AMD: R7xx Speculation

Status
Not open for further replies.
Hmm, there's now mention of Rv770 having a split clock domains ala G8x and G9x?

Considering that rumors now are that it has 480 SPs and 32 texture units on 16 ROPs, wouldn't a higher shader clock just keep the bottleneck in the texturing units?

Regards,
SB

Probably, but on those titles that are shader limited it could do nothing but help.
 
One of the interesting comments in that article is "According to our sources, AMD is no longer clocking the cards towards performance, but towards power efficiency instead." .

It seems to me a subtle change that would come from the "AMD" portion of the company where the importance of being the absolute fastest at whatever costs, is now being directed to make the best card for a Platform. ATI as a separate company competed against nVidia only and as such the nature of the competetion always had the goal of being the fastest. With AMD at the helm it seems that there is more emphasis on price and balance.

While for some this is bad in that it does not seem to push the envelope for speed, the new direction is aiding all of us in pricing. If we had kept going in the old direction, the costs of the top end graphics card would have been astronomical whereas its getting reasonable. Those of us that never would have bought the top end are now first the first time in ages allowed to buy at reasonable price pretty performant cards from both nVidia and AMD.

At least from my perspective, that is good.
 
I'm dubious about a clock domain for the ALUs because of the use of the register file. Both ALUs and TUs appear to work directly with the register file which makes me sceptical about the timing of reads and writes if ALUs and TUs are clocked independently.
What if the TUs would also work at shader domain clock? Maybe instead of 6 clusters with 16x5 shader units and 32 tmus, 4 clusters with 24x5 shader units and thus 24 tmus? The texel fillrate numbers would be different, but close at the rumoured clock speeds. Sure the 6 cluster version sounds better, but also significantly more complex, and as you pointed out would be quite a bit more different to rv670.
Aren't baseless rumours just fun :). Oh and are AMD seriously thinking about a 256MB version of a 4850? Now they don't have the serious problems when under video memory pressure as nvidia's G80/G92, but still it would look quite seriously underequipped for a card of that caliber, no matter what the final specs will be...
 
One of the interesting comments in that article is "According to our sources, AMD is no longer clocking the cards towards performance, but towards power efficiency instead." .

It seems to me a subtle change that would come from the "AMD" portion of the company where the importance of being the absolute fastest at whatever costs, is now being directed to make the best card for a Platform. ATI as a separate company competed against nVidia only and as such the nature of the competetion always had the goal of being the fastest. With AMD at the helm it seems that there is more emphasis on price and balance.

While for some this is bad in that it does not seem to push the envelope for speed, the new direction is aiding all of us in pricing. If we had kept going in the old direction, the costs of the top end graphics card would have been astronomical whereas its getting reasonable. Those of us that never would have bought the top end are now first the first time in ages allowed to buy at reasonable price pretty performant cards from both nVidia and AMD.

At least from my perspective, that is good.

If you consider sub-par "top-end" performance by AMD/ATI, then yes the price cut matches nicely. However, Nvidia is truly top-end, and their prices show it as well. So far nothing has really changed in that playing field when you consider performance/price ratios.
 
Well "sub-par "top-end" performance by AMD/ATI" maybe is known by enthusiasts, but I am not sure the masses really are that keyed into that measurement. We are talking about the same people that look at the memory size '512MB' and will buy the card ignoring that the card is a low end card.

Thus when we see 'high-end cards from AMD/ATI and they are at a lower price, nVidia has to match the target price regardless of where we may think the price-performance is. So the price gets lower for both vendors for those now main stream cards.

The high-end cards then are priced relative to those. Many times the reviews will say that x-card is the fastest but is a 10% increase in performance worth a 100-200 price difference. So I really think there is pressure on the top end cards now to be within a reasonable amount of the main stream cards.

For some the fastest will override the cost, but for many others saving by getting the next best thing is what happens. I don't see a lot of people saying they own the Ultra from nVidia but tons have the 8800GTS. A comprimise in speed and cost.
 
Well who knows how the temptations of the masses flow anyway? One day it's Ipods, and the next day it's GeForce MX4000, and next thing you know we're back to those '90s Snap Bracelets!!! Or something!

Personally, I am very happy to see idle power use taken into consideration. My 8800GTX is ridiculous, with something like 100W idle power draw, and would never be left on 24/7. I was so impressed with R38x0's draw being less than a 8600GT that I bought one. I needed decent 3D for my comp that is on always. 9600GT is nice as well, and I hope RV770 and NV's next GPUs are as efficient.
 
I agree 100% on power.

I am the payer of the Hydro bill in my home and with 5 computers in the house we have been starting to get very careful about turning off stuff when not in use. Power efficiency is becoming more and more important.

A friend of my son who has built his super power machine, 1000Watt power supply and nVidia 8800GTX over clocked made me laugh when he said that he can't leave his computer on anymore overnight because his room gets so heated up that it kept waking him up. Hard to argue for air conditioning when the rest of the house is fine and you have the only Tropical climate zone in house.

I hope we get more and more efficient in power but with decent speed available for our toys.
 
What Shaidar means is it's the same pic, cropped and rotated, but the original pic (that Arnold linked) is itself a fake, apparently.
 
Hmm, there's now mention of Rv770 having a split clock domains ala G8x and G9x?

Considering that rumors now are that it has 480 SPs and 32 texture units on 16 ROPs, wouldn't a higher shader clock just keep the bottleneck in the texturing units?

Regards,
SB

ATI has had split clock domains for a long time in current chips.
 
A friend of my son who has built his super power machine, 1000Watt power supply and nVidia 8800GTX over clocked made me laugh when he said that he can't leave his computer on anymore overnight because his room gets so heated up that it kept waking him up.
Heh, that reminds me, when I saw one of those at Newegg's site the other day, I thought "that's not a powersupply, that's a heater."

I know I haven't kept up with the "up to par" performance segment, but sheesh, 1000Watt PS. There is no way this segment can keep going in this direction. If that's what it takes to play the latest PC games, then PC gaming might as well be dead.
 
Once again...

The latest top notch PC system (Intel) does not need over 500 watts. The newer CPUs use less power than the system listed below.

From XBIT Labs 1000 Watt PSU Review:

Intel-based system (Power Draw 442 Watts):
  • Quad-core Core 2 Extreme QX6700 CPU (Kentsfield) overclocked to 3.5GHz
  • Two Foxconn GeForce 8800GTX graphics cards in SLI mode
  • ASUS Striker Extreme mainboard (LGA775, NVIDIA nForce 680i SLI)
  • 2GB DDR2-800 SDRAM (Mushkin XP2-6400PRO, 4 x 512MB)
  • Two Western Digital WD1500AHFD hard disk drives in a RAID0
  • Various trifles like a DVD-ROM, fans, etc

We installed Windows XP SP2 on these systems and ran Stress Prime 2004 / Orthos for the CPU and 3DMark 2006 for the graphics card; these two programs were running simultaneously in the third test mode. Here are the PSU power consumption numbers (using a Tagan TurboJet TG1100-U96; we measured its power draw from the wall outlet and multiplied the result by this PSU’s efficiency factor, about 0.83):

p1.png
 
Netiquette, shmetiquette--that's just old news! :p The thread you linked links the original source at the bottom of the post, that German (some would say French) site.
 
Status
Not open for further replies.
Back
Top