Nvidia in 2003

[bennyboy123]

Hi,


From what I have seen on this forum, there are a lot of really technically proficient people on this forum. I am curious about what Nvidia could do in 2003. After NV30 (GeForce Fx), could Nvidia possibly have the next generation NV4X (FX2) out by the end of 2003 or would it just be a higher clocked NV30 with maybe faster memory if available? Would the delay in NV30 be so significant that it would also puch back the generation after in a domino effect?


Thanks

 

[Reverend]

There are no definite answers to your questions (I'd doubt even NVIDIA employees can answer definitively) regarding timeframe. If you're looking for opinions, mine is that NV4x could possibly be available end-2003 and that it would not be a higher-clocked NV30 (that'd probably be a NV35) and that NV4x would most probably be DX10. My opinion of course.

 

[Ante P]

There are no definite answers to your questions (I'd doubt even NVIDIA employees can answer definitively) regarding timeframe. If you're looking for opinions, mine is that NV4x could possibly be available end-2003 and that it would not be a higher-clocked NV30 (that'd probably be a NV35) and that NV4x would most probably be DX10. My opinion of course.

DX10 won't be availible for quite some time though?
I've heard several reports for quite some time now that MS will save DX10 for NT6.0/Longhorn which isn't due until late 2004/early 2005.

"Upgrading" the Shader versions to 3.0 which seems to be resident in DX9? And perhaps adding better support for "dynamic" Displacement Mapping and some sort of HOS together with bumped core and memory seems more likely to me.

Earlier on there was some talk about DX9.1, in early 2003 allready, with PS/VS 3.0 but I'm under the impresion that its allready there in the current DX9 version?

 

[Bigus Dickus]

Going on nVidia's release record in the past two or three years, we would expect to see the NV35 by Christmas 2003, and the NV40 late fall of 2004.

Previous record may not be a good indicator of future performance though. YMMV.

 

[RM. Andersson]

They will probably have to release the NV40 as soon as they can.
They will not delay it on purpose til they think the time is right to release it IMO.
They dont have time to think like that because of the products ATI plan to release.
As soon as possible will be the best option for Nvidia.

They want to regain market shares and earn more money and NV40 will help them do that if it´s better compared to what ATI can offer.

If they have a development team that is working with NV40 only I dont think they have told them to delay the projekt.
That would be a big surprise IMO.

Regards!

 

[JD]

ps3.0 is in dx9. Ms will update hlsl compiler on regular basis and will release dx updates for shaders imo. I also think ms will concentrate on d3dx lib more since core api is so well defined now.

I have no doubts that nvidia will compete with ati tech wise and will probably sell at higher prices like they do now. I think most people will be satisfied with slower cards because the gap between mid and high end is widening imo. As chips get more complex, r&d costs increase thus chip prices will too to recoup r&d costs.

 

[KimB]

NV35 would more likely be out by August, and NV40 by early next year.

Now I'd like to say something else. In the recent past, ATI's and nVidia's product generations (and model numbers) have coincided fairly well. This may not hold true into the future. Going into the next couple of years, we will likely see far less change in graphics cards as far as programmability is concerned. That means that the only big differences between generations will likely be performance and image quality. Depending on how things go, we may see, for example, an NV45 that is roughly on par with an R600 in terms of technology.

Currently, there are rumors floating around about a programmable primitive processor. This is most certainly the next big step in GPU's. We also have the VS 3.0 specs that include a texture read in the VS. I don't see any reason why both of these can't make it into the NV35.

Anyway, what I suppose I'm trying to say is that don't attempt to figure out how nVidia and ATI will compete next fall until the chips are announced, at the very least. This time around, the model numbers may not mean as much as they have in the past.

 

[Ilfirin]

Note that VS3.0 only sets a MINIMUM instruction count of 512, the upper bound is entirely up to the vendor.

It is the same with PS3.0 except that there is a cap at 32,768 instructions.

I expect the next few years (until DX10) to be little more than each card being faster and faster (with of course progression with features like anisotropic filtering and FSAA), with more and more supported instructions. As far as programmer centric feature sets go (i.e. stuff that requires developer support), I don't see much of a change for a while, at which point there will likely be huge changes.

 

[megadrive0088]

Hi,


From what I have seen on this forum, there are a lot of really technically proficient people on this forum. I am curious about what Nvidia could do in 2003. After NV30 (GeForce Fx), could Nvidia possibly have the next generation NV4X (FX2) out by the end of 2003 or would it just be a higher clocked NV30 with maybe faster memory if available? Would the delay in NV30 be so significant that it would also puch back the generation after in a domino effect?


