Not seeing any performance penalty by quick synthetic benchmark. Will let you know later this week when I run another batch of RNASeq analysis. That is some pretty memory I/O heavy workload.
I look forward to your report, resident cat geneticist.
From what I've read on the performance numbers and the Meltdown mitigation code, the penalty shows up when you have lots of context switches, since the mitigation forces a full TLB flush on context-switches rather than just carry on like previously. This is also why AMD isn't affected - they apparently use a tagged TLB design that prevents the attack from working since a process/VM can't peek at another process/VM thanks to having different tags for their TLB entries (an attempt to peek results in an access error). Spectre for the most part seems to be a very small performance hit, but then again, it's mitigations are nowhere near as effective as Meltdown.
With good code sitting inside a single process, there should be almost no performance changes. Looking forward to your results, xkm1948!
Time to dig up some old Pentium II or K6-2
Those are superscalar, out of order, speculative architectures too. You have to go down to the original P5 (Pentium) or AM5x86 (predecessor to K5, itself preceding K6, both of which are speculative) architectures to get a non-speculative x86 chip. For those wondering: Pentium Pro to Nehalem/Westmere are more or less P6 in various states of evolution, with NetBurst and Sandy Bridge (the basis of everything from Sandy Bridge all the way up to Coffee Lake and beyond) being the succeeding ground-up x86 architectures. To not leave ARM out in the cold: ARM7 jumped on the pipelined, superscalar, speculative bandwagon circa 2000, and then the out of order bandwagon with the ARM Cortex-A9 around 2010. And finally we have POWER, which had all of those since the first gen in 1990.
For the most part, since the Pentium Pro showed up, all high-performance chips have been at least pipelined, superscalar and speculative, and very often out of order too, which is why Spectre is gonna be such a pain for us for the next 2-4 years.
At work right now so I can't access most game sites but the one I could ran before and after benches on some games and of the 6 games tested it looks like only Assassin's Creed: Origins is affected much but that's probably because Ubisoft loaded it up with DRM. Not only Denuvo but VMProtect too.
http://www.dsogaming.com/news/windo...triple-a-games-tested-in-cpu-bound-scenarios/
NFS: Payback uses Denuvo as well, but no VMProtect. Despite it's name, VMProtect doesn't use hardware virtualization (
read more here) and shouldn't be affected performance-wise since it runs within the same process (same process => no context-switch => doesn't get affected). I'm gonna suspect something else unless VMProtect has changed since 3.5 years ago (which I don't think has happened)
