Firstly, thanks go to The_Asgard for his input and his encouragment to try this setup!
As reported in another thread, I've purchased a HIS X1900GT to Crossfire with my Connect3D X1900GT and successfully got it to work!
Both cards are of the 575/600 Core/Mem variety, and this on an Asus P5WD2 motherboard with a PCI-Ex Config of 16x/4x lanes. The rest of my spec is in my sig.
At present, I haven't had a lot of time to play, but here's the initial feedback with my original single card results.
Single Card under 16x PCI-Ex, Omega 6.7 Drivers, 600/625 Clocks:
2005 - 8984
2006 - 4576
Single Card under 4x PCI-Ex, Omega 6.7 Drivers, 600/625 Clocks:
2005 - 7738 = 13.87% Output loss over 16x
2006 - 4122 = 9.92% Output loss over 16x
Crossfire setup with 16x/4x PCI-Ex Bandwidth, Catalyst 6.9, 575/600Clocks:
2005 - 10751 = 20% Increase over Single Card in 16x Lanes.
2006 - 7402 = 61% Increase over Single Card in 16x Lanes.
Observations:
These were taken with the same Quality settings as the Single card results, and Catalyst A.I set to 'Standard'.
Its interesting the disparity between the two benchmarks. Although Driver differences and clock differences may acount for around another 5% improvement, as advised by many people the PCI-Ex is limiting the potential here. If we take the base '10% loss' for the PCI-Ex 4x slot and the fact Crossfire is transferring images across the bus to the first card for output, I think the system does pretty well.
The good news here really is that 72 Pixel Shaders do help, and will help with Next-Gen games
Whats more interesting though, is game performance. Again, this was done with Catalyst A.I set to 'standard', and thus using Supertiling/Scissor mode where appropiate. I don't have anyway of testing for framerates, but I was able to take both FEAR and Oblivion for a spin last night, increasing the quality settings.
FEAR
In Fear, I have moved from 1024x768 at 2xAA/4xAF to 1280x960 at 2xAA/16xAF and HQAF. The initial stuttering that I used to suffer from once the level had initially loaded has now all but dissapeared, and I get no 'bog downs' in speed at critical moments. It feels as if the second card can takes the strain when numerous grenades go off and massive firefights breakout, where before I got the odd glitch.
Oblivion
In Oblivion, I've gone from 1024x768 with 0xAA/4xAF, HDR and quality sliders set to default (I'd say medium to high in-game settings) to 1280x1024 at 0xAA/16xAF, HDR and all in-game detail settings are to the max. I get no stutter in outdoor area's and is very smooth unlike before hand. Before the game was certainly playable, but the experience now is a lot better as well.
I haven't tried a 'chuck' patch yet to get AA with HDR working yet, as soon as this is integrated into the main driver I'll give it a spin.
Next Steps:
I'm going to try forcing AFR mode by selecting Catalyst A.I to 'High', and re-run the benchmarks.
If anyone can recommend a way of benchmarking my games to see the improvements, please let me know.
Additionally, the clocks on the single card were higher than the standard clocks on the Crossfire setup, so I'm going to 'crack the nut' of Crossfire overclocking and see just how far this can go.
Conclusion
Now for the (un)surprising bit - I wouldn't recommend anyone plump for software crossfire. I've only indulged because the setup only cost me an extra £150 to implement, and the returns don't justify the expense.
Luckily when I upgraded my sempron to this current PenD 805, I bought a half decent PSU as well that supports SLI/Crossfire with enough amps. At 580 Watts sustained and 38 Amps on the combined 12v rails, its pushing it, but its stable. So if you take the cost as being £270 for two cards and possibly a PSU upgrade, then I'd say get a 512MB XT instead.
Also, if you chuck the fact in as well that their is no 'true' 16x/16x PCI-Ex motherboard available right now for any Intel platform, then this combination isn't going to live up to most peoples expectations - when we have this platform available (RD600 .... hopefully) then maybe we'll see software crossfire work to its potential.
