Article.aspx?newsid=7043) memory clock speed ,than why the XTX have only 800mhz (http://www.dailytech.com/Overclocking+the+R600/article7044.htm)?
Something wrong :smile:

[Sound_Card]

RD790 just got PCI Sig certified..

http://www.pcisig.com/developers/compliance_program/integrators_list/pcie/

It looks more and more like a completely new family testbed


Nice catch!:razz:

[Robin B]

If i read it right he said it was the XT he overclocked, and and not the XTX.

[Kaotik]

If i read it right he said it was the XT he overclocked, and and not the XTX.

But when did that, was it 10 or 11k? change to 13k on XT? :???:

[PatrickL]

I hope we will soon see tests with resolutions we buy theses cars for....

[Karma]

Performance question:

Based on the R600's unified architecture, and the ability for the 320 "processors" to do different types of work (if I understand that right), then wouldn't there be more consistency in the frame rate?

e.g. instead of "spikes" of the frame rate going down to below 30 due to a fixed architecture being bottlenecked at certain times redering a scene, would the R600's unified architecture perhaps keep the average frame rate higher?

Bueller?

[neliz]

But when did that, was it 10 or 11k? change to 13k on XT? :???:

The first 10k's we hear about were on a E6600... drivers maybe? they both focus on the driver being RC4.

[Skinner]

I hope we will soon see tests with resolutions we buy theses cars for....

Yes of maybe, games? Nah, that would be unusual ;)

Happy to see someone put the info on the net though.

[neliz]

e.g. instead of "spikes" of the frame rate going down to below 30 due to a fixed architecture being bottlenecked at certain times redering a scene, would the R600's unified architecture perhaps keep the average frame rate higher?

Bueller?

Does G80 solve all of your framerate dip issues?

[Arnold Beckenbauer]

XTX at 745? fudo says the XT runs at 742. interesting..

Fudo or one of the Inquirer gang said, that the only difference between XT and XTX will be the size of VRAM: 512 MB vs. 1024 MB.

€dit: Hm wait. He said:
I'm currently benchmarking the Radeon HD 2900 XTX, though I'll revist the XT if anyone has any particular requests.
He didn't say, these are HD2900XTX's results.

[Arty]

Note that the tests were done with 4xAA and 16xAF, although in several cases the cards were not very stressed, no doubt. That said, if performance was as good as the 8800 GTX in general, then don't you think the 512MB XT model would sell for higher than $449-499 MSRP?
Back in November '06? Yes. Now? Not exactly. Maybe AMD thinks along the same lines, they are late to the market and the reason they delayed (apparently BOM) will enable them to offer more competitive products. A more "value" product is one of the keys that could help them to get back market share.

Again, its all my speculation as to why the XT could launch $100 lower than the GTX. I havent ruled it out neither have I seen enough to classify the XT as a GTS-class product.

[IbaneZ]

I hope we get some XTX news today before AMD nukes dailytech. :lol:

[TOAO_Cyrus]

http://www.dailytech.com/Overclocking+the+R600/article7044.htm

Updated with 1600x1200 3dmarko6 scores. Anyone have a link to 8800GTX scores at that resolution?

[Cuthalu]

The first 10k's we hear about were on a E6600... drivers maybe? they both focus on the driver being RC4.

He's probably talking about this: http://www.dailytech.com/ATI+Radeon+HD+2900+XT+Performance+Benchmarks/article7043.htm :roll:

3DMark06 = 11447pts. And dailytech OC-test shows 13547pts in 3DMark06. That's too big difference in points with jump from QX6700 to QX6800, would you say?

[russo121]

It's interesting that with a core overclock of 100Mhz (745Mhz to 845Mhz), and memory overclock of 200Mhz (1.6Ghz to 2.0 Ghz)

That's an overclock of 400MHz for memory. My hope is that they test the hd2900xtx with barcelona - that would be a freaking test.

