Intel Raptor Lake 0x129 CPU Microcode Performance Impact On Linux

[phoronix]

Phoronix: Intel Raptor Lake 0x129 CPU Microcode Performance Impact On Linux

Motherboard vendors have begun releasing updated BIOS versions for Intel Core 13th/14th Gen motherboards that offer the new "0x129" CPU microcode that is intended to address the Raptor Lake stability issues that have been causing instability problems and crashing errors for a growing number of Intel Core 13th/14th Gen processors. Intel reported in their (Windows) testing that the 0x129 CPU microcode should offer negligible performance impact but I was curious to run my benchmarks under Linux of this new CPU microcode.

https://www.phoronix.com/review/intel-raptor-lake-0x129

[Linuxhippy]

Initially Intel claimed no negative impact, now at least the tested CPU loses a single percent.
Not nice, but also not that bad considering the costs of all those side-channel attack mitigations introduced in the past.

[Adarion]

One might wonder what it actually does. And how they manage to have this semi-moderate imact. Higher frequencies usually require more voltage to have the transistors still switching correctly, but iirc. the problem was caused partially by voltage issues.

And how long the benchmarks were running. Maybe they go earlier into throttling to prevent issues once the thing gets really warm?

[espi]

One might wonder what it actually does. And how they manage to have this semi-moderate imact. Higher frequencies usually require more voltage to have the transistors still switching correctly, but iirc. the problem was caused partially by voltage issues.

And how long the benchmarks were running. Maybe they go earlier into throttling to prevent issues once the thing gets really warm?

Apparently the new microcode just clips the voltage requested to 1.55v. Apparently Intel was trying to predict vdroop and requesting insanely high voltages, and if there was no vdroop (load wasn't heavy enough or the motherboard was doing something weird) then it actually got that insane voltage.

This probably means that in some situations where the CPU goes from idle to very heavy loading, the voltage drops below the safe for a few cycles and the CPU has to clock stretch to avoid crashing.

[aerospace]

By definiton, if this microcode applies to a cpu, that very cpu has been subjected to permanent, difficult to state damage, and should be RMA'ed. It freezes the degradation to the current level. So basically the goal of to this microcode update is to get back as few damaged cpus as possible, leaving million of cpus in the "damaged, but working" state in the hands of the customers.

EDIT: Has any reviewer pointed this out?

[ms178]

I am even thinking of downgrading the microcode or disabling late microcode loading on Linux alltogether due to Intel's unvervolting restrictions with newer Raptor Lake microcode. As it seems, my BIOS loads older microcode on purpose that allows more undervolting which is also stable on Windows but Linux overwrites that microcode with a newer one at startup and crashes soon after when using more than a mere -30 mV. Thanks, Intel! What a silly problem to have.

[WannaBeOCer]

The % of performance lost is larger than 1%. If you go back to comparing it to the original bios that didn’t enforce “Intel defaults.” For example my 12700K with an Asus board using stock(MCE) would all-core boost to 4.7Ghz now with the new stock bios which uses microcode 35 on 12th gen force Intel Defaults which cause all-core to go up to 4.4Ghz. Previously 2 cores would hit 5Ghz now only 1 core hits 5Ghz. Behavior seems the same as before if I re-enable MCE manually.

[Hibbelharry]

I don't really know but my thoughts on the matter are like this:

I'm somewhat skeptical the new microcode really fixes anything. I watched some tests on youtube and there are still core voltages seen above 1,45v which is still pretty high. I also saw quite some discussions where people suggested the cpu ringbus degrading due to excessive voltage is the main culprit. If thats right I guess really nothing will change and the new microcode is a placebo to keep lawyers away until next gen is launched...

[bearoso]

I am even thinking of downgrading the microcode or disabling late microcode loading on Linux alltogether due to Intel's unvervolting restrictions with newer Raptor Lake microcode. As it seems, my BIOS loads older microcode on purpose that allows more undervolting which is also stable on Windows but Linux overwrites that microcode with a newer one at startup and crashes soon after when using more than a mere -30 mV. Thanks, Intel! What a silly problem to have.

I would make sure this patch is there, even if it does limit undervolting.

If your BIOS in fact does something special with the microcode and helps, it's pretty easy to disable the late microcode loading. I think you'll find it's the same in the end, unfortunately.

*edit*: I believe Windows also does microcode loading, but secretly. It's usually part of their updates.

[WannaBeOCer]

I don't really know but my thoughts on the matter are like this:

I'm somewhat skeptical the new microcode really fixes anything. I watched some tests on youtube and there are still core voltages seen above 1,45v which is still pretty high. I also saw quite some discussions where people suggested the cpu ringbus degrading due to excessive voltage is the main culprit. If thats right I guess really nothing will change and the new microcode is a placebo to keep lawyers away until next gen is launched...

1.45v is low for Intel CPUs, I've run my 6700k Skylake, 9900k Coffee Lake and 12700k Alder Lake at 1.5v for years without an issue under water cooling. It's always been recommended to stay under 1.5v for 24/7 use. From reviewers they clearly show CPUs aren't hitting anywhere near 1.6v anymore which is great.