Matrox Parhelia-512 from Impress Watch (Japanese site)

[Dave Baumann]

P10 main target is the the professional market, right?

The P10 architecture will be pushed into the consumer space by Creative, so its target is both markets.

 

[pascal]

Dave
You are saying the same thing.
It will be pushed but next year, right?

Also P10 is a product or the architecture?

 

[Nappe1]

wheeh... a long one but I'll tried to be as objective as I can be.

Both are above all that we seen so far. Matrox has more tradional way than 3DLabs, but only time shows both good and bad sides.

From what I’ve heard so far I think the biggest issue 3Dlabs will have is raw performance, which may be an issue for gaming cards – its going to be tricky to take an architecture as flexible as this and make it perform optimally in the limited gaming environment we have at the moment. So, IMO, the biggest concern for 3Dlabs will be sheer performance, and this will not get answered until we see products.

I agree. Also Parhelia's pratical performance is hard to predict so I am waiting some tests.

The other problem we have is that both these products are ‘tweener’ products – they are both .15um and are limited to being half DX9 and half DX8 parts. However, P10’s flexibility should mean that much more can be programmed for it and allow much of the functionality of DX9, possibly the only thing that would prevent it from being fully DX9 compliant is the lack of floating point texture stages.

both really are tweener products. And P10 is something that PixelFuzion tried over 3 years ago. afaik, their chip was even more programmable, but in that case, I think it was too programmable and became too complex. 'P10 way' really is the future, but I am fearing same as you: does it have too much programmability that is left unused because no one else has it? 3DLabs has also earlier showed the way. (for example everyone still remember Permedia2 and it's integrated GLINT geometry engine which was some kind of early stage of Transform part of HW T&L. correct me if I am wrong.) In OpenGL it gave significant boost for programs that used it, but for DX it was too much too early.

In terms of performance I would be highly surprised if both these products are not surpassed by NV30 and R300, and they will probably turn up within a few months of Matrox’s part. And timing is going to be a big factor on the perception of this part – if this is ready for release now then it will be an impressive feat, however if this won’t be available until the end of summer and the others are not that far behind then its not quite as impressive.

I am sure that at least feature crown will be taken from 3DLabs and Matrox, but performance is another thing. it would be logical that P512 and P10 would be passed in performance, but there is some interesting points...

3DLabs said in some interview that fully DX9 compliant version of P10 would need smaller than 0.13µm process. so it could be possible that DX9 graphics pipeline really needs a lot of more transistors. (at least doing every phase in floating point sounds some kind of resource and transistor hog to me.)

still it will be some sort of suprise, if Matrox and or 3DLabs are able to be fastest even in some small area after the R300 and NV30 are benched.

if NV30/R300 would be available as chips in August, how long it will take before we start seeing some cards?

at least for Matrox, it will be critically important that they are able to deliver in July and with at least with fully working drivers. (they don't have afford for Smoothvision kind of mockups.)

Another issue is the architecture. Matrox’s part still seems to be mired in fixed functionality, whereas API’s such as OpenGL2 and DX9/10 are more and more moving away from fixed functionality – something that 3Dlab’s part seems a far better starting point for future development; 3Dlabs probably won’t have to completely reinvent their architecture for future revisions, whereas Matrox may well have to, and considering the time it took them to come up with a new gaming architecture how long will it be before they manage another significant architecture change?

Only time shows that. I think Matrox has planned this somehow. Or maybe they just make 3D part all over again for next card. They did it twice during G-series (G100 ->G200 was big step and G200 -> G400 was smaller one), so I don't see any problems with that. But still, it will be more or less traditional way.

Perhaps 3Dlabs have been a little too forward looking in their part, which could impact on performance in the short term, but if you ask me I’d say that parts such as NV30/R300 and onwards are probably going to be moving closer to P10 than they are the fixed function nature that we’ve seen so far.

yes and they should be. Parhelia is actually pretty much late. I know when it was planned to come out, but I don't have any proof to show that, so I leave it as a secret. Still it will be product good enough to bring Matrox back to spotlight.

Another question is software support. Matrox has very aggressive Dev Rel. Just look at their gaming catalog / updates / patches section. I don't recal any other company doing such aggressive tech "push thru" since voodoo1/Mystique/rage II/S3d times.

there still so many questions that will be answered before the christmas... this time will be full of excitement I must say.

 

[Dave Baumann]

pascal,

It will be pushed but next year, right?

Also P10 is a product or the architecture?

I think we’ll see a Creative consumer part this year.

As for the P10 naming, that a good question – I’d initially thought the chip but some comments from Tim at 3Dlabs indicate that P10 may be also the name for the architecture.

Nappe1,

In OpenGL it gave significant boost for programs that used it, but for DX it was too much too early.

Its no coincidence that 3Dlabs are pushing openGL2 in the fashion they are, and this should be along at pretty much the right time. OpenGL2 is also a little beyond what can be managed on current (even P10) hardware as well, AFAIK.

3DLabs said in some interview that fully DX9 compliant version of P10 would need smaller than 0.13µm process. so it could be possible that DX9 graphics pipeline really needs a lot of more transistors.

The real issue there is the transistor cost of floating point texture/pixel stages.

if NV30/R300 would be available as chips in August, how long it will take before we start seeing some cards?

They are probably already sampling, so it wouldn’t surprise me if we see silicon reasonably soon after announcement. This will likely certainly be the case with NVIDIA and ATi recently said they wouldn’t announce until its close to being ready as a product.

Or maybe they just make 3D part all over again for next card. They did it twice during G-series (G100 ->G200 was big step and G200 -> G400 was smaller one), so I don't see any problems with that. But still, it will be more or less traditional way.

I think there is a question mark over the development resources they have at the moment, and one that probably has some merit given how long this has been in the making and the fact you say its late.

Another question is software support. Matrox has very aggressive Dev Rel. Just look at their gaming catalog / updates / patches section. I don't recal any other company doing such aggressive tech "push thru" since voodoo1/Mystique/rage II/S3d times.

But seeing as they haven’t been prevalent in the gaming market for so long how much of this do they have left? Is there any point in keeping that level of support when you have nothing to support?