MP5,1 MDS Mitigation - Real World Impacts & Results

[bsbeamer]

All results below are:
authentic MP5,1 mid-2012
dual 3.46 GHz with 128GB RAM
Sapphire Pulse RX 580 8GB
BootROM 144.0.0.0.0
Mojave 10.14.5 18F132
(SMC 1.39f11)

WITHOUT FULL MITIGATION (HT Enabled)
Geekbench Single-Core Score: 2941
Geekbench Multi-Core Score: 22195
Geekbench OpenCL Score: 131963
Adobe Premiere Pro CC 2019 13.1.2 Build 9 H.264 Pre-Rendered Encode via Export: 4:59.23 (min)

WITH FULL MITIGATION (HT Disabled)
Geekbench Single-Core Score: 2914
Geekbench Multi-Core Score: 22401
Geekbench OpenCL Score: 133345
Adobe Premiere Pro CC 2019 13.1.2 Build 9 H.264 Pre-Rendered Encode via Export: 5:30.71 (min)

-----

Mitigation was applied via the two Terminal commands in Step 3:
https://support.apple.com/en-us/HT210108

Note for Step 1: wait at least 3 minutes holding Command (⌘)-R if you do not have boot screens (non-EFI GPU like RX580)

I'm personally noticing approximately a 10% real world hit with HT disabled right now. It's not enough to slash and burn this machine, but it is enough to feel like I'm being forced to drink burnt coffee. Personally am angry with Intel for not releasing microcode updates to patch this properly, but at the same time angry with Apple for not providing a serviceable MP6,1 replacement or making information on MP7,1 available. Icing on the cake - it's still not 100% clear disabling HT actually provides full mitigation on MP5,1.

Worth noting - several Geekbench scores actually increased with HT disabled. Approximately 1% for both OpenCL and Multi-Core scores. I personally do not rely on these at all, but many here swear by those numbers for everything. Geekbench 4.3.3 (401459) if it matters to anyone.

Will be opening a ticket with Apple to hopefully get some clarity (or correction) on this list:
https://support.apple.com/en-us/HT210107

My machine is TECHNICALLY not part of the list and would default to assume it should be supported. I'm sure this is an oversight on Apple's part, or intended to make this look like it impacts 2010-era machines only. Looks much better on paper since those are all about 10+ years old. My personal mid-2012 is not quite 7 years old right now. Despite the lack of Apple's Pro machine focus these past several years, I would still prefer to upgrade to another Mac Pro in the next 12-18 months if it's a worthy and appealing workstation.

 

[axrst]

All results below are:
authentic MP5,1 mid-2012
dual 3.46 GHz with 128GB RAM

You maybe have done something wrong here: Is there 5.1 mac pro mid-2012? Do you mean mid-2010 ? Or I'm missing something?

 

[bsbeamer]

You maybe have done something wrong here: Is there 5.1 mac pro mid-2012? Do you mean mid-2010 ? Or I'm missing something?

What exactly do you think was "done wrong"?

This is an authentic mid 2012 Mac Pro with upgraded processors to dual 3.46 GHz.

 

[axrst]

Yes.. but not 5.1, it's 6.1 i think. Mine (mid-2010) it's the 5.1.

Please go to system report and see in the hardware overview / model identifier...

 

[bsbeamer]

Yes.. but not 5.1, it's 6.1 i think. Mine (mid-2010) it's the 5.1.

Please go to system report and see in the hardware overview / model identifier...

MP5,1 is mid 2010 and mid 2012. MP6,1 is late 2013. Please get your facts straight.

 

[axrst]

Oh, sorry. I was almost sure that 2009 -> 4.1 , 2010 -> 5.1 and 2012 -> 6.1 I was wrong! I checked it now...

edit:
I apologize for my mistake and the off topic. Maybe a moderator could delete these off topic posts.

 

[bsbeamer]

Oh, sorry. I was almost sure that 2009 -> 4.1 , 2010 -> 5.2 and 2012 -> 6.1 I was wrong! I checked it now...

There is also no Mac Pro 5,2 or Mac Pro 5.2 model. Mac Pro early 2009 models are 4,1.

This is ENTIRELY off topic for this thread.

