The LAST R600 Rumours & Speculation Thread

[jamesnmandy]

Yeah I think its better to release product later with working drivers than have product released but 5 months without drivers

of course redemption in the form of new drivers delivering performance on par with the late to market R600 would be possible still....they've had a long time to develop a driver, perhaps waiting for R600/G81 release to steal some thunder

 

[_xxx_]

Ahem, "Singlefire"?

 

[Robbitop]

If you make a decision like that in the design process, i'd be willing to bet the engineers would find ways to hide the latencies - maybe through minuscule caching, high frequency RAM (what matters is time, not clock cycles) or whatever.

As you know a clock cycle is a unit for time if you know the dependancy to the clockrate of the ASIC.
An of course for things like texture caching and for storing terms for arithmetical operations latency is a problem, which cannot be hidden so easily.

Thanks for thinking i'd be foolisch enough to think you could take any given GPU and substitute its caches with VRAM.

I'd not call it foolish but creative combined with a bit of crazyness. X-D
Don't sweat it, I never ment to implicate such things. I just wanted to answer your question properly.

 

[trinibwoy]

Are you being sarcastic?

Nope, lol I mean, what possible excuse could Nvidia have if ATi produces cleaner drivers without the benefit of user feedback that Nvidia has had for the last 4 months?

 

[KimB]

Nope, lol I mean, what possible excuse could Nvidia have if ATi produces cleaner drivers without the benefit of user feedback that Nvidia has had for the last 4 months?

But I don't see how it's pie on nVidia's face given that they've had the hardware out for people to use for the last four months (or however long it's been). You should also know by now that WHQL does not mean bug free.

 

[trinibwoy]

Yeah, they certainly get their props for getting hardware out five months earlier, but that actually makes it worse since their software guys presumably have had more time with final hardware. What Nvidia is going through now is exactly what I see happening in projects at work where there was poor planning and execution. Excuses and promises. It just reflects on them poorly, no matter how you look at it.

There's also the confidence factor. If ATi were to produce relatively good, stable drivers from day one it would certainly impact consumer confidence in making a stress free purchase especially under Vista. Now, Nvidia does have time to get their act in order but I don't see how scrambling for a couple weeks can make up for months of shoddy work.

 

[Robbitop]

Silent_guy. The numbers of clocks are dependant to the access time in ns relative to the core clock of the ASIC. In case of R600 I took the value of 800 MHz.

You are only looking at the access time of a single memory cell. But it takes time to allocate adresses, to find the right data, and so on. Espeacially to external memory with very bad CL rates (GDDR3 has a CL of 7!!) suffers from higher latency.
The real latency is pretty hard to determine because it's more than just a single access to a single RAM cell. (as we remember the K8 has much much worse access time to its L1 caches than the C2D has, the L2 and L3 latencies are much worse).

The values I have given are from a made who programs a lot with ASM. (of course they are different from other ASICs, but the difference between the sots of RAMs and flipflops are pretty high)

 

[stevem]

You should also know by now that WHQL does not mean bug free.

Is that old (XP) or new (Vista) WHQL? There's bugs, then there's compliance...

 

[KimB]

Is that old (XP) or new (Vista) WHQL? There's bugs, then there's compliance...

Why would anything change under Vista? The new OS adds whole new layers of complexity, meaning more possible software configurations need to be tested. So many more that improving the testing method cannot hope to cope with the increased complexity required.

 

[Chameleoki]

Ahem, "Singlefire"?

Crossfire on a single card = SingleFire????

 

[silent_guy]

Silent_guy. The numbers of clocks are dependant to the access time in ns relative to the core clock of the ASIC. In case of R600 I took the value of 800 MHz.

You are only looking at the access time of a single memory cell. But it takes time to allocate adresses, to find the right data, and so on. Espeacially to external memory with very bad CL rates (GDDR3 has a CL of 7!!) suffers from higher latency.
The real latency is pretty hard to determine because it's more than just a single access to a single RAM cell. (as we remember the K8 has much much worse access time to its L1 caches than the C2D has, the L2 and L3 latencies are much worse).

The values I have given are from a made who programs a lot with ASM. (of course they are different from other ASICs, but the difference between the sots of RAMs and flipflops are pretty high)

I don't doubt that a CPU read or a GPU read from external memory takes 150 cycles, but that doesn't mean you can do much better than that. As you say yourself, CL is only 7 cycles, so it's clearly not the fault of the RAM that the total read takes 150 cycles. It's the clock domain synchronization, the arbitration etc. And, yes, there are a lot of practical cases, where you can go much lower than 150 cycles, if you have very predictable coherent streaming patterns.

Now take internal SRAM: like I wrote the large CPU caches are major exceptions because of their size. But in a typical chip, pretty much all SRAM's run at speed and they have no problem with allocation etc. So to assume that an R600 needs 10 cycles to access an SRAM is excessive.

It's not very helpful to make generalized claims without context and you're only now providing it and then only for the SDRAM.

