Had some fun today with my Matrox M3D.
The point was to compare performance with period correct primary cards + PowerVR 3 (because it's fun).
All tests have been done with Final Reality.
All default, same PC, in a row.
PC Specs :
Mainboard : NMC-6BEX (440BX, Slot1)
CPU : Pentium III Coppermine 733 @842MHz
FSB : 153MHz - AGP : 102MHz - PCI : 38MHz
256mb SD-RAM
AGP aperture : 64mb
Palette snoop : enabled
OS : Windows 98SE
Tested cards :
The attachment Cards.jpg is no longer available
In a reader's order :
Videologic PowerVR PCX2 itself (Matrox)
Imagination PowerVR 3 (Powercolor Kyro II) (AGP)
Cirrus Logic Laguna3D (GD5465) (Chaintech) (AGP)
Tseng ET6000 @100/100MHz (NMC) (PCI)
Intel 740 (Real3D Starfighter) (AGP)
SiS 6326 @100/100MHz (Diamond SpeedStar A50) (AGP)
Planned : ATi 3D Rage Pro Turbo.
I know my setup won't make my data usable as a reference. I know FR isn't the best way to bench a PowerVR. I know that OCing some of the cards makes it even worse as a reference (even though it's from factory in 6326's case). I was just having fun and I'm sharing results here with no pretension.
Results
Imagination PowerVR 3 :
Overall : 4.61
3D : 2.35
2D : 7.94
Cirrus Logic Laguna3D
Politely declined to ride my AGP 102MHz slot.
Tseng ET6000 @100MHz
Overall : 4.10
3D : 2.28
2D : 8.19
Intel 740
Overall : 4.79
3D : 2.37
2D : 8.52
SiS 6326
Overall : 4.20
3D : 2.20
2D : 8.13
I fail to explain why the Starfighter performs so well. It was the first card to be benched, I benched it once more after the others to make sure it was consistant. It was. Typically, the i740 isn't the best card to be paired with a PCX2.
It's among the worst in other benchs I came accross (including one here, with Tomb Raider).
I'm a bit disappointed by the Kyro II's performance here. Might try other early 2000's cards.
The ET6000 performs slightly better than the 6326 (at the same freq.) even though it shares PCI's bandwidth. The gap between the i740 and these two ones is definetely noticeable in 3D.
In Robots :
6326 @100MHz : 24.21fps
Starfighter : 30.18fps
In City Scene :
6326 @100MHz : 21.55
Starfighter : 25.66
More data attached.
My conclusions :
Even though my data can't be a reference to others, it shows (once more) that the primary card is a factor in PCX2's speed. More suprisingly, in these tests with Final Reality, the fastest wasn't the one I expected.
According to Gona's tests on Tomb Raider, the 740 is among the slowest when paired with a PCX2. Maybe it's related to the benchmark itself (didn't find a way to benchmark TR for now). Or maybe it's related to my OC'ed platform : the i740 might be better at using system's RAM and since the FSB and the RAM itself are faster than in a stock system, it would become a serious advantage. I might test again with @133 FSB. It would strange though, since the Starfighter card features 8mb of video RAM, which should be enough without the need of the System RAM. Also it wouldn't explain why it does better than the Kyro II card (with its 64mb).