Does Windows disable the software mitigations on the i9-9900K where hardware mitigations are present? Did you try disabling them using the registry to see if that makes any difference in performance?
Ian, can you comment on lack of hyper-threading in these results? As I understand, "disable hyper-threading" was one of the early (if incomplete) mitigation recommendations for these flavors of attack. I wonder if results would tell a different story with hyper-threading enabled. Is there something about these hardware fixes that would make them agnostic to whether hyper-threading is enabled?
Disabling HT is common for microarch perf comparisons, as it means that statically partitioned bits of the core microarch no longer pose a potential perf bottleneck. But you make a good point, and a couple of people emailed me about it. Will retest when I get home in a couple of weeks.
Ian, in addition to enabling HT, which is most essential. Both with and without HT, since customers have large numbers of both types. Can you please also remove the software patches on the 8086K. Otherwise we cannot clearly see which, if any of your specific performance tests were actually touching / affected by spectre and meltdown... And whilst this might be more difficult and time consuming. It would nevertheless really be a great help to be certain and prove unambiguously within the same test conditions, that you were actually hitting and covering scenarios affected by these mitigations.
i remember when linpack testing was criticized. but linpack runs better with HT\SMT off anyway on both amd and intel. its the correct way to test it. not sure if this is the same situation but its worth mentioning
"So basically disabling all major benefits from speculative memory accesses to begin with"
??? last time I looked, that's been part of multiprogammed machines since the 60s.
"The STRETCH was not a complete success. It guessed which path would be taken from a branch instruction and obeyed instructions along that path. If it had taken the wrong branch, it had to undo the effects of those instructions executed which really compromised its overall performance. Much time was spent at Aldermaston writing compilers that would allow the STRETCH to guess correctly more often." here: http://www.chilton-computing.org.uk/ca/technology/...
Speculation occurs whether HT is enabled or not, it's HT agnostic. If anything, HT is used to hide memory latency when sufficient resources allow. See my comment above
I know your site relies on advertising. But you need to find a way to keep yourselves from being a vehicle for frauds and scams. For example on this page "thousands of people from the us cancel Netflix..."
This links to a fraud site that true to get people to turn over their credit card to a fake streaming service.
I keep getting: "Millionaire Mom from Helsinki Exposes 3-step Secret to Work from Home" "The Rich in Finland Are Trying to Ban This Bitcoin Video" "Finland Company Fires Manager After Making Thousands Rich By Mistake"
Ryan, is you can successfully convince your web overlords to ditch revcontent, you will be doing a major service to your readers. The "ads" are clickbait that often use NSFW images of women that I'd categorize as misogynistic, many of the "ads" lead to scams and malware.
This is tech site, I suggest find an ad provider that advertises tech.
please retest with 8700k and earlier. If I'm losing 10% I want to know when the class action suit starts as I want some money back! I didn't pay for a chip that only next 90% of it's perf or less. My 4790k hasn't had any fixes yet due to this. Again, they owe ALL OF US MONEY, and more on older stuff clearly according to intel themselves.
I hope you ask AMD what is wrong (or not) with their 7nm chips upcoming. I'm going to buy one probably anyway assuming they are NOT as bad as Intel (weren't already, so...why would next gen be worse). I wouldn't even patch HOME pc's for anything you have to sit at my PC to exploit (doesn't happen at home). But I'd still want some rebates or something to shut up.
It's broke, slowing me down is still broke even if mitigated. And testing without HT is useless as others noted. Whether it masks stuff or not, who buys an Intel chip and turns off HT? NOBODY. If that's Intel's fix, I want a LOT of my money back...LOL.
https://www.zdnet.com/article/beyond-spectre-fores... "Red Hat strongly recommends to mitigate the problem, you manually enable Linux specific kernel parameters or potentially disabling features like Intel hyperthreading, after the available updates have been applied."
Yeah, not an option IMHO. Oracle in that article says disabling HT doesn't fix it all anyway, and just slows you down of course. "The real fix to all these problems, Intel admits, is by replacing today's processors. " Ahh, so, did you do this on purpose? Whenever you have a major slowdown in PC sales (eh, last 5 years declining), you release the bugs (much like fbi/doj leaking right and left on fake news) and claim you need to fix them? Er, Uh, just buy our crap again, we promise we'll get it right this time, but you ALL have to replace every chip we've ever made...LOL. I want my freaking money back. :(
I'll buy AMD next due to 7nm anyway, but this is the last time I buy Intel even if they win 2020-2021. Color me pissed for some time. I have at least two chips to buy at 7nm, and I'm thinking AMD wins my vote for the next one or two after that by default. Intel now has to EARN my purchase and that will take a LONG while.
Good luck winning a lawsuit with the complaint "Intel owe me a refund because they didn't forecast a major security flaw being discovered in their products", you fucking dumbass.