Hope you find this useful,
Arthalen
As reported in another thread, I've purchased a HIS X1900GT to Crossfire with my Connect3D X1900GT and successfully got it to work!
Both cards are of the 575/600 Core/Mem variety, and this on an Asus P5WD2 motherboard with a PCI-Ex Config of 16x/4x lanes. The rest of my spec is in my sig.
At present, I haven't had a lot of time to play, but here's the initial feedback with my original single card results.
Single Card under 16x PCI-Ex, Omega 6.7 Drivers, 600/625 Clocks:
2005 - 8984
2006 - 4576
Single Card under 4x PCI-Ex, Omega 6.7 Drivers, 600/625 Clocks:
2005 - 7738 = 13.87% Output loss over 16x
2006 - 4122 = 9.92% Output loss over 16x
Crossfire setup with 16x/4x PCI-Ex Bandwidth, Catalyst 6.9, 575/600Clocks:
2005 - 10751 = 20% Increase over Single Card in 16x Lanes.
2006 - 7402 = 61% Increase over Single Card in 16x Lanes.
Observations:
These were taken with the same Quality settings as the Single card results, and Catalyst A.I set to 'Standard'.
Its interesting the disparity between the two benchmarks. Although Driver differences and clock differences may acount for around another 5% improvement, as advised by many people the PCI-Ex is limiting the potential here. If we take the base '10% loss' for the PCI-Ex 4x slot and the fact Crossfire is transferring images across the bus to the first card for output, I think the system does pretty well.
The good news here really is that 72 Pixel Shaders do help, and will help with Next-Gen games
Whats more interesting though, is game performance. Again, this was done with Catalyst A.I set to 'standard', and thus using Supertiling/Scissor mode where appropiate. I don't have anyway of testing for framerates, but I was able to take both FEAR and Oblivion for a spin last night, increasing the quality settings.
FEAR
In Fear, I have moved from 1024x768 at 2xAA/4xAF to 1280x960 at 2xAA/16xAF and HQAF. The initial stuttering that I used to suffer from once the level had initially loaded has now all but dissapeared, and I get no 'bog downs' in speed at critical moments. It feels as if the second card can takes the strain when numerous grenades go off and massive firefights breakout, where before I got the odd glitch.
Oblivion
In Oblivion, I've gone from 1024x768 with 0xAA/4xAF, HDR and quality sliders set to default (I'd say medium to high in-game settings) to 1280x1024 at 0xAA/16xAF, HDR and all in-game detail settings are to the max. I get no stutter in outdoor area's and is very smooth unlike before hand. Before the game was certainly playable, but the experience now is a lot better as well.
I haven't tried a 'chuck' patch yet to get AA with HDR working yet, as soon as this is integrated into the main driver I'll give it a spin.
Next Steps:
I'm going to try forcing AFR mode by selecting Catalyst A.I to 'High', and re-run the benchmarks.
If anyone can recommend a way of benchmarking my games to see the improvements, please let me know.
Additionally, the clocks on the single card were higher than the standard clocks on the Crossfire setup, so I'm going to 'crack the nut' of Crossfire overclocking and see just how far this can go.
Conclusion
Now for the (un)surprising bit - I wouldn't recommend anyone plump for software crossfire. I've only indulged because the setup only cost me an extra £150 to implement, and the returns don't justify the expense.
Luckily when I upgraded my sempron to this current PenD 805, I bought a half decent PSU as well that supports SLI/Crossfire with enough amps. At 580 Watts sustained and 38 Amps on the combined 12v rails, its pushing it, but its stable. So if you take the cost as being £270 for two cards and possibly a PSU upgrade, then I'd say get a 512MB XT instead.
Also, if you chuck the fact in as well that their is no 'true' 16x/16x PCI-Ex motherboard available right now for any Intel platform, then this combination isn't going to live up to most peoples expectations - when we have this platform available (RD600 .... hopefully) then maybe we'll see software crossfire work to its potential.
Hope you find this useful,
Arthalen
