Windows Systems Performance Impacts from Spectre and Meltdown

Discussion in '[H]ard|OCP Front Page News' started by Kyle_Bennett, Jan 9, 2018.

  1. Kyle_Bennett

    Kyle_Bennett El Chingón Staff Member

    Messages:
    48,674
    Joined:
    May 18, 1997
    There has been a lot of discussion about Spectre and Meltdown since last week and now Microsoft has something to say about Windows Systems performance after patching.

    In general, our experience is that Variant 1 and Variant 3 mitigations have minimal performance impact, while Variant 2 remediation, including OS and microcode, has a performance impact.

    Here is the summary of what we have found so far:

    * With Windows 10 on newer silicon (2016-era PCs with Skylake, Kabylake or newer CPU), benchmarks show single-digit slowdowns, but we don’t expect most users to notice a change because these percentages are reflected in milliseconds.
    * With Windows 10 on older silicon (2015-era PCs with Haswell or older CPU), some benchmarks show more significant slowdowns, and we expect that some users will notice a decrease in system performance.
    * With Windows 8 and Windows 7 on older silicon (2015-era PCs with Haswell or older CPU), we expect most users to notice a decrease in system performance.
    * Windows Server on any silicon, especially in any IO-intensive application, shows a more significant performance impact when you enable the mitigations to isolate untrusted code within a Windows Server instance. This is why you want to be careful to evaluate the risk of untrusted code for each Windows Server instance, and balance the security versus performance tradeoff for your environment.


    So in summation, with processors older than Skylake (launched in August 2015) changed the way branch prediction is being handled so that it is not more specific to indirect branches. There will be some penalty but not like what we will see in previous CPU architectures. Windows 7 and 8 is going to be a bigger loser than Windows 10. Thanks cageymaru.
     
  2. dgingeri

    dgingeri 2[H]4U

    Messages:
    2,389
    Joined:
    Dec 5, 2004
    They missed a generation. The don't mention anything about Broadwell, which is newer than Haswell and older than Sky Lake. Broadwell is much like Haswell, so it is likely in the "older" category, but I find it odd they completely missed that one.
     
  3. Lothar

    Lothar Limp Gawd

    Messages:
    294
    Joined:
    Jun 7, 2004
    It's all a conspiracy to get people to upgrade to newer processors. ;)
     
  4. sfsuphysics

    sfsuphysics I don't get it

    Messages:
    12,238
    Joined:
    Jan 14, 2007
    wanted to say it explains why my old i3 laptop runs like shit now... but to be fair it always has. Always on the fence of whether or to put an SSD into a $400 budget laptop to speed it up.
     
  5. Trigneus

    Trigneus n00bie

    Messages:
    2
    Joined:
    Feb 11, 2016
    Unknown benchmarks loosely quoted for another "calm the typical user" story. Do any "typical" users visit [H]?
     
    Johan Steyn likes this.
  6. Az Syndicate

    Az Syndicate Limp Gawd

    Messages:
    382
    Joined:
    Feb 15, 2002
    Will this be the nail in my 2600K?....
     
  7. Dregan73

    Dregan73 Limp Gawd

    Messages:
    178
    Joined:
    Mar 10, 2007
    what if all this was planed, to force us to buy new cpus, conspiracy theories go......
     
    LightsOut41 likes this.
  8. Cmdrmonkey

    Cmdrmonkey Gawd

    Messages:
    883
    Joined:
    Jul 19, 2004
    Seems like a big scam to sell new hardware to people that probably don't need it based on something that may not even be a threat in the real world

    The conspiracy theory about Apple was proven correct. I'm very skeptical of this. CPUs had gotten to a point in recent years where they basically didn't need to be upgraded. I had no plans to upgrade my 4770K unless my PC suffered a hardware failure. This is awfully convenient for Intel.
     
  9. lollerwaffle

    lollerwaffle Limp Gawd

    Messages:
    458
    Joined:
    Feb 3, 2008
    I temporarily stopped Win10 updates and I'm not going to install BIOS updates for a while yet either.

