Testing Windows 10 Performance Before and After the Meltdown Flaw Emergency Patch

[Julio Franco]

The IT world was caught by surprise this week when Meltdown and Spectre hardware flaws were disclosed. With OS patches coming in, including an emergency fix for Windows 10, we've conducted a set of tests in the last 12 hours to see what impact this update has on performance for desktop users, if any at all.

Read the full article here.

https://www.techspot.com/article/1554-meltdown-flaw-cpu-performance-windows/

 

[ZackL04]

So much for the death of intel and rise of amd.

Fake news created by AMD (jk)

 

[Setnom]

Hey, thanks for this. I've been worried about the decrease in performance, but thankfully it's not that serious.

 

[Capaill]

Some graphs are difficult to understand as it's not clear if they are measuring seconds or throughput. Also "Measured in seconds, Higher is better" appears on a few graphs - I'm pretty sure that's not the case
But good to see the overall message is that casual users / gamers are ok. Enterprise users might be concerned though.
Not that I care either way as I'm on Ryzen now.

 

[ZackL04]

Some graphs are difficult to understand as it's not clear if they are measuring seconds or throughput. Also "Measured in seconds, Higher is better" appears on a few graphs - I'm pretty sure that's not the case
But good to see the overall message is that casual users / gamers are ok. Enterprise users might be concerned though.
Not that I care either way as I'm on Ryzen now.

Ryzen hasnt been proven to not be affected, its a much smaller user base and hasnt been tested.

Ifit does or not, looks to be a negligable affect anyway

 

[m4a4]

Yeah, when I read "30% reduction in speed" from certain articles, I interpreted it as "up to 30% reduction in speeds, but probably nothing for consumers".
Predictably it was fear-mongering for a few more page views...

 

[Puiu]

Some graphs are difficult to understand as it's not clear if they are measuring seconds or throughput. Also "Measured in seconds, Higher is better" appears on a few graphs - I'm pretty sure that's not the case
But good to see the overall message is that casual users / gamers are ok. Enterprise users might be concerned though.
Not that I care either way as I'm on Ryzen now.

Ryzen hasnt been proven to not be affected, its a much smaller user base and hasnt been tested.

Ifit does or not, looks to be a negligable affect anyway

Ryzen has been proven to not be affected by Meltdown at all.
As for Spectre, Variant One (Bounds Check Bypass) affects all CPUs but will receive a software patch with no real kernel overhead issues. Variant Two (Branch Target Injection) might affect AMD, but until now researchers haven't been able to do it (it was done on Intel and some ARM CPUs).

As far I understand, the reason AMD isn't that concerned about Spectre is because of how it works in relation to their architecture. From the white paper it seems that it exploits the cached data from a branch mispredict. That will rarely contain sensitive data and is highly dependent on the target app.

 

[Puiu]

Yeah, when I read "30% reduction in speed" from certain articles, I interpreted it as "up to 30% reduction in speeds, but probably nothing for consumers".
Predictably it was fear-mongering for a few more page views...

It wasn't really fear-mongering, just a reaction to early benchmarks on server applications. You can see this in the random 4k reads in this article, which is an important metric in many server workloads. We've know from the start that regular desktop applications would see at most a 5% reduction. Most desktop applications are not IO bound or make a lot of syscalls which is why games are fine with very few exceptions.
For example PostgreSQL seems to be hit quite hard by this patch. Older chips are also hit harder than the newer Kaby/Coffee Lake.

This is an official statement from Intel which put oil on the fire: "performance impacts are workload-dependent, and, for the average computer user, should not be significant and will be mitigated over time."

We just need to wait for Steve to finish the rest of the tests, although as he said he won't do server stuff since it is outside of his expertise.

 

[m4a4]

It wasn't really fear-mongering, just a reaction to early benchmarks on server applications. You can see this in the random 4k reads in this article, which is an important metric in many server workloads. We've know from the start that regular desktop applications would see at most a 5% reduction and that games should have no issues.
For example PostgreSQL seems to be hit quite hard by this patch. Older chips are also hit harder than the newer Kaby/Coffee Lake.

