AMD: R7xx Speculation

[leoneazzurro]

I just stumbled upon this little site with some information about R700. Real or not?

3Dmark Vantage 12,5K score in Extreme settings?
If this is true, assuming even a 90% scaling, each R700 should score a little less than 7000 at the same settings, and 3800-4000 for the single 4870. Nice if true.

EDIT: what CPU is in this system? Four quad Opterons?
EDIT2: no, CPU score pretty "normal"... hmmm

 

[Jawed]

I just stumbled upon this little site with some information about R700. Real or not?

Damn.

Tri-SLI GTX280 at 702/1511/1196 (overclocked 17%/17%/8%) is 16% faster than R700

http://www.xtremesystems.org/forums/showthread.php?t=191313

That's X14556 for 3 GT200s and X12526 for 2 RV770s.

Jawed

 

[fellix]

now comes the MGPU-revolution?

It's their secret O.C.S.P.G. (One CrossFire Setup Per Gamer) program!

 

[AlphaWolf]

Damn.

Tri-SLI GTX280 at 702/1511/1196 (overclocked 17%/17%/8%) is 16% faster than R700

http://www.xtremesystems.org/forums/showthread.php?t=191313

That's X14717 for 3 GT200s and X12526 for 2 RV770s.

Jawed

So 16% faster with a 17% overclock, does that mean they equal at default.

 

[leoneazzurro]

Damn.

Tri-SLI GTX280 at 702/1511/1196 (overclocked 17%/17%/8%) is 16% faster than R700

http://www.xtremesystems.org/forums/showthread.php?t=191313

That's X14717 for 3 GT200s and X12526 for 2 RV770s.

Jawed

It's 2 R700, anyway

Edit: anyway, what's a little strange is that in the slide is said that R700 is a 2 Teraflop borad. Now, if it's 2X4870, we should have 2,4 Teraflop per board (and marketing rarely misses these numbers). So it's possible that R700 is 2x4850 + GDDR5 instead...

 

[AnarchX]

Damn.

Tri-SLI GTX280 at 702/1511/1196 (overclocked 17%/17%/8%) is 16% faster than R700

http://www.xtremesystems.org/forums/showthread.php?t=191313

That's X14717 for 3 GT200s and X12526 for 2 RV770s.

Jawed

R700 @ Crossfire X are 4 RV770.

 

[Jawed]

R700 @ Crossfire X are 4 RV770.

Hahahaha I was thinking it was fishy implying ~X7000 for one RV770.

Jawed

 

[Sunday]

It's 2 R700, anyway

Edit: anyway, what's a little strange is that in the slide is said that R700 is a 2 Teraflop borad. Now, if it's 2X4870, we should have 2,4 Teraflop per board (and marketing rarely misses these numbers). So it's possible that R700 is 2x4850 + GDDR5 instead...

well with proper price - say $399 who cares what counts is that is faster and cheaper then GTX280, and with two 4850+GDDR5 it would be more energy efficient than GTX280!

 

[trinibwoy]

The AA performance seems amazing.

Yep, maybe AA performance on RV670 was shader limited. But whatever it was, it's damn well fixed now. What makes it even more impressive is the relatively low bandwidth available to the chip.

 

[Pete]

Does Cat 8.5 drive RV770 properly?

Didn't even cross my mind, though I guess they seem to have made enough changes that even that functionality might be affected. Still, I thought Damien would at least know of it and so mention it.

No, even 8.6 deosnt. No PowerPlay eg and so on...

Thanks.

 

[Love_In_Rio]

Excellent in game tests:

http://www.xtremesystems.org/forums/showthread.php?t=191574

Great great performance. Can play crysis ( HD 4850 CF ) in very high at 1600x1200 without problem ( only minor stuttering in strange parts of last level ).

 

[Jawed]

Yep, maybe AA performance on RV670 was shader limited.

I'm betting it's just Z rate...

But whatever it was, it's damn well fixed now. What makes it even more impressive is the relatively low bandwidth available to the chip.

Yeah, 64GB/s versus 70.4GB/s in 9800GTX or 103.7GB/s in 8800U or 111.9GB/s in GTX260.

It appears Damien used adaptive/transparency AA throughout the test, which I think means this review is going to look quite different to others. I also expect the drivers for GT200 are prolly "rusty" in transparency AA modes simply because it's not tested very much (X-bit always uses it).

Jawed

 

[trinibwoy]

So are we gearing up for R600 in reverse? Now it seems Nvidia's big chip is going to be upstaged. And by a much smaller one to boot. Even if the 280 is somewhat safe in the single-GPU stakes, the 260 seems to be in mortal danger at $450 vs the 4870.

 

[mczak]

Looks like GTX260 is going to have to hit $200 once HD4870 arrives. People are calling it good value for money at $400

Yes, if the HD4870 really scales with more than core clock (e.g. assuming the HD4850 is quite a bit bandwidth limited), seems it could keep up with it which would be amazing...
Even if not, it would only be marginally slower overall... Heck even the HD4850 can keep up with the GTX260 in some titles - that's got to hurt...
Though the GTX260 probably could have a big advantage in some ultra-high-res + AA situations (thanks to almost twice the ram, even if the HD4850 doesn't really seem to suffer from "only" 512MB ram in all of those benches from hardware.fr), that advantage would of course be gone very fast with a 1GB version (that assumes though gddr5 1gbit chips would be available - are they or only 512mbit chips?)

 

[Jawed]

that advantage would of course be gone very fast with a 1GB version (that assumes though gddr5 1gbit chips would be available - are they or only 512mbit chips?)

Rumours indicate HD4870 will come in 512MB and 1GB variants. I presume the clamshell configuration will be deployed for 1GB, using double the count of chips as on the 512MB board.

Jawed

 

[A.L.M.]

I'm betting it's just Z rate...

It is, actually. According to W0mbat, RV770 will have twice the Z fill rate of R600 (but it is easy to figure it out just looking at the gpu diagram).

 

[mczak]

Rumours indicate HD4870 will come in 512MB and 1GB variants. I presume the clamshell configuration will be deployed for 1GB, using double the count of chips as on the 512MB board.

Ah ok. So 1gbit gddr5 chips will, in fact, not be available. I think generally graphic card manufacturers try to avoid the twice-as-many-chips configuration - different board layout (or if only one it must be designed for twice as many chips hence a bit more complicated), signal noise problems etc. (so can't achieve quite the same memory clock). Much easier if you can just swap in the same number of larger parts. Not that this is really a disavdantage compared to GTX260, which needs 14 chips (with much more complicated pcb routing) anyway .

 

[mczak]

It is, actually. According to W0mbat, RV770 will have twice the Z fill rate of R600 (but it is easy to figure it out just looking at the gpu diagram).

So is this confirmed to be real now?
Oh and since everybody seems to have forgotten about them: what about the other RV7xx chips? When will they appear? What configuration?

 

[trinibwoy]

So is this confirmed to be real now?
Oh and since everybody seems to have forgotten about them: what about the other RV7xx chips? When will they appear? What configuration?

I doubt they'll show up anytime soon. RV670 can easily accomodate the sub $150 market for now.

 

[Mat3]

Hopefully they'll have something good for a 400W PSU because the 3650 is crap.