    I mean, we have to yet see what the BIOS updates bring in terms of slowdown when they install new microcode. That'll be ontop of any Windows slowdowns, or at least compound them somewhat. So I'm taking all those benchmarks as too early. Microsoft is not saying whether they measured with udpated microcode or not.

    My BIOS has been great and I don't see a need to patch it. And the Windows patches, even once I finally have to install them to get other updates, can be disabled with PowerShell scripts. So there's path forward with no slowdowns. The worst exploit so far were the browser exploits, and FF as well as Edge has already plugged them. The rest of the exploits, I'm not juicy enough of a target for them.
     
  10. Grimlaking

    Grimlaking [H]ard|Gawd

    Messages:
    1,792
    Joined:
    May 9, 2006
    stefan231 likes this.
  11. SixFootDuo

    SixFootDuo [H]ardness Supreme

    Messages:
    4,303
    Joined:
    Oct 5, 2004
    Starting to think this was all an elaborate plan to get people to upgrade from their PC's.
     
  12. Cmdrmonkey

    Cmdrmonkey Gawd

    Messages:
    883
    Joined:
    Jul 19, 2004
    Notice how it conveniently affects PCs that are 4+ years old, even though there have been effectively no big performance changes in mainstream Intel CPUs since Sandy Bridge in 2011. This stinks of planned obsolescence.
     
    LightsOut41 likes this.
  13. Methadras

    Methadras [H]ardness Supreme

    Messages:
    6,098
    Joined:
    Dec 19, 2000
    Wonder what this means for 3770 i7's.
     
    John721 and admiralperpetual like this.
  14. TrailRunner

    TrailRunner Limp Gawd

    Messages:
    266
    Joined:
    Feb 15, 2012
    Wow, 1 hour and 12 comments, several of those comments about this being a ploy to force CPU upgrades on people, and not ONE comment about this being a ploy to force people to upgrade from 7 to 10.
    [H], I am disappointed in you.
     
    qb4ever, haste., Meeho and 3 others like this.
  15. Cmdrmonkey

    Cmdrmonkey Gawd

    Messages:
    883
    Joined:
    Jul 19, 2004
    It's both. Intel and Microsoft are colluding on this one.
     
  16. Spidey329

    Spidey329 [H]ardForum Junkie

    Messages:
    8,344
    Joined:
    Dec 15, 2003
    I know, right?
     
  17. heatlesssun

    heatlesssun Pick your own.....you deserve it.

    Messages:
    46,726
    Joined:
    Nov 5, 2005
    Pretty amazing. Considering that some or all of these flaws effect Linux, iOS, macOS, AMD & ARM CPUs, etc.
     
    juanrga and BigJayDogg3 like this.
  18. ol1bit

    ol1bit [H]ard|Gawd

    Messages:
    1,182
    Joined:
    Jan 15, 2007
    I could care less. I don't run any anti-virus or IE, and seem to be fine for the last 6 years. I have Windows 7 and Windows 10 PCs. One Windows 10 PC is running Bloomfield, If is slows down too much, back to Windows 7 I go. And for Corp sites its a crazy emergency, with thousands of servers, and BIOS updates, this will be Hell.
     
  19. MV75

    MV75 Gawd

    Messages:
    814
    Joined:
    Nov 13, 2007
    Nope, it's to get people to 10. How can it possibly really be about upgrading hardware when the bug is still there? Release new hardware free of the bug, hell yes to upgrading.

    See there is no windows 7 / 8.1 with newer silicon listed? We already know why, that was back in their forced to update to win 10 round #566 last year to forcibly end support for a currently supported os. I wonder how it really fares against 10 on same hardware? I'm guessing exactly the same.

    All I know is that I'm not updating to the latest monthly update, that's for sure. And what about consoles..... ? lol. Systems that already run to the 9th just to barely run what they run, they'll enjoy a slow down too.
     
