NV35 Reviews

[demalion]

What a rosy picture for misinformation this presents as people hear from other sources how the NV35 is "2x as fast as the NV30 for shaders" and think that validates this conclusion.

You were talking about misinformation here, and then later in your post you brought it up again. I was pointing out that {H} had benches that showed there was a doubling of shader performance in relation to the 5800U(although you have to go back to the prior review to cross reference it). The NV35 is showing ~2x shader performance compared to the NV30. Considering the trouble {H} goes through to level the playing field(Splinter Cell....) I doubt they are rigging the test up

Ah, that makes sense, but including this quote would have made things a lot clearer the first time.

Well, you're free to consider that equivalent for yourself, if you wish, but I'll simply point out to you that fp24 precision processing is hardly a secret we had to go digging for, and I think it is a pretty thin justification for the nature of Anand's commentary.

I didn't say it was justified in the least, I was simply posting the misinformation that Anand was talking about.

Well, with your quote selection, I thought you were implying that HardOCP actually validated his commentary. I don't even think he meant fp24 as the implied misinformation, as I deem the explanation "creditable" except for the rather markedly dedicated slant of usage to equate ATI and nVidia as a pretext for making results outside the context of the hardware capabilities..i.e., results are determined by driver "mojo", rather than the hardware's actual floating point processor resources, and, of course, nvidia has the most mojo ("yeah, baby").

I made no commentary whatsoever as I am in no way supporting anything about the claims, you made mention that you didn't see an example of ATi's misinformation stated by Anand, I clearly remembered the quote so I posted it as is.

Heh, I dispute the term misinformation being equally applicable to the items mentioned. If you want ATI misinformation, fp24 isn't it...I'd vote for ATI listing displacement mapping as a feature that the R300 has and the NV30 doesn't (it is this last that I view as a falsity), which should only be true (AFAIK) if the nv30 has driver issues related to successful implementation (I don't know of such a thing). It is more like the "UltraShadow" type of misinformation where wording like "double" is used in relation to "other architectures" in regards to "stencil pixel" output (quite true when comparing to the RV350 ).

 

[Bjorn]

Yeah, nVidia driver code:

If PathRequset = NV30 then
UsePath = NV30
ElseIf PathRequest = ARB2 then
PathUse = NV30
End If

Maybe. But i would guess that we would hear from Carmack if this turns out to be true.

 

[demalion]

Well id don't know how Nvidia does the trick but there's no more IQ issues:

Be careful...the teapot doesn't seem to be associated with any performance results, and we already know the cards are capable of high quality. It only indicates that for standard usage, the actual quality for the NV35 is repaired, not that it is fixed to not using integer while performing quickly as well. However, I don't understand why the other NV3x shots are as bad as they are for a simple teapot render, unless it is from dropping precision arbitrarily from what was requested and introducing significant errors that way.

The 3dmark 03 shot...it appears that is something that remains to be discussed and analyzed in more detail.

 

[Evildeus]

So what? Before it wasn't correctly rendered when we found issues in 3DMark, it's no more the case! Are you saying that Nvidia tweakes its driver for ALL PS2/VS2 applications? Then how can you say it's tweaking or normal? :? Perhaps it's the NV30-4 series that have too much broken features...

Well id don't know how Nvidia does the trick but there's no more IQ issues:

Be careful...the teapot doesn't seem to be associated with any performance results, and we already know the cards are capable of high quality. It only indicates that for standard usage, the actual quality for the NV35 is repaired, not that it is fixed to not using integer while performing quickly as well. However, I don't understand why the other NV3x shots are as bad as they are for a simple teapot render, unless it is from dropping precision arbitrarily from what was requested and introducing significant errors that way.

The 3dmark 03 shot...it appears that is something that remains to be discussed and analyzed in more detail.

 

[demalion]

So what? Before it wasn't correctly rendered when we found issues in 3DMark, it's no more the case! Are you saying that Nvidia tweakes its driver for ALL PS2/VS2 applications?