We just need to wait for Steve to finish the rest of the tests, although as he said he won't do server stuff since it is outside of his expertise.

The way articles I saw were titled and other things were done poorly then (not saying TS did). A lot didn't really mention servers up front, and definitely would lead less tech-savvy people into believing it would significantly affect them.

Now I get to look forward to a minimal 30% slowdown.

Earlier this week, some security researchers said any fix -- which would need to be handled by software -- could slow down computer systems, possibly by 30 percent or more.

The good news is that there is a fix. But unfortunately the security patch will slow down processing power by as much as 30 per cent, hitting personal computers and cloud services around the world.

But a few of the quotes I found off of FB news (yes FB news, because I get to see how stupid the uninformed reactions are lol)...

 

[Puiu]

The way articles I saw were titled and other things were done poorly then (not saying TS did). A lot didn't really mention servers up front, and definitely would lead less tech-savvy people into believing it would significantly affect them.




But a few of the quotes I found off of FB news (yes FB news, because I get to see how stupid the uninformed reactions are lol)...

That's because everything was under NDA and people working on the issue could no disclose any details (even the Linux public code has the comments redacted). All we knew was that everybody was scrambling to create patches for some serious security flaws.

For those that want more details on Linux/server benchmarks should go to phoronix and check some of the early tests done there. There will also more tests/results in the near future. (the Xeon scalable VM tests are really interesting)

Here's a nice quote: "While any performance slowdown is unfortunate, at least for the workloads tested thus far they've basically been manageable. I would equate it similarly to the cost of enjoying full-disk encryption. Page table isolation makes sense in increasing kernel security: it's not that this is some outright shoddy code hack to workaround a bug but something that does make sense."

 

[hahahanoobs]

And this is why I didn't jump on the bash Intel bandwagon...

 

[m4a4]

That's because everything was under NDA and people working on the issue could no disclose any details (even the Linux public code has the comments redacted). All we knew was that everybody was scrambling to create patches for some serious security flaws.

For those that want more details on Linux/server benchmarks should go to phoronix and check some of the early tests done there. There will also more tests/results in the near future. (the Xeon scalable VM tests are really interesting)

Here's a nice quote: "While any performance slowdown is unfortunate, at least for the workloads tested thus far they've basically been manageable. I would equate it similarly to the cost of enjoying full-disk encryption. Page table isolation makes sense in increasing kernel security: it's not that this is some outright shoddy code hack to workaround a bug but something that does make sense."

Yeah, I get that. My point is that a lot of what I saw with articles was immediately "the death of Intel" and whatnot just to get the views.

 

[Puiu]

And this is why I didn't jump on the bash Intel bandwagon...

Indeed, the big performance drops have been mostly limited to synthetic tests. In reality, for servers in falls somewhere between 0 to 15% and for desktops under 5%.

But that doesn't mean we should be forgiving. It's still a major security flaw. God knows how long it will take for many companies to update their software. (we all know some will never do)

Yeah, I get that. My point is that a lot of what I saw with articles was immediately "the death of Intel" and whatnot just to get the views.

Yeah, Intel will be fine, although this doesn't mean that AMD can't use this opportunity to sell some servers Business as usual.

 

[Evernessince]

And this is why I didn't jump on the bash Intel bandwagon...

The performance hit was just an additional reason to be angry, not the main reason. A critical security flaw that the CEO "cached" (lol) in on should have you angry enough.

The performance hit does not abate any of that or the fact that this is going to still affect other online services.

This is only a one processor test mind you, so it isn't completely conclusive on the performance end of things.

 

[noname]

https://support.microsoft.com/en-us...ive-execution-side-channel-vulnerabilities-in
[[
Important
Customers who only install the Windows January 2018 security updates will not receive the benefit of all known protections against the vulnerabilities. In addition to installing the January security updates, a processor microcode, or firmware, update is required. This should be available from the device manufacturer.