[vertex_shader]

He's probably talking about this: http://www.dailytech.com/ATI+Radeon+HD+2900+XT+Performance+Benchmarks/article7043.htm :roll:

3DMark06 = 11447pts. And dailytech OC-test shows 13547pts in 3DMark06. That's too big difference in points with jump from QX6700 to QX6800, would you say?

13547 is not OC'ed result, this is why something looks wrong :smile:

[vertex_shader]

I'm currently benchmarking the Radeon HD 2900 XTX, though I'll revist the XT if anyone has any particular requests.


Looks like Sven benched the XT, but than where the big difference come?
Anh: D975XBX2 BadAxe2, Intel Core 2 Extreme QX6700, 2x1GB DDR2-800 11447 (http://dailytech.com/Article.aspx?newsid=7043)
Sven: ASUS P5N32SLI-E Intel Core 2 Extreme QX6800, 2x2GB DDR2-800 13547 (http://www.dailytech.com/Overclocking+the+R600/article7044.htm)

Driver is the same, 8.361 RC4.

[Kaotik]

It's not the only thing wrong there, the Core2Quad has to be OC'd a LOT for the numbers, it's the only possible explanation.
He claims 8800GTX gets 14k+ on that system, however on ORB, the highest score for any 8800GTX on Core2Quad in 2800-3000MHz range (2970 in fact, which is slightly higher than XQ6800's default clock) was just 12.6k (even though it's different drivers, but drivers don't make that near 2k jump there)

[Cuthalu]

13547 is not OC'ed result, this is why something looks wrong :smile:

Yes, I know that. I said OC-test, not OC'd result, to be exact. ;) Overclocked QX6800 or the card being XTX would explain the difference between scores.

[neliz]

3DMark06 = 11447pts. And dailytech OC-test shows 13547pts in 3DMark06. That's too big difference in points with jump from QX6700 to QX6800, would you say?

Yep

[vertex_shader]

Yes, I know that. I said OC-test, not OC'd result, to be exact. ;) Overclocked QX6800 or the card being XTX would explain the difference between scores.

Reading the article again, from what i understand Sven tested the XT, he is tesing the XTX now.
I think its a OC'ed QX6800 Sven forget to write this :smile:

I can say one for sure, the 2 dailytech article testing the r600 very unprofessional.

[Razor1]

Looks like Sven benched the XT, but than where the big difference come?
Anh: D975XBX2 BadAxe2, Intel Core 2 Extreme QX6700, 2x1GB DDR2-800 11447 (http://dailytech.com/Article.aspx?newsid=7043)
Sven: ASUS P5N32SLI-E Intel Core 2 Extreme QX6800, 2x2GB DDR2-800 13547 (http://www.dailytech.com/Overclocking+the+R600/article7044.htm)

Driver is the same, 8.361 RC4.


Like Anh, I was able to get pretty close to a 2.0GHz memory clock while still keeping the system stable (http://dailytech.com/Article.aspx?newsid=7043). For reference, my GeForce 8800 GTX (core clock at 650 MHz, memory at 2.0 GHz) scores 14128 on the same system with ForceWare 158.19.


Think its the cpu thats making a difference, man 14k on 3dmark at stock for a g80 gtx, thats alot, I don't remember it going past 12.5 k at stock.

[Kaotik]

The CPU can make the difference, but that QX6800 has to be overclocked a LOT for those numbers like I said in my post above

[Cuthalu]

Think its the cpu thats making a difference, man 14k on 3dmark at stock for a g80 gtx, thats alot, I don't remember it going past 12.5 k at stock.

That gtx is not at stock values.

core clock at 650 MHz, memory at 2.0 GHz

So, 575->650MHz (13% increase) and 1.8GHz->2GHz (11% increase), and if the scores correlate well with core clock increase, it gives us about 14125 (1.13*12500), which coincides very well with 14128pts.

[leoneazzurro]

Think its the cpu thats making a difference, man 14k on 3dmark at stock for a g80 gtx, thats alot, I don't remember it going past 12.5 k at stock.

