A note on percentages for you guys here.

[Sherlock]

None of the online databases go back far enough to make a valid assumption for a brand new architecture. Ok the the int part may be based on the GF4 core, bu the FP part is brand new. To me it seems like either a) the hardware isn't up to it, b) The hardware is bust or c) The drivers are bust and not just unoptimised.

Some might say the need to clock the new architecture so high and the blowdryer fan might give some hint as to this chip's futre potential.
Unless some redeeming features come to light, things look bleak for this architecture.

 

[KimB]

At the same time the idea that it can usually beat the Radeon 9700 Pro (even at 4x AA...up to 1280x1024) is promising for the FX's architecture, though it does still need major improvements in the AA area (but don't hold your breath...improvements here are just not likely until the NV40...).

 

[Joe DeFuria]

At the same time the idea that it can usually beat the Radeon 9700 Pro (even at 4x AA...up to 1280x1024) is promising for the FX's architecture,

No relly given the fill-rate advantage (higher clocks) afforded by the 0.13 process.

The "apples to apples" architecture comparison results of the Radeon 9500 Pro vs. the downclocked NV30, does not bode particularly well for the FX architecture.

 

[KimB]

That downclocked picture only looked at 1600x1200 w/ 4x AA. The massive drop at that resolution seems to indicate a memory size limitation on the FX's part. Since a lower-cost version (NV31/NV34) probably won't ever run at that setting anyway, this isn't a good indication of low-end performance in the least.

As for higher-end performance, there is a very good chance for future drivers to significantly improve this particular score. That and 256MB video cards aren't far off.

 

[Joe DeFuria]

That downclocked picture only looked at 1600x1200 w/ 4x AA. The massive drop at that resolution seems to indicate a memory size limitation on the FX's part. Since a lower-cost version (NV31/NV34) probably won't ever run at that setting anyway, this isn't a good indication of low-end performance in the least.

Interesting spin, though nonsensical.

So what you're saying is that these lower cost chips won't be running in GPU bound situations? Or that some "mysterious FX memory limitation" is only relevant at some obscure setting, that doesn't matter?

As for higher-end performance, there is a very good chance for future drivers to significantly improve this particular score.

So you keep telling everyone.

That doesn't change the fact that right now, the NV30 architecture, apples to apples with the R300, doesn't look so good.

 

[RussSchultz]

And the athlon is better than the P4 clock per clock, but that isn't the fastest processor around, now is it?

A year or so ago, when the P4 was introduced, it was given the relative thumbs down. Now? It seems like its shown its merit by heading upwards of 3ghz.

Not that that's any indication at all as to what the GFFX architecture can do, but still: clock to clock comparisons are interesting, but don't mean much in the end.

 

[KimB]

Joe, wait until some performance comparisons at lower resolutions/FSAA levels are displayed before proclaiming that the R300 architecture is faster clock for clock.

 

[overclocked]

A little of the topic but does the newest ATI driver´s have fixed the problems seen with the fps dropping too 1-2fps or something like that in games and for you who has the 9700pro does it accure often or??

 

[Althornin]

Joe, wait until some performance comparisons at lower resolutions/FSAA levels are displayed before proclaiming that the R300 architecture is faster clock for clock.

Unless a 40% downclock on results in a 10-15% performance decrease, is there even a question?
And if the GFFX scales so poorly, what is wrong with it?

 

[KimB]

A little of the topic but does the newest ATI driver´s have fixed the problems seen with the fps dropping too 1-2fps or something like that in games and for you who has the 9700pro does it accure often or??

The stuttering problems are much less common now, but some still have the issues.

 

[KimB]

Joe, wait until some performance comparisons at lower resolutions/FSAA levels are displayed before proclaiming that the R300 architecture is faster clock for clock.

Unless a 40% downclock on results in a 10-15% performance decrease, is there even a question?
And if the GFFX scales so poorly, what is wrong with it?

As I said, I believe it's a memory size limitation. If this is true, then this can easily explain why underclocking doesn't change performance much: most of the performance hit would be from AGP texturing. The GeForce FX downsampling at scanout can easily explain this significant drop in performance at 1600x1200x4 compared to the other resolutions.

If this is the reason for the lower performance, then it seems very possible for nVidia to later move the downsampling to during the buffer swap when running at this resolution, which should significantly decrease the performance hit (on a 128MB board).

In the meantime, most people will not consider either the FX or Radeon 9700 Pro playable at 1600x1200x4, except with older games. These older games are much more likely not to show problems with memory size (since they don't use that many textures).

 

[overclocked]

overclocked wrote:
A little of the topic but does the newest ATI driver´s have fixed the problems seen with the fps dropping too 1-2fps or something like that in games and for you who has the 9700pro does it accure often or??

Chalnoth

The stuttering problems are much less common now, but some still have the issues.

