The Performance Impact Of Spectre Mitigation On POWER9

[Maddo]

Yeah, i guess my English got in the way, what i meant is right now there is no hardware acceleration for cross architecture emulation/virtualization hence the IPC penalty will be massive enough to make any game test worthless.

Also i'm not sure if it is even possible to emulate x86_64 on PPC with hardware acceleration since IBM don't have an x86 license deal, hence it can open the doors for Intel to sue the living bejesus out of them(i read this several times on forums searching around for x86 emulation on PPC but i'm not 100% this is actually legit but sounds about right as your usual technology being held back due to legal crap)

Not having a license doesn't mean you can't emulate a different architecture.

[jrch2k8]

Not having a license doesn't mean you can't emulate a different architecture.

as stated "i read this several times on forums searching around for x86 emulation on PPC but i'm not 100% this is actually legit but sounds about right as your usual technology being held back due to legal crap" is not an assertion but a common response to why IBM don't add hardware acceleration for x86 in their CPU, also we are not talking about emulation since qemu is capable of so on software but hardware acceleration since software emulation have a massive performance hit which make it useless for games(this is the context of the response)

[Weasel]

Yikes... and I thought Intel got hit hard...

That's what happens when you listen to clueless people on this forum instead of those who actually know what they're talking about.

[schmidtbag]

That's what happens when you listen to clueless people on this forum instead of those who actually know what they're talking about.

Meaning what? Benchmarks showed that Intel had some pretty hefty performance hits from all of the mitigations. It's not a matter of opinion.

[s_j_newbury]

That's what happens when you listen to clueless people on this forum instead of those who actually know what they're talking about.

Intel would be hit even harder than they are already if they enabled full userspace-userspace Spectre protection in-kernel. I'm not sure why this is the default on POWER9, it doesn't make them look as competitive compared to Intel/AMD and presumably POWER9 users know whether they need it or not. Just last month there was an article here reporting on Intels efforts to avoid such a feature and instead provide a "Lite" mitigation approach: https://www.phoronix.com/scan.php?pa...-App-To-App-V3

In that article there's a quote from one of the developers on the penalties of full protection:

Tim Chen summed it up in the patches, "leaving STIBP on all the time is expensive for certain applications that have frequent indirect branches. One such application is perlbench in the SpecInt Rate 2006 test suite which shows a 21% reduction in throughput. Other application like bzip2 in the same test suite with minimal indirct branches have only a 0.7% reduction in throughput. IBPB will also impose overhead during context switches...Application to application exploit is in general difficult due to address space layout randomization in applications and the need to know an application's address space layout ahead of time. Users may not wish to incur performance overhead from IBPB and STIBP for general non security sensitive processes and use these mitigations only for security sensitive processes. This patchset provides a process property based lite protection mode that applies IBPB and STIBP mitigation only to security sensitive non-dumpable processes."

[Weasel]

Meaning what? Benchmarks showed that Intel had some pretty hefty performance hits from all of the mitigations. It's not a matter of opinion.

I mean people who think that all the speculative execution exploits are Intel-only and keep on babbling about it, just like the Nvidia haters.

[schmidtbag]

I mean people who think that all the speculative execution exploits are Intel-only and keep on babbling about it, just like the Nvidia haters.

Sure, it isn't Intel-only, but Intel has a hell of a lot more of them than AMD and ARM (not sure about MIPS, SPARC, or PPC). Each additional exploit has an impact on performance, when mitigated.