EDIT: already posted.

[Kaotik]

That gtx is not at stock values.



So, 575->650MHz and 1.8GHz->2GHz.

Yes but still, I just checked ORB for 14k+ numbers, they all have C2Q's in the 3600MHz+ range, XQ6800 is just 2933MHz, and who knows what kinda clocks the 8800GTX's in those benches have

[INKster]

That gtx is not at stock values.



So, 575->650MHz and 1.8GHz->2GHz.

That R600 isn't at stock speeds either.

[jimmyjames123]

They've "fixed" the Half Life 2: Episode 1 score for the 8800 GTS. It went from a clearly wrong 57.4 FPS to 119.2 FPS, vs 112.0 FPS for the XT. Really too bad that they haven't run more benchmarks at settings that stress the cards more (and it's not like they can't do it, since they seem to have produced 3dmark06 scores for an overclocked XT at higher resolutions than their initial test).

[Cuthalu]

I did some checking in the orb, and there's several results with QX6800(stock) and 8800GTX which have over 14000 points, so it shouldn't be impossible without overclocking the processor.

EDIT: now the Dailytech's article says following:
The card used was a Radeon HD 2900 XT 512MB.

That settles it.

[Kaotik]

I did some checking in the orb, and there's several results with QX6800(stock) and 8800GTX which have over 14000 points, so it shouldn't be impossible without overclocking the processor.

EDIT: now the Dailytech's article says following:


That settles it.

Where did you find those QX6800 at stock @ 14k results? All the 14k+ results I found had C2Quads (it doesn't say wether it's normal or "extreme") in the 3.6GHz+ range, while QX6800's stock clocks are 2933MHz

[TOAO_Cyrus]

so anyone have any GTX/GTS comparisions for the 1600x1200 results?

[Razor1]

That gtx is not at stock values.



So, 575->650MHz (13% increase) and 1.8GHz->2GHz (11% increase), and if the scores correlate well with core clock increase, it gives us about 14125 (1.13*12500), which coincides very well with 14128pts.


ah thx for the correction, didn't read what was inside the parenthesis ;)

[Cuthalu]

Where did you find those QX6800 at stock @ 14k results? All the 14k+ results I found had C2Quads (it doesn't say wether it's normal or "extreme") in the 3.6GHz+ range, while QX6800's stock clocks are 2933MHz

There's one: http://service.futuremark.com/compare?3dm06=1583360

[Kaotik]

There's one: http://service.futuremark.com/compare?3dm06=1583360

Noticed the "Co-operative adapters Yes"? It's 2 8800GTX's in SLI

[Cuthalu]

My mistake. :oops:

So, the best single score is 12782 @ 590/1840.

[INKster]

Noticed the "Co-operative adapters Yes"? It's 2 8800GTX's in SLI

8800 GTX had no SLI support yet in the drivers (97.92) they used in that test, i think...

[Kaotik]

8800 GTX had no SLI support yet in the drivers (97.92) they used in that test, i think...

Well something has to explain how it's that much higher than other scores with the same CPU and clocks, and that one has SLI on according to ORB, the other's don't.

[INKster]

Well something has to explain how it's that much higher than other scores with the same CPU and clocks, and that one has SLI on according to ORB, the other's don't.

But if the driver has no 8xxx SLI support, they could have had 4 cards installed, and it still wouldn't make any difference, now would it ? ;)
My bet is that that score is rigged, somehow, but not due to the hardware used.

[Kaotik]

But if the driver has no 8xxx SLI support, they could have had 4 cards installed, and it still wouldn't make any difference, now would it ? ;)
Of course it wouldn't, but got any better ideas to explain the score?
I don't have 97.92 release notes here so can't confirm the SLI support or lack of it, it was 8800-only release though, I think?

edit: I just checked the official release notes, it doesn't specificly state for or against supporting SLI on a quick read, but it does have SLI-related notes and it is 8800-only release, which would imply it does support SLI on them.