When is it most seen/if you can give any examples and has ATI described
this problem/ or the source too the problem in their driver-updates realesenotes?
Is this a purely a driver issue or can it be som glith in the hardware that been fixed in the refresh soon cooming? :?

 

[Doomtrooper]

The stuttering appeared to worsen with DX9, I also noticed the Nvidia forums were complaining about stuttering after DX9 release, I played with the Cat 2.5 on my 9500 NP and DX 8.1 and stuttering was minor to none, then tried DX9 and Cat 3.0a and it worsened.
Its not a hardware issue, its a driver issue IMO..Vsync related not sure but can be frustrating as it doesn't happen on all titles.

I got a whole new setup coming with Nforce2/9700 combo so I will give the setup a run and see how it goes, new drivers are supposed to be coming soon to fix the stutter bug.

 

[Althornin]

As I said, I believe it's a memory size limitation. If this is true, then this can easily explain why underclocking doesn't change performance much: most of the performance hit would be from AGP texturing. The GeForce FX downsampling at scanout can easily explain this significant drop in performance at 1600x1200x4 compared to the other resolutions.

No, you miss my point:
You use the low memory argument to say "GFFX may not be (per clock) slower than the R300" because we only have downclocked benchies available at 1600x1200.

I said:

Unless a 40% downclock on results in a 10-15% performance decrease, is there even a question?
And if the GFFX scales so poorly, what is wrong with it?

Meaning: To compare clock by clock, GFFX must be downclocked about 40%. (35% to be exact). Considering it does NOT lead most benchies (of a full speed GFFX at lower res, while still being GPU limited with AA/aniso) by more than 10-15% (if indeed it leads at all), what makes you think it can match the R300 clock for clock?

 

[K.I.L.E.R]

what makes you think it can match the R300 clock for clock?

It doesn't look like it does.

 

[KimB]

When is it most seen/if you can give any examples and has ATI described
this problem/ or the source too the problem in their driver-updates realesenotes?
Is this a purely a driver issue or can it be som glith in the hardware that been fixed in the refresh soon cooming? :?

It appears to be purely a driver issue, and it also appears to be machine-specific. Your best bet would be to head over to Rage3D and look under the tech support section. You'll notice a number of posts about stuttering. Most people are able to fix it in one way or another (if they encounter the issue), but a few manage to only move the problem around. I myself noticed it with just one driver set, not in any others.

 

[KimB]

Meaning: To compare clock by clock, GFFX must be downclocked about 40%. (35% to be exact). Considering it does NOT lead most benchies (of a full speed GFFX at lower res, while still being GPU limited with AA/aniso) by more than 10-15% (if indeed it leads at all), what makes you think it can match the R300 clock for clock?

If you're going to be comparing it on a clock for clock basis, it only makes sense to compare it to an R300 that also has the same memory bandwidth. Anyway, I did misunderstand you. I thought you were referring to the FX only performing 10%-15% slower on the 16x12x4 bench, and I wasn't able to load up the review to check. Makes a whole lot more sense that this isn't what you meant.

Anyway, if you are concerned about clock-for-clock performance, the best way to compare would be against a Radeon 9500 Pro (with 8 pipes and a 128-bit interface) clocked at the same speed. With this comparison, at a resolution lower than 16x12x4, I think the FX will win (most of the time). Would be nice if somebody could post some benches...

 

[jb]

And the athlon is better than the P4 clock per clock, but that isn't the fastest processor around, now is it?

Russ it was stated that the FX architecture was promising since it could beat the R300 in some cases. What Joe brough it was the fact that it may not really be the architecture to thank for this FX advatange but more of her fill rate thanks to a higher clock. Thus Joe's comment was valid to see what part of that advatage is coming from architecture and what part from brute speed. Some of us here find that interesting....

Chalnoth,
I have heard your theory on why the GX does so bad at 16x12 x4. I dont by them. I think its just running out of MB at that point. Sorry but from what we have seen that looks to be the case.

So for the smarter folks out there...what is the amount of MB you need for a MSAA method at 16x12 @ 32 bit colors?

 

[Sherlock]

No matter if we speculate about downclocking, memory limitations, and driver updates one fact remains: this showing of Nvidia's super chip seems hardly a speedbump for the r300/r350. Who will buy stock in Nvidia based on this new architecture showing significant limitations? Would anyone here invest in Nvidia after seeing it's new card?

To be fair, if they only fixed the heat/noise problem and sold this card at a reasonable price then they would have had a competitor, though 6 months late( that's alot of lost sales).

 

[Sxotty]

]
I am not saying Nvidia is the *Devil* or anything. Simply that there is a rather large group of people with rosey colored glasses on.

Actually they are probbably just people who in the past were victims of ATI's sh itty drivers. I know I have been. It is good that ATI has gotten their act together, but in the past they were worse, and that is all there is too it, even if you are a fanATIc.