NVIDIA GT200 Rumours & Speculation Thread

[Arun]

Heh, let's not go overboard either!

 

[trinibwoy]

Another CJ-provided Nvidia's GTX 200 series slide turns up at vr-zone.
This time it's a performance chart:

http://www.vr-zone.com/articles/GeForce_GTX_280_%26_GTX_260_Gaming_Performance/5817.html

Well that doesn't bode well for Crysis performance if it's only 80% faster than the X2 since the X2 is far from playable.

There's a pretty recent review with the latest drivers at computerbase (gotta love their graphs) with some of these games on there.

http://www.computerbase.de/artikel/...eon_hd_3870_x2_atomic/3/#abschnitt_benchmarks

 

[Lukfi]

About the ALU:tex ratio, I think Arun is referring to the lower shader clocks on the GT200. That kinda offsets their high number and drives the ALU:tex ratio back to G80 numbers.

 

[trinibwoy]

About the ALU:tex ratio, I think Arun is referring to the lower shader clocks on the GT200. That kinda offsets their high number and drives the ALU:tex ratio back to G80 numbers.

Relative to G80 the clocks haven't changed much at all.....

 

[Lukfi]

G80 had 1500 MHz shaders IIRC, GT200 has just 1300, not a big deal but there's 240 of them, with those numbers 200 MHz could make a difference. Besides, GT200 has faster core clock, 601 vs. 575 MHz.

 

[CarstenS]

The original G80 (GTX) had 1350 MHz on the shaders while GTS was only 1188.

 

[trinibwoy]

G80 had 1500 MHz shaders IIRC, GT200 has just 1300, not a big deal but there's 240 of them, with those numbers 200 MHz could make a difference. Besides, GT200 has faster core clock, 601 vs. 575 MHz.

G80 was 575/1350. GT200 is 600/1300. Not enough of a difference to have a material impact on ALU:TEX ratio. Besides, if it was something so trivial Arun wouldn't have mentioned it as it's easy enough with the info we have today to make that comparison.

 

[Lukfi]

=>Quasar: My mistake. In that case, the ALU:tex ratio has risen from 9,4:1 to 13:1. I calculated ( SP count × SP clock ) ÷ ( TMU count × core clock ) to get the numbers.

 

[leoneazzurro]

The original G80 (GTX) had 1350 MHz on the shaders while GTS was only 1188.

1200

 

[CarstenS]

1200

Depends on if you believe Tools like Rivatuner or a PDF.

 

[leoneazzurro]

Depends on if you believe Tools like Rivatuner or a PDF.

Well, Rivatuner on my GTS 512 ( A POV ) says 648/1648 but on another would-be-identical card (XFX) says 652/1653. So I think I will stick to the official specs

 

[XMAN26]

1200

its 1188, 1200 was a marketing thing. The 8800GTS 320/640 bios were set to 1188 not 1200.

 

[XMAN26]

Well, Rivatuner on my GTS 512 ( A POV ) says 648/1648 but on another would-be-identical card (XFX) says 652/1653. So I think I will stick to the official specs

NvBiToR is your friend. It will allow you to see the cards bios while your machine is on. Use that to see exactly what it is.

 

[leoneazzurro]

OK, I am only wondering why no review I read pointed this out...

 

[XMAN26]

OK, I am only wondering why no review I read pointed this out...

A few I have seen where they show RT shots I believe do point it out, atleast I'm sure I've read atleast 2 that have.

I used NvBiToR to change the BIOS on my 8800GTS 640 from BFG clocks of 550/1350/750(BIOS set) 580/1500/850. I should note that even thought the core was set to 580, RT reported 576 core clock.

 

[Arty]

Didn't they do that already for G92 (with G94) ?
Why would this time be any different ?

My point was in comparing, not price segmentation, since we know about the 55nm shrink of G92 (which, BTW, would be conflicting with G94, no ?), but how much die size and transistor savings it could achieve by cutting down the chip (therefore hinting at the GT200 midrange derivatives' performance, something i doubt we'll see this year).

No, they didnt.

9600GT and 8800GT are in different leagues. Even if you dont share that opinion, the motivation for Nvidia there was die size (27%) and transistor savings (33%). There was no such savings in the gpu you were suggesting.

 

[trinibwoy]

Yeah I also don't see the point of a half-GT200. Nvidia probably isn't going to adopt GDDR5 until their next architectural overhaul anyway (if ever) so G92b can fill that role quite nicely till then.

Their most pressing concern would be to shrink GT200.......

 

[ZerazaX]

I wonder why they didn't show the 9800GX2 in those slides? 9800GX2 is what, about anywhere from 20-40 or 50% faster than 3870X2 in certain games? I guess the saying that GTX280 was 50% faster than previous generation wasn't in shader operations itself, but in actual performance?

At least those Vantage scores seem to line up with the slides

 

[chavvdarrr]

A few I have seen where they show RT shots I believe do point it out, atleast I'm sure I've read atleast 2 that have.

I used NvBiToR to change the BIOS on my 8800GTS 640 from BFG clocks of 550/1350/750(BIOS set) 580/1500/850. I should note that even thought the core was set to 580, RT reported 576 core clock.

Well, setting BIOS is like "set freq to 20 times the base frequency... so if this freq-cy is not 29 but 28.8 ...which is the right value - the one you want or the one that you get ?

 

[Ailuros]

I don't think the ALU-TEX ratio calculation, relative to G80, is as simple for GT200 as it might seem...

Either way I count the ratio is higher on GT200 than on G80.