AMD, a company that knows a thing or two about building powerful graphics processors, has whipped out a card that has more flops than either Nvidia's K20 high-end GPUs or Intel's x86-based Xeon Phi coprocessors. That's one sure way to cut through all the noise that Nvidia and Intel are making about their just-released offerings …
$ per DP teraflops per watt
Is that an SI measurement?
My guess is
it will run crysis.
"And yes, it can play Crysis"
at max settings?
"And yes, it can play Crysis."
But not all that well since it's a workstation card and not tuned for gaming.
Specs comparison... (double precision)
K20, 225W, 1.17 TFDP, $3199 estimated -- 5.2 GFDP/Watt, 0.37 GFDP/$ estimated
K20X, 235W, 1.31 TFDP, >$3199 estimated -- 5.57 GFDP/Watt, <0.41 GFDP/$ estimated
5110P, 225W, 1.011 TFDP, $2,649 -- 4.49 GFDP/Watt, 0.38 GFDP/$
3120A, 300W, 1.003 TFDP, ~$2,000 -- 3.34 GFDP/Watt, 0.50 GFDP/$
S9000, 225W, 0.806 TFDP, $2,499 -- 3.58 GFDP/Watt, 0.32 GFDP/$
S10000, 375W, 1.48 TFDP, $3,599 -- 3.95 GFDP/Watt, 0.41 GFDP/$
Unlike the K20s and the Phi boards, these boards do not appear to be targeted at super-heavy DP processing and are more along the lines of the K10 which was released earlier in the year. The S10000 tops the K10 (released earlier in the year) on SP albeit with a significantly higher TDP, and at the same TDP the S9000 has less SP. Both beat the snot out of the K10's comparatively sickly 0.19 teraflops of DP... so these might be the most balanced cards depending on your processing needs.
Now just wait until SSLShader is released and you can build yourself a loadbalancer that scales into hundreds of thousands of connections per second.
And AMD does well in supercomputers also
Maybe they do know what they are doing at AMD?
Re: And AMD does well in supercomputers also
A few things to consider: the Titan is the only machine they have in the top 10, they are there because of their CPUs not their GPUs, the Titan is a Cray machine and it will be one of the last Cray Opterons (for the foreseeable future at least) as Cray is switching to Xeon CPUs (presumably due to PCIE 3.0 support for their next Gen interconnects, which AMD does not have on their Opterons).
IMHO... the short term to midterm does not look great for AMD in the HPC space, although on the long term the rumor (or at least one rumor) seems to be that they will be bundling their CPUs and GPUs on the same die for the server space like they have done for the desktop with Fusion. If true, that would be an interesting gambit - i.e. screw the coprocessor model and just use our all-in-one CPGPs.
Just one AC's $0.02 of course... FWIW
Doesn't matter how "crazy-fast" the card is, the drivers will still be crap.