Eh, I've seen cases with less merit. If it gets big enough, Intel will probably settle and mail everyone with a proof of purchase of the affected part a five-dollar coupon for a future Intel processor purchase.
This would have zero merit. Intel never advertised their processors as having N performance with 0 security flaws. The whiner is free to run his system the same way as the day he bought it. Nobody is forcing him to install the patches. In fact, since Intel's claims are generally in the category of "X% faster than the previous generation," comparing patched to patched, the generational improvements have exceeded their claims.
I think it would be way, way more interesting, and far more useful, to test CPUs that people actually buy, rather than the 9900K which will be purchased by the 1%.
Also, I gather that the results will not be directly applicable to older architectures. I would bet that most people (even geeks reading this article) will still be using something between Sandy Bridge and Skylake in their computers. One more reason to test other, older CPUs as well.
Still, an interesting read. The kind that only Anandtech does. ;)
I did manage to figure all but one of those out myself, but there's no reason readers should have to go to that effort. CFL-R is the only one of the four you explain anywhere in the article, and that explanation comes after the chart. Please put some kind of legend or labels in or before the chart.
"margin of error"; Typically if random performance varies test to test by X, and no change is expected, statistically you would expect some results to be faster and some slower on average. Which is why Enron always beating predictions set off alarm bells to statisticians and why adjustments in historical temperatures that almost always result in lower temperatures in past and/or higher modern temperatures does the same to logical people.
It looks like this "hardware" fix is just the microcode being updated, instead of having UEFI patching it during POST.
If there's no performance or security improvement, it'd be better to not have added it to hardware at all. Because a software patch can be updated easier than a hardware one.
Late to the scene, but good because we were able to look at a few reviews on YouTube.
9900k on a z370 board will have a 15% increase in performance IF you use a BIOS version that does not include the microcode update + disabling the meltdown/spectre mitigation in Win10. The same cannot be said for a 9900k running on a Z390 board because the microcode is present on all BIOS versions, from 1st release to current.
The 10700k on a Z490 (pretty much the same CPU) will run slightly slower than a 9900k on a z370 because they are baking in the mitigation (same as z390 BIOS).
On YouTube, there are vids TRYING to demonstrate the differences in performance, but they are using z390. Only one person attempted this on a z370 and the outcome was as predicted, the gains were around 10-15%.
I have the 9900k refresh, and I believe the OS side gets disabled, but as the article says its still the same performance hit, so effectively if you have the newer chip the mitigation becomes compulsory.
It confuses the inspectre took, the inspectre tool will report (in default configuration) that it thinks the protection has manually been disabled, but also states above it that there is no manual override. The tool was developed to assume it seems all intel cpu's are vulnerable and its not been updated since the new cpus came out.
The powershell script reports that OS mitigation is not needed, the same as AMD processors.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
35 Comments
Back to Article
warpsharp - Monday, December 3, 2018 - link
Does Windows disable the software mitigations on the i9-9900K where hardware mitigations are present? Did you try disabling them using the registry to see if that makes any difference in performance?warpsharp - Monday, December 3, 2018 - link
Or is that not applicable to the variants where the hardware mitigations are present?Alexvrb - Tuesday, December 4, 2018 - link
You answered your own question. It's strictly firmware vs "hardware" fixes.I'll reserve judgement on the efficacy of their "hardware" (non-architectural) fix until Ian retests with HT.
Ryan Smith - Monday, December 3, 2018 - link
"Does Windows disable the software mitigations on the i9-9900K where hardware mitigations are present?"Yes, it does. Here is what Speculation Control reports on a 9900K:
http://images.anandtech.com/doci/13659/SpecControl...
bobhumplick - Thursday, December 6, 2018 - link
thats too bad. i run with the windows patches disabled and now we cant disable them. too badravyne - Monday, December 3, 2018 - link
Ian, can you comment on lack of hyper-threading in these results? As I understand, "disable hyper-threading" was one of the early (if incomplete) mitigation recommendations for these flavors of attack. I wonder if results would tell a different story with hyper-threading enabled. Is there something about these hardware fixes that would make them agnostic to whether hyper-threading is enabled?Ian Cutress - Monday, December 3, 2018 - link
Disabling HT is common for microarch perf comparisons, as it means that statically partitioned bits of the core microarch no longer pose a potential perf bottleneck. But you make a good point, and a couple of people emailed me about it. Will retest when I get home in a couple of weeks.dreamcat4 - Wednesday, December 5, 2018 - link
Ian, in addition to enabling HT, which is most essential. Both with and without HT, since customers have large numbers of both types. Can you please also remove the software patches on the 8086K. Otherwise we cannot clearly see which, if any of your specific performance tests were actually touching / affected by spectre and meltdown... And whilst this might be more difficult and time consuming. It would nevertheless really be a great help to be certain and prove unambiguously within the same test conditions, that you were actually hitting and covering scenarios affected by these mitigations.bobhumplick - Wednesday, June 26, 2019 - link
i remember when linpack testing was criticized. but linpack runs better with HT\SMT off anyway on both amd and intel. its the correct way to test it. not sure if this is the same situation but its worth mentioningpeevee - Monday, December 3, 2018 - link
"Each processor was set to four cores each with no hyperthreading, with the frequency set to 3.0 GHz for all workloads at all times. "So basically disabling all major benefits from speculative memory accesses to begin with, thus invalidating the whole test.
