Whaddaya think of the Kyro PowerVR 3?

[The Rock]

This chip sounds like it has quite a bit of potential. There are 4 or 5 previews floating around (www.3dforce.com). Anyone care to share their thoughts on this?

The Rock
www.3dforce.com

[dZeus]

PowerVR series has a nice design.... driver suck á la S3.

If they can come up with GOOD drivers, this is a very nice card... I don't think it will have goood drivers, until I see the opposite stated in a review.

[MadCat]

The PVR 2 was an impressive chip. Clocked at a lowly 125MHz, it outperfromed TNT2 ultras at low resolution, all thanks to infinite planes technology. Its only downfall was the 64-bit path to memory that killed performance above 800x600x16, even the tilied architecture couldnt make up for that. It wasnt a driver problem, the chip was simply already a year old when released, and they didnt update it appropriately to make up for that.

I havnt read the article, but if this new chip has the memory bandwidth improved and even half the theoretical fillrate of the GF2, I would be interested.

------------------
This Signature Space FOR SALE / RENT

[Nuno]

Could we expect the Matrox m3D-II ?

[Himself]

It uses normal sdram/sgram, it's currently not clocked very high, can't recall what, less than 150MHz, it has comparable results to a GeForce 1 in Q3 if you can believe the reviews. It seems to solve the ram bandwidth problem without having to rely on bleeding edge ram or segmented ram designs.
It could very well lead the charge in how to fix the ram bandwidth problem for the next few years, especially considering 3DFX's acquisition of GigaPixel.

But, the proof is in the results, I want more testing, especially with the FSAA.

[Technoid]

I had a M3D and the only bad with it was the lack of even DX6 compliant drivers!

[MadCat]

Well, one can expect that from a video accelerator developed in 1996. Yes folks, the PCX2 upon which the M3D was based is quite old, around the same release time as Voodoo Graphics. The PCX1 was followed a few months later by the PCX2. The only difference was the PCX2 supported billinear filtering.

If you want some drivers for your m3d, try these out:

<a href="ftp://ftp.matrox.com/pub/mga/m3d/4.1.1.5.002.zip">M3D Feb 1998 drivers</a>

<a href="http://pvr.gamestats.com/files/pvr412gen.exe">M3D "patched" drivers</a>

Never used em, but theyre from pvr-net.com, one of the the last of the great vestige of PoverVR sites.

------------------
This Signature Space FOR SALE / RENT

[paulcs]

The GeForce SDK was clocked at a lowly 120 MHz, and it's a lot faster than this new chip. I'm always concerned when new products are released they are not competitive with existing products.

The driver upgrades that were to make the Viper II and the ATI Rage Fury Maxx never came. And the PowerVR boys seem to have distinguished themselves by delivering no product or late to market product for four years and somehow remaining in business.

Paul
paulcs@flashcom.net

[MadCat]

Four years? where have you been?

The PowerVR PCX2 made a sizeable market for itself, it had all the most important things to gain marketshare in 1997: solid drivers and a native Quake port. But this chip had one downfall: processor dependency. The same display lists and infinite plains that make the Kyro so good hurt the PCX2. The Kyro and the PVR2 chips both handled these things in hardware, but the PCX2 and PCX1 made the CPU calculate display lists and visible objects. As a result, people have discovered that PCX2 chips are quite powerful...on todays procssors!

As for not having a product for two years after the PCX2, have you ever heard of the Dreamcast paulcs?

The reason the PowerVR2 PC version was delayed was because They were chosen at the last minute to make the video accelerator for the Dreamcast. 3dfx was in negotiations with Sega to use the Voodoo2 to power Dreamcast, but Sega dropped out and went to NEC at the last minute. In order to meet the tight schedule for the Dreamcast, NEC had to drop all PC version work.

So lets see, PCX2, Dreamcast PVR2/PVR2 PC, PVR3...thats as many chips as ATI has released in the last three years (Rage Pro 1997, Rage 128 1999, Radeon 2000), and theyre still afloat.

Also, what benchmarks have you been reading? This chip is almost as fast as the GF2 and the Voodoo 5 5500, it would run circles around a GeForce SDR. Now, if youre talking PVR2 vs GeForce SDR, thats another story, but an unfair comparison.

------------------
This Signature Space FOR SALE / RENT



[This message has been edited by MadCat (edited 09 June 2000).]

[paulcs]

I was specifically talking about the PC market. I have gone through every post in this thread, and consoles have not been mentioned once. I'm pretty sure I could go through every thread posted posted to the MURC forums, Soap Box excepted, and with rare exception find the same thing. I considered briefly citing the PC videoboard division of their operation, but then I thought it was obvious. My mistake.

