Do you know if AMD would have a shitfit if I just happened to run these tests with... umm....
With a fglrx driver I have brought back with me on a flash drive, in my time machine (TARDIS-esque) from about May 2012.
Edit: Also, do you think it will throw off the results terribly much that the rest of my system is somewhat older than the 7970?
Core i7 920 (first gen Nehalem)
16 GB DDR3 1066 in dual channel config (because I'm stupid and bought the wrong thing lol)
ASUS P6T Deluxe v2 motherboard (X58 chipset -- also first gen Nehalem)
SATA 3gb/s HDDs (2TB)
(Psst: due to my motherboard, the 7970 is stuck in PCI-E 2.0 configuration, so it might be bandwidth-limited somewhat, though I read on anandtech or somewhere like that, that the GPU isn't even maxing out PCI-E 2.0 bandwidth for games, only for compute)
Last edited by allquixotic; 03-30-2012 at 09:11 AM.
I ran Unigine Heaven 3.0 benchmark (too many of the others were not running at all / running incorrectly) using... er... "interesting" drivers on KDE 4.8 with KWin acceleration enabled. On Fedora 16 with hand-built 3.3 stable kernel
For the lazy: my HD7970 turns out to get an average of less than 1 FPS per second higher than Michael's HD7950 overclocked to 900/13975 MHz (which is the same clock frequencies used by the HD7970 ... HMMMMMMMMMMMMMMMMMMMMM)
So basically if you're going to buy a HD7900 series, get the HD7950 and save a few bucks. The 7970 doesn't appear worth it because you can just OC the 7950 the 7970's clock level and will get basically the same performance.
For those who think you can apply this logic down the line to lower-end cards: no. The physical hardware available on the cheaper cards (7800 series, 7700 series, etc) is / is going to be heavily cut-down. My observations are that if two cards have the same first two numbers (79xx, 78xx, etc) they are probably going to be identical if you overclock the cheaper one to equal the more expensive one's clock rate. I didn't know / believe this until now but it seems like I get zero advantage from the 7970 over the 7950 with a software tweak.
Then again, my CPU and memory bandwidth is vastly inferior to that on Michael's test system (as you can see from my test results), so that may have something to do with it. Or the differences between fglrx performance on KWin vs unity, or the differences between the fglrx versions we're running (I'm using... an interesting version )
I'm going to be getting a 7850 possibly today and am afraid to get anything with custom cooling or clocks.
My thinking is staying close to the reference design will ensure that things like fan/voltage/clock adjustments work in linux.