Comprehensive Threadripper tests - memory vs cpu freq at capped power

Matthew Dillon dillon at backplane.com
Sun Aug 19 17:58:40 PDT 2018


I've created a web page with some fairly comprehensive compiler performance
testing of the 2990WX.  If you've already seen this page, hit reload.  I
added a whole new section on Memory frequency selection and how it effects
the concurrent-compile test when power is capped (and when it isn't).

This particular test is fairly memory-intensive, running 64 concurrent
gcc's over and over again, so these tests give us an idea of just how much
we can ratchet-back the power consumption without impacting performance
(for this test) much.

The test is important, because the threadripper only has 4 memory channels.
  Consumer Intel and AMD parts have 2 memory channels, AMD's threadripper
has 4, Intel's Xeon's have 6, and AMD's EPYC has 8.  Since threadripper
only has 4, memory-intensive workloads essentially cap performance.  It's
interesting just how low in frequency we can go with only a minimal impact
on the compile workload.  Memory speed trades-off against frequency, so if
you don't want to pull 330W at the wall in the stock configuration, then
buying 3000 MHz memory for the threadripper is not necessarily the best
choice.

http://apollo.backplane.com/DFlyMisc/threadripper.txt

We can also get an idea of how much power AMD's infinity fabric eats at
various memory speeds.  When power is severe limited, what power that
doesn't go to the infinity fabric can be used by the BIOS to boost CPU
frequency, so the trade-off is particularly telling and informative when I
cap power to lower wattages.  It shows up in the test results in a very
obvious way.

-Matt
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.dragonflybsd.org/pipermail/users/attachments/20180819/c28c8158/attachment.htm>


More information about the Users mailing list