|AMD FX-8150 Bulldozer Processor|
|Reviews - Featured Reviews: Processors|
|Written by David Ramsey|
|Tuesday, 11 October 2011|
Page 15 of 18
AMD FX New Instructions
The basic "x86 instruction set" has been with us since the eight-bit Intel 8080 processor was introduced in 1974. Since then, both Intel and AMD have introduced extensions to the x86 instructions like MMX, SSE, 3D Now, and so forth. In the Bulldozer, AMD has added these instructions:
The problem with new instructions is that existing programs won't benefit, since the instructions didn't exist at the time the programs were built. New instructions require support from compiler vendors, who must upgrade their products to generate the new code when appropriate (or directed by the programmer). Current benchmarks and applications will not benefit from these new instructions, but new ones will.
To demonstrate the performance improvements possible with these new instructions, AMD provided recompiled versions of x264HD using the new AVX and XOP instruction features. Both the 2500K and the FX-8150 support AVX instructions, but the 1100T does not.
For the first two runs, Intel wins, although if you compare the FX-8150's Run 1 scores against its scores in the standard x264HD benchmark, you'll see a 61% improvement, from 75fps to 120.61fps. Intel improves too, from 93.42 to 150.7fps, which is also a 61% improvement. In Run 2, the performance deltas are similar.
AMD pulls ahead in Runs 3 and 4, but the improvements in frame rates over the standard x264 benchmark are minimal. Now, let's look at the results from x264HD coded with the XOP instructions.
Since the XOP instructions are currently unique to the Bulldozer, the Sandy Bridge processor and the 1100T cannot run this benchmark at all. However, the scores returned for the FX-8150 are virtually identical to the ones returned by the AVX version of the benchmark in the first two runs.
The pattern repeats itself in Runs 3 and 4. The scores for the FX-8150 are about the same as they were for the AVX version of the benchmark.
AMD says that current Microsoft compilers incorporate the switches to enable code generation for these instructions. Based on this single benchmark, they can provide significantly improved performance that dramatically narrows Intel's performance lead.