[ERK]

Sorry if it's been posted already, but ixbt (http://www.ixbt.com/news/hard/index.shtml?08/27/51) has a comparison up between HD2900XT and 8800GTS.

EDIT: Heh, my bad. They're just linking to the dailytech. (I dont read russian) ;)

[Rangers]

Reading the article again, from what i understand Sven tested the XT, he is tesing the XTX now.
I think its a OC'ed QX6800 Sven forget to write this :smile:

I can say one for sure, the 2 dailytech article testing the r600 very unprofessional.

Well, Dailytech doesn't sign NDA's.

It's most likely just odd editors who got hold of R600's, and post to their blogs as they wish.

It's not a proper reviews, but yeah, they could be better.

[Rangers]

Beyond 3d has really died off since they made the forum software changes.

I cant believe we get confirmed R600 benches after all this time and it generates very little discussion.

[Martin Eddy]

Probably because we've been talking about it for about 6 months. :wink:

[BRiT]

I blame the extensive delays and it falling within our expected performance range.

I'm more interested in what IQ improvements it offers than raw benchmark numbers.

[Rangers]

Probably because we've been talking about it for about 6 months. :wink:

But these are the first ever confirmed benches. Dont get me wrong the discussion of the benchmarks has been somewhat helpful.

Also it's true, hard to be more boring than "almost as fast but cheaper" which is what this version of R600 so far appears to be. So of course the benchmarks (these, so far) are boring I'll grant that.

new inq

http://www.theinquirer.net/default.aspx?article=39145

[Bjorn]

But these are the first ever confirmed benches. Dont get me wrong the discussion of the benchmarks has been somewhat helpful.

Also it's true, hard to be more boring than "almost as fast but cheaper" which is what this version of R600 so far appears to be. So of course the benchmarks (these, so far) are boring I'll grant that.

The benchmarks are also pretty limited at this point. I'd like to see really high FSAA modes (i'm using a 19" 1280*1024 LCD monitor), high resolution benchmarks and video engine tests (need a new graphics card for my htpc) before making any judgements.

[hoom]

I'm waiting on confirmed benchies at 1920*1200 or higher with AA&AF on before getting excited.

[IbaneZ]

I'm waiting on confirmed benchies at 1920*1200 or higher with AA&AF on before getting excited.

The settings they used, 4xAA/16xAF at 12*10 is fine by me. *Kicks my old 19" LCD*. :lol:

Fuad says the XTX 1024 MB is clocked at 800/2200 MHz. It's gonna be fun to see if that extra bandwidth will make a big difference. Hopefully DT will post some benchies later today. :)

[_xxx_]

Beyond 3d has really died off since they made the forum software changes.

I cant believe we get confirmed R600 benches after all this time and it generates very little discussion.

That's because we learned not to believe every chap who comes around the corner with some "special secret preview benchmark numbers". If it was some more competent/credible source, it would surely look different.

[nicolasb]

Fudo says:

HD2900 XTX with 1GB scores 13,5k+

R600XTX or the faster iteration of Radeon HD 2900 series clocked at 800 MHz core and 2200 MHz GDDR4 memory will score 13.500+ in 3Dmark06 default settings. You will achieve this from QX6800 quad core and 8.36 drivers.

Geforce 8800 Ultra (at 650 – 2160 MHz memory will score about 500 faster or about 14.000 marks. It would not with the old driver but with Forceware 158.19 it scores much better in this test.

The ATI's final driver the one that fixes the Audio bit should make Radeon HD 2900 XTX to perform at the same level as G80 Ultra. Not all hopes are lost. Our reader reminded us that we claimed before that Radeon HD 2900 XTX performs slower than the XT card and we have to say that these tests where done with the latest revision of the 1 GB card and therefore the scores are now much better.

[chavvdarrr]

very strange...

R600 G80
320 128 scalar PU
512 384 bit wide bus


Yet...R600 is not significantly faster... why ?! Not again "16 ROPs are enough" eh?!

