After running some Intel Ironlake Mesa 9.1 benchmarks on Tuesday, I ended up running some new UXA vs. SNA 2D benchmarks to see where xf86-video-intel Git is today for the Intel "Gen5" Arrandale/Clarkdale processors. The benchmarking in this article is just looking at this Ironlake configuration and for just the very latest Git tip as of yesterday, but an upcoming article will also incorporate Sandy Bridge and Ivy Bridge results too. Plus it will look at the SNA/UXA performance over the course of several xf86-video-intel driver releases to see how the performance has evolved.
Chris Wilson at Intel's Open-Source Technology Center has done impressive work on SNA since introducing it in mid-2011 when it was already a crazy fast acceleration architecture. Chris has been pretty much the lone Intel developer still advancing the Intel DDX driver to the point of putting out 19 point releases in the xf86-video-intel 2.20 series. Just yesterday he began optimizing some hot-paths within the SNA code-base by hand-writing code for AVX and SSE 4.2 within the GPU driver. He's been making SNA acceleration tweaks on a near daily basis.
For this end of the month benchmarking, embedded below are some of the results for this UXA vs. SNA 2D benchmarking with the Intel Core i3 330M system. While Ironlake graphics aren't the latest and greatest hardware, even with SNA it's a big step forward.
The Intel benchmark results in full can be found in this OpenBenchmarking.org result file.
There's many more benchmarks -- 21 in total -- on OpenBenchmarking.org within 1302262-FO-INTELSNAI70. Of all the UXA vs. SNA benchmarks, there were only two cases where SNA wasn't the clear winner:
The performance of Intel SNA continues to appear fantastic and the margins of SNA's lead over UXA continues to widen, even for older hardware. More concrete data will be offered in the 2D driver testing over several Intel DDX releases plus incorporating the newer (and more interesting) Sandy Bridge and Ivy Bridge hardware. Again, head on over to OpenBenchmarking.org for the rest of the results.