NV40: Surprise, disappointment, or just what you expected?

[KimB]

Speaking of Doom 3...i am curious to see what type of gains the Ultrashadow II technology will have on the NV cards. Is this a feature that can be turned on/off? Have you heard any new rumors about this game being bundled with the Geforce 6 series?

"Ultrashadow II" is a programmer-side feature, so whether or not it can be turned off is up to the developer.

 

[Dave Baumann]

I thought we'd heard of rumors of .11 micron with low-k.

110nm is a cost optimised shrink of the 130nm FSG node. Being optimised for cost will mean there is not a low-k variant from TSMC. This I know for a fact.

Since I don't think ATI will be releasing some mid-low R4xx parts for a while yet, it seems to me that the NV4x parts for these markets will be left almost unchallenged, at least for a little while.

As yet, NVIDIA's NV4x committment with the rest of the line is "by the end of Q3". I would suggest that you don't yet have a clear picture of things.

 

[lwells]

I am generally pleased that Nvidia (err.. nVidia) is back in the game, more so because competition is good than an particular fondness for NV. I still think I will probably buy ATi's next-gen part even if its a little slower due to nVidia's duplicity when it came to drivers, etc, etc, but that's a personal (and economically irrational) choice I guess.

LW.

 

[Dave Baumann]

The 0.11 micron general-purpose technology is expected to enter risk production in the first quarter of next year.

Could just be me not having any idea what the exact meanings of "general-purpose technology" and "risk production" are, but it sounds like 0.11 micron isn't ready yet.

Calender or financial?

 

[Tim Murray]

Crap, that never occurred to me (probably because I have no idea when those dates are, meh. you'd think I'd know, but I can't even remember the number of days in each month). Oh well, Dave, you win. Getting used to EDT, or are you back?

 

[kemosabe]

Since I don't think ATI will be releasing some mid-low R4xx parts for a while yet, it seems to me that the NV4x parts for these markets will be left almost unchallenged, at least for a little while.

Seems the last roadmaps we got indicated that not much would change in the mid-range in the immediate future, since both ATI and NVDA are essentially sticking to PCI-E compatible versions of RV360 and NV36. But RV370 is about to make a spring entrance in the low end, and Dave can correct me but early indications are that it is a cost-effective 0.11u DX9 core based on RV350 (?). Haven't heard anything to suggest that NV34 won't continue to be NVDA's low-end part for the next several months. For what it's worth, Orton has recently stated that they have secured several top-to-bottom OEM design wins for the new lineup.

 

[Rolf N]

I was pleasantly surprised! I expected about 10 to 15 frames more than the current cards and instead, got to see double the performance at 16x12!

The image quality is (imho) better on the nVidia side now which is just what I was hoping. While looking at the tail shots comparison on Tech Reports review, I was completely happy with not having the blurring that is being shown on the 9800XT. AA goes to nVidia as far as I am concern because of the non blurring. (I can handle mere difference in horizontal AA vs. blurring.)

Tech Report Tail Section

Um hate to blow your bubble but if you read the review he stated that in the non aa shots both cards were blured so the radeon was rendering it correctly. It was the 6800 that was not rendering the shot correctly.

They were both rendering the shot "correctly". Supersampling has an effect similar to AF. 2x2 OGSS is equivalent to 2xAF. If you wonder why this is so, you just need to look at what happens to textures in any game if you increase resolution. There's more area to cover, so the chip will select larger mipmap levels, which means more detail.
The downfilter destroys this additional detail again for surfaces parallel to the near plane (just like AF does nothing for parallel surfaces). For surfaces not-so-parallel, you'll get more resulting texture detail than without supersampling. There are no detrimental side effects (such as texture shimmering caused by a negative LOD bias).

PS: this doesn't work on ATI R(V)2xx chips, because ATI "counters" supersampling with LOD bias adjustments with a net result (in texture quality) of, err, nothing
At least this was the case when I last looked for it.

 

[Pete]

Hey, who can read French. Not me. Somebody translate the damn thing already.

Already done.

We thus had resolved to test PS3.0 via some small shaders written in assembly by our care (since the HLSL compiler doesn't yet manage PS3.0). We however didn't have time to test this point in-depth. We concentrated on the cost of the conditionals while avoiding ransacking the performances in advance. For example we carried out a conditional carried out on a value which depends on the triangle. The 4 pipelines of each quad engine can thus function at the same time. We have, of course, avoided any texture use. This test resembled this:

If xxxx
Red screen
Else
Green screen
Endif

It's the simplest conditional we could realize. It took not less than 9 passes in the pipelines, which is enormous. We were expecting 2 passes in hoping that nV envisaged its architecture so that that could be done in only one. This result thus disappointed us, but it is possible that it is ascribable merely to immature drivers: time will tell!

I translated "branchement" as "conditional," but I suppose I should've left it as "branching."

The only thing I'm really disappointed about with GF6 is the AA. PS3.0, we'll just have to see if that's ever an issue since R420 probably won't support it and developer support will probably be limited at best.

I'm surpised no one's raised (more of) a fuss about their AF and its potential texture shimmering. NV3x's nicer AF was about the only thing it had over R3xx. I hope it'll be revealed in a later driver, once the benchmark frenzy has worn off, but I suspect nV's move to ATi's AF (given Derek's talk of Manhattan and Euclidean weighting/distance at AT) means they ditched the transistors spent on their previous type.

 

[Evildeus]

Welll the article also state that they don't know if the disapointing result on branching is due to the drivers or not. I think we need other tests, other drivers and perhaps another DX.

 

[Quitch]

The only thing I'm really disappointed about with GF6 is the AA. PS3.0, we'll just have to see if that's ever an issue since R420 probably won't support it and developer support will probably be limited at best.

Until we can start running extensive PS 3.0 tests, I think it's a little early to starting singing the praises of nVidia supporting it and ATI (probably) not. nVidia supported PS2.0, and look what happened there. Not that I think nVidia's PS3.0 support will be quite so bad, but will it really be a killer feature, or something that benefits in 5% of pixel shaders (we must remember the difference between what PS3.0 can do, and what the nv40 can do with PS3.0)?

 

[martrox]

Overall.....I'm very impressed. Only "real" complaints are lack of better AAand the power requirements, as many here have stated. The real suprises are lack of noise & lack of real heating issues.. A job well done.

 

[Unknown Soldier]

My concern for the NV40 isn't now .. it's for the future and those OC'ing freeks.

That big Die generates alot of heat. We've all heard that. When the NV40 was shown at GDC it had a clock of 475Mhz?? Maybe more? So the scalabity of the NV40 is there to maybe hit 500Mhz.. but how much heat will get generated when the card is running so high?

Lets say if the R42x does come out with the high clock(600Mhz) and the engine can preform and the R42x does beat the NV40 in most benchmarks.. how far can we expect Nvidia to push the NV40 before the heat issue becomes critical? What i've read is the NV40 can do 550mhz. How much further can it be pushed?

US

 

[Unknown Soldier]

Ok .. seems like the 6800 is actually cooler than the 5950 Ultra.

I also see on NvNews, MikeC did a few Room temperature tests.

Interesting.

US

 

[Malfunction]

I was pleasantly surprised! I expected about 10 to 15 frames more than the current cards and instead, got to see double the performance at 16x12!

The image quality is (imho) better on the nVidia side now which is just what I was hoping. While looking at the tail shots comparison on Tech Reports review, I was completely happy with not having the blurring that is being shown on the 9800XT. AA goes to nVidia as far as I am concern because of the non blurring. (I can handle mere difference in horizontal AA vs. blurring.)

Tech Report Tail Section

Um hate to blow your bubble but if you read the review he stated that in the non aa shots both cards were blured so the radeon was rendering it correctly. It was the 6800 that was not rendering the shot correctly.

They were both rendering the shot "correctly". Supersampling has an effect similar to AF. 2x2 OGSS is equivalent to 2xAF. If you wonder why this is so, you just need to look at what happens to textures in any game if you increase resolution. There's more area to cover, so the chip will select larger mipmap levels, which means more detail.
The downfilter destroys this additional detail again for surfaces parallel to the near plane (just like AF does nothing for parallel surfaces). For surfaces not-so-parallel, you'll get more resulting texture detail than without supersampling. There are no detrimental side effects (such as texture shimmering caused by a negative LOD bias).

PS: this doesn't work on ATI R(V)2xx chips, because ATI "counters" supersampling with LOD bias adjustments with a net result (in texture quality) of, err, nothing
At least this was the case when I last looked for it.

Thank you for that explaination. You cleared up alot of things for me that others hadn't considering their approach. Again, Thanks.

 

[Malfunction]

Ok .. seems like the 6800 is actually cooler than the 5950 Ultra.

I also see on NvNews, MikeC did a few Room temperature tests.

Interesting.

US

I was just as surprised about those details. Still, I don't see SFF: Prescott/NV40 or 9800PR0 256MB or 9800XT or R420 happening... lol.

 

[Khronus]

I don't see the big deal with the NV40. Theres no doubt its fast but with 2x the pipelines and another 12GB of memory bandwith than a 9800XT I'm not surprised it smokes a 9800XT at high resolutions. You can compare specs http://www.tomshardware.com/graphic/20040414/geforce_6800-02.html there. I'm definently taking a wait and see approach since by all reports the R420 has got another 50Mhz faster memory and at least another 100Mhz GPU clock. The lack of PS3.0 means jack to me, PS2.0 can do 95% of what PS3.0 can do and devs will write all games to take advantage of PS2.0 because of the installed hardware base. My Shuttle mini pc has neither the room nor the power for an Nvidia 2-slot furnace.

 

[jb]

For myself, the surprise is not that NV40 has delievered this performance, because I knew the specification for some time beforehand, and looking at the specification (in the assumption it was correct) then this is where I would have expected NV40 to be. The surprise, for me, is that this time the majority of what we were told beforehand has actually come to pass (which wasn't necessarily the case previously) and they have been fairly free and forthcoming with interesting architectural information - this is refreshing and welcome, and I hope this is something that is adopted long term and not just a one off.

Me two. However could it be that this time the have a part that they are really proud off? Kind of like a mother showing off her honnor role son? It seems when you have a winner on your hands you will want to talk about. But if you have that "loser" kid then you really don't talk about him much...or make up stories about him Anyways I am glad to hear about this and hope NV continues.

Also would like to congrads NV on a really top notch product. The only thing holding me back is I want to compare it to the R420 to see who gets my money this round

 

[Bjorn]

..The lack of PS3.0 means jack to me, PS2.0 can do 95% of what PS3.0 can do and devs will write all games to take advantage of PS2.0 because of the installed hardware base. My Shuttle mini pc has neither the room nor the power for an Nvidia 2-slot furnace.

Nothing stops the devs from also taking advantage of PS3.0. Though it remains to be seen if the difference will be that noticeable (Iq or speed).

 

[Dave Baumann]

Welll the article also state that they don't know if the disapointing result on branching is due to the drivers or not. I think we need other tests, other drivers and perhaps another DX.

In the initial round of NDA breifings I'd asked Kirk how the branching was supported on the PS of NV40 and then his answer was "Oh, we use the parallelism of the pipeline to figure out the answer". To me, this sounded very much like they have just exposed the conditional write masks for NV3x as a PS3.0 "dynamic branch" capability.

 

[Quitch]

Can you say that again, but slowly?