Intel Sandy Bridge Acceleration On Non-SNB Hardware

[Bonivita]

I did not know about this before.. But Now when I read it, I found it really interesting...
I hope to know more on this topic ...
Thank you so much in advance!

[Michael]

Something else to consider whilst benchmarking is the effect of cpufreq. Some of these tests are GPU bound with virtually zero CPU usage which causes the conservative governors to put the CPU to sleep, which has the side effect of limiting memory bandwidth as well. If you see tests gradually get faster and faster, this is likely the cause and they should max out under the performance governor.

True. I can run another test after Oktoberfest to look at this, since PTS can monitor the real-time CPU frequency via simply setting the MONITOR=cpu.freq environmental variable so can see how the CPU frequency is being toggled on a line graph for each test. I have some SNA tests under Unity, Unity 2D, and GNOME Shell I'll have up this weekend, but this will be my last new tests I run before Oktoberfest. (I also have some other SNB Core i3 and Core i5 system tests and other things in the queue for the next few weeks, but don't recall any of them looking at SNA).

[ickle]

Beware the cpu governor

Something else to consider whilst benchmarking is the effect of cpufreq. Some of these tests are GPU bound with virtually zero CPU usage which causes the conservative governors to put the CPU to sleep, which has the side effect of limiting memory bandwidth as well. If you see tests gradually get faster and faster, this is likely the cause and they should max out under the performance governor.

[darkbasic]

(It would also be rather nice if the Git version could be reported to the DDX version string in a similar manner to Mesa.)

For god's sake, please send an e-mail to Martin Gr??lin before doing such a thing

[Michael]

@Michael, what was the git id for the ddx you tested? Your blurb suggests git, the logs suggest 2.16.0 - I'm just trying to reproduce your results...

Should be as of 695e711. If I recall correctly, the version is still set at 2.16.0 in Git master and hasn't been bumped yet to reflect development. (It would also be rather nice if the Git version could be reported to the DDX version string in a similar manner to Mesa.)

[Michael]

As an aside, it would be very interesting to show just how big an effect running under a compositing manager has on 2D throughput...

I can have some results out in a few days.

[ickle]

@Michael, what was the git id for the ddx you tested? Your blurb suggests git, the logs suggest 2.16.0 - I'm just trying to reproduce your results...

[ickle]

And what is the impact of Unity?

As an aside, it would be very interesting to show just how big an effect running under a compositing manager has on 2D throughput...

[phoronix]

Phoronix: Intel Sandy Bridge Acceleration On Non-SNB Hardware

Last week benchmarks were published of Intel's New Sandy Bridge Acceleration architecture (SNA) that showed several performance improvements for 2D and 3D, but the new acceleration architecture still wasn't mature with a few regressions compared to the normal UXA back-end. While the focus of this SNA support is on speeding up operations for Sandy Bridge (SNB) and forthcoming Ivy Bridge (IVB) hardware, SNA is supported for older Intel graphics processors too. Here are some benchmarks of the Sandy Bridge New Acceleration architecture when using the Ironlake and Gen3 back-ends.

http://www.phoronix.com/vr.php?view=16394