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

And for the 512 MB card, is there any chance of frequency upgrades?

 

What's interesting is that the NV40 really appears like it was made for OpenGL 2.0. With essentially zero resource limits in hardware, one should be able to run any GLSL program on the NV40. Hopefully the driver support will be there at launch.

 

Perhaps the rest is ok, but that part is BS

 

Well, lukar's post is pretty laughable, especially since he said the card can do FP24 AND FP32. What would be the point of supporting both? Not only that, but making such a pipeline work effectively would be really difficult, and the engineer's time would be more adequately spent working on other parts that would actually help performance and visual quality.

 

What is the latest version of ogl out there(implemented), and in what game?

 

Version 1.5 is the latest out. It exposes the GLSL language as an ARB extension. Version 2.0 is expected to include GLSL into the core of the API, so when I say "OpenGL 2.0" I'm really mostly talking about GLSL.

No games as of yet appear to be making use of GLSL. It'll be a bit.

 

Assuming that was the case and higher cards supported PS 3.0, then why would ATi be evangelising the use of the ps_2_b profile in their presentations to developers?

 

Has that actually been confirmed? I believe this theory originated from Dave's quote of a conjecture by an ATI engineer in regard to NV40=200 Million transistors rumor.

Also, what is the "correct" way to count transistors? CPU manufactures do seem to include cache in their transitory budget numbers, as does pretty much everyone from Sony (EE) to IMB (PPC5) to (surprise) ATI itself in the case of Flipper. I personally have hard time believing that memory cells are not being counted as part of transistor budget.

Dave, did you ever get to the bottom of this?

 

1)Well I don't know that they actually did, do you? It was in some leaked presentation notes, was anyone actually present for one of those presentations that could clear that up?

2)They could have the SM3.0 featureset, but still not think its really ready to be really used.

I am still just playing devils advocate, I don't believe that r420 will fully expose SM3.0. However, I am not convinced that anyone outside of NDA knows this for certain either, so it is all just guesses.

 

Doesnt seem like ATI will support 3.0, i guess its best to keep a open mind
about it cause they might just want people to belive it so they can jump out
and yell "suprise".. but it seems very unlikly in any case..

The big question in all this to me is, will PS3.0 support matter in the time between now, and the launch of NV50&R500?
and if so, how much?

Now, im not very knowledgable with this stuff, so anything i say is just
speculations and mostly unbased, just thoughts, which should be kept in mind.
But If we look at this Farcry "PS3.0 mod" as they called it, its basicly updated
content..
NV40 can prolly render the same thing in PS3.0 as with PS2.0 there, but slightly faster in PS3.0.. producing the same visuals...

Which means inturn that R420, can produce the same image, and if it turns
out faster then NV40 it could negate what ever speed NV40 gains on its
PS3.0 mode...

And i think this is the sort of PS3.0 implementations that will be dominant
for a good while, we might not se full on use of it, doing stuff that would
severely hamper peformance in PS2.0 mode until a few years ahead..

Like UE3.0, that prolly used some PS3.0 stuff in the demoing in the video,
and reason available PS2.0 hardware cant run it at decent rates is prolly
due to overall shader performance, mem bandwidth and that sort of stuff.
not due to PS3.0 vs PS2.0, again, NV40 could prolly run it almost as fast
in PS2.0.. as with R420...

So, there is alot of word twisting from nvidia i think, they want to over
emphasize the importance of using it now, while I would think ATI and specially ATI fans prolly underestimates it and tries to dissmiss it as
either "too demanding" or other reasons...

While in truth it is a step forward in evolution, not a revolution and not of
no importance at all either...

Similar to AMDs 64bit efforts, it wont play a vital role in a while, it doesnt
now, but its a step forward that would have been taken sooner or later..
and now that AMD has it, Intel will use it, but not until software is available.
Which it wouldnt be if AMD never took the step in the first place..

Nvidia will break the ground for PS3.0, ATI will follow when it wil start to
play a important role..

Well, that is the way Im looking at things, but im not a "man of the business" and Im well aware that my own impessions might be very
faulty, based on missconceptions and disinformation and stuff like that..
so take it for what it is, and correct me if im wrong..

 

I don't think PS3.0 support will matter that much between now and Fall 2004.

 

I expect PS 3.0 support to give the NV4x parts an edge in nearly every game that makes heavy use of shaders that arrives this year. In other words, I expect the support of SM 3.0 in the NV4x to improve performance (not by a huge amount, mind you....I expect 5%-15%, depending on the situation).

Larger gains will come later, after games start to built for PS 2.0 as the minspec. Once we're that far, the added efficiency from SM 3.0 will start to become more apparent.

More important, however, is the support for FP16 textures and blending. This support will make HDR just that much easier to design. We may see a few games start to come out that will only support HDR on the NV4x, not because it's not possible with SM2 hardware, but because it's easier on the NV4x (branching and long programs similarly make things easier: unless we see developer tools to make multipass and the like easier, we may well see many shaders soon that, while they are possible on SM2 hardware, aren't exposed because it's easier to do that particular effect with SM3).

 

Do you mean a edge NV40 @ PS2.0 vs NV40 @ PS3.0, or a edge vs R420..

 

but PS2.0 was a somewhat big step, ps3.0 is more of a fine tuning..
which in turn means that PS3.0 support is coming alot faster in games then
PS2.0 did, its even out already in Farcry 1.1, even tho you need DX9.0c
to expose it on a FX6800..

between the release of NV30 and the NV40 there wasnt alot of PS2.0 titles,
Farcry, Tomb raider.. and.. well there is prolly some more, but not alot..
Which was lucky for Nvidia cause they made the mistake of arrogance..
From what i understand, they developed the Nv30 expecting it to be the
standard for DX9, of some reason they left the Dx dev group for a while
and during that time they progressed developing DX9 standard, and ATI
was there participating and building the R300 to the DX9 specs..
When Nvidia came back, things had progressed, and DX9 wasnt what
they wanted it to be, and they were left with the FX shaders not conforming
to the standard, and doing something they didnt design it for..

Ultimatly it was their own fault, but it also means that Nv30 wasnt in itself
bad, just poorly adapted to the task it is being given in DX9..
Thus a bad buy if you want to run DX9 games...

Well, if that story is true anyway, its just something ive read...

It should also mean that the NV40 wouldnt be repeating the same mistake
and have redesigned shaders to work as it should...
And if so, they have gotten a card out that can run Dx9 well, before the
games started showing up for real..

The ones who got the short end of the stick are those who bought FX5
cards...

 

An edge on NV40 at PS 2.0 vs. NV40 at PS 3.0.

 

Regarding shader model 3.0 with respect to Unreal Engine 3:
http://www.beyond3d.com/forum/viewtopic.php?t=11124&postdays=0&postorder=asc&start=60

 

By the way, according to the video released by DemoCoder, while they may have not used PS 3.0, they did use FP16 for the lighting (I assume this means that they used an FP16 framebuffer). Also stated during that video was that the 6800 is the first video card capable of running the demo at decent framerates.

Oh, and don't forget that the Unreal Engine 3 showing that Mark Rein was talking about took place before the NV40 launch.

 

just before that video starts, he also sais the scene is about 1 milion polys..
in the video from the launch..
and then he continues with the source art is 2 milion worth, but half of that
is converted into normal maps and such "simulations" of geometric detail..

Dunno how current cards handle 1 milion poly scenes, but with the added
use of Shaders and HDR it sounds pretty heavy...