3-Way SLI and 5760x1080 BF3 benchmarks using GTX 580 1.5GB

[oorah]

I havn't seen any reviews on 3-way SLI performance and scaling for BF3, so I slapped some numbers from Fraps together to make a few charts. It will at least give us an idea on 3-way SLI scaling in BF3...I'm glad to see very good scaling results as of yet, albeit at 1920x1080.

Finally some 2d surround goodness. 4xMSAA was unplayable due to most likely hitting the VRAM wall, where the menu/cursor instantly becomes sluggish, which carried over into gameplay as indicated in the unplayable framerates.

I played on Caspian Border with 64 slots and the only setting I could find consistently smooth was 0xMSAA. I am a nitpick when it comes to framerates not dipping below 60fps, so I was pretty pleased with the peformance. IMO, the sequence benchmarked stresses the GPU's about 80% as much as a 64 player map with a lot of action going on. Nothing beats playing the actual game and getting a good feel for playable settings.

I used the same canned benchmark as tomshardware, using the same sequence as seen in the following video...
http://www.youtube.com/watch?v=wAA-NJj0Z0k&feature=player_embedded
2-way SLI scaling numbers were consistent with Tom's.

Test setup:
asus maximus iv extreme, 2500k @ 4.8ghz, XSPC rs240 kit, evga gtx580 1.5gb tri-sli, 128gb m4, 2tb hitachi, 8gb mushkin redline 2133mhz, antec tpq-1200, storm sniper

 

[cmdrdredd]

I've played Bf3 quite a bit and below 60fps is not gamebreaking for this title. Anything above 40 is perfectly playable for me. Opinions vary but it's not so much a twitch game that you need every frame IMO.

Nice to see some tri SLI numbers though. Hardly ever see this.

 

[Termie]

I played on Caspian Border with 64 slots and the only setting I could find consistently smooth was 0xMSAA. I am a nitpick when it comes to framerates not dipping below 60fps, so I was pretty pleased with the peformance. IMO, the sequence benchmarked stresses the GPU's about 80% as much as a 64 player map with a lot of action going on. Nothing beats playing the actual game and getting a good feel for playable settings.

I used the same canned benchmark as tomshardware, using the same sequence as seen in the following video...
http://www.youtube.com/watch?v=wAA-NJj0Z0k&feature=player_embedded
2-way SLI scaling numbers were consistent with Tom's.

Test setup in sig

Agreed - canned benchmarks really aren't close to actual MP matches.

Also, your sig isn't showing up - I think you need 20 posts or something like that for it to appear. Might want to just post your system details in the OP.

 

[cmdrdredd]

Agreed - canned benchmarks really aren't close to actual MP matches.

Also, your sig isn't showing up - I think you need 20 posts or something like that for it to appear. Might want to just post your system details in the OP.

It depends on what you're trying to do. As I've always said the "canned benchmarks" you refer to are used not to stress the GPU 100%, but to get a repeatable baseline with which to test various hardware in an environment that is the same each time so the results are repeatable. MP varies game to game so FPS would vary as a result.

 

[Termie]

It depends on what you're trying to do. As I've always said the "canned benchmarks" you refer to are used not to stress the GPU 100%, but to get a repeatable baseline with which to test various hardware in an environment that is the same each time so the results are repeatable. MP varies game to game so FPS would vary as a result.

Absolutely, but IMO reviewers should clearly note not just that they are benching the SP game, but that the MP game results will be lower. That way people don't get too optimistic about the performance of their cards.

For instance, for the hypothetical HD6950 buyer (know any???), actual performance at high settings will be 46-48fps, not the 60fps that was widely reported. That's a huge difference if someone is shopping for a card to play BF3 multiplayer.

 

[cmdrdredd]

Absolutely, but IMO reviewers should clearly note not just that they are benching the SP game, but that the MP game results will be lower. That way people don't get too optimistic about the performance of their cards.

For instance, for the hypothetical HD6950 buyer (know any???), actual performance at high settings will be 46-48fps, not the 60fps that was widely reported. That's a huge difference if someone is shopping for a card to play BF3 multiplayer.

Yeah, Hardocp does a good job of it by posting separate MP from the SP benches. In their latest BF3 preview they noted briefly how the MP played separately from the SP. I'm sure they will have a more in depth article down the line with separate numbers.