I'm saying the general PS2 tweaks are indicated to be broken for the nv30 in those other shots, as it is known that the hardware is capable of not exhibiting such problems within its capabilities (the problem is how quickly it does it). I'm also saying that the general PS 2 tweaks seeming to be fixed for a screenshot someone else is evaluating as identical to the REFRAST for you, without performance figures available, does not give any useful information with regards to the applicability of the NV35's behavior for 3dmark 03 and floating point shader performance.

Then how can you say it's tweaking or normal? :? Perhaps it's the NV30-4 series that have too much broken

It is pretty clear the drivers up to 43.51 change shader behavior uniquely for 3dmark 03, with visible discrepancies with DX 9 REFRAST and Radeon output apparent to my own observation of shots (not cropped) I've seen posted. If you wish to consider this established NV30 behavior a "So what?", disregard comments made already here in the forums, and focus instead only on the cropped comparison from parts of a 43.45 screenshot and what someone else is saying is an image that is identical between the 9800 output and 43.51 screenshots for the NV35 (without provided screenshots for your own comparison), I'm afraid I'll simply have to disagree with your evaluation, and we'll have to discuss in further detail at a future date.

 

[T2k]

C'mon HL2 benchmarks...

MuFu.

Yep. Second that... this is what I'm waiting for.... 8)

 

[Joe DeFuria]

C'mon HL2 benchmarks...

MuFu.

Yep. Second that... this is what I'm waiting for.... 8)

Well, if this is any indication:

http://www.ati.com/companyinfo/press/2003/4644.html

Half-Life 2 is pushing the envelope with contextual AI, simulated physics, characters that rival any digital actors in Hollywood, and advanced rendering capabilities,” said Newell. “When we reviewed all the platforms available for use in the public unveiling of HL2, ATI’s RADEON 9800 PRO was unquestionably the choice for showing Half-Life 2 at its best.

If you take away almost guaranteed PR "tendency to overexaggerate a litte bit", I'd guess that the 9800 Pro is probably a little faster. (Either that, or the NV3x drivers had some rendering issues that weren't fixed in time for the decision to be made.) It's also "possible" that the NV35 wasn't one of the cards "available for use", though with both the cards publically launched and benchmarked, and both companies eager to show off thier stuff, I'd find that unlikely.

 

[saf1]

I wonder if we will see anything demo'd on a NV35?

Maybe something from Tim Sweeney and company?

 

[Evildeus]

Demalion, i'm talking of 44.** drivers, seems you prefer to talk about 43.**. Sure someone else is evaluating it for me, isn't that the meaning of REVIEW? Are you saying that i should not accept the evaluation of a reviewer? I've confidence in HFR till now, and your comments won't change it. Sure there are issues in NV30-4, and none on NV35. And i'm asking why would it be the NV35 which should be at fault and not NV30?
"So what" disregard your comment on 44.** drivers not the others, don't make me say what you want to here. There are issues with other drivers, it doesn't seems with 44.** aka det FX. Performance figures are available, Screenshots are available, what do you want more? A testimony on the bible that Nvidia is not cheating? :?

So what? Before it wasn't correctly rendered when we found issues in 3DMark, it's no more the case! Are you saying that Nvidia tweakes its driver for ALL PS2/VS2 applications?

I'm saying the general PS2 tweaks are indicated to be broken for the nv30 in those other shots, as it is known that the hardware is capable of not exhibiting such problems within its capabilities (the problem is how quickly it does it). I'm also saying that the general PS 2 tweaks seeming to be fixed for a screenshot someone else is evaluating as identical to the REFRAST for you, without performance figures available, does not give any useful information with regards to the applicability of the NV35's behavior for 3dmark 03 and floating point shader performance.

Then how can you say it's tweaking or normal? :? Perhaps it's the NV30-4 series that have too much broken

It is pretty clear the drivers up to 43.51 change shader behavior uniquely for 3dmark 03, with visible discrepancies with DX 9 REFRAST and Radeon output apparent to my own observation of shots (not cropped) I've seen posted. If you wish to consider this established NV30 behavior a "So what?", disregard comments made already here in the forums, and focus instead only on the cropped comparison from parts of a 43.45 screenshot and what someone else is saying is an image that is identical between the 9800 output and 43.51 screenshots for the NV35 (without provided screenshots for your own comparison), I'm afraid I'll simply have to disagree with your evaluation, and we'll have to discuss in further detail at a future date.