    Last edited: Jan 9, 2018
  20. heatlesssun

    heatlesssun Pick your own.....you deserve it.

    Messages:
    46,726
    Joined:
    Nov 5, 2005
    In terms of raw performance running Win32 sure. There are other considerations like battery life, which is a much more important concern now than over 8 years ago when 7 launched as most PC at the time sold were still desktops. But we are talking about flaws that predate even Windows 7. Had this been discovered 8 years ago no doubt we'd be hearing the argument of Microsoft using this to force people to upgrade to Windows 7.
     
  21. admiralperpetual

    admiralperpetual Limp Gawd

    Messages:
    195
    Joined:
    May 7, 2015
    or us 3570k's...
     
    qb4ever, hexamon and gaphiltfish like this.
  22. maddude

    maddude 2[H]4U

    Messages:
    3,076
    Joined:
    Sep 6, 2006
    I doubt this is a master plan by Intel, as it would drive consumers straight to AMD.

    If anything, AMD was the real discoverer of these exploits and published them to driver their own sales! *tinfoil hat tightens*
     
  23. lollerwaffle

    lollerwaffle Limp Gawd

    Messages:
    458
    Joined:
    Feb 3, 2008
    It looks like it won't be the same. The reason being, the Windows 7 display driver model (DDM) is very, very different from the Windows 10 one. The Win7 one has font rendering in the kernel code, and the Win10 one does not. What that means is, whenever text is rendered on your desktop on in any desktop application, Win7 will make a LOT more kernel code transitions, which trigger the workaround of clearing certain CPU states and slowing down things compared to pre-patch. So in many 'regular Joe User' cases, Win7 will definitely be slower. How much, that remains to be seen.

    Source: https://www.theverge.com/2018/1/9/16868290/microsoft-meltdown-spectre-firmware-updates-pc-slowdown

    As far as 'it's to get people to 10', I don't buy it. Microsoft is already pushing VERY hard to get people to 10, I don't think they need this issue to push it further. They are patching Win7/8, it's just there are architectural differences that will make 7/8 slower compared to 10, that's not some sort of planned obsolescence. I mean the amount of long shot for this to be true is looney tunes conspiracy level, which I am not accusing you of, given you may not know the above.

    But seriously ask yourself, if you are staying on Microsoft platforms, don't you slowly but surely see the final last gasp of Win7 in view? It's coming and it's reasonable at this point, IMO.
     
    The Lamb and Lazorz_Go_PewPew like this.
  24. MV75

    MV75 Gawd

    Messages:
    814
    Joined:
    Nov 13, 2007
    Mmmmm maybe. Back then they weren't aggressive or as bastardly about it all like the forced windows 10 upgrade. Plus you still had full control of everything in 7 as well as full features that weren't constantly being eroded. Upgrading was an entirely different landscape back then without the bastard precedents that have created this backfire distrust of the company.
     
  25. MV75

    MV75 Gawd

    Messages:
    814
    Joined:
    Nov 13, 2007
    What I'm getting at is what about my setup, win 8.1 on kaby lake, not haswell?
     
  26. lollerwaffle

    lollerwaffle Limp Gawd

    Messages:
    458
    Joined:
    Feb 3, 2008
    I have no authoritative answer to that. I'll give you my guess: Kaby Lake has instructions that mitigate the performance impact of the workarounds. There is no reason to assume that Win8.1's patch won't take advantage of those instructions just the same as Win10. The same for Win7. Reason being, a LOT of corporations still run Win7. They will be pissed to all hell if there is a larger performance impact than needed because Microsoft develops different patch code for different OS's just to sell Windows 10. They could not afford that stunt, and frankly, I personally don't think they are that evil.

    I know 8.1 also had DDM changes, but I do not know if that moved font rendering code from kernel to user space like Win10 did. MS' official statement doesn't say one way or the other and I am too lazy to deep dive into DDM changes for 8.1 ;) Your question is a good one though.
     
