Test Setup and SiSoft Sandra 2013 Results
The tests were done on the Gigabyte GA-F2A85X-UP. The BIOS was updated to F4 which is dated 2013-03-13. This BIOS also had the interesting ability to unlock the heretofore unknown DDR-3 2133 multiplier in older Trinity parts. I tested the A10 5800K with both 1866 and 2133 memory settings.
AMD does not officially support more than 1 DIMM per channel when running at 2133 speeds. Though AMD provided a 4 x 4GB 2133 kit, I only used two of the DIMMs. The AMP profile was used with the 6800K, 6700, and the 5800K at 1866 speeds. The Intel XMP profile was used with the 5800K at 2133 speeds. Interestingly enough, the AMP profile did not set the DIMM voltage to 1.6 when running at 2133 speeds. These had to be set manually to insure stability with these particular DIMMS at that speed. The A10 6700 also supports a maximum memory speed of 1866 rather than the full 2133 of the 6800K part.
Ryan was able to procure a Core i5 4670K for a handful of tests to compare to the Richland APUs. Obviously AMD will be at a disadvantage in overall CPU performance as compared to this latest Haswell part, but the graphics portion should still be superior. Due to some logistical problems we were unable to get results on every test with the 4670K. This model does not support HyperThreading, so it is a quad core part that supports up to four threads. This particular processor retails around $249.
Windows 7 64 Bit Ultimate was used with the necessary Bulldozer threading updates/fixes. For these tests I did not utilize a standalone GPU. Resolution was set at 1280×1024 on the desktop, and all games were run at 1280×960.
Gigabyte GA-F2A85X-UP4 Motherboard
AMD RG2133 2 x 4GB DDR-3 2133 DIMMS
WD Caviar Black 7200 RPM 1TB Drive
Corsair TX750W Power Supply
Windows 7 64 Bit Ultimate
SiSoft Sandra 2013 SP3
It seems like this benchmark is constantly updated to reflect the very latest changes in processor design. Nearly every aspect of computer performance is contained from standard integer and floating point performance to OpenCL and GPGPU applications. For this test I use a limited selection of tests to get a good feel about where these processors fit.
Yes, the 4670 is significantly faster in the CPU tests, but I find the OpenCL results to be just as interesting. Intel has yet to put its full might behind improving its OpenCL performance while utilizing the much improved graphics unit of the Haswell series. Interestingly enough, the double precision performance is still slightly above that of the competition from AMD. The AVX2 unit is a monster on the Intel side, and AMD cannot come anywhere close to keeping up with the Piledriver design. The dual 128 bit floating point/SSE/AVX unit is shared between “cores”, so each module houses one of these units. In theory, even though the AMD part can handle four threads, it essentially has only two floating point units available.
When comparing AMD parts, both the 6700 and 6800K are quite a bit faster than the older top-end Trinity model, the 5800K (at both 1866 and 2133 speeds). Again I am quite impressed that AMD is able to match (and exceed) the performance of the 5800K with the 6700. Remember, the 6700 is a 65 watt TDP part while the 5800K is 100 watts.