[Kaotik]

very strange...

R600 G80
320 128 scalar PU
512 384 bit wide bus


Yet...R600 is not significantly faster... why ?! Not again "16 ROPs are enough" eh?!

Where did you get that R600 would have scalar alus?
Also, straight comparison like that would require the alu's to be identical, which they surely are not.

[INKster]

Also, straight comparison like that would require the alu's to be identical, which they surely are not.

Yes, but if we leave the ALU design aside, there's still the bandwidth issue.

R600 XTX: 140.8 GB/s
G80 Ultra: 112.8 GB/s

Why can't a 512bit, 2200MHz memory card clearly beat a 384bit, 2350MHz memory competitor ?
That huge 28 GB/s difference suggests that R600 might not have the same bandwidth-saving capabilities as G80.

[Kaotik]

Yes, but if we leave the ALU design aside, there's still the bandwidth issue.

R600 XTX: 140.8 GB/s
G80 GTX: 112.8 GB/s

Why can't a 512bit, 2200MHz memory card clearly beat a 384bit, 2350MHz memory competitor ?
That huge 28 GB/s difference suggests that R600 might not have the same bandwidth-saving capabilities as G80.

Well for the memorybandwidth alone to make any sort of real difference, the one with less bandwidth should be limited by that memory bandwidth of it's at the said settings.

[PatrickL]

very strange...

R600 G80
320 128 scalar PU
512 384 bit wide bus
Yet...R600 is not significantly faster... why ?! Not again "16 ROPs are
enough" eh?!

Maybe wait for more serious tests at higher resolutions before judging the cards :)

[Evildeus]

Want an XT for 380€?

http://www.sallandautomatisering.nl/?redirect=/content/productinfo.php?pid=34021 ;)

Chipset ATi Radeon HD2900 XT
Geheugen 512 MB
Type geheugen GDDR3
GPU snelheid 750 MHz
Geheugen snelheid 828 MHz
Interface PCI-e 16x
RAMDAC 400 MHz
TV-out HDTV / HDCP / HDMI
Video-in Ja
DVI 2 x
D-sub Nee

[INKster]

Want an XT for 380€?

http://www.sallandautomatisering.nl/?redirect=/content/productinfo.php?pid=34021 ;)



I think you didn't include the 19% Dutch BTW tax (VAT equivalent). ;)

edit
scratch that, it's there.
A nice price indeed.

[Evildeus]

I think you didn't include the 19% Dutch BTW tax (VAT equivalent). ;)

edit
scratch that, it's there.
A nice price indeed.:p Indeed it's 380 VAT included ;)

[neliz]

Wow.. And you can image salland isn't one of the cheapest here either.. I feel a pre-order comming up :)

€ 388,99 inc. shipping fee! :) w00h00 (oh.. 381 when paying with iDeal..)

[Love_In_Rio]

Based on the pics of the cooler system could this be 65nm finally ?

[erick]

Dailytech overclocks 2900 XT: http://www.dailytech.com/Overclocking+the+R600/article7044.htm

Memory goes up to 1995 MHz apparently, while GPU reaches 845 MHz (745 MHz by default).

[INKster]

Dailytech overclocks 2900 XT: http://www.dailytech.com/Overclocking+the+R600/article7044.htm

Memory goes up to 1995 MHz apparently, while GPU reaches 845 MHz (745 MHz by default).

http://forum.beyond3d.com/showpost.php?p=975257&postcount=2654

[Kaotik]

Based on the pics of the cooler system could this be 65nm finally ?

It's still 80nm, and has always been, it was known for ages that the full-length card wouldn't be hitting the retail shelves, but was a OEM model, and the fan was placed after the PCB itself to get it to full-length specs, which enables OEM's to secure the card from both ends to the case, which is safer for transportation.
When the 65nm models come, they're R650's, not R600's, and probably named HD 2950's or something

[erick]

http://forum.beyond3d.com/showpost.php?p=975257&postcount=2654


Oops, didn't think I was so much behind the times :P

[neliz]

Oops, didn't think I was so much behind the times :P

Beyond3D, when you blink, the rest is next-gen.

