What are your temperatures under load like, as a matter of curiosity?
Well there is “typical gaming” and there is furmark+prime95.
I played a bit of Bulletstorm at max settings, 2048x1152 and got:
CPU temp max 59c
GPU temp max 64c, fan duty 49%
That’s without turning the fan knob up at all. To be fair, Battlefield 3 loads it maybe 2x as much as this, but there’s still a TON of headroom with the fan knob turned up.
With prime95 + furmark – which I consider to be utterly absurd but a good representative of the “you will never ever see it this bad in real life” absolute crazy possible maximum:
CPU temp 76c
GPU temp 84c, fan duty 86%
I don’t like running this test for too long because it’s … ridiculous, but it does trend toward stabilization at about those numbers. Note that the video card fans don’t even get to 100% which indicates the cards are not totally stressed, temperature wise, even in the dumb furmark case.
Per the above, extended play in Battlefield 3, which is VERY gpu intensive:
with fans on “low”, GPU temp 71c
with fans on “medium” (audible, but not irritating and a lot safer for exhausting all that 400w+ heat buildup over time), GPU temp 69c
As you can see here …
… it’s pretty much all GPU, the game is really not that CPU intensive, as long as you have a dual core CPU of reasonable vintage as a minimum.
and do note that actual multiplayer gameplay will have much lower framerates than what is listed there, as it’s a SP benchmark. Also surprising:
MSAA is a waste of time IMHO on this engine because of all the fancy image effects and post-processing; smoothing polygon edges adds virtually nothing. But it’s 30%+ perf loss on ATI, and only 10-15% on Nvidia
Nvidia’s SLI scaling is a bit lower than ATI’s crossfire scaling here. 1.9x versus 1.65x - 1.75x. That’s enough to more than negate their MSAA advantage, provided you have a dual card setup.
I’m suspicious of Tom’s Hardware’s findings regarding CPU performance. They’re using a pretty short segment in the single player game where most things will be scripted or canned. I can imagine the dynamic mayhem of the multiplayer would be more demanding on the physics simulation, but that’s not a repeatable, easily tested scenario. To get something I’d trust for comparison purposes you’d need to basically record performance over a very long time in multiplayer. But I don’t see them playing for 6 hours with each CPU on an active 64 player server.
If they want to loan me their Crossfire 6990 rig and Battlefield 3 for a little while, I’d be more than happy to play for six hours to get them a data point.
Brad, I agree. The reports are that multiplayer is a LOT more CPU intensive, which makes total sense.
For example on my i7-2600k, overclocked to 4.0 Ghz, I see peaks of 50% CPU usage in multiplayer. But to be honest I almost never see it higher than that. So this means you’d need a 2.0 Ghz quad core CPU to be safe in multiplayer, perhaps?
I took the average load of all 8 “cores” to get the CPU load, and the average of all the GPU temp sensors (there are three per GPU) to get the GPU temps.
I’d be curious to see results from a 64 player server, too. I mostly took umbrage to the suggestion on Tom’s Hardware that you could get by with a single, hyperthreaded core just fine. I doubt that’s even true of the single player if you look at more than just that 90 second sequence they used.
Same specs as above, 64 player servers on Seine Crossing (2 rounds) and Firestorm (1 round). note that this is on all HIGH settings with shadows at MEDIUM, so I am not sure how having anything on Ultra would affect it.
So… yeah. Spiking up to 70% of a quad core 4.0 GHz CPU is no joke, looks like 64 player servers is where you want to be calling for a quad.
For completeness, CPU and GPU1 / GPU2 temps as well:
