nforce4max wrote:Matrox did at one point tried to market the Parhelia as a gaming card but they were very late to getting the card on the market and worse it was weak let alone the drivers. They could have done ok with it had they made it on a smaller process and scaled up the overall design but it still would have been slower than a GF 6600-6800 however it would have been beastly in CAD. There isn't much else out there so you are screwed unless suffer through with a bottlenecked pci card or bottleneck a newer pci-e or agp card with an adapter which likely isn't cheap.
What do you mean smaller process for the Parhelia? Parhelia, GeForce 4 Ti, and the Radeon 9700 all used the 150nm process node. Anything smaller would've been unfeasible... 130nm wasn't available until 2H 2003 (Parhelia was Q2 2002). I think the Parhelia's problem was the fact that Matrox decided to abandon the gaming GPU market in 1999, after the G400. G400 wasn't the best, but it was certainly a viable choice, depending on its price. Coming back after a 4 year hiatus out of practice doesn't bode well. Late to market is subjective, because on paper it should've rivaled Ti4600/R9700... and it launched between those 2 cards.
Technical issues are covered by Anandtech ( http://www.anandtech.com/show/936/6 ).
1. No Occlusion Culling. It had the highest theoretical MTexel fill rate, but that was entirely wasted because of this. This probably contributed to the wasted memory bandwidth and the over complicated 512-bit memory design. This in turn drove up costs.
2. T&L is incredibly weak in certain situations (1 light) and subpar at best.
3. Subpar shader performance (matching only a GeForce 3 era card).