'Kernel memory leaking' Intel processor -- a serious cpu bug!?!

[Distorted Horizon]

http://www.game-debate.com/news/24334/m ... it-hardest

Lots of noise for nothing. I just noticed that I already have that update and I haven't noticed anything in the performance.

[jdnz]

the test machine is too old to receive an updated BIOS with the new intel microcode so there was NO spectre mitigation active (just the meltdown fix), I'll hunt out a newer machine with win7 on and re-run the tests with both the spectre and meltdown fixes active next

found an old HP eliteone 800-g1 which had an i5-4570s cpu and for which HP has already release an updated bios with the new spectre mitigation microcode, so tested that.

https://www.3dmark.com/compare/pcm10b/1 ... 10b/179170#

it's a VERY slow machine (the combination of the slow 4570s cpu and a sluggish 5400rpm hdd see to that), but it was measurable that disk thruput took a hammering, both the application load tests and a lot of the file i/o tests show a real and perceptible hit

[donkey tugger]

the test machine is too old to receive an updated BIOS with the new intel microcode so there was NO spectre mitigation active (just the meltdown fix), I'll hunt out a newer machine with win7 on and re-run the tests with both the spectre and meltdown fixes active next

found an old HP eliteone 800-g1 which had an i5-4570s cpu and for which HP has already release an updated bios with the new spectre mitigation microcode, so tested that.

https://www.3dmark.com/compare/pcm10b/1 ... 10b/179170#

it's a VERY slow machine (the combination of the slow 4570s cpu and a sluggish 5400rpm hdd see to that), but it was measurable that disk thruput took a hammering, both the application load tests and a lot of the file i/o tests show a real and perceptible hit

Thanks for posting this. I've got a n hp 800 g1 (small form factor, i7 4770) so am interested.

I was looking at the results and it states the post update turbo frequency is only 798 MHz, as opposed to 2,894 MHz. Is it possible the update has switched off the turbo setting in the BIOS?

Also, on the HP downloads site, the release notes make no mention of any meltdown or spectre fix;

"Version: 2.30 Rev.A

Fixed in this release: - Updated graphics Firmware
- Updated PXE ROM to fix some network compatibility issues
- Added fixes for Remote Manageability
- Added Intel Reference code updates for chipset, memory and CPU tosupport introduction of Haswell Refresh Processors

So are you sure it is in fact related to the current issue?

[jdnz]

I was looking at the results and it states the post update turbo frequency is only 798 MHz, as opposed to 2,894 MHz. Is it possible the update has switched off the turbo setting in the BIOS?

don't think so, just mis-reporting

I actually went and disable both the spectre and meltdown mitigations via registry to see if the speed went back to original levels (since the new kernel could have different issues even without the overhead of the fixes), sure enough speed came back up to close to original (though not exactly - so looks like there is possibly some speed loss in the new kernel even without the protection active)

https://www.3dmark.com/compare/pcm10b/1 ... 10b/179153

however does mean that you can always turn on the protections for when you need to access the web, and then turn them off when you want max speed for your DAW (does of course require a reboot)

Also, on the HP downloads site, the release notes make no mention of any meltdown or spectre fix;

Version: 2.30 Rev.A

that's off the 'revision history' page - however the actual BIOS update is 2.72 - if you look back on the main product support page it says

"Fix and enhancements:

Due to security changes, after loading this BIOS, older versions cannot be installed.
- Adds new BIOS Setup policy to allow Power On Password prompt when Waking On LAN.

NOTE: By default this new setting will be enable, so it will prompt for Power On Password on a Wake on Lan event.

- Updates to latest Intel microcode 0x23.

- Updates to latest UEFI PXE version 0018"

without it the powershell tool for checking shows NO spectre support, after applying it the tool shows BIOS/microcode support

[jdnz]

however does mean that you can always turn on the protections for when you need to access the web, and then turn them off when you want max speed for your DAW (does of course require a reboot)

also of interest that the powershell too was showing no PCID use, despite the 4570 supporting it (same as the other 4xxx series).

A bit of googling showed other people reporting that on PCID capable hardware if they booth win8 (or win10) with the patches PCID is used, but it looks like the patched win7 kernel just will not use it (which is a shame as it’d offset some of the slowdown - looks like microsoft really is out to kill win7)

going to put win10 on the eliteone on monday and re-run the tests to see if the slowdown is really that much less on win10 than win7 or it’s just the hardware

[Kaine]

Now the patches and updates are rolling out, I've thrown one on the bench for a before and after performance comparison.

http://www.scanproaudio.info/2018/01/12 ... kstations/

[Daags]

Now the patches and updates are rolling out, I've thrown one on the bench for a before and after performance comparison.

http://www.scanproaudio.info/2018/01/12 ... kstations/

I think what would be very useful is to see Win 7 benchmarks on newer silicon ("Skylake, Kabylake or newer CPU"), before and after patch ...and also VS the same hardware running Win 10 before and after patch.