 

[jb]

But I don't see how it's pie on nVidia's face given that they've had the hardware out for people to use for the last four months (or however long it's been). You should also know by now that WHQL does not mean bug free.

How can it not be pie on their face? Vista did not come out over night. If you compettior is able to come out with drivers that work and have the same level of profromance as the older OS, and you not.... Remember nV has aways been thought of as having better drivers. Here is one case where they are behind.

 

[KimB]

How can it not be pie on their face? Vista did not come out over night. If you compettior is able to come out with drivers that work and have the same level of profromance as the older OS, and you not.... Remember nV has aways been thought of as having better drivers. Here is one case where they are behind.

ATI hasn't done that yet, not for the R600. Have they done it for their older hardware? Have people done tests on Vista with nVidia's older hardware?

Edit: Did a Google, and found this:
http://www.pcper.com/article.php?aid=354&type=expert&pid=1

From this article, at least, I see no reason to believe that ATI's Vista drivers are any better than nVidia's. Some games nVidia does as well or better under Vista, while ATI registers large performance drops. Other games are the exact reverse. Anyway, nVidia's got WHQL Vista drivers out today. I'm going to check 'em out and see if they fix my outstanding issues (flat panel scaling doesn't work properly, and the mouse is not visible in CoH/CoV, under 100.64).

 

[pjbliverpool]

ATI hasn't done that yet, not for the R600. Have they done it for their older hardware? Have people done tests on Vista with nVidia's older hardware?

Edit: Did a Google, and found this:
http://www.pcper.com/article.php?aid=354&type=expert&pid=1

From this article, at least, I see no reason to believe that ATI's Vista drivers are any better than nVidia's. Some games nVidia does as well or better under Vista, while ATI registers large performance drops. Other games are the exact reverse. Anyway, nVidia's got WHQL Vista drivers out today. I'm going to check 'em out and see if they fix my outstanding issues (flat panel scaling doesn't work properly, and the mouse is not visible in CoH/CoV, under 100.64).

Any chance of a report on the sleep function? It didn't work properly in previous drivers.

 

[Mintmaster]

From this article, at least, I see no reason to believe that ATI's Vista drivers are any better than nVidia's.

Does any of this bickering have anything to do with performance?

I was under the impression that it was things like the late availability of initial drivers, incompatibility, broken features, lack of SLI, developers seeing problems, etc. Namely, the usual driver rants.

 

[Monty]

Any chance of a report on the sleep function? It didn't work properly in previous drivers.

Working for me. Shocked at how quick it got out of sleep mode, came on quicker than my monitor

 

[rwolf]

ATI hasn't done that yet, not for the R600. Have they done it for their older hardware? Have people done tests on Vista with nVidia's older hardware?

Edit: Did a Google, and found this:
http://www.pcper.com/article.php?aid=354&type=expert&pid=1

From this article, at least, I see no reason to believe that ATI's Vista drivers are any better than nVidia's. Some games nVidia does as well or better under Vista, while ATI registers large performance drops. Other games are the exact reverse. Anyway, nVidia's got WHQL Vista drivers out today. I'm going to check 'em out and see if they fix my outstanding issues (flat panel scaling doesn't work properly, and the mouse is not visible in CoH/CoV, under 100.64).

Average Frame rate difference between Vista and XP
BattleField 2 - Nvidia: -19.9 FPS ATI: 0 FPS
Fear - Nvidia: -11 FPS ATI: -8.5
Lost Coast: - Nvidia: -17.7 ATI: +.5

http://www.beyond3d.com/forum/showthread.php?t=38642

 

[Razor1]

you have to look at the 7900 drivers not the g80, since the g80 is a new GPU its going to have a compounding effect of this and Vista drivers.

 

[trinibwoy]

Yeah I agree, ATi shouldn't get a free ticket on the new architecture ride until R600 comes out and proves itself. But how does R580 compare to G71 on Vista?

 

[Pete]

That PCPer link Chal contributed has 8800GTX, 7900GTX, and X1950XTX #s at 16x12, 20x15, and 25x16, all 4xAA 8xAF. rwolf seems to be using the 25x16 #s (with a typo or two). If we compare 7900GTX and X1950XTX at that same res, and throw in CoD2 and Prey with their caveats:

[U]Game[/U]  [U]7900GTX / X1950XTX[/U]
BF2:     -2.1 / 0
Fear:    -2.8 / -8.5
HL2:LC:  -6.0 / +0.5
CoD2:   -3.7* / -0.5     * ctrl. panel AA
Prey:    -1.3 / n/a**    ** ATI OGL installer error, no score

Compared to the 7900, the 8800 takes a much bigger hit in BF2, HL2:LC, and Prey; loses about the same in FEAR; and actually gains some frames in CoD2.

Anandtech has a Vista Perf. Guide with both GPU- and CPU-bound (20x15 and 8x6, I think) resolution charts for a wide range of cards with Oblivion (D3D), Q4 (OGL), and HL2:LC (x64, albeit only 8800 and X1950 #s there). Overall, AT shows a wash.