Whaddaya think of the Kyro PowerVR 3?

[Himself]

Actually, it sounds like does so much internally that the normal speed up you get from going to 16 bit modes wouldn't apply as much to this architecture anyway. It always redering with 24 bit accuracy, the intermediate results stay on the chip itself, and it only renders visable bits of objects. The only slowdown would be in reading half as many pixels per 32 bit texture as a 16 bit texture, but it only has to be done for a depth level of 1 (or more if you count transparency) and intermediate results stay on the chip. I don't think the drivers would have much to say about it from the sounds of it, all they could do is get the textures to the card faster.

What I like is that they get GeForce numbers with ram being accessed at 110MHz, this is at TNT 1 speeds, not DDR at 166MHz. Obviously there is something to tile based rendering.

Price well, the $200 card is a 64MB deal apparently and the 32MB more like $150. My personal options are one of these or an ATi Radeon, which is likely to be at GeForce like pricing at $300 US. I could buy an NVIDIA card, but I could also shoot myself in the head, it would hurt less.

[MadCat]

Im sorry if my tone came off as offensive paulcs, that was not my intention.

Let me clear up a few of my statements with a few of yours:

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

It can be fatal when the average CPU on the market is a Pentium 133 and highend machines boasted Pentium MMX chips. You see, the trouble was not CPU scalability, but its dependancy; the PCX2 was up against the Voodoo Graphics and Glide, and the Rendition v1000 + native RRedline, which are both very low CPU-dpendancy combos. The problem was the PCX2 got marginally slower framerates on the same CPU vs these other two cards, in fact its CPU dependancy was so high the minimum system requirement was a P166! Even with this kind of opposition, the PCX2 sold well to highend system owners who could appreciate the scalability of these chips, since the VG and v1000 reached their framerate limits in the P200 range.

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.

Right you are, nobody in their right mind would still be using a PCX2 board, and no one in the right mind would have bought the castrated PVR2. However, I would like to point out one significant difference between the PVR3 and the PVR2. I was considering the PVR2 before I bought my G400, and even during Alpha testing the PVR2 performed terribly at high res/color depth. This card, on the other hand, performs well even at the Alpha level. This review board is also clocked at a meager 115 MHz core and memory, the clock speed is supposed to be 10MHz higher at release. And with this architecture, a small increase in clock speed means a sizeable increase in performance. I personally think this board will meet the proposed specs, after all this is the same .25mu process used for the PVR2, and they clocked that thing at 125MHz.

I would also like to put up my personal opinion on something: It doesnt faze me that the GeForce DDR beats this card at 640x480x16, this just proves the GeForce has more brute fillrate but less effective bandwidth. I play at 800x600x32 and up, and thats where the Infinite Planes and Tile Rendering architectures really make a difference. If The 32mb version of this card can hit $150 (around $75 cheaper than the cheapest DDR), I would consider buying it, but I have to let some time go by and find out more about this potential gem. Im not saying its the be-all and end-all, but its a nice chip with a potential market.

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

[paulcs]

You might want to seriously consider shooting yourself in the head before picking up that ATI board. It'll hurt less, create less of a mess, and it'll be a lot less stressful.

Seriously, I have experience with an ATI "gaming" board, and you might want to have a "friend" try it before you take the plunge. The experience was simply awful, and I found myself using fan modified drivers over ATI's for most of the board's life.

The guy who modified those drivers, and helped save ATI's butt, runs a popular ATI fansite. He and many ATI users are enraged with ATI over MAXX driver support.

ATI screwed up with the Rage Fury, the Rage 128 Pro, and now their MAXX. Just when you think they've done everything possible to make users' lives miserable, they come up with something else.

I'm sure the Radeon's DVD performance will be great. But take my word for it. Let a "friend" buy one before you do.

Paul
paulcs@flashcom.net

[paulcs]

Madcat, you have a point about CPU dependancy and older CPU's. I considered editing that comment out after posting it, but if I'm going to stick my foot in my mouth, I might as well leave it there.