[Kaine]

I don't disagree, but I also don't have benches set-up and a real lack of time to do so at the moment. Also, it's pretty hard to justify testing anything with Windows 7 to my boss now that it's at the end of its lifecycle and we haven't sold it for over a year.

However, the patch can be installed and uninstalled. We could get a thread running where people add in their own benchmarks, should anything untoward happen with the results, just roll it back again. All the testing software is available either open source or with time-limited trail, I'd certainly be keen to see some crowd sourced results to help flesh out the data.

[jdnz]

However, the patch can be installed and uninstalled. We could get a thread running where people add in their own benchmarks, should anything untoward happen with the results, just roll it back again.

you don’t even need to uninstall it - just set the appropriate registry keys and windows reverts to it’s old behaviour - see the registry keys section of
https://support.microsoft.com/en-us/hel ... ilities-in

note also that you can enable/disable specte and meltdown protections individually via the key - I’ll be running tests monday to see which one has the biggest hit on speed (as well as then retesting the same hardware running win10).

[Zombie Queen]

you don’t even need to uninstall it - just set the appropriate registry keys and windows reverts to it’s old behaviour

Interesting, does it really affect performance drop?

[Zero dB]

you don’t even need to uninstall it - just set the appropriate registry keys and windows reverts to it’s old behaviour

Interesting, does it really affect performance drop?

Not for me. I installed the patches (Windows and firmware/UEFI) on my Surface Pro 4 M3 and the thing has slowed down like hell, just doing very basic stuff - opening/closing browser tabs and opening native native apps. It's unbelievable. A lagfest. This is Skylake/Win 10 - apparently shouldn't be particularly affected, so MS say. I'm calling bullsh!t! I simply can't imagine how bad this would be on my (currently blazing fast) Haswell/Win 7 desktop DAW system - holding off for now.

I tried those registry keys on my Surface to undo the patches but I noticed no improvement at all. Maybe they've undone the security patching without a performance improvement?! Great! My SP4 currently performs worse than my 2010 HDD laptop - madness.

I'm currently in shock about quite how bad this thing is. I can't believe everyone isn't in more of a panic about this stuff! Feel like we're being taken back years here!

I've found no way to roll back the firmware of my Surface, sadly.

ZdB

[jdnz]

you don’t even need to uninstall it - just set the appropriate registry keys and windows reverts to it’s old behaviour

Interesting, does it really affect performance drop?

I ran tests yesterday, and yes with the patches and BIOS fix applied but the registry set to disable both the spectre and meltdown fixes, benchmarked speed is back to what it was prior to the patch (well within the typical +/-2% you get between pcmark10 runs normally anyway)

[jdnz]

I tried those registry keys on my Surface to undo the patches but I noticed no improvement at all. Maybe they've undone the security patching without a performance improvement?! Great! My SP4 currently performs worse than my 2010 HDD laptop - madness.

grab the powershell module and check whether it IS reporting the fixes as disabled

https://support.microsoft.com/en-nz/hel ... powershell

not sure if you’ll need to install WMF5.1 on win10 or not (had to on win7 as the version of powershell was too old)

the keys are meant to work on all versions of windows (I tested on win7) - you did check for typos and also that you had the setting right way around?

[Zombie Queen]

Thanks, it's something to ponder... Changing registry values in w10 can be not quite straightforward, so maybe it has not been applied properly. Me, I'm on W7/XP, at the moment I'm more afraid of Microsoft than hackers, so I'll just skip updating Windows core at least for some time and hell, I'm not gonna touch firmware. Worst case scenario, I might just dust off the ancient Athlon board and set it as net browser. It becomes guerrilla times for computing.

[Zero dB]

I tried those registry keys on my Surface to undo the patches but I noticed no improvement at all. Maybe they've undone the security patching without a performance improvement?! Great! My SP4 currently performs worse than my 2010 HDD laptop - madness.

grab the powershell module and check whether it IS reporting the fixes as disabled

https://support.microsoft.com/en-nz/hel ... powershell

not sure if you’ll need to install WMF5.1 on win10 or not (had to on win7 as the version of powershell was too old)

the keys are meant to work on all versions of windows (I tested on win7) - you did check for typos and also that you had the setting right way around?

Thanks jdnz. I had a go with this but it's a little beyond my knowledge - I can't make out what it all actually means! Here's a screengrab: https://imgur.com/a/hrrhY

Any idea what this means?

Thanks again.