The Combined Impact Of Mitigations On Cascade Lake Following Recent JCC Erratum + TAA
Blender, of course, misses out big when not having Hyper Threading active.
For cases like Python and PHP, the only performance hit is from the new CPU microcode.
If looking at the geometric mean across dozens of benchmarks carried out, the new CPU microcode update lowered the performance by 1~2%. With the new CPU microcode, the cost of the default mitigations on Cascade Lake with this dual Xeon Platinum server was just ~1%. Having TSX enabled and thus TSX Async Abort in place meant losing a few percent on performance while TAA combined with Hyper Threading is where the performance was about 15% lower compared to the best results. So at least out-of-the-box there is just a small (but still noticeable, particularly from the new CPU microcode -- though again in the future ideally to be addressed via toolchain updates to offset most of those hits) performance hit from these mitigations except when disabling Hyper Threading or having TSX support enabled where TAA then introduces a new and real performance penalty.
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.