LLVMpipe Scaling With Intel's Core i7 Gulftown

Written by Michael Larabel in Processors on 1 November 2010 at 08:03 AM EDT. Page 4 of 4. 15 Comments.

For a convenient way to see overall how LLVMpipe scales on this six-core Intel Core i7 processor, here's this overview graph below generated by the Phoronix Test Suite that we normally do not include within articles, but it illustrates the per-core scaling for every test.

To see just how much of the Core i7 CPU was being used by LLVMpipe, we used the Phoronix Test Suite to monitor the CPU usage when running OpenArena for four test runs at 1920 x 1080. We ran this test when using two cores, six cores, and then six cores with Hyper Threading. The Phoronix Test Suite was set to poll the average CPU usage across all available cores using the system_monitor PTS module.

To not any real surprise, the CPU usage was the highest with two cores when the average CPU usage wound up being 83.6%. With all six cores the CPU usage was 75.4% and then when Hyper Threading was flipped on it averaged out to 50.4%.

Even with the Core i7 970 "Gulftown" being a nearly $900 USD processor (NewEgg.com and Amazon.com), which is only outdone by the Core i7 980X Extreme Edition for now, the LLVMpipe performance still isn't up to the point you would want to use it for OpenGL gaming but it is certainly the best performance we have seen out of this open-source driver that heavily uses LLVM. With some games at some resolutions the speed was 30+ FPS, which does show that LLVMpipe has potential and that it is a much better alternative to Mesa's classic software rasterizer and Gallium3D's Softpipe driver, which would only run at a small fraction of the speed found with LLVMpipe.

These results also exhibit hope that once LLVMpipe is able to implement the GLX_EXT_texture_from_pixmap extension and meets all the technical requirements for running a compositing window manager, that we can at least find an acceptable level of performance -- without chewing up an entire CPU -- in a composited environment with Compiz, KWin, and Mutter, which will become more important with the GNOME 3.0 Shell and also Canonical's Ubuntu Unity Desktop. This is important for providing a semi-decent out-of-the-box Linux desktop experience for scenarios where Linux graphics drivers are not yet available or other issues barring proper GPU acceleration.

It was also good to see that LLVMpipe is capable of scaling with multiple threads effectively and was an interesting experiment in general. Also, as would be expected, as the resolution increased, so did the efficiency of the per-core scaling. Coming up in the next few weeks will also be GCC vs. LLVM benchmarks on the Gulftown. If there are any other Linux (or BSD / OpenSolaris) tests you are interested in seeing from this 32nm 12-thread Intel processor, let us know.

If you enjoyed this article consider joining Phoronix Premium to view this site ad-free, multi-page articles on a single page, and other benefits. PayPal or Stripe tips are also graciously accepted. Thanks for your support.


Related Articles
About The Author
Michael Larabel

Michael Larabel is the principal author of Phoronix.com and founded the site in 2004 with a focus on enriching the Linux hardware experience. Michael has written more than 20,000 articles covering the state of Linux hardware support, Linux performance, graphics drivers, and other topics. Michael is also the lead developer of the Phoronix Test Suite, Phoromatic, and OpenBenchmarking.org automated benchmarking software. He can be followed via Twitter, LinkedIn, or contacted via MichaelLarabel.com.