[Unknown Soldier]

:D

Whahhaa .. ye, miss a day and you almost half a year behind.

US

[Lux_]

G80's ALU's are clocked twice the core. Therefore it's more like 320 vs 256.


R600 G80
320 128 scalar PU
512 384 bit wide bus

[INKster]

G80's ALU's are clocked twice the core. Therefore it's more like 320 vs 256.

Actually, i think the core/shader clockspeed ratio is closer to 1: 2.35, but who's counting ? :)

[Sound_Card]

When you are OC'ing G80, are you also OC'ing the shaders?

[Anon Lamer]

Any 3dmark03 scores? Its the only 3dmark where the score isnt influenced by cpu. I dont understand why they abandoned this approach.

[Razor1]

When you are OC'ing G80, are you also OC'ing the shaders?


They can go with the core overclock, depends, the shader domain has its own clocks so they can be overclocked seperately.

[nicolasb]

G80's ALU's are clocked twice the core. Therefore it's more like 320 vs 256.

Actually, i think the core/shader clockspeed ratio is closer to 1: 2.35, but who's counting ? :)He was comparing the clockspeed of Nvidia's shaders with clockspeed of ATI's shaders. You're comparing Nvidia's shaders with Nvidia's core. Not the same.

[INKster]

He was comparing the clockspeed of Nvidia's shaders with clockspeed of ATI's shaders. You're comparing Nvidia's shaders with Nvidia's core. Not the same.

No, he wasn't.
Take a look at what he said about G80's ALU's in your post above...
Since R600's shaders have the same clockspeed as the main core, the G80 reference and the double clockspeed is pretty obvious.
Besides, even comparing the shaders of the R600 XTX at 800MHz, versus the G80 Ultra's at 1600MHz isn't really fair, since the fundamental design of each of them is completely different.

[Jawed]

Since R600's shaders have the same clockspeed as the main core
That's an unknown, since the architecture itself isn't known...

Jawed

[INKster]

That's an unknown, since the architecture itself isn't known...

Jawed

Yes, but i was speculating based on what is known now about R600.
No news/rumors about a separate shader clock domain so far in the game points to no domain at all, although i could be wrong, of course.

[CJ]

Wow.. And you can image salland isn't one of the cheapest here either.. I feel a pre-order comming up :)

€ 388,99 inc. shipping fee! :) w00h00 (oh.. 381 when paying with iDeal..)

Yep, pre-order inc! It's been confirmed by someone else that even the HIS HD2900XT will cost around €300-310 excluding 19% VAT. So add the 19% to the price and you get around €368,90. Supposedly none of the AIBs have plans for an XTX yet, but production for the XT is going on at full speed at the moment. AIBs are busy allocating stock as we speak. There might be a small shortage in the beginning since a lot of people are interested in the HD2900XT, but there will be enough for everyone.

[ants]

EDIT: Sorry it is a repost...

Didn't see this posted yet...

http://www.sallandautomatisering.nl/?redirect=/content/productinfo.php?pid=34021

Pre-order page for a Sapphire 2900XT

Before the page is pulled the specs shown are...

Chipset ATi Radeon HD2900 XT
Memory 512 MB
Memory Type GDDR3
GPU clock 750 MHz
Memory clock 1660 MHz
Interface PCI-e 16x
RAMDAC 400 MHz
TV-out HDTV / HDCP / HDMI
Video-in Yes
DVI 2 x
D-sub No

[pakotlar]

Actually, i think the core/shader clockspeed ratio is closer to 1: 2.35, but who's counting ? :)

More like 216 vs 320 if we are using 1350mhz ALU clock and 800mhz ALU clock for x2900xtx (at 745mhz it becomes 232). Obv, this does not account for architec. differences, but on power, the x2900xt (at 745mhz) should have ~ 38% more shading power than the 8800gtx.