 

[bsbeamer]

Full screenshots of Geekbench and system information for HT enabled (MDS mitigation not applied):

 

[MisterAndrew]

Disabling SMT (HT) doesn’t seem to result in much of a performance loss, at least on 12 core machines. I agree it would be good to get more clarification on this issue, but here’s what I think.

Disabling SMT is Apple’s method of “full mitigation,” but I think we are interpreting it wrong. Apple lists these machines as not supporting the mitigations due to lack of microcode updates. I think disabling SMT provides full mitigation for machines that are receiving microcode updates, but disabling SMT on these machines doesn’t provide full mitigation because Intel states that disabling SMT alone does not provide protection against MDS attacks.

 

[bsbeamer]

Full screenshots of Geekbench and system information for HT disabled (MDS mitigation applied):










[doublepost=1558461687][/doublepost]

Disabling SMT (HT) doesn’t seem to result in much of a performance loss, at least on 12 core machines. I agree it would be good to get more clarification on this issue, but here’s what I think.

Disabling SMT is Apple’s method of “full mitigation,” but I think we are interpreting it wrong. Apple lists these machines as not supporting the mitigations due to lack of microcode updates. I think disabling SMT provides full mitigation for machines that are receiving microcode updates, but disabling SMT on these machines doesn’t provide full mitigation because Intel states that disabling SMT alone does not provide protection against MDS attacks.

Still have my single 3.33GHz W3680 tray. May try to swap and perform these identical tests later this week or next week to see the difference.

 

[amedias]

All results below are:
WITH FULL MITIGATION (HT Disabled)
Geekbench Multi-Core Score: 22401

typo? your screenshot says multicore result with HT disabled was 21401, which would also be more in keeping with the HT disablement, rather than the increase which your original post suggested. Although GB scores do fluctuate on each test within a small amount so I never take them as gospel anyway.

 

[bsbeamer]

typo? your screenshot says multicore result with HT disabled was 21401, which would also be more in keeping with the HT disablement, rather than the increase which your original post suggested. Although GB scores do fluctuate on each test within a small amount so I never take them as gospel anyway.

There is an absolute chance it's a typo. Here are several additional that I've just run with same setup and HT disabled, all showing different top-line numbers:








All multicore tests with HT disabled appear to be in the 20500-22000 range. As noted above, I'm personally not a fan of Geekbench scores. Too much value is put into them and they are not consistent.

This machine is taking around a 10% real world performance hit with HT disabled for most of my common tasks. I'm sure I could press this further to make that difference even greater, but those are usually tasks run overnight.

Will also note, initial boot times appear to be SLIGHTLY faster with HT disabled and NVME system drive.

 

[bsbeamer]

All results below are:
authentic MP5,1 mid-2012
single 3.33 GHz with 48GB RAM (original single CPU tray)
Sapphire Pulse RX 580 8GB
BootROM 144.0.0.0.0
Mojave 10.14.5 18F132
(SMC 1.39f11)

WITHOUT FULL MITIGATION (HT Enabled)
Geekbench Single-Core Score: 3086
Geekbench Multi-Core Score: 15262
Geekbench OpenCL Score: 136186
Adobe Premiere Pro CC 2019 13.1.2 Build 9 H.264 Pre-Rendered Encode via Export: 6:14.95 (min)

WITH FULL MITIGATION (HT Disabled)
Geekbench Single-Core Score: 3084
Geekbench Multi-Core Score: 13396
Geekbench OpenCL Score: 136146
Adobe Premiere Pro CC 2019 13.1.2 Build 9 H.264 Pre-Rendered Encode via Export: 6:38.08 (min)

Full screenshots of each will be posted shortly.
[doublepost=1558533360][/doublepost]SINGLE CPU - HT ENABLED (no mitigation)

 

[bsbeamer]

SINGLE CPU - HT DISABLED (mitigation applied)

 

[Zwhaler]

Does disabling hyperthreading with CPUSetter have the same protection as the terminal method? I ran Geekbench before and after using CPUSetter w shutdown and restart and my scores in 32-bit mode (free license) went up after disabling hyperthreading (21,541 before vs 21,757 after disabling hyperthreading). Both times Geekbench listed my machine as having 2 processors and 24 threads. Is it supposed to list 24 threads even though I disabled with CPUSetter and restarted before running the second test? My Bios is 141.0.0.0.0 on 10.14.5 and Software Update in Sys Prefs doesn't list any available updates.

