Originally Posted by Mad Cat
Hey guys, update from the [H] side since Kyle is probably busy. He has finished the SB rig and is stress testing at a 4.7GHz OC. He says it's been stable so far, and shouldn't have to go lower than 4.6GHz in a worst case scenario. I think he intends to ship today so Brent can begin testing next week. Link
Also, to address the concerns over VRAM: Brent checked the VRAM usage of the 580s and made sure to keep the settings low enough across both cards so that the 580s never maxed out their VRAM usage. This has two effects. First, this should test the processing power of both cards fairly since neither card is capped on VRAM. This is reflected in the FPS charts. If AMD is beating NVidia in these charts, then it should be due to more processing power and better scaling, not more VRAM. Additionally, this setup means that if the chart says a certain setting range, the AMD setup may actually still have the performance headroom to push the settings higher while still remaining at acceptable framerates due to more VRAM. The max playable settings are noted in the text.
As for the resolution, [H] tends to look at max playable settings. Game dependent, they consider this as 30-45 FPS. These setups would go well beyond those frame rates on lower resolutions, so [H] does not consider it worth the time to review the lower resolutions for multi-card setups. By running high resolutions, they try to see what kind of max playable settings you can get at acceptable framerates. Bear in mind that Brent has to play through entire levels multiple times, at different settings, on several different games, and across multiple hardware configurations. This takes a long time and they can't cover everything. They do they best they can to give us the most informative reviews possible.
Not trying to start a war over whose reviewing methods are better. Just trying to shed some light on why [H] tests the way they do. Hopefully you guys can appreciate their work and add them to your short list of places to look when making a buying decision.
Well thats nice, but I don't think it will make much difference. I am predicting ATI still wins at these resolutions, its just the way it is.
I am very curious as to how you enforce the max GPU VRAM usage. Do you try settings and just see how close you can get to 1528MB? Because I assure you, it goes up and down by 100s of MB depending on where in the benchmark / game you are at. These spikes are detrimental to any type of "performance" monitoring you may be doing.
Originally Posted by Levesque
24/7 with 2X590? Are you sure?
Not afraid of the VRMs on your 590s a little bit? Those are my 24/7 settings btw. Come on. Be honest. You keep your 590s OC 24/7? You wrote somewhere that it took you 6 HOURS tweaking the OC on your 590. 6 HOURS! So be honest here...
Only thing on my computer that has a 24/7 OC is my CPU. Nothing else needs a 24/7 overclock in my rig. You tell me how long it takes to run 25~30 benchmarks of 3DMark11 and a dozen or so Heaven runs, and dealing with the occasional crash/reboot/unstable overclock. Plus I like to roll with caution
6 hours is nothing, I generally spend a week.
Then do it, be my guest. 22K GPU score 3D Mark 11 with your 24/7 settings. BTW, I don't care about the CPU score, since we are talking about GPUs here. Not the subject of the thread
No. I'm ''beating'' you (like I care for synthetic benchmarks) on GPU scores. Look again. Subject of the thread: GPUs.
And here's what I paid for me GPUs. 6990 = 700$, XFX 6970 325$ (with manufacturer rebate), Powercolor 6970 = 325$ (with manufacturer rabate).
Total: 1350$. How much $ for 2X590? And how much for 580 Quad-SLI?
A better score is a better score
Its all good though, I paid more money for mine, but I am happy with it. I would prefer to run Heaven, as I have no power draw limitation in Heaven. I am beating your overall score with 13% overclock in 3DMark11 as opposed to what I can do in Heaven 30~35%. So its possible for me to catch up to you GPU score wise. I truly hope so, someday.
I was commenting on all the results I have seen with SandyBridge show low Combined score (which is important) which effectively lowers your 3DMark11 score. Looks like it is bottlenecking those cards. Combined score is the most realistic portion of the benchmark, it greatly taxes both the GPUs/CPUs at the same time. Yours looks too low to me.Edited by RagingCain - 4/29/11 at 3:25pm