AMD: R7xx Speculation

[ZerazaX]

Not sure but I saw this:

http://en.expreview.com/2008/04/26/800gtx-scores-5851-in-3dmark-vantage/

9800GTX scored P5851 with GPU score 5636

If the CPU was relatively similar, then thats an impressive P score for the 4850

 

[ZerazaX]

Did anyone notice the core runs at 625 MHz? (Now where does that lead to?)

Yes, I thought that was what the rumor for the 4850 was..

 

[Lukfi]

So, higher-clocked shader domain is now accepted as the most probable possibility? That's what the rumour also says.

 

[ZerazaX]

Any confirmation on shader clocks or count CJ?

 

[annihilator]

Wasn't there an official AMD slide confirming the 1-something GFLOPS that was calculated from the rumors?

Or were the rumors made up to meet that processing rate?

 

[CarstenS]

Thanks.

With various AI settings R6xx IQ varies less than R5xx, but in terms of shimmer R600 is never as good as R5xx at its best?

AI only affects sharpness/shimmer (and performance). Angle-dependency is not affected by AI, is it?

Jawed

That's what I'd also get from it, yes.

 

[leoneazzurro]

(from the GT200 thread)

Very low consumption in idle is something I deem almost certain for RV770-based cards. I think idle power draw around 30 watts should be doable with R700.

There were rumors saying that RV770 would draw 10W in idle, but there were no confirmation about this.

 

[Pete]

Arun, I'm not sure how to read that wink b/c I'm not sure how CPU speed affects P scores. Are there big differences? Big enough to affect our guesses as to how the rumors tally with these scores?

HW.fr says doubling the CPU score bumps the X score by 1%, so totally winging the calc by the %ages shown there should mean a similar doubling shouldn't affect P scores by more than 5%. Dummy #s show that doubling or halving a CPU score of 10k nets a 3-8% change in total score. So, not much, it seems, but I'm guessing your wink means the 4850 was tested on a Phenom which scores less than the QX9770 and so the 4850's score is more impressive relative to the 3850. Then again, the 4850 was OCed.

Lukfi, you referring to Arun's 625*800*2 = 1 Tflop? Ah, nevermind, you're talking about a separate shader clock. Assuming only 4870 (not 4850) will hit 1Tflop, it'd only need 1.04GHz to reach it with 480 ALUs (or whatever we're calling them). Only intensifies my head-scratching over majorly improved Vantage #s with similar texturing power.

 

[ZerazaX]

Except I think it was the 4870 that was rated at 1 TFlop because I doubt the 4850 is. And the 4870 supposedly has clocks at 800 (or was it 850) for core so that would be incorrect as well unless we're suggesting the 4870 has > 1TFlops performance

 

[Tchock]

I think there's one CCC setting that I find dubious overall.

 

[annihilator]

You mean x16 AF, right? What's Vantage's Performance default for AF?

 

[ZerazaX]

Performance is Trilinear i think? High is 8 and Extreme is 16

 

[Silent_Buddha]

What I'm going to be most interested in seeing when sites can run and post extensive benchies...

How Rv670 compares to Rv770 with no-AF and with 16xAF.

I'm wondering if the gap will widen significantly once 16xAF is enabled, showing that ATI did address that failing of R(v)6xx.

Regards,
SB

 

[annihilator]

OK, thanks.

By the way, why is ALL information/pics etc. of the new ATI series about 4850? And now it has surfaced on Computex.

What about 4870? Why still no confirmed visuals or information? And what's that with HD3950? I can't understand.

 

[ZerazaX]

4870 simply won't be released yet til end of June / mid July at least it seems due to GDDR5 or they're still trying to keep it a secret. 4850 has already been shipped to OEM's and AIB's and it looks like they're going to retail now as well, so it's already ready to be displayed and thrown into the wild.

And after hoom posted those 3dMark scores which showed how much AF hurt compared to AA.... if ATI fixed the AF problem, then the AA + AF scores + improved clocks and shaders can make RV770 quite potent finally and vindicate the R600 architecture

 

[AlphaWolf]

OK, thanks.

By the way, why is ALL information/pics etc. of the new ATI series about 4850? And now it has surfaced on Computex.

What about 4870? Why still no confirmed visuals or information? And what's that with HD3950? I can't understand.

I'd bet because 4870 is still in limited quantity due to GDDR5.

 

[ZerazaX]

8800GTX was P6417 at 625 core / 1500 shader / 1000 mem clocks from a user on [H] w/ a quad at 3.2

So in Vantage at those settings, 4850 ~ 8800GTX

 

[Lukfi]

=>leoneazzurro: I can't confirm it either since I don't know for sure, but something tells me it might be true.

=>ZerazaX: They sure aren't keeping 4870 a secret since they made a press release about their next-gen boards using GDDR5.

 

[Tchock]

@ZerazaX:

Yup, no aniso, just tri. It does seem 2 steps back when AF is getting enforced in general settings nowadays.

Oh, and notice the absence of shader clocks? That's a given. Changing the oh-so cheap ALUs, perhaps required in a radical way, to run 200Mhz higher is deep into ludicrous territory.

 

[leoneazzurro]

@ZerazaX:

Yup, no aniso, just tri. It does seem 2 steps back when AF is getting enforced in general settings nowadays.

Oh, and notice the absence of shader clocks? That's a given. Changing the oh-so cheap ALUs, perhaps required in a radical way, to run 200Mhz higher is deep into ludicrous territory.

Also on Nvidia control panel there's no setting for the shader clock, only for the core (at least in the drivers I have). You can set (and see) shader speed only with other tools.