 

[bsbeamer]

My Bios is 141.0.0.0.0 on 10.14.5 and Software Update in Sys Prefs doesn't list any available updates

This is your first problem. Need to update to 144.0.0.0.0 BootROM. Download the FULL INSTALLER to install the firmware update. See this thread for more info:
https://forums.macrumors.com/thread...requisite-to-disable-hyper-threading.2132317/

 

[Zwhaler]

This is your first problem. Need to update to 144.0.0.0.0 BootROM. Download the FULL INSTALLER to install the firmware update. See this thread for more info:
https://forums.macrumors.com/thread...requisite-to-disable-hyper-threading.2132317/

Thanks, it's all done now. Does this mean that CPUSetter is not a safe way to shut down hyperthreading, even with BIOS 144? Before and after scores are 21,541 vs. 20,281 (Note: before with hyperthreading was measured with Geekbench v2 and after without hyperthreading was measured with Geekbench v4)

 

[bsbeamer]

Does this mean that CPUSetter is not a safe way to shut down hyperthreading, even with BIOS 144?

Apple's own documentation is not 100% clear that the "fix" implemented by the two Terminal commands entered via Recovery Boot even truly fully addresses the MDS vulnerability on MP5,1. Everything I've read so far seems to suggest this will prevent an MDS attack, however.

I personally would not suggest any third party tool (like CPUSetter) for MDS mitigation at this time. Stick with Apple's methods unless there is provided documentation suggesting something else. If you're already taking the performance hit, best to stick with what they recommend with macOS Mojave at this time. From what I'm reading, mitigation for MP5,1 will depend on up to date software in addition to the HT disable.

Unless your Hardware Overview shows this:



MDS mitigation has not been applied.

-----

Just an update - ticket was opened with Apple earlier this week for MP5,1/mid 2012 not being on "the list" but there has been no followup or clarification provided. Still only cites 2009/2010 machines on the list as of right now.

 

[Lycestra]

There's a third option that lays between CPUSetter and Apple's "full mitigation" nvram method that requires bootrom 144. In Instruments (part of Xcode), preferences, CPUs, theres a way to disable "Hardware Multi-Threading", and seems to behave similar to CPUSetter. Doesn't require a reboot, but also doesn't stick past reboot either. My non-thorough testing seemed to show that System Information doesn't show Hyper-Threading as disabled for either CPUSetter or Instruments. But I contend that Instruments' checkbox is an Apple method, and like CPUSetter, doesn't require a rom upgrade (which requires mojave compatibility, which is a pretty high bar). Is it worth running a test to see if geekbench gets similar results with the Instruments method?

 

[bsbeamer]

Is it worth running a test to see if geekbench gets similar results with the Instruments method?

Geekbench scores or tests with HT disabled via this method may not be very helpful, but if you have the chance go ahead and run before/after. Have a feeling methods of mitigation for HT disable via software post-boot do not actually provide mitigation as intended. What we really need is further documentation about MDS mitigation on macOS Mojave 10.14.5+ with MacPro5,1 and proper methods to do so... or microcodes from Intel to simplify this entire thing.

 

[treekram]

Intel has a document which states what processors have had (or will have) microcode updates, and those processors where they're not planning at this point to do updates. I didn't go down the list of all the 2012 Mac Pro variations but the ones I looked at were all in the "not planned" category.
https://www.intel.com/content/dam/w...A00233-microcode-update-guidance_05132019.pdf

 

[startergo]

Products not planned 3/4:

 

[flaubert]

Well, I finally managed to disable Hyperthreading on my X5690 cMP 5,1. I ran test encodes of a simple half hour segment of a DVD using Handbrake. Surprisingly, despite have half as many computing units in play, the performance loss was not dramatic: the time to encode went from 6 min 15 secs to 6 min 46 secs, an increase of about 8%. I think in my everyday use the performance hit will be imperceptible; I don’t even rip DVD’s that often.

 

[Synchro3]

Products not planned 3/4:
View attachment 838940

The W3690 is not on the list?

 

[bsbeamer]

The W3690 is not on the list?

Also on the list of products NOT having microcode updates.