Thanks

Going by Nvidia's history, The GeForce FX2 should be the NV35. much like the GF2 was NV15 after GF1 was NV10. GeForce FX2/NV35 will most likely be out late summer or early fall, with the absolute latest being Christmas 2003. 'FX2/NV35 should be a refresh of FX/NV30, more than a speed bump. An updated core, PLUS speed increase. more transistors.
135-150 million transistors is my guess. It's not going to be a whole new generation, like NV30 is over NV20/NV25. FX2 should be to FX what the NV25 was to the NV20. I'm saying basicly the same thing in different ways so you understand Perhaps NV35/FS2 will support VS and PS 3.0
and DX9.1 or 9.2 but DX10 will be the next generation

Nv40 will be a whole new architecture, the next generation, most likely with DX10 and PS/VS 4.0 or 5.0 (depending on how much DX9 is stretched) NV40 will be a 2004 product. If Nvidia is agressive, they'll have NV40 out by March 2004, but i highly doubt they'll have NV40 out in late 2003, even though they should since NV30 is nearly a year late. (6 months to almost a year actually.)

 

[MDolenc]

Note that VS3.0 only sets a MINIMUM instruction count of 512, the upper bound is entirely up to the vendor.

It is the same with PS3.0 except that there is a cap at 32,768 instructions.

VS_3_0 and PS_3_0 have instruction slot count between 512 and 32,768. however since both pixel shaders and vertex shaders now have static and dynamic flow control actual number of executed instructions can be much higher.
You'll have to wait for DX10 to see unlimitited instruction slots...

 

[KimB]

I'd just like to add one more thing:

Since the TNT2, each refresh that nVidia has released has progressively added more to the architecture:

TNT2: Efficiency improvements, particularly for 32-bit color
GeForce2: Supported two bilinear textures per pixel pipeline per clock instead of the one available with the GeForce, as well as other performance improvements.
GeForce4: FSAA improvements, significant efficiency improvements, some added shader instructions.

What more will the NV35 add over the NV30? Will some of those improvements be seen in the NV34 (The supposed low-end NV3x part to be available around April)?

 

[WaltC]

Hi,


From what I have seen on this forum, there are a lot of really technically proficient people on this forum. ...

Thanks

Oh, this is definitely not true. What you have stumbled upon here is nothing less than the Worldwide Consortium of Technical Fiction Writers, and we're proud of it! Basically, it's our job to invent fictional technical scenarios which we then attempt to put down on epaper, with our colleagues making friendly suggestions on how we could substitute arcane technical words and acronyms at various points in our fiction, all for the noble purpose of writing Better Technical Fiction. As you've noticed, some of us have become extremely proficient. It is indeed nice to be noticed, though.

 

[bennyboy123]

256 bit memory interface

Hi,

Would it be easy for Nvidia to implement a 256 bit memory interface on either the NV35 or NV40 to be in stores next American Thanksgiving?


Thanks

 

[nooneyouknow]

Chalnoth, I think you have it slightly miscalculated:

R300 ~= NV30
R400 ~= NV40
R500 ~= NV45 / NV50

The problem is no one really knows how they will line up but I think the above, up to the R400 timeframe, represent what is the reality. Now, in saying that, if DX does not get a 9.1 update, the true feature set of R400 / NV40 will not be shown / none unless you are using OGL 2.0. Sure we expect PS / VS 3.0 to be supported in R400 but the truth is that the R400 might be able to do lots more, who knows...

 

[megadrive0088]

big deal if Nvidia has a part (NV35) by next fall with a 256-bit bus. they will have been around 15+ months late compared to the rest of the industry (Parhelia, P10, R300)


that said, i would be horrified if NV35 has a 128-bit bus. that would mean Nvidia wouldnt move to 256-bit until NV40 in 2004!

 

[KimB]

The bus width is not the important part. What's important is final performance. If nVidia can still deliver in performance with a 128-bit bus, why move to the more expensive 256-bit bus?

 

[nooneyouknow]

Chalnoth makes a good point. What will be interesting is how well the NV30 / R300 match up considering the bus width. You would think that if R350 was still 256-bit plus adding DDR II, final speed being still up in the air, what kind of performance delta can be acheived vs. the original R300. I think that will show the efficiency of the architectures when the clock speeds are closer matched on both GPU / Mem.

 

[Althornin]

The bus width is not the important part. What's important is final performance. If nVidia can still deliver in performance with a 128-bit bus, why move to the more expensive 256-bit bus?

Well, so instead of moving to the more expensive bus, they can use the more expensive memory (exotic high speed ram).

 

[bennyboy123]

Hi all,

It looks like the GF FX lags behind the 9700 in 1600x1200 with AA and AF filtering in real world games. R400 will have more than 30 GB/s bandwidth for almost double the NV30 when they use a 500Mhz memory clock. If NV35 still has 128 bit, the R400 will rule back to school if in time and Christmas. Given the pressing need for Nvidia to implement 256 bit for the 2003 holidays, is it easily and quickly feasible?


Thanks

 

[Xmas]

VS_3_0 and PS_3_0 have instruction slot count between 512 and 32,768. however since both pixel shaders and vertex shaders now have static and dynamic flow control actual number of executed instructions can be much higher.
You'll have to wait for DX10 to see unlimitited instruction slots...

Can anyone explain to me why DirectX, ie the software, sets an upper limit for the number of instructions? I can't exactly see the sense behind this. Shouldn't it be limited by the hardware/driver alone?