On Fri, 22 Jul 2016, Henrique de Moraes Holschuh wrote: > > On Mon, 28 Mar 2016, Stuart Bennett wrote: > > > On 28/03/16 19:00, Henrique de Moraes Holschuh wrote: > > > >That said, what microcode comes with this new BIOS? Is it 0x36 already, > > > >or something earlier? > > > > > > It is indeed 0x36. > > The intel-microcode upstream releases for 2016 have microcode 0x38 for > the Xeon E5-v3 (signature 0x306f2). The intel-microcode package > recently uploaded to jessie-backports (3.20160607.2~bpo8+1) has it. > > Maybe you could test this new microcode and report back whether it fixed > the regression or not?
ping? Extra data: very similar throttling was reported to happen on Haswell-E i7-5820K. That processor is somewhat similar to the Xeon E5-v3 26xx, and uses the same microcode. Report: https://communities.intel.com/thread/105416 The i7-5820K processor reports it is hitting a current limit (incorrectly), and throttles down to 900Mhz in 90% of the boots. This can be worked around by setting a very very high current limit. This would explain why only some motherboards/configurations show the issue, as different motherboards will set those limits to different values, and even for the same motherboard, the limits set it can change depending on DRAM configuration. This could indeed realistically be fixed in microcode 0x38. Please test the intel-microcode packages in unstable/testing or stable-backports when possible, and report back... -- Henrique Holschuh
