NEC PC 3DEngine (PCX1) vs Matrox m3D (PCX2)

Today we compare the PowerVR PCX1 vs PCX2 for #GPUJune2
For the Half-Life 1 benchmark you can download from here www.vogonswiki...
#gpujune2
CHAPTERS
---------------------------------------------------
0:00 Intro
0:42 About PowerVR
2:28 PCX2 Improvements
2:59 Test System
4:22 Looking at the NEC PC 3DEngine & Matrox m3D
5:46 Quake 2
7:00 Quake 2 Benchmarks
7:53 Unreal
8:41 Unreal Benchmarks
9:48 Mechwarrior 2
13:00 Half-Life 1
15:18 Half-Life 1 Benchmarks

[@PixelPipes]

Very nice overview! The PCX2 was a really good improvement, and in my experience, slightly faster than the Voodoo 1 in most things. The way it works is also totally unique compared to everything else of the time, and all that together makes it one of the most interesting (and coolest imo) cards of the era.

[@anasevi9456]

As someone with a dreamcast [second hand with a big box of games from a bootsale after Sega had left the market] as a sprog... it's neat seeing the more ancient roots of PowerVR.

[@dkrbk]

I remember that for a long time a lot of enthusiasts deplored the choose of the PowerVR 2 over the "Voodoo 2" (most probably Voodoo Banshee) for the Dreamcast. They couldn't be more wrong. The PowerVR equals or outperforms the Voodoo 1 for a much simpler and cheaper design and the PowerVR2 is as fast as Voodoo 2's in SLI...

[@ScanLinesAU]

Great video as always, the effort put in does not go unnoticed!

[@Aeduo]

I kinda prefer Quake 2 without texture filtering. It has a kinda overly smudgy N64 look with loss of details, and the textures seemed to have a lot of details drawn in that are maybe intended to have sharp edges. The lighting does look rather bad though. Half Life definitely benefits from the filtering though. Probably was made with it in mind.

[@chrisducati26]

Great video! You forgot to test native tombraider version and ultimate Race 3track version

[@compufood]

Another great video! Glad to see my card going to good use instead of sitting in a box. ;)

[@michaeldale837]

Got anything else interesting to sell me? :D

[@compufood]

@@michaeldale837 maybe! I've got a Diamond Edge 3D NV1 to sell at some point, but I'm working on something and still need it for the moment - I'll keep you posted if you're interested.

[@stevec00ps]

Nice video thanks :)

[@PROSTO4Tabal]

Nice research. Subscribed

[@SinnerX69]

Nice! :) Now please compare PCX2 vs. PMX1 :D

[@r4z4m4t4z]

"shutdown sequence initiated, shutting down" i heard that a lot. the gold days when a gpu came with video games!

[@aublak7492]

I have a Midas 3. I found it at an ewaste site, but its missing its power regulator though. Never got around to repairing it yet.

[@Spear_n_Jackson_Number3]

At 7:09 you have benchmark figures, and I was thinking if we just calculate the raw pixels per second being output we get: 640x480: PCX1: 5.71 MP/s PCX2: 6.91 MP/s 1024x768: PCX1: 8.1 MP/s PXC2: 9.4MP/s The ratio of those latter figures seem pretty close to the ratio of the clock speeds. Out of curiosity, do you know if it's running with a 16bpp or 24bpp framebuffer? It's just that If the MP/S of the 1024x768 cases were closer, I'd be concerned about the 2D graphics card's PCI transfer rate. It's just that we found some 2D graphics card chips 'skimped' on the PCI write support (cough-S3 Virge-cough) so they rather throttled the rendering rate. There was a little benchmark/diagnostic tool (VidBand IIRC) that tried to assess the 2D card. I think Matrox Millennium's were about the best at the time

[@Spear_n_Jackson_Number3]

Hmm. Alternatively, is it doing double buffering with vsync enabled? If so, can you disable vsync? (It should make little difference visually as PVR is, to a certain extent, already double buffered)

[@modlabs]

Nice review! But it would be nice if you could compare PCX2 not only at 66 MHz clock, but on 83 MHz too. Now you can’t say what is really affecting performance. It could be clock speed or PCI bus.

