Matrox Productiva G100 review

G100

First I should say this is again not exactly a gamers card. The whole point of Productiva was to be good cheap office card and therefore 3d acceleration is even less desirable than with Millennium. But Matrox pulled this one out after press announcement claiming G100 is a powerful DirectX 6 accelerator and quickly drop such boasting later. Despite its release just one month before G200 and being pin compatible with it, those two have little in common. Playing with G100 I can see why, it looks like inside is just a respin of Mystique's 3d engine even if feature set is closer to G200 on paper. For its release date in spring 1998 clocks are quite low (according to rumors and blind tools as low as 41,5/62,8 MHz), yet performance clearly exceeds previous MGA chips. Matrox still could not be bothered with OpenGL driver. G100 also support 24 bit color 3d rendering. Data Processing Unit can now calculate vertex fog and bilinear interpolator was added to Texture Mapper. AGP configuration is a bit perplexing. 1x speed is supported even with sideband addressing, but without DMA modes. Pipeline depth is whole two steps :)


While Productiva reports usage of AGP aperture, Incoming is dropping textures in 32 bit mode.

Reference specification sets clock at 71,5/143 MHz, but it does not seem anybody was so crazy to waste blazing fast memory on this chip. Alright, forget what Matrox said about powerful 3d accelerator and welcome to the domain of corporate computers. This Hewlett-Packard card will be your guide in the amazing world of stipple alpha blending.


Very simple board of cheapest Matrox product.

Gaming experience

Good news is that unlike Mystique and Millenium II this card displays textures in all games, only world texturing of Half Life and Thief resists. Last totally failing titles are Populous:TB and TNP. Also G100 can properly render fog of distance. And it finally has a bilinear filter. Well, this is one of the ugliest filters I saw, but it still helps. Baseline per polygon mip-mapping is implemented as well. The last big question is then the blending and G100 will attack your eyes with worst implementation ever.


Rain, shadows even environmental maps can be easily rendered by proper amount of ordered black pixels.

Those checkboards are even more abusive and more frequent too. It is hard for me to imagine how Matrox would not fix this old weakness and instead made it worse. The patterns are defined for 4x4 pixel squares by four bit register, meaning 16 options. Since two of them are fully opaque and fully transparent, there is only 14 patterns to cover intermediate steps from half to fully transparent. That is right, no steps between half transparent and opaque! This is the oldest Matrox chip you can play Unreal with and it shows. Like so often with early 3d cards you better disable mip-mapping though, or you will face some random mip level switching. I did that for my benchmark, as it decrease performance a bit and somewhat compensates for that low IQ. Unfortunately G100 suffers from mip-mapping errors in more games, even those there it cannot be disabled. In most cases games on G100 looks better than on Mystique, but there are situations when that horrible blending ruins it all. Pictures cannot tell how it looks in games with intensive blending. In words, let just say it looks to me like somebody wanted to enhance 3d games by adding ASCII art. Compatibility with MSI is broken, so for OpenGL games Techland's d3d wrapper was used.


I am saving screenshots at this position to compare texturing of the fence. The "winner" is clear already.
View Matrox screenshot gallery

Overclocking

G100 have a huge potential, Matrox really put the chip clock unnecessarily low. Basic PLL clock is 124,2 and according to other sources usual 3/2 dividers are used. I found it difficult to believe such low chip clock in 1998, personally I would expect divider of 2. My Productiva climbed to reported 56/84 before memory begun to fail. That is unimaginable for 8ns chips, another reason to reject old dividers.

35% percent clock bump has to show up right? These are the times when vsync on slow display makes me a sad panda. Only 18% percent average fps increase after excluding some results which totally hit frame limiters/refresh speed.

At those speeds even minimal fps won't save me, only 23,5% gain and lots of values at 25 and 37 Hz.

Surprise

Since Matrox quickly remove any gaming bells from G100 (just like with Millennium II) criticism of 3d capabilities would be off the place. In the office I could very well be happy with such a card. The big geeky question is what is the 3d engine exactly. It is faster than Mystique even with extra features, but basic 3d rendering looks similar. It is an evolution of Fast Texture Architecture, but what else could change? I found out one big thing, G100 supports multitexturing. All the specifications you can find around web say otherwise, but the numbers speak for themselves.

Shogo and Unreal run 40-50% faster and 3dmark 99 and 2000 shows more than three times texture fillrate increase. Watching this puppy drawing 100 mega texels per second back in the days when I believed it runs at 41 MHz filled my head with associations with Parhelia's quad texturing. Anyway, this should mean Matrox was second to the gaming market with multitexturing. Why didn't they advertise this? Don't tell me they were scared of 3dfx's patents. I am more inclined to believe that Matrox realized how silly it was to come out with this stipple alpha madness in 1998 and decided to pretend they were not even trying to be good in 3d. Multitexturing speed aside, would Matrox develop another Fast Texture Architecture 3d core and release it just one month before much better 3d chip? I have no idea. Other games tested do not support single pass multitexturing, which means there are still more performance enhancements in G100 compared to Mystique.

Later I got this board equipped with 8 MB SDRAM clocked at nice 99 MHz and it still has the same OC potential. This raised my expectations regarding gaming speed, but surprisingly it did not exceed tested 4 MB G100 unless memory capacity is of concern. At least with this card 32 bit depth becomes realistic possibility and allowed me to confirm mipmap selections of G100 are indeed problematic. In fact, these two cards often trade blows which implies chip clock is compensating memory clock differences. With system clock of 198 for 8MB card, it could be divided by 5 for ~40 MHz. But if the real chip clock is higher, then integer dividers would get me nowhere near the other board. Finally, I rejected the usual assumed clockrate of ~40 MHz, G100 has more options. Matrox register documentation claims 3 as only divider, but at the same time specifies reference clock of G100 at 71,5/143. That optional divider of 2 would explain how can my two G100 cards with very different system and memory clock perform as they do. With small 3d polygons faster chip wins, large polygon fills are more limited by memory bandwidth. I don't see any other reasonable explanation, so cards tested are now considered to have a lot more rational clocks of 62,5/125 and 66/99 MHz.

Epilogue

So much about a chip that should have play games until release date arrived and Matrox turned on a dime. Soon the prices dropped to 100 dollars and G100 was presented as office card- and it was excelent at that. G200 got all the attention. Despite being the slow one among new architectures it was very different from previous MGA chips because Matrox became for first time 3d image quality leader. The shocker came next year when G400 with all the traditional advantages of Matrox showed framerates on par with hyped TNT2 and Voodoo3. Devoted fans were waiting for the next big thing which would show GeForces and Radeons what Matrox can do, but waiting took too long. Following chips were about making G400 more cost efficient. After some time Matrox attempted a comeback to the high performance market with Parhelia in summer 2002. But the core was not very well balanced for games. It offered more of geometry performance and color precision which is good for professional applications, however, games suffered from poor pixel fillrate. Parhelia appealed only to small niche and Matrox left this arena to concentrate on graphics technologies for other uses. Somehow the company goes on with less than 1% percent market share, althought their last chips were released in 2008 and now offers products powered by AMD.

Continue to performance summary