  27. LMT MFA

    LMT MFA Limp Gawd

    Messages:
    186
    Joined:
    Jun 11, 2014
    Could you? Cuz it sounds like you couldn't.
     
  28. GoodBoy

    GoodBoy Gawd

    Messages:
    902
    Joined:
    Nov 29, 2004
    So where does Broadwell fit into this? It's right between Haswell and Skylake... 5th Gen. I have a Broadwell-E cpu. so hoping it's closer to the Skylake side of the performance..
     
  29. GoodBoy

    GoodBoy Gawd

    Messages:
    902
    Joined:
    Nov 29, 2004
    That's pretty brave. I would patch up if I was you.. But to answer your question: https://www.techspot.com/article/1556-meltdown-and-spectre-cpu-performance-windows/page4.html

    and the larger performance drops was in write times on SSD's, vs read times, which in some but not all situations/tests had some large performance drops, in other situations the read times took a smaller performance hit.
     
    lollerwaffle likes this.
  30. WhoMe

    WhoMe [H]Lite

    Messages:
    108
    Joined:
    Jan 3, 2018
    So will it be the same for Linux?
     
  31. aokman

    aokman Limp Gawd

    Messages:
    447
    Joined:
    Jan 3, 2012
    So does this mean I can RMA my CPU as faulty lol
     
    chappedstick, SFB, Mav451 and 5 others like this.
  32. unfortunateson

    unfortunateson Limp Gawd

    Messages:
    149
    Joined:
    Jul 17, 2004
    My result set of a quick Novabench test on an i5-Skylake PC running Windows 7 before/after patching shows a slight boost in CPU score (2%) but my SSD write speeds are lowered 17% and reads are lowered 9%.
     
  33. MV75

    MV75 Gawd

    Messages:
    814
    Joined:
    Nov 13, 2007
    Why brave? Run a safe browser, FF with ad block, don't visit dodgy sites (open spam mail), or use crappy applications, seems like there is no avenue for attack, nor any reason to patch.

    I do now wonder if it's per device the slow down happens? Would a (just say at the basics level for now), raid 0 mitigate any slow downs compared to a single device as you're splitting the throughput between at least two devices. Or would it increase cpu usage even further handling both at the same time?
     
    jtmcclain and EODetroit like this.
  34. umeng2002

    umeng2002 Limp Gawd

    Messages:
    502
    Joined:
    May 23, 2008
    You'll get a $35 check in the mail in 5 years from a class-action suit.
     
    mullet likes this.
  35. Frobozz

    Frobozz [H]ard|Gawd

    Messages:
    1,660
    Joined:
    Feb 15, 2002
    revisiting MSSQL server licensing is going to be painful for some. Isn't it per core?
     
  36. Derangel

    Derangel [H]ardForum Junkie

    Messages:
    15,156
    Joined:
    Jan 31, 2008
    You should go look at all the times malicious code has been injected into "safe" websites. Visiting "safe" sites won't protect you if happen to visit during a time that something has been injected into it.
     
  37. Grimlaking

    Grimlaking [H]ard|Gawd

    Messages:
    1,792
    Joined:
    May 9, 2006
    The impact would be the same and the additional latency would be per transaction. So a raid zero would feel it less. I don't really notice it myself with an nvme drive.
     
  38. Grimlaking

    Grimlaking [H]ard|Gawd

    Messages:
    1,792
    Joined:
    May 9, 2006
    Yes... Yes it is at 25k per 4 cores. That same 25k would buy a dual cpu server at 16 cores each with 384gb of ram dual dual port 10gb nics and 3 independent fc cards... Not that I've checked.
     
  39. rgMekanic

    rgMekanic [H]ard|News Staff Member

    Messages:
    3,363
    Joined:
    May 13, 2013
    Working great to get me to want to upgrade to Threadripper
     
  40. bugleyman

    bugleyman Limp Gawd

    Messages:
    420
    Joined:
    Oct 27, 2010
    Yup.