The G92 Architecture Rumours & Speculation Thread

[INKster]

Did Rivatuner 2.06 just made a pre-announcement about other G92 products besides the 8800 GT ?

are available on NVIDIA 8800GT and other upcoming G92 based NVIDIA products

Source

 

[flopper]

Seems ati/amd are second position again now at middle end.

I have a x1950xtx, its run great for a year.
I am gonna upgrade soon, since ati/amd havent released any info on the upcoming nov release it seems that nvidia is the way to go.

8800GT seems like a sweet middle end card to get.
For me it simply are such a great deal that I wonder, what can amd/ati came up with if any?

 

[2900guy]

i hope that means soon and not next year.

 

[Kaotik]

It's funny how people already seem to know how RV670's perform since they've judged them to be slower no matter what
Yes, I know, it's probable, but do we actually know there hasn't been any architectural enhancements, which among with the higher clocks of XT model might actually make it faster too?

 

[INKster]

It's funny how people already seem to know how RV670's perform since they've judged them to be slower no matter what
Yes, I know, it's probable, but do we actually know there hasn't been any architectural enhancements, which among with the higher clocks of XT model might actually make it faster too?

But you're also assuming that there hasn't been any architectural enhancement to G92 from the G8x...
It works both ways, you know ?

 

[2900guy]

so even though 8800gt is not as efficient in using its fillrate, based on the 3dmark06 test, it still has considerable more texel fill rate than the GTX. so are its lower pixel fillrate and 16 less SP the main reason its still not beating the GTX?

 

[Kaotik]

But you're also assuming that there hasn't been any architectural enhancement to G92 from the G8x...
It works both ways, you know ?

I didn't mean that, nor said anything about changes in G92, just that at least some people here have apparently decided already that there's no way RV670(XT) could beat G92(GT) no matter what, even though we don't really know anything about the RV670's for sure but the amount of stream processors and memorybus width and the fact that it has gone through necessary changes to support DX10.1 / SM4.1 requirements

 

[HAL]

http://www.tweaktown.com/articles/1211/crysis_demo_performance_analysis/index.html

 

[AlexV]

I didn't mean that, nor said anything about changes in G92, just that at least some people here have apparently decided already that there's no way RV670(XT) could beat G92(GT) no matter what, even though we don't really know anything about the RV670's for sure but the amount of stream processors and memorybus width and the fact that it has gone through necessary changes to support DX10.1 / SM4.1 requirements

Some things point to the fact that the RV670 won't be faster than the R600(Crytek's config file for Crysis seems to suggest that, and they may have an inkling on what's what). The 8800GT seems to fare a tad bit better than the R600, even in its 2900XT incarnation(see the above mentioned Tweaktown review, and bear in mind that Tweaktown tend to like ATi a lot). Putting those two data points together, it's not that far-fetched to think that the RV670 won't smack the G92.

 

[ShaidarHaran]

Some things point to the fact that the RV670 won't be faster than the R600(Crytek's config file for Crysis seems to suggest that, and they may have an inkling on what's what). The 8800GT seems to fare a tad bit better than the R600, even in its 2900XT incarnation(see the above mentioned Tweaktown review, and bear in mind that Tweaktown tend to like ATi a lot). Putting those two data points together, it's not that far-fetched to think that the RV670 won't smack the G92.

Everything I've seen points to RV670 XT being faster than R600. The only cases where this may not be true would be @ extreme resolutions & AA levels, due to the bandwidth difference. Given the identical functional unit count and higher clock though, it doesn't make any sense to think that RV670 XT is slower than R600, except in the special cases I just mentioned.

 

[Kaotik]

Everything I've seen points to RV670 XT being faster than R600. The only cases where this may not be true would be @ extreme resolutions & AA levels, due to the bandwidth difference. Given the identical functional unit count and higher clock though, it doesn't make any sense to think that RV670 XT is slower than R600, except in the special cases I just mentioned.

And if they've improved the "achilles heel" of R6xx family, it might even do better with AA than R600

 

[ShaidarHaran]

And if they've improved the "achilles heel" of R6xx family, it might even do better with AA than R600

I'm trying not to get my hopes up too much

 

[vertex_shader]

Some things point to the fact that the RV670 won't be faster than the R600(Crytek's config file for Crysis seems to suggest that, and they may have an inkling on what's what).

0x1002, 0x9400, 3 // ATI Radeon HD 2900 XT
0x1002, 0x9401, 3 // ATI Radeon HD 2900 XT
0x1002, 0x9402, 3 // ATI Radeon HD 2900 XT
0x1002, 0x9403, 3 // ATI Radeon HD 2900 PRO
0x1002, 0x9405, 3 // ATI Radeon HD 2900 GT
0x1002, 0x9500, 2 // RV670
0x1002, 0x9501, 2 // RV670
0x1002, 0x9505, 2 // RV670
0x1002, 0x9507, 2 // RV670
0x1002, 0x9580, 3 // RV630

Sience when rv630 is in the same performance level as r600?

 

[ShaidarHaran]

Sience when rv630 is in the same performance level as r600?

Seems like a typo, or perhaps those values don't mean what we think they mean.

 

[satein]

Sience when rv630 is in the same performance level as r600?

Are you trying to say that RV630 is RV670 from that INF definition?

Well, when will we hear that RV670 will have the same shader units as the RV630

I feel very confuse nowwww.

 

[ShaidarHaran]

Are you trying to say that RV630 is RV670 from that INF definition?

Well, when will we hear that RV670 will have the same shader units as the RV630

I feel very confuse nowwww.

He's referring to the fact that both R600 and RV630 are given the same value for whatever characteristic that defines.

 

[satein]

He's referring to the fact that both R600 and RV630 are given the same value for whatever characteristic that defines.

It came from INF file.

3 of the R600 and RV630 mean that they might require the same set of device driver files loaded (it was grouped in the INF file), but the RV670 is set as 2. It may require something different to what loaded for R600 and RV630

 

[vertex_shader]

He's referring to the fact that both R600 and RV630 are given the same value for whatever characteristic that defines.

That numbers are the performance scores, GT/GTX/ULTRA score is 4, GTS 3, 8600GT/GTS 2, its Crytek BS, just skip it

 

[ShaidarHaran]

That numbers are the performance scores, GT/GTX/ULTRA score is 4, GTS 3, 8600GT/GTS 2, its Crytek BS, just skip it

Yeah, makes no sense to lump HD 2900 XT in the same category as HD 2600 XT, and then have HD 3850/3870 performing lower than either

 

[Mintmaster]

Things are even worse for the 8600GTS, for which your equation yields only 5.4 GT/s to the 7.6 GT/s it achieves in 3DM06. Could tex. coord. reuse be so high?

If they redesigned the test, you could fetch all textures with the same two interpolated values by using shader math to adjust the fetch coordinates. AFAIK multitexturing tests don't work that way, as it makes the texture lookups dependent on shader math, which slows down speed on older architectures like the GeForce FX.

Rightmark's fillrate #s are closer to your equation, as they saw 5.8GT/s max for a 675/1350 8600GTS (though they're doubting their test considering the discrepancy with NV's GPUs). RightMark's fillrate test yields lower rates than 3DM06's, and ATI's GPUs score closer to their theoretical tex. units * clock #s.

Ahh, yes, I remember seeing that. RightMark's results do indeed make more sense (note that for a 1350 MHz shader clock, 5.4 GTex is the theoretical max). Note that G80 performs at theoretical speeds too, so it's not all NVidia processors.

Not sure what 3DMark is doing.