[@Spear_n_Jackson_Number3]

The NEC silicon libraries turned out to be quite conservative. I heard from colleagues that the CLX2 in dreamcast could have been clocked at 133MHz, except it had to run at half the speed of the SH4 which was limited to 200Mhz.

[@dkrbk]

@@Spear_n_Jackson_Number3 Damn, what a beast. It was extremely fast at 100MHz already, compared to a Voodoo 2 or a Riva TNT. PowerVR PCX2 can be overclocked to 90MHz with proper cooling.

[@Banzeken]

7:53 Unfortunately, this is not using SGL. Both instances of Unreal are running in *software mode* and clearly do not look like SGL is supposed to look in Unreal.

[@3dfxvoodoocards6]

Interesting video :)

[@john_ace]

Just a bit of personal experience: The normally used S3 is not a perfect fit for the PowerVR. The PowerVR needs an extremely fast 2D graphics card for the best fps. In the 90s i had an "Imagine 128" 8mb PCI professional 2D graphics card that gave me about 10% more fps (if i remember correctly) with the Matrox m3D. I have could not test this since i no longer have either card for the PC. The Imagine 128 series of cards could be accessed sometimes faster than system ram and was perfect for DMA on the PCI. I don't know about the V3 though. The PCI bus can sustain about 80-100 MiB/s 1024x768 at 16bit will need 45 MiB/s at 60Hz, which should be possible over PCI (even with other cards installed). Having tested the Apple Quickdraw 3D accelerator which uses a similar rendering method (released oct. 95), i have found that using 3 of these cards at the same time is possible without saturating the PCI-bus if the CPU can feed the cards (>400Mhz for 3 cards). The Apple 3D accelerator uses 16x1 'tiles' instead of the 32x32 tile of the PowerVR. The PowerVR will therefore be more efficient since it can transfer 2KiB blocks at a time via DMA against the 32 byte blocks of the Apple card. Apple went with trilinear filtering that could not be turned off. When comparing the Apple card to the PowerVR shows that it only had 10 MTex/s and 120KPoly/s which is about half to a third of the PowerVR. Test on the Macintosh have shown the Apple card to be around half as fast as a Voodoo 1 in real world scenarios. BTW: Does anyone know why was it not possible to use multiple PowerVR cards at the same time as it was possible with the Apple card?

[@dkrbk]

I don't think this is a matter of how fast the 2D chip is. For example, I have no improvement at all by overclocking the primary card whatever it is. And newer cards don't provide better speed than old ones. I think this is more a matter of "compatibility", especially in drivers. You are right though, performance with S3 cards is terrible. In my experience the best cards to be paired with the PCX2 are late ATI Rage cards. Radeons are equally fast but show compatibility issues in some games. docs.google.com/spreadsheets/d/1p3G-lBQMQGcxKP_2Y-ZxiqDyHHrVI3A-yrkbkw1g2k4/edit?usp=sharing I don't know why it's not possible to use more than one PowerVR card since it's supposed to be scalable. I suppose this is related to drivers again. What you said about Imagine128 cards is very interesting, I always wanted to try them with PowerVR's but it's expensive. I'll try Imagine128 and Imagine 128 II sooner or later and update my datasheet.

[@wishusknight3009]

It would be interesting to see how each card scales with clocks...

[@PROSTO4Tabal]

Similar comparison can be done between riva 128 and tnt

[@shaneseward7139]

Hello! Thanks for the content, very informative. Do you run Tomb Raider 1 on the PCX2 out of pure DOS? I connot run it on mine, I just get a black screen. Thanks!

[@michaeldale837]

Yes I believe so, I don't 100% remember.

[@MILOGA17]

Hi, do you know where i can get the driver for the pcx2 Matrox M3D?

[@josephfrye7342]

well pcx 2 graphics card is suitable for resident evil and biohazard pc game though besides the pcx 1 though so there you have it.

[@slckb0y65]

20 full FPS in 640x480 under unreal ??? no mere mortal shall be allowed to wield such power ... ;)

[@Banzeken]

Not being rendered by the PCXs, sadly. This bit in the video is misinformation.

[@intel386DX]

PCX look like software rendering 😂