[HN Gopher] Intel Raptor Lake 0x129 CPU Microcode Performance Im...
___________________________________________________________________
Intel Raptor Lake 0x129 CPU Microcode Performance Impact on Linux
Author : dangle1
Score : 24 points
Date : 2024-08-11 21:00 UTC (2 hours ago)
(HTM) web link (www.phoronix.com)
(TXT) w3m dump (www.phoronix.com)
| debo_ wrote:
| > On a geo mean basis for 188 benchmarks run, the performance
| overall was flat from this new BIOS / CPU microcode... For the
| most part the performance was the same but there were some
| exceptions observed...
| 0x000xca0xfe wrote:
| Data movement/branch intensive code seems to be affected most
| (WireGuard -10%, some Python benchmarks -8%).
| BearOso wrote:
| The branching and the scripting language hit was unexpected.
| It slows single-threaded programs the most. That means this
| definitely wasn't caused by an out-of-spec high power limit
| set on the motherboards. So the common theory was wrong.
|
| Games seem to be unaffected. There was suspicion regarding
| the timing of this patch and AMD's new processors coming out.
| That now seems to be unfounded, with only a minimal effect on
| consumer applications.
| Dalewyn wrote:
| >It slows single-threaded programs the most. That means
| this definitely wasn't caused by an out-of-spec high power
| limit set on the motherboards. So the common theory was
| wrong.
|
| No, low thread workloads draw more volts than high thread
| workloads. This is easily observable using anything that
| measures CPU voltages. Single threaded use cases seeing
| more impact, if any, is completely within expectations.
|
| That said, Intel said performance drops should be
| negligible if any.
|
| I haven't updated my system yet since ASUS's releases are
| currently marked as beta, but I don't expect there to be
| notable performance drops either. Granted, my 14700K in my
| daily driver has been working perfectly fine since I bought
| it and so have all the other Raptor Lake machines I care
| for.
| BearOso wrote:
| Earlier reports believed the motherboard was setting the
| total power limit too high, like 300W. That would have
| affected multithreaded workloads.
|
| The voltage spiking during mispredicted vdroop was
| actually the cause.
| PaulKeeble wrote:
| I have seen a bigger hit on my 13700k on an MSI Z790 board. About
| 2% on single threaded workloads due to the boost now not going
| past 5.3Ghz but 20% on multithreaded like Cinebench. The
| performance hit is really quite significant.
|
| Its made even worse by the fact the pre microcode benchmark was
| on a hand application of Intels Default power recommendations to
| reduce damage, so 125W and limitations on amps and boost times.
| Its now running on performance mode which can use 250W. I don't
| understand how some people see no difference, others seeing a
| little and I am seeing a lot. Something is going on here.
___________________________________________________________________
(page generated 2024-08-11 23:01 UTC)