The only benchmarks I've seen were on Anandtech. I probably should have looked at more, but, in this case, the GeForce 256 is beating the KYRO board in Quake 3 performance:
http://www.anandtech.com/showdoc.html?i=1253&p=9

Given these numbers, I don't see KYRO's new board running circles around the GeForce 256.

The situation improves for the KYRO board with UTBench, but UTBench is an extremely problematic benchmark, and it's difficult to imagine anyone buying a board based upon UTBench benchmarks. I just don't see people lining up to buy the Rage Fury Maxx because of its great UTBench performance.

CPU dependency or scalability, at least on these forums, is not generally thought of as a fatal flaw.

Madcat, I don't think it was necessary to adopt the tone you did. I've had cordial relations with all the forum regulars, and I think this was a little unnecessary.

The truth of the matter is, I don't know of a single person using a PC card based upon a chip designed by this manufacturer. Not one. As far as I'm concerned, this is a bad sign. I didn't consider the PVR 2 a viable solution or a competitive product when it was released, and I'm suspicious of this chip as well.

While I've seen performance shortcomings corrected in drivers in the past, the claim has become a cliche, and more often than not, turns out not to be true.

I think all new PC gaming boards will be and should be compared to the GeForce2. It's the top dog as far as speed goes, and it has everyone's attention.

Let's talk about this (hopefully, in a civil manner) in three months. If KYRO seriously challenges nVidia, 3dfx and Matrox in the retail market or ATI, Matrox, and nVidia in the OEM market, well, then maybe I was wrong to be skeptical.

Paul
paulcs@flashcom.net

[Himself]

Let's talk about this (hopefully, in a civil manner) in three months. If KYRO seriously challenges nVidia, 3dfx and Matrox in the retail market or ATI, Matrox, and nVidia in the OEM market, well, then maybe I was wrong to be skeptical.

Exactly.

Here are a few more links:
http://boomgames.com/dyndoc.php/kyro/4
http://www.pvr-extremist.com/PC/kyro...asp?Section=PC

This pdf file on tiled rendering is a good read too:
http://www.pvr-extremist.com/downloads/kyro/tbr.pdf

[paulcs]

Well, the 3DMark2000 numbers in the PVR-Extremist article are impressive. I'd like to see more realworld numbers, and that, for better or worse, means Quake 3.

I'd like to see someone come along and clean nVidia's clock, if only to shut up some of their louder avocates. Matrox doesn't seem to be in a rush to get out a thunderously fast 3D accelerator, and the Voodoo5 would have looked a lot better if it shipped on schedule.

Paul
paulcs@flashcom.net

[Himself]

Erm, the first link I gave is Q3 arena benchmarks.

Basically it gets 60fps in 1024x768, 70 in 800x600 and 80-90 in 640x480. Bit depth doesn't matter, nor cpu speed it seems. Nothing wrong with scores like that, they are certainly up there.

[Narayan]

Is it true that Matrox was sort of forced to buy PCX2 chips from NEC, in order to renew the deal they had for memory chips?

[paulcs]

Himself, I saw them. The Quake 3 numbers aren't great compared to nVidia's last two offerings. I ran 3DMark2000 with the same settings as the guy in the second link did, with a GeForce2 GTS and a similar system. The 3Marks scores were very similar.

The KYRO board isn't scaling with bit depth, although it's scaling fine from 800x600 to 1024x768. It should be interesting to see the 16-bit scores with a driver upgrade.

I'm looking at the Athlon 850 scores and comparing them to scores I got with a PIII @ 800/VIA Apollo Pro 133A-based system, and nobody's clock is getting cleaned. The GeForce 256 DDR beats it soundly everywhere but 1024x768/32-bit.

Now the GeForce2 may be having some driver issues of it's own (flat scores between 800x600 and 1024x768), but it's numbers are much better across the resolution and bit depth spectrums.

If this board is being marketed as competition for the GeForce 256, as opposed to the GeForce2, *I think* KYRO has to get those 800x600/16-bit numbers over 100 FPS and get the thing out before the price on the DDR board drops to the KYRO board's projected price point.

If the KYRO board is set to compete against the GeForce2, well, they have a ways to go. I don't think this is the case, however. At $200.00, the KYRO board will most likely be marketed as a low cost solution, given that the GeForce2 will probably maintain its current pricing for the next several months.

Paul
paulcs@flashcom.net