FunBunny2 - Monday, December 3, 2018 - link
"So basically disabling all major benefits from speculative memory accesses to begin with"??? last time I looked, that's been part of multiprogammed machines since the 60s.
"The STRETCH was not a complete success. It guessed which path would be taken from a branch instruction and obeyed instructions along that path. If it had taken the wrong branch, it had to undo the effects of those instructions executed which really compromised its overall performance. Much time was spent at Aldermaston writing compilers that would allow the STRETCH to guess correctly more often."
here: http://www.chilton-computing.org.uk/ca/technology/...
Ian Cutress - Monday, December 3, 2018 - link
Speculation occurs whether HT is enabled or not, it's HT agnostic. If anything, HT is used to hide memory latency when sufficient resources allow. See my comment aboveUrthor - Monday, December 3, 2018 - link
Interesting you report as "within the margin of error"Will Anand ever report on its P-values???
Ratman6161 - Monday, December 3, 2018 - link
I know your site relies on advertising. But you need to find a way to keep yourselves from being a vehicle for frauds and scams. For example on this page "thousands of people from the us cancel Netflix..."This links to a fraud site that true to get people to turn over their credit card to a fake streaming service.
Ryan Smith - Monday, December 3, 2018 - link
Hey Ratman, I haven't seen that one. Could you please send me an email with screenshots and other info?LJM - Tuesday, December 4, 2018 - link
I keep getting:"Millionaire Mom from Helsinki Exposes 3-step Secret to Work from Home"
"The Rich in Finland Are Trying to Ban This Bitcoin Video"
"Finland Company Fires Manager After Making Thousands Rich By Mistake"
It's ridiculous.
LJM - Tuesday, December 4, 2018 - link
Pretty much all ads in the "Ads by RevcontentFrom The Web" section look like a scam every time.
Hixbot - Tuesday, December 4, 2018 - link
Ryan, is you can successfully convince your web overlords to ditch revcontent, you will be doing a major service to your readers. The "ads" are clickbait that often use NSFW images of women that I'd categorize as misogynistic, many of the "ads" lead to scams and malware.This is tech site, I suggest find an ad provider that advertises tech.
TheJian - Tuesday, December 4, 2018 - link
please retest with 8700k and earlier. If I'm losing 10% I want to know when the class action suit starts as I want some money back! I didn't pay for a chip that only next 90% of it's perf or less. My 4790k hasn't had any fixes yet due to this. Again, they owe ALL OF US MONEY, and more on older stuff clearly according to intel themselves.I hope you ask AMD what is wrong (or not) with their 7nm chips upcoming. I'm going to buy one probably anyway assuming they are NOT as bad as Intel (weren't already, so...why would next gen be worse). I wouldn't even patch HOME pc's for anything you have to sit at my PC to exploit (doesn't happen at home). But I'd still want some rebates or something to shut up.
It's broke, slowing me down is still broke even if mitigated. And testing without HT is useless as others noted. Whether it masks stuff or not, who buys an Intel chip and turns off HT? NOBODY. If that's Intel's fix, I want a LOT of my money back...LOL.
https://www.tomshardware.com/news/disable-intel-hy...
OpenBSD disabled it, can't get to the article right now on vpn (toms hates my vpn..ROFL) too lazy to turn it off.
https://www.zdnet.com/article/beyond-spectre-fores...
"Red Hat strongly recommends to mitigate the problem, you manually enable Linux specific kernel parameters or potentially disabling features like Intel hyperthreading, after the available updates have been applied."
Yeah, not an option IMHO. Oracle in that article says disabling HT doesn't fix it all anyway, and just slows you down of course.
"The real fix to all these problems, Intel admits, is by replacing today's processors. "
Ahh, so, did you do this on purpose? Whenever you have a major slowdown in PC sales (eh, last 5 years declining), you release the bugs (much like fbi/doj leaking right and left on fake news) and claim you need to fix them? Er, Uh, just buy our crap again, we promise we'll get it right this time, but you ALL have to replace every chip we've ever made...LOL. I want my freaking money back. :(
I'll buy AMD next due to 7nm anyway, but this is the last time I buy Intel even if they win 2020-2021. Color me pissed for some time. I have at least two chips to buy at 7nm, and I'm thinking AMD wins my vote for the next one or two after that by default. Intel now has to EARN my purchase and that will take a LONG while.
