Most Significant Graphics Technology for 2005

[ANova]

Hell, it should be clear that the NV4x is already faster on a clock-for-clock basis than ATI's offerings in the majority of benchmarks, and that's before making use of PS/VS 3.0.

Huh? What planet are you living on?

 

[ninelven]

erased

 

[Blade]

I think what Chalnoth meant to say was that the GF6800U is clocked at 400MHz. Rx800XT-PE is 520MHz.

Performance per MHz on the 6800U is much better than the XT-PE, across the board. (including benchmarks that the XT wins)

 

[MuFu]

Hell, it should be clear that the NV4x is already faster on a clock-for-clock basis than ATI's offerings in the majority of benchmarks, and that's before making use of PS/VS 3.0.

Huh? What planet are you living on?

Per-clock.

Edit - Blade beat me to it.

 

[ANova]

Per clock yes, but at the same time the R420 is able to achieve those clockrates, while the NV40 is not. Until the NV40 can, then I consider it's ability to achieve more performance per clock cycle a mi-nute matter. The R420 is also capable of much higher overclocks then the NV40.

 

[Acert93]

Per clock yes, but at the same time the R420 is able to achieve those clockrates, while the NV40 is not. Until the NV40 can, then I consider it's ability to achieve more performance per clock cycle a mi-nute matter. The R420 is also capable of much higher overclocks then the NV40.

Per clock is what his post stated though... which is true. The NV40 is faster than the R420 performance per clock. I would agree that performance/clock comparisons may not be a great way to evaluate things unless we know NV may be able to scale, but I am not sure that was his point. His point was per clock performance. Oh well.

As the R420 having better overlocks than the NV40, I have not heard this (actually heard the reverse). My 6800GT OCs more than 20% w/on issue (350HMz stock; OCs to 421MHz w/o issue). Do R420 chips OC 70MHz and/or 20%?

 

[ANova]

Per clock yes, but at the same time the R420 is able to achieve those clockrates, while the NV40 is not. Until the NV40 can, then I consider it's ability to achieve more performance per clock cycle a mi-nute matter. The R420 is also capable of much higher overclocks then the NV40.

Per clock is what his post stated though... which is true. The NV40 is faster than the R420 performance per clock. I would agree that performance/clock comparisons may not be a great way to evaluate things unless we know NV may be able to scale, but I am not sure that was his point. His point was per clock performance. Oh well.

As the R420 having better overlocks than the NV40, I have not heard this (actually heard the reverse). My 6800GT OCs more than 20% w/on issue (350HMz stock; OCs to 421MHz w/o issue). Do R420 chips OC 70MHz and/or 20%?

An X800 Pro can overclock to XT PE speeds fairly easily in most cases. If you throw in a phase change system I've seen the X800 XT reach speeds up to 730 MHz with a score of over 7500 in 3dmark05. The highest I've seen the 6800U reach was around 600 MHz with a score of ~5800. Before nvidia introduced SLi the top 20 over at Futuremark were all X800 XTs with the exception of one or two 6800Us near the bottom.

 

[hovz]

3dmark05 looks nice a shell, but it has zero relvence in regards to using it to predict how future games will perform. no game in the near future will make any real use of vertex shaders.

 

[ANova]

I figured someone would say that. At any rate, my point was that Futuremark provided a place to take a look at what kind of clocks people were achieving on cards from both IHVs and ATI was and still is clearly ahead in that aspect.

 

[kemosabe]

Certainly the per-clock performance advantage has been a godsend to NVDA in this generation, but the relevant question going into 2005 is whether they'll be able to achieve significantly higher clocks than NV40/45 if the presumably more shader-efficient R520 architecture is indeed clocked at least as high as R480 (if not higher)?

Assuming an equivalent number of pipes for the sake of argument, will NV47/48/whatever be capable of reaching 550MHz+ with decent yields on whatever foundry process they have it slated for?

 

[hovz]

im pretty sure nvidia could add more pipes on 09nm without much trouble. a 24 pipe nv40 shouldnt have any trouble matching or beating the r520 if current rumors are correct.

 

[kemosabe]

im pretty sure nvidia could add more pipes on 09nm without much trouble. a 24 pipe nv40 shouldnt have any trouble matching or beating the r520 if current rumors are correct.

I didn't specifically say, "Assuming an equivalent number of pipes for the sake of argument" for nothing.

 

[hovz]

thats about as useful as the old "the p4 takes so many more mhz just to match the athlon at xx mhz"

 

[chavvdarrr]

Certainly the per-clock performance advantage has been a godsend to NVDA in this generation, but the relevant question going into 2005 is whether they'll be able to achieve significantly higher clocks than NV40/45 if the presumably more shader-efficient R520 architecture is indeed clocked at least as high as R480 (if not higher)?

Assuming an equivalent number of pipes for the sake of argument, will NV47/48/whatever be capable of reaching 550MHz+ with decent yields on whatever foundry process they have it slated for?

What makes you think that assuming equivalent number of pipes, R520 can reach 550MHz + ?
NV4x - fp32, SM3, 400MHz
R420 - fp24, SM2, 500MHz
R520 - fp32, SM3, MHz ?
Going to fp32 and SM3 increases number of transistors with say 20%, that increases heat and limits clock. Sure going to 0.09 will improve the clock (hopefully), but same arguement works for NV. But ATi have to do several things in same time:
shrink to 0.09 process. Intel and AMD had problems with it
add fp32 support - that should be easyto implement, still can need tweaks so wider parts work at ~ same speed as fp24
add SM3 support
In contrast Nv "only" has to go shrinking. Anyone who is aware how mature is TSMS's 0.09 process? That's a big IMO

 

[Bjorn]

Certainly the per-clock performance advantage has been a godsend to NVDA in this generation, but the relevant question going into 2005 is whether they'll be able to achieve significantly higher clocks than NV40/45 if the presumably more shader-efficient R520 architecture is indeed clocked at least as high as R480 (if not higher)?

Assuming an equivalent number of pipes for the sake of argument, will NV47/48/whatever be capable of reaching 550MHz+ with decent yields on whatever foundry process they have it slated for?

The NV40 vs R420 are on different processes. Compare the 6600 vs X700 and Nvidia is able to reach higher clocks. So i think that it's a bit to early to say that Ati probably will be able to reach higher clockspeeds then Nvidia.

 

[_xxx_]

no game in the near future will make any real use of vertex shaders.

???

 

[Simon F]

no game in the near future will make any real use of vertex shaders.

???

You see they're going to be quantum theory based. All games in the near future will send all objects through the pipeline rotated and translated to all possible positions. The pixel shader then will make all rendered pixels belonging to an incorrect position fully transparent. This thus eliminates the need for vertex shaders.
QED.

 

[hovz]

no game in the near future will make any real use of vertex shaders.

???

which games in the near future will be vertex heavy to a degree even close to that of 3dmark05?

 

[_xxx_]

What do you mean by "real" use? Being used at all is real use, ain't it? Doesn't need to be heavy.

 

[DeanoC]

no game in the near future will make any real use of vertex shaders.

???

which games in the near future will be vertex heavy to a degree even close to that of 3dmark05?

All of them?

Even if you have a game thats shifts all processing to the pixel shader (which would be extremely inefficient as many things can be linearly approximated with no major loss in quality i.e. diffuse irradience and atmospheric approximation) something still has to setup the pixel shaders. Blending N bones, morph targets, rotating into tangent space, etc all have to be in the vertex shaders.

Character vertex shaders are easily 50+ instruction (100 is more realistic), particle vertex shaders can be even higher.