[Bob]

More like 216 vs 320 if we are using 1350mhz ALU clock and 800mhz ALU clock for x2900xtx (at 745mhz it becomes 232).
If you're going to go though these contorsions, you might as well just count GLFOPS.

[Kaotik]

Actually, i think the core/shader clockspeed ratio is closer to 1: 2.35, but who's counting ? :)

There's no fixed ratio, don't remember the exact mechanism, but for example GTS = 500/1200 (2.4), GTX = 575/1350 (~2.358), Dell-clocked Ultra = 650/1566 (~2.409)

[INKster]

More like 216 vs 320 if we are using 1350mhz ALU clock and 800mhz ALU clock for x2900xtx (at 745mhz it becomes 232). Obv, this does not account for architec. differences, but on power, the x2900xt (at 745mhz) should have ~ 38% more shading power than the 8800gtx.

Scalar architectures are usually more efficient than vector ones because they can work on more than one instruction (at different stages) per clock cycle per ALU.
The greater amount of stages also has the interesting side-effect of boosting clockspeeds at the cost of latency, but i'm no expert in these matters :D.

[Arty]

Want an XT for 380€?

http://www.sallandautomatisering.nl/?redirect=/content/productinfo.php?pid=34021 ;)

Chipset ATi Radeon HD2900 XT
Geheugen 512 MB
Type geheugen GDDR3
GPU snelheid 750 MHz
Geheugen snelheid 828 MHz
Interface PCI-e 16x
RAMDAC 400 MHz
TV-out HDTV / HDCP / HDMI
Video-in Ja
DVI 2 x
D-sub Nee

Yep, pre-order inc! It's been confirmed by someone else that even the HIS HD2900XT will cost around €300-310 excluding 19% VAT. So add the 19% to the price and you get around €368,90. Supposedly none of the AIBs have plans for an XTX yet, but production for the XT is going on at full speed at the moment. AIBs are busy allocating stock as we speak. There might be a small shortage in the beginning since a lot of people are interested in the HD2900XT, but there will be enough for everyone.

I think someone here is squirming because of the impending bet loss .. :lol:

[Frank]

Scalar architectures are usually more efficient than vector ones because they can work on more than one instruction (at different stages) per clock cycle per ALU.
The greater amount of stages also has the interesting side-effect of boosting clockspeeds at the cost of latency, but i'm no expert in these matters :D.
Yes, you can utilize the ALUs better, because you have finer instruction granulary. And because they're simpler, you can clock them higher. But you need a lot more decoding and scheduling logic, and more buffers to keep more (but smaller) threads in flight.

[Evildeus]

I think someone here is squirming because of the impending bet loss .. :lol:
lol :lol:

[Silent_Buddha]

Those prices are encouraging. At this rate, I wouldn't be surprised to see HD 2900 XT street for below 399 USD withing 2-3 weeks of launch at some of the places that historically price for volume rather than margin. However, I expect Newegg to price the HD 2900 XT above 399 USD as they love to wring every last cent when demand is high.

As to what's currently been benched so far. What's there to generate buzz on? Absolutely nothing but FPS at rather low res with settings that would hardly tax a high end card.

When I start seeing comparisons at 1920x1200 and up feating 16x CSAA (NV) vs 24x CFAA (ATI), max AF, and all eye candy on. Then I'll start to get either excited or disappointed. :P

Plus in depth IQ comparisons. It's a given that edge AA should be basically equivalent between the two at this point. But I'm wondering if either does anything to further address transparency AA and shader AA artifacts.

What I'm absolutely not interested in is benches of G80 or R600 with 4x or less AA and 8x or less AF. Any site that benches with no AA and no AF immediately will go onto my ignore and never visit again list. :P

Regards,
SB

[Arty]

So there is a difference in clocks between the XT & XTX:

XT : 750/830
XTX : 800/1100

Also now Fuad has changed his tune saying that XTX might not be delayed after all. :lol: