Full NV35 official feature list inside

[Uttar]

I've recieved, of course not in an official manner and I'm personally not under any NDA, the full NV35 feature list.
It's mostly PR talk, since it's pretty much the PR you'll get at launch, but it's still quite interesting.

I've used bold for everything I find noteworthy.

 CineFX ( TM ) Shading Architecture
 Support for the Microsoft DirectX® 9.0 Pixel Shader 2.0
 Support for the DirectX 9.0 Vertex Shader 2.0
 Long pixel programs up to 1,024 instructions
 Long vertex programs up to 256 static instructions with up to 65,536 instructions executed
 Dynamic, conditional execution and flow control
 256-bit advanced memory interface combines a wider memory data path with next generation controller technology for superior performance and throughput.
 Full 128-bit, studio-quality floating point precision through the entire rendering pipeline with native hardware support for 32 bpp, 64 bpp and 128 bpp rendering modes
 Accelerated pixel shaders allow for up to 12 pixel shader operations/clock
 Up to 16 textures per rendering pass
 Support for sRGB texture format for gamma textures
 IntellisampleTM performance technology, a Hi-Res compression technology (HCT), increases performance at higher resolutions through advances in compression and anti-aliasing technology.
 Shadow Volume Accelerator – accelerates shadow volumes by maintaining them more accurately and discarding useless information.
 Full nViewTM multi-display technology capability
 Integrated NTSC/PAL TV encoder supporting resolutions up to 1024x768 without the need for panning with built-in Macrovision copy protection
 Microsoft® Video Mixing Renderer (VMR) creates support for multiple video windows with full video quality and features in each window
 DVD and HDTV-ready MPEG-2 decoding up to 1920 × 1080i resolutions
 Dual, integrated 400 MHz RAMDACs for display resolutions up to and including 2048 × 1536 @ 85 Hz
 Dual DVO ports for interfacing to external TMDS transmitters and external TV encoders
 Dual internal TMDS encoders (one single and one Dual link) able to drive next-generation flat panel displays with resolutions greater than 1600 × 1200
 Digital Vibrance Control 3.0
 0.13 micron process technology for higher levels of integration and higher operating clock speeds
 40 mm × 40 mm, BGA 1309 flip-chip package
 Comprehensive Microsoft DirectX® 9.0 (and lower) and OpenGL 1.4 (and earlier) support
 DirectX and S3TC texture compression

EDIT: I corrected one or two things which are probably typos in the info I recieved - beside that, it's as-is.
EDIT 2: It's dual DVO ports, not DVI ports!

Please also note nVidia is not marketing any DDR-II in this feature list. This seems to imply the NV35 *does* use DDR-I.

The 12 PS ops/clock thing is rather vague, because you could actually describe the NV30 that way too. It's still interesting to see nVidia won't market it as a "8 pipelines" architecture, though.

Sounds like the "accelerated shadow volumes" stuff is what Carmack was rumored to be asking about.

Also, I'd like to thank my source for giving me all of this very tasty info! And no nVidia, I won' tell you who it is. Don't be too angry though - not like I see a good reason to be angry since it doesn't really harm anyone's business :)


Uttar

[DSC]

Very dissappointing, it seems we'll have another "4x2 or 8x1" debacle/fiasco on our hands.... Why can't Nvidia just clear up this issue once and for all...... :(

No mention of improvements in AA, guess we're still stuck with crappy AA implementations until NV40/45.... :(

Everything else seems in order. Wonder what is the delta of performance improvement is over NV30... :D :p

Once again, GREAT find, Uttar. Keep up the good work of getting all tasty infos. ;) :angel:

[Uttar]

No mention of improvements in AA, guess we're still stuck with crappy AA implementations until NV40/45.... :(

Well, nVidia didn't insist too much the GF3 had better 2x AA patterns than the GF2, IIRC. But yes, we'll probably have to wait until the NV40 for better AA patterns... No mention of AA Gamma Correction, either...


Once again, GREAT find, Uttar. Keep up the good work of getting all tasty infos. ;) :angel:

Well, it really isn't me who should be thanked - it rather should be the sources :) I'm really not much more than a rumor poster here - just an intermediary.


Uttar

[mikechai]

Well done Uttar and the anonymous hero !! ;)

That's why I visited NFI:GPURW 10 times a day.

:D

[mikechai]

Accelerated pixel shaders allow for up to 12 pixel shader operations/clock

So can we estimate its performance to be 3 times of FX 5600 ultra ?

[Uttar]

Originally posted by mikechai
So can we estimate its performance to be 3 times of FX 5600 ultra ?

Nope. It's an awful lot more complex than that, because the FX 5600 is actually capable of 6 ( 2 FP, 4 FX ) pixel shader instructions.

It's really hard to figure out what nVidia truly meant to say. How much of it is FP, and how much of it is FX? No idea... None at all.
It's quite likely, sadly, than that part will be about the same as on a NV30... Beside that texturing could be done as the same time as FP operations.
So it would be faster than the NV30, maybe by a factor of about 1.5x per-clock in many real-world cases. But is that sufficent against the R350? Only if you use FX intermixing - and DX doesn't allow it, only nVidia's OpenGL extensions.


Uttar

[mikechai]

Now the main difference between NV30 and NV35 is:-
1. 256-bit advanced memory interface combines a wider memory data path with next generation controller technology.
2. Accelerated pixel shaders allow for up to 12 pixel shader operations/clock.

Anand had mentioned before that the NV31 memory controller is more like NV35 than NV30. What is the main difference between memory controller of NV31 and NV30?

Will it be 32bit x 8 or 64bit x 4 crossbar?

With double memory bandwidth and 1.5x more pixel throughput, I think NV35 performance will be 1.5x in non bandwidth situation and 2x in bandwidth limited situation.
What do you think?

[mikechai]

But is that sufficent against the R350? Only if you use FX intermixing - and DX doesn't allow it, only nVidia's OpenGL extensions.

It seems possible with Cg.

In balanced mode, NV35 will definitely faster than R350.
What I'm thinking is whether NV35 will beat R350 in application mode.

[Uttar]

Actually, we don't know for sure if the 12 operations/clock thing is a change or not.
The NV30 is actually capable of doing that in very specific cases! Really. For all we know, they could be claiming something already in the NV30...

Also, Cg cannot force intermixing of FP & FX in DX9. It simply can't - Cg isn't an API, and it's limited by the API.

Most rumors suggested a 64x4 memory bus.

I'd guess, *overall*, 60% faster in PS limited cases, 10% faster in VS limited cases ( +5% in most cases, +25% when branching ) , and 90% faster in bandwidth limited cases.


Uttar

[Philibob]

 Comprehensive Microsoft DirectX® 9.0 (and lower) and OpenGL 1.4 (and earlier) support

hmm, does this mean that NVIDIA have no OpenGL support (like ATI's F-buffer)

[Dazz]

I was thinking the same thing, i was expecting them to go with OpenGL 2.0 also like ATi has done.

[zakelwe]

It seems like they have fixed the main issue, bandwidth, but I do not know enough about marketing speak to judge whether the shader performance is improved over nv30 to a large degree.

You'd assume "next generation controller technology " means that it is a new and improved memory controller over nv30, but who knows? Could be bollockese.

I guess it will be 5-10% faster than the 9800 Pro. If needs to be if it is to stop the " fair weather " nvidiots from running away to Ati :D

[Ratchet]

The only real reasons why I consider the NV30 a complete failure (compared to the R300 series) are:
1) FlowFX - no properly designed piece of hardware should need something that drastic to cool it (and still runs hot)
2) bad AA and AF implementations
3) poor pixel-shader performance when run at DX9 level precision
4) 128bit bandwidth limiting bus

so, we know they've addressed #4, but what about the other 3?

[mikechai]

OMG! How can I make such mistake!

I actually misread it as "12 pixels/clock" instead of "12 pixel shader operations/clock".

By the way, how many pixel shader operations/clock for NV30 ?

[Paul]

Originally posted by Ratchet
The only real reasons why I consider the NV30 a complete failure (compared to the R300 series) are:
1) FlowFX - no properly designed piece of hardware should need something that drastic to cool it (and still runs hot)
2) bad AA and AF implementations
3) poor pixel-shader performance when run at DX9 level precision
4) 128bit bandwidth limiting bus

so, we know they've addressed #4, but what about the other 3?

The FlowFX is gone, or at least it *was*. If they've decided to use it again, it'll only be on the top end model, probably the one using DDR2 in a few months.

AA/AF won't change too much, aside from the speed boost. I don't really think the implementation is bad anyway, but it's a personal thing i guess.

No idea about PS with DX9. That's more of a driver issue, isn't it?

[Luminescent]

Mikechai:
Actually, we don't know for sure if the 12 operations/clock thing is a change or not.
The NV30 is actually capable of doing that in very specific cases!

[SnakeEyes]

Honestly, if the performance is good enough to enable 4xS AA and 8x aniso (application mode, not that flakey quality / performance stuff) in my UT2003, then I'd probably be very happy with it. (Since they've already eliminated the FlowFX noisemaker)

I found 4xS and 8x aniso to make for a stunning image in the game with my GF4Ti4600, but at terrible framerates. My 9700Pro does nicely with 4xAA and 4x quality aniso (I've actually used 8, but I play in a clan, and there are a few situations where I wasn't happy with some slowdowns during match play..), but I personally felt that the IQ of the GF4 with the settings I mentioned was on par. It was the framerate thing that finally did in my GF4. ;)

[gravioli]

Looks pretty good. If it's the fastest thing out there this fall, then I may have a Nvidia chip in my system again. :)

[Solomon]

All this info is just salt and rocks. I ain't believing anything until I actually see the card and sites review it. Everything else coming from Nvidia is just hot air until proven otherwise.

Regards,
D. Solomon Jr.
*********.com

[Ratchet]

Originally posted by Solomon
All this info is just salt and rocks. I ain't believing anything until I actually see the card and sites review it. Everything else coming from Nvidia is just hot air until proven otherwise.

Regards,
D. Solomon Jr.
*********.com
Agreed, they haven't done anything to earn the trust of their consumers that's for sure.

However, they do seem to have learned from the disaster that was the nV30 launch so hopefully this time around they will manage to not only put together a decent card, but also get it on the market.

[Moose]

Originally posted by gravioli
Looks pretty good. If it's the fastest thing out there this fall, then I may have a Nvidia chip in my system again. :)

If they can match ATI's image quality AND provide faster speeds and do it quietly I will be interested.

[savyj]

Originally posted by Moose
If they can match ATI's image quality AND provide faster speeds and do it quietly I will be interested.

But what if it only has 85% the image quality but 140% the performance and is quiet? Can the mainstream-hardcore gamers (oxymoron, I know) really be happy with a significantly slower somewhat better looking card? I honestly don't know if I believe that...times are changing however so I'll keep an open mind.

[Ratchet]

Originally posted by savyj
But what if it only has 85% the image quality but 140% the performance and is quiet? Can the mainstream-hardcore gamers (oxymoron, I know) really be happy with a significantly slower somewhat better looking card? I honestly don't know if I believe that...times are changing however so I'll keep an open mind.
Of course, "signifigantly slower" is a relative term in this context, the R300/350 is still the best overall card on the market. I'm thinking most people won't mind the trade-off in performance if the image quality is noticably better.

[GlowStick]

First of all, Thanks Uttar for spilling the beans, so we can know first : D

The NV35 looks very promising, i want to see how it run on Doom3 (final release), i get the feeling that shadow stuff may come in very handy.

I like how all these new products are comming out in the summer, its when i get all my extra cash from!!

[savyj]

Originally posted by Uttar
Up to 16 textures per rendering pass

So does this mean it's 8X2 in some cases and 8X1 in most others? I can live with that :)