Intel: CPU patch has minimal impact in the real-world

Quote:

---

But Intel released Coffee Lake, and its CEO sold millions of shares, knowing of issues.

---

Read more.

Not even close to "minimal impact".
Look at all 4 statements.

Apple - listing cherry-picked artificial benchmarks, which are simply ignoring everything that those benchmarks don't test.
Microsoft - forced patching of the affected areas to mitigate the issue.
Amazon - what about the "minority".
Google - again if the majority are only affected negligibly, what about the minority.

Seems odd for Microsoft to say nothing to see hear as their customers are reporting problems with their virtual machines.

Quote:

---

Originally Posted by ByteMyAscii

Not even close to "minimal impact".
Look at all 4 statements.

Apple - listing cherry-picked artificial benchmarks, which are simply ignoring everything that those benchmarks don't test.
Microsoft - forced patching of the affected areas to mitigate the issue.
Amazon - what about the "minority".
Google - again if the majority are only affected negligibly, what about the minority.

---

They are clearly protecting their own Server/Cloud/Other businesses

I honestly expect that ~90% of their customers will notice no difference whatsoever. Those running specific workloads will have problems but hopefully once the hole is sealed there will be optimisations released which will mitigate the impact. At the moment everyone is panicking for no justifiable reason. Wait and see what the real world impact actually is. At the moment we've got Which magazine putting crap all over Facebook implying that people will all experience around a 30% performance hit rather than explaining this was a synthetic benchmark on an OS that hardly any of their readers will use and employing a workload in a setting they're even less likely to use. Once the hole is sealed, they can work on optimisations for those people who are really impacted. For gamers, this just appears to be a non issue for the time being. We just need to be less dramatic about this - the security hole is clearly a major issue. The fallout should be manageable although there are some people who are going to have to use the opportunity to switch to Ryzen. In that case, I would say they should be reimbursed by a class action or something similar.

EDIT: What is going to happen to Intel is that people will probably stop buying their CPUs until new architecture comes out fixing the flaw. By the sounds of it, AMD won't have such issues.

EDIT EDIT: I do wonder if the CEO will be investigated for selling his shares with insider knowledge??

I thought this was the emergency fix and not the full one which is expected to cause a bigger difference... Also Hardware unboxed noticed fairly big differences in their m.2 speeds post patch.

Quote:

---

Originally Posted by philehidiot

I honestly expect that ~90% of their customers will notice no difference whatsoever. Those running specific workloads will have problems but hopefully once the hole is sealed there will be optimisations released which will mitigate the impact. At the moment everyone is panicking for no justifiable reason. Wait and see what the real world impact actually is. At the moment we've got Which magazine putting crap all over Facebook implying that people will all experience around a 30% performance hit rather than explaining this was a synthetic benchmark on an OS that hardly any of their readers will use and employing a workload in a setting they're even less likely to use. Once the hole is sealed, they can work on optimisations for those people who are really impacted. For gamers, this just appears to be a non issue for the time being. We just need to be less dramatic about this - the security hole is clearly a major issue. The fallout should be manageable although there are some people who are going to have to use the opportunity to switch to Ryzen. In that case, I would say they should be reimbursed by a class action or something similar.

---

I agree most people won't notice a difference however it's going to take a long time to seal some, or all, of the holes and until they are sealed the workarounds are going to have a noticeable impact on performance in some workloads, that will improve overtime as currently the fixes/workarounds are probably taking a scorched earth approach, that will take time to change as software developers will need to rewrite their code to excluded certain data from speculative execution.

Quote:

---

Originally Posted by outwar6010

I thought this was the emergency fix and not the full one which is expected to cause a bigger difference... Also Hardware unboxed noticed fairly big differences in their m.2 speeds post patch.

---

I read it the other way around - this patch has the biggest impact, which will be mitigated more in future patches. Noticing an impact in m.2 speeds isn't the same thing as seeing much of a real-world impact.

Quote:

---

Originally Posted by outwar6010

I thought this was the emergency fix and not the full one which is expected to cause a bigger difference... Also Hardware unboxed noticed fairly big differences in their m.2 speeds post patch.

---

If the emergency fix fills the hole, and has less performance impact than the full fix, why bother developing the full fix? The rushed-out-in-six-months fix we've got now can only be slower than any further fixes.

CEO sells up, smells like insider trading !

Quote:

---

Originally Posted by kalniel

Noticing an impact in m.2 speeds isn't the same thing as seeing much of a real-world impact.

---

It would if your real world application is heavily disk I/O bound.

Interestingly heavy disk I/O seems to be all up in the air, PCPer did some testing and found both losses and oddly some configurations saw gains, it's all very confusing.

So much for only Meltdown causing a slowdown though.

This user on Reddit ran Realbench with the Windows patch and again with the patch and BIOS microcode update (which is AFAIK only for the Spectre bugs):
https://np.reddit.com/r/pcmasterrace...atch_and_bios/
https://i.imgur.com/35xae2F.png
The imaging results look rather bad at over -21%.
Interesting but RealBench is hardly a good benchmark.


Huge CPU usage increase when Epic Games upgraded one of their servers