The_Assimilator - Tuesday, December 4, 2018 - link
Good luck winning a lawsuit with the complaint "Intel owe me a refund because they didn't forecast a major security flaw being discovered in their products", you fucking dumbass.Lord of the Bored - Tuesday, December 4, 2018 - link
Eh, I've seen cases with less merit.If it gets big enough, Intel will probably settle and mail everyone with a proof of purchase of the affected part a five-dollar coupon for a future Intel processor purchase.
DominionSeraph - Wednesday, December 5, 2018 - link
This would have zero merit. Intel never advertised their processors as having N performance with 0 security flaws. The whiner is free to run his system the same way as the day he bought it. Nobody is forcing him to install the patches.In fact, since Intel's claims are generally in the category of "X% faster than the previous generation," comparing patched to patched, the generational improvements have exceeded their claims.
Jimios - Tuesday, December 4, 2018 - link
I think it would be way, way more interesting, and far more useful, to test CPUs that people actually buy, rather than the 9900K which will be purchased by the 1%.Also, I gather that the results will not be directly applicable to older architectures. I would bet that most people (even geeks reading this article) will still be using something between Sandy Bridge and Skylake in their computers. One more reason to test other, older CPUs as well.
Still, an interesting read. The kind that only Anandtech does. ;)
jensend - Tuesday, December 4, 2018 - link
Please do more to label or explain your chart. "CFL-R CL-AP WHL-U SKX-RAML-Y" is gobbledygook that takes readers a lot of effort to decode properly.
velanapontinha - Tuesday, December 4, 2018 - link
thisRyan Smith - Wednesday, December 5, 2018 - link
CFL-R: Coffee Lake RefreshCL-AP: Cascade Lake-AP
WHL-U: Whiskey Lake-U
SKX-R: Skylake-X Refresh
AML-Y: Amber Lake-Y
jensend - Wednesday, December 5, 2018 - link
I did manage to figure all but one of those out myself, but there's no reason readers should have to go to that effort. CFL-R is the only one of the four you explain anywhere in the article, and that explanation comes after the chart. Please put some kind of legend or labels in or before the chart.HStewart - Thursday, December 6, 2018 - link
I am curious on why we see increases on some of test - are these application program better - or do these test based on hardware related fixes.So is possible that performance is effected on how the software is designed.
ironargonaut - Monday, December 17, 2018 - link
"margin of error"; Typically if random performance varies test to test by X, and no change is expected, statistically you would expect some results to be faster and some slower on average. Which is why Enron always beating predictions set off alarm bells to statisticians and why adjustments in historical temperatures that almost always result in lower temperatures in past and/or higher modern temperatures does the same to logical people.TesseractOrion - Monday, September 16, 2019 - link
Sad attempt at trolling by ironargonaut SMHTom01 - Sunday, December 9, 2018 - link
I want to disable all those useless Patches.I prefer Speed.
ironargonaut - Monday, December 17, 2018 - link
Slow down, speed kills. :)HikariWS - Friday, December 14, 2018 - link
Well that's really sad.It looks like this "hardware" fix is just the microcode being updated, instead of having UEFI patching it during POST.
If there's no performance or security improvement, it'd be better to not have added it to hardware at all. Because a software patch can be updated easier than a hardware one.
balrogmad - Saturday, August 15, 2020 - link
Late to the scene, but good because we were able to look at a few reviews on YouTube.9900k on a z370 board will have a 15% increase in performance IF you use a BIOS version that does not include the microcode update + disabling the meltdown/spectre mitigation in Win10. The same cannot be said for a 9900k running on a Z390 board because the microcode is present on all BIOS versions, from 1st release to current.
The 10700k on a Z490 (pretty much the same CPU) will run slightly slower than a 9900k on a z370 because they are baking in the mitigation (same as z390 BIOS).
On YouTube, there are vids TRYING to demonstrate the differences in performance, but they are using z390. Only one person attempted this on a z370 and the outcome was as predicted, the gains were around 10-15%.
chrcoluk - Monday, February 1, 2021 - link
I have the 9900k refresh, and I believe the OS side gets disabled, but as the article says its still the same performance hit, so effectively if you have the newer chip the mitigation becomes compulsory.It confuses the inspectre took, the inspectre tool will report (in default configuration) that it thinks the protection has manually been disabled, but also states above it that there is no manual override. The tool was developed to assume it seems all intel cpu's are vulnerable and its not been updated since the new cpus came out.
The powershell script reports that OS mitigation is not needed, the same as AMD processors.