I've never benchmarked the GeForce DDR at 640x480 at any bit depth or with any test. I was only talking about resolutions starting at 800x600. I'm writing the second part of a rather involved look at the last three generations of nVidia products. I tested a TNT2U, a GeForce DDR, and a GeForce2 on five different platforms using six different tests. I'm swimming in nVidia numbers. I avoided testing in 640x480 because I think its archaic. It's becoming increasingly rare to see anyone playing in that resolution, and the numbers themselves seem to tell you more about CPU peformance than anything else.

Again, I would love to see someone come out of the blue and decisively beat nVidia at its own game: sheer speed. (And I'd prefer it to be Matrox.) But I've noticed that speed increases tend to occur incrementally, and we've yet to see a Voodoo2 SLI-like step from one generation to the next since the Voodoo2.

Videoboard manufacturers always seem to announce products with impressive specs and features that sound great in theory, and I am almost always disappointed when the product comes to market and is tested by objective parties. I become doubly suspicious when prerelease boards post really impressive numbers on synthetic tests, and their realworld numbers are less impressive. (I don't really count UT-generated numbers. It's a really strange test, and the results are often bizarre.)

I just can't help but be skeptical. At the early stages of a board's life, the marketing people really come into play. I just can't help but think that every word out of their mouths is at best hyperbole and every number is somehow tainted.

And that's my rant for the day.

Paul
paulcs@flashcom.net

[This message has been edited by paulcs (edited 10 June 2000).]

[The Rock]

Not to drag this off topic, but listen to what paulcs said if your foolish enough to consider buying anything made by ATI. The fan site that Paul speaks of (I'm guessing) is http://www.rageunderground.com. It's run by the same guy that runs www.3dforce.com, the site that I work with. Lightspeed is the man, and he has done a LOT to help ATI users with problems in the past, and was also an ATI fan in the past, going so far as to create his own driver sets from various versions of ATI's driver components. But this foolish Rage Fury Maxx has stained ATI like Monica Lewinsky's blue dress. Lightspeed got a Rage Fury Maxx free from ATI for review purposes, and it's a TERRIBLE card. I speak with him regularly, and let me tell ya, that card only lasted 2 days in his machine. Also, in case you haven't heard, ATI has given up on producing a Maxx driver for Windows 2000, after they PROMISED one to users. They lie through their teeth. They also offer no refund to users who bought the card expecting Win2K drivers in the future; hell, they offer NO compensation of any kind. That just STINKS. /rant off

Anyhoo, back on topic, the Kyro whoops a 64MB DDR GeForce in high res 32bit benchmarks, even with alpha drivers. No one in their right mind is gonna play in 16bit color when 32bit can be had with ZERO performance penalty. Tile based rendering is a helluva lot more innovative than T&L if you ask me, since T&L is still useless and does nothing to address memory bandwidth problems. Imagination Technology is tackling the memory bandwidth issue in a very smart manner. Instead of taking the Nvidia approach, which basically says "well, if we have memory bandwidth issues at high res, just slap more expensive RAM on the card and drive the price up", Imagination Tech's approach actually attempts to cure the bandwidth problem by tackling one of the main causes of it. Imagine 3d games in the future with extremely high poly counts, most of which do NOT need to be rendered most of the time. Kyro's approach, IF implemented properly, is bullet-proof, and seems to be much more of a boon than T&L is. Just my $0.02.

The Rock
www.3dforce.com

[This message has been edited by The Rock (edited 10 June 2000).]

[paulcs]

OK, I give up! I'm just too suspicious, and I want this board to give nVidia a run for it's money. If it succeeds, it'll up the ante for all graphics chip manufacturers and shut the nVidia trolls up.

Lightspeed *is* the man. I'm very much in his debt as are many others. What an awful mistake I made when I replaced my OpenGL-challenged G200 with a Rage Fury. Never again!

Paul
paulcs@flashcom.net

[This message has been edited by paulcs (edited 10 June 2000).]

[Himself]

Paul, et al,

Don't worry, I never jump the gun on buying video cards.

[Himself]

Besides, if I had listened to all the nay sayers about the G400 and it's crappy drivers, I would have missed out on a great card, something to think about.

[The Rock]

Paulcs:

Oh the irony. I "downgraded" to an ATI card from my G200 just like you did. And after I got it, did I EVER feel like a cheap slut for cheating on Matrox. I won't make THAT mistake again. ATI = video card crabs.

The Rock
www.3dforce.com