I'm buying a HD6950 myself, actually it shipped today. It will be one of two things for me.

1) Unlocked to 6970 shaders and overclocked and I grab a 2nd one and hope for unlock too at which point I crossfire them as what would basically be 6970s.
2) An interim upgrade until I snag another 6950 for crossfire.

You might wonder why, when I have a GTX 295, I would do this. The 295 is barely getting 40fps online on high settings and medium terraign detail because of memory limitation. Plus I can't get DX11 on it which I kind of want. I want to play through some games with DX11 on and experience the whole thing that way.

 

[tviceman]

You are getting excellent scaling going from single, to SLI, to 3-way SLI. Why are you using msaa at all? The Frostbite engine doesn't work well with MSAA, and from what I've read it might actually decrease image quality. Regardless, is there noticeable microstutter when playing with all 3 GPU's enabled?

 

[Termie]

Finally some 2d surround goodness. 4xMSAA was unplayable due to most likely hitting the VRAM wall, where the menu/cursor instantly becomes sluggish, which carried over into gameplay as indicated in the unplayable framerates.

I played on Caspian Border with 64 slots and the only setting I could find consistently smooth was 0xMSAA. I am a nitpick when it comes to framerates not dipping below 60fps, so I was pretty pleased with the peformance. IMO, the sequence benchmarked stresses the GPU's about 80% as much as a 64 player map with a lot of action going on. Nothing beats playing the actual game and getting a good feel for playable settings.

I used the same canned benchmark as tomshardware, using the same sequence as seen in the following video...
http://www.youtube.com/watch?v=wAA-NJj0Z0k&feature=player_embedded
2-way SLI scaling numbers were consistent with Tom's.

Test setup:
asus maximus iv extreme, 2500k @ 4.8ghz, XSPC rs240 kit, evga gtx580 1.5gb tri-sli, 128gb m4, 2tb hitachi, 8gb mushkin redline 2133mhz, antec tpq-1200, storm sniper

Nice rig, and thanks for contributing your benchmarks. From the above it looks like surround plus ultra hits the memory limit. Not surprising, as 1920/ultra uses about 1.3GB. You'd be way past that at 5760. Amazing, however, that 2xMSAA doesn't go through the memory limit. Very nice!

Yeah, Hardocp does a good job of it by posting separate MP from the SP benches. In their latest BF3 preview they noted briefly how the MP played separately from the SP. I'm sure they will have a more in depth article down the line with separate numbers.

I'm buying a HD6950 myself, actually it shipped today. It will be one of two things for me.

1) Unlocked to 6970 shaders and overclocked and I grab a 2nd one and hope for unlock too at which point I crossfire them as what would basically be 6970s.
2) An interim upgrade until I snag another 6950 for crossfire.

You might wonder why, when I have a GTX 295, I would do this. The 295 is barely getting 40fps online on high settings and medium terraign detail because of memory limitation. Plus I can't get DX11 on it which I kind of want. I want to play through some games with DX11 on and experience the whole thing that way.

Yeah, you're going to do way better than that with an HD6950. I don't recall which features are DX11 exclusive, but you'll see more of something, and it will perform better. Win-win. Just don't expect 60fps!

 

[cmdrdredd]

Nice rig, and thanks for contributing your benchmarks. From the above it looks like surround plus ultra hits the memory limit. Not surprising, as 1920/ultra uses about 1.3GB. You'd be way past that at 5760. Amazing, however, that 2xMSAA doesn't go through the memory limit. Very nice!



Yeah, you're going to do way better than that with an HD6950. I don't recall which features are DX11 exclusive, but you'll see more of something, and it will perform better. Win-win. Just don't expect 60fps!

I don't expect that, not until I crossfire and overclock em.

 

[happy medium]

uh oh the gtx580 runs out of memory at ultra high res at 4xmsaa, I bet suddenly we all need 4x msaa?

My guess 2xmsaa is not enough now?

"How about ULTRA w/ FXAA high but no MSAA (you realize by now MSAA is shit for this game, right?). A 6950 can run at 6970 speed or faster."

http://forums.anandtech.com/showpost.php?p=32501957&postcount=34