Note Surface customers will receive a microcode update via Windows update.
]]

Also they provide a PowerShell script there to verify the protections are enabled.

TECHSPOT must update the article with the PowerShell Verification Script Output so we can see if the patch is applied or not on the system, not just installed.

 

[Flebbert]

There are a few pieces of this story that I think a key.

1. Intel have been aware of the issue since June but only now that the embargo is lifting in a few days have they addressed the issue. I know that MS etc would have been working on patches etc but Intel definitely waited to the 11th hour to ensure sales wouldn't be affected.

2. CPU's releasing ahead of schedule seems a little convenient especially with the IME issue dropping in December as well. Wanted to maximize sales.

3. The CEO selling shares is dodge, $24M to $250K just before. If he had around 510,000 shares and the difference is between $47 and $43 each that is a $2M difference. Reality is however that give it 4 weeks and Intel's shares will have recovered.

4. The performance is 10% of this issue, the security flaw that allows you to rent a server on AZURE and give you full access to what is stored in memory for other VMs on that CPU is way worse. If that could not be fixed that would have been a different story.

5. This has been around for 20 years of CPUs and the only reason we are hearing about it is that it can be fixed. If this couldn't be fixed with software we would have never heard about it until there was a critical incident.

6. AMD and Intel have been talking long before the start of this week, AMDs statement was way to fast and they have obviously completed alot of testing to be as confident as they were.

Pretty interesting week really.

 

[antiproduct]

Any chance we'll get to see benchmarks on older systems? Since there have been so many "do you need to upgrade your sandy bridge CPU" articles on this site, it'd be nice to revisit some of those old systems with this patch before and after. Personally, I'm hoping my ivy bridge CPU can hang on for awhile longer till the 2nd gen Ryzen. Upgrading can be a pain.

 

[Kenrick]

Now Intel, im waiting for the final
blow, give us the coffeelake discounts that will undercut the competition. That will make things right and back to normal.

 

[Nobina]

I won't rest easy until I see older gen CPUs in these tests.

 

[Potato Judge]

It was pointed out by another tech channel that there is a chance that the upcoming patch could be reverse engineered by hackers. So performance drop is the least to worry about if that would happen. It's Y2K all over again.

 

[infiltrator]

I wonder, if there is going to be a lawsuit against Intel?
Nearly in all cases nowadays, someone ends up being suited.

 

[Humza]

Can I be a bit selfish here and ask for gaming benchmarks for a 4690k/GTX 1070 combo?

 

[SamuraiSamson]

I won't rest easy until I see older gen CPUs in these tests.

Guru 3D has an article where they check the performance impact of the patch on an i7 5960X. There is a drop of about 2-3 % on average.

 

[SamuraiSamson]

I won't rest easy until I see older gen CPUs in these tests.

Guru 3D has an article where they check the performance impact of the patch on an i7 5960X. The drop is around 2-3% in most applications.

 

[Adobo]

And this is why I didn't jump on the bash Intel bandwagon...

Indeed, the big performance drops have been mostly limited to synthetic tests. In reality, for servers in falls somewhere between 0 to 15% and for desktops under 5%.

But that doesn't mean we should be forgiving. It's still a major security flaw. God knows how long it will take for many companies to update their software. (we all know some will never do)

Yeah, I get that. My point is that a lot of what I saw with articles was immediately "the death of Intel" and whatnot just to get the views.

Yeah, Intel will be fine, although this doesn't mean that AMD can't use this opportunity to sell some servers Business as usual.

I woke up today to a 32% reduction in performance after Amazon patched one of the AWS instances we use. This is severely impactful for IO Heavy processes which are commonplace in a microservice oriented organization.

My team and I will have to spend multiple workdays scaling our systems and reforecasting hosting costs due to this issue.

So no, it's not just synthetic tests. There are multiple real world scenarios that show crippling decreases in performance. I surely hope no onr ever has to wake up their car suddently grtting 30% less miles per gallon.