Meltdown and Specter patch performance hit tested on Optane and VROC.

Discussion in 'SSD and HDD storage' started by nosirrahx, Jun 14, 2018.

  1. nosirrahx

    nosirrahx Master Guru

    Messages:
    450
    Likes Received:
    139
    GPU:
    HD7700
    The Meltdown and Specter patches were rumored to hit Optane pretty hard and since I am using a 905P in my new workstation as an OS/app drive I decided to do some testing. I tested 4 way VROC 0 as well. The top pic has the OS set to 1709 (with no updates) with the last BIOS I could find before the microcode updates started being mentioned in the notes. The bottom pic has the OS set to 1803 + kb4284835 + kb4100347 and the current BIOS for this board. In both pics the 905P is on the left and the 4 way VROC 0 900P is on the right.

    [​IMG]

    [​IMG]
     
    fantaskarsef likes this.
  2. Fox2232

    Fox2232 Guest

    Messages:
    11,808
    Likes Received:
    3,371
    GPU:
    6900XT+AW@240Hz
    That's painful to watch.
     
    Venix, Angantyr and Dragondale13 like this.
  3. RealNC

    RealNC Ancient Guru

    Messages:
    4,893
    Likes Received:
    3,168
    GPU:
    RTX 4070 Ti Super
    Can you test with everything fully updated, but with the mitigations disabled? You can use the InSpectre tool to easily disable and re-enable Meltdown and Spectre mitigations:

    https://www.grc.com/inspectre.htm

    It would be interesting to see if just disabling the mitigations on a fully updated system restores I/O perf to pre-update levels.
     
  4. Dragondale13

    Dragondale13 Ancient Guru

    Messages:
    1,525
    Likes Received:
    243
    GPU:
    GTX 1070 AMP! • H75
    It won't, but you'll see for yourself.
     

  5. nosirrahx

    nosirrahx Master Guru

    Messages:
    450
    Likes Received:
    139
    GPU:
    HD7700
    I could take a look at that.
     
  6. 321Boom

    321Boom Member Guru

    Messages:
    122
    Likes Received:
    12
    GPU:
    GTX980 Ti
    I'm interested in this too if you're willing to post results with the patches enabled and disabled.
     
  7. Agent-A01

    Agent-A01 Ancient Guru

    Messages:
    11,628
    Likes Received:
    1,119
    GPU:
    4090 FE H20
    The answer is yes, it does restore performance by disabling spectre protection.
     
    Smough likes this.
  8. 321Boom

    321Boom Member Guru

    Messages:
    122
    Likes Received:
    12
    GPU:
    GTX980 Ti
    Thanks for pitching in Agent-A01. Glad to know that's the answer since I've already been disabling these patches before a gaming and recording session, wouldn't want to find out I've been doing that for nothing. Appreciate your insight :)
     
  9. nosirrahx

    nosirrahx Master Guru

    Messages:
    450
    Likes Received:
    139
    GPU:
    HD7700
    I have some free time today, I will take a look.
     
  10. nosirrahx

    nosirrahx Master Guru

    Messages:
    450
    Likes Received:
    139
    GPU:
    HD7700
    Well.......that was interesting. Right up front I will need to stipulate that I ran the VROC bench 4 times and the 905P test 6 times just to convince myself that the results were consistent.

    It seems that the OS and BIOS improvements (once the mitigations are disabled) really did a lot to improve VROC 0 4KQ1T1 speed. On the down side this also means that what looks like a big drop in performance when going from 1709 + old BIOS to 1803 + new BIOS is actually far worse once disabling mitigation reveals the actual potential performance. VROC 0 compared to RAID 0 brings 4KQ1T1 to levels I would have assumed impossible. Of course you cant have this performance without turning off potentially critical security measures so I am not sure this actually good news.

    I have no idea what was going on with the 4KQ2T2 numbers on the 905P. They were very consistent and lead me to believe that perhaps my original testing was bad.

    All in all disabling mitigation clearly has positive results, far better than I expected actually. My expectation was that the numbers would end up between the unpatched and patched numbers.

    [​IMG]
     
    fantaskarsef likes this.

  11. nosirrahx

    nosirrahx Master Guru

    Messages:
    450
    Likes Received:
    139
    GPU:
    HD7700
    After the next BIOS update and cumulative update I will being making my final build of my OS. I will be ripping a bunch of stuff out with the MSMG toolkit. It will be interesting to see if any removals (like defender and system restore) have any impact on performance. If anything interesting turns up I will post it here.
     
    fantaskarsef likes this.
  12. nosirrahx

    nosirrahx Master Guru

    Messages:
    450
    Likes Received:
    139
    GPU:
    HD7700
    I did some testing on my NUC8i7HVK with 2 800P in RAID 0. Updates and BIOS are current, these are the results with protection enabled and disabled (pretty sure you can tell which one is which):