 

[Johnny Rotten]

If PathRequset = NV30 then
UsePath = NV30
ElseIf PathRequest = ARB2 then
PathUse = NV30
End If

God bless the internet! Bug fixes and optimisations are dead, long live cheats and hacks!

 

[demalion]

Demalion, i'm talking of 44.** drivers, seems you prefer to talk about 43.**.

Look again at the 3dmark 03 screenshots you provided, and please note the correspondence between what I talk about and their captions. It doesn't do any good to quote my post in its entirety after your response when your response doesn't do anything to address the points I raise.

If you want to accept the presentation of evaluations and disregard the issues I've raised, I already recognized we have nowhere to go with this conversation, there is no need to continue establishing that. All the rest of your 44.xx/43.xx comments and your choice to accept the evaluation as presented are recognized in my discussion, and I have nothing new to add besides the above paragraph.

 

[T2k]

I wonder if we will see anything demo'd on a NV35?

Maybe something from Tim Sweeney and company?

Anything is possible - but that's not gonna be UT2003, for sure...

 

[Bjorn]

I wonder if we will see anything demo'd on a NV35?

Maybe something from Tim Sweeney and company?

Anything is possible - but that's not gonna be UT2003, for sure...

I think we already mentioned that this is with the NV35 using 8X FSAA which also uses supersampling = very low perfomance.

 

[Lezmaka]

Then they should just remove the xS support from the drivers so they can be like ATI and not worry about the performance hit from the supersampling.

 

[Bjorn]

Then they should just remove the xS support from the drivers so they can be like ATI and not worry about the performance hit from the supersampling.

I think that a lot of people that plays older games wouldn't be happy if they did that. Same type of people that are complaining to Ati with regards to them not supporting it.

 

[demalion]

Wow, Tom's did a pretty amazing job of mangling.

The PR presence is strongly evident, and provide an interesting data point for illustrating what points nVidia PR material is dictating to "reviewers" for emphasis by correspondence to Anandtech's article and the Avault "effort".

The full gammut of "Z-rate", "Stencilrate", and "Texturerate" are out, including the new(?) invention of multiplying the single texture fillrate by 16x (? I guess) to get "~27.2 Billion AA Samples/s". I suppose it is possible that NV35 has a purely multisampling 16x AA mode with functioning color compression, but I'm guessing "marketing math" for now (strange, because they list "4x" as the max AA mode right beneath the figure).

Textures per texture unit: 8 for R350, 16 for NV30/35.

They listed both driver version 43.00 and 44.03, but I don't recall "before and after" driver comparison.

They don't seem to mention memory for the NV35...the price listed is $399, but the context of other reviews and the article commentary, as well as no actual statement that it was 128MB that I noticed, seem to establish that it was just another failure of information on their part.

Not a pretty picture, IMO.

 

[FUDie]

Then they should just remove the xS support from the drivers so they can be like ATI and not worry about the performance hit from the supersampling.

When were you hired as spokesperson for ATI?

-FUDie

 

[psurge]

Bambers,
Do you have link to the tommti systems shader scores for the NV35?

Thanks,
Serge

 

[demalion]

I was just taking a gander at the Tech-Report review and further confirmation that nVidia is emphasizing this: "They say these new drivers take a "motion-based approach" to eliminating common texture artifacts like sparkles."

Recalling the "4x" AA discussion from some time ago, and then my own post processor theories and that nVidia has established that they have to provide driver code to "accurately represent" post processing for screenshots, it occurs to me that the image quality comparison morass may become a full time job for technical review sites.

I can't help thinking of parallels to 22-bit color and how some of the concepts of that might apply extensively to the NV3x in unexpected (to me) ways.

 

[T2k]

I wonder if we will see anything demo'd on a NV35?

Maybe something from Tim Sweeney and company?

Anything is possible - but that's not gonna be UT2003, for sure...

I think we already mentioned that this is with the NV35 using 8X FSAA which also uses supersampling = very low perfomance.

And? Who cares?
Noboy gives a flying frog about it if this is the only one which comparable to ATI's 6xAA.