    [​IMG]
     
    321Boom, RealNC and DLG like this.
  13. 321Boom

    321Boom Member Guru

    Messages:
    122
    Likes Received:
    12
    GPU:
    GTX980 Ti
    @nosirrahx Thanks for taking the time to run these tests. Really appreciate it, very useful information.
     
    nosirrahx likes this.
  14. olenbr

    olenbr Guest

    Messages:
    4
    Likes Received:
    0
    GPU:
    GT 730
    I know this is an old thread but i need some help. planning on upgrading but only if I can avoid performance penalty on both CPU/storage side. I do not plan to update windows so it'll stay on 8.1, never updated once since initial install.

    where does the performance penalty comes from the list below?:

    - built in silicon in the newer batch of CPUs from intel? 9900K, KF, upcoming KS
    - windows OS update within the kernel
    - bios microcode

    I will need to update the bios in order to take more ram which includes the microcode so is this hit unavoidable?
     
  15. user1

    user1 Ancient Guru

    Messages:
    2,746
    Likes Received:
    1,279
    GPU:
    Mi25/IGP
    most of the mitigations require both microcode + os updates, if you lack one of them, it will not enable any performance reducing mitigations( to my knowledge), I sure hope your pc is offline though, bare minimum i would have the meltdown patch(the really scary one) it has virtually no performance impact.

    also as far as fixed silicon goes, i believe meltdown+l1tf(9900k and newer) and MDS are patched on the latest chips(might be patched on kf , will likely be on newer chips), not sure about some of the other exploits. ill see if i can find the intel document on the subject.
     
    Last edited: Sep 17, 2019
    olenbr likes this.

  16. olenbr

    olenbr Guest

    Messages:
    4
    Likes Received:
    0
    GPU:
    GT 730
    hey thanks a lot for the details breakdown man, really helps.

    I do plan to have HT turned off because cooling issue which solves one security issue but I forgot which one, is that the meltdown one?

    also for the newer CPU, so it is 100% no avoiding it. basically going for 8 cores over my current 6 cores will have some small benefits on muti-thread scenario but for single-thread I am to lose performance is that how it works? hope you can help me answer these too.
     
  17. user1

    user1 Ancient Guru

    Messages:
    2,746
    Likes Received:
    1,279
    GPU:
    Mi25/IGP
    meltdown is a different bug , it functionally allows an unprivileged program to access all system memory, the bugs remedied by disabling smt are l1tf and MDS class exploits (and spectre to a more limited degree).

    as for performance , the compounded impact of the various exploits does reduce single threaded performance, though not by much, primarily workloads that use alot of IO (things like disk reads and writes) suffer the most. I wouldn't worry too much about single threaded performance, as the newer chips clock higher.

    I would wait and see what the newer 9900KS revision has, I suspect that more of the exploits will be patched in silicon , not needing as many software patches/microcode changes that hurt performance.


    just to clarify the original 9900k is patched against meltdown and l1tf, and the newer revisions like the 9900kf are patched against meltdown l1tf and possibly MDS, all older 8000(and before) desktop chips are vulnerable to most if not all the exploits.

    The reason im not completely sure about the 9900KF is because this https://software.intel.com/security...-enumeration-and-architectural-msrs#MDS-CPUID document does predates its launch, and shows that several mobile chips and some desktop chips are completely or partially patched against mds.
     
  18. olenbr

    olenbr Guest

    Messages:
    4
    Likes Received:
    0
    GPU:
    GT 730
    so the biggest one being meltdown, does that impact disk i/o or is that for spectre and MDS?

    which security issue and patch gives the highest hit to storage performance in this case?
     
  19. user1

    user1 Ancient Guru

    Messages:
    2,746
    Likes Received:
    1,279
    GPU:
    Mi25/IGP
    spectre/mds and its related variants cause the greatest performance impact, they are the ones that hit disk io . meltdown is the most serious interms of breaching security but has a minimal to non-existant performance impact when patched

    this is because they(spectre/mds) cannot be fixed via software patches or microcode, the only thing intel can do to mitigate them is disable speculation(big performance hit) when performing certain tasks and flush caches every time they touch something sensitive(also a performance hit), now they some tricks to minimize the performance impact, but it will always be there for affected chips. the chips themselves are pretty fast already, so you typically wont see much impact on slower io like sata, but m.2 drives and nvme drives are noticeably affected.
     
    olenbr likes this.
  20. olenbr

    olenbr Guest

    Messages:
    4
    Likes Received:
    0
    GPU:
    GT 730
    thanks for clearing that up for me.

    I do have plans to get a 9900KF/KS whichever better binned especially they come with built in silicon fixes for meltdown. aside from that should be good, I just dont wish to have storage performance hit and from the looks of it, CPU performance reduction is guaranteed but just a small percentage of it, while storage performance wont be lost if it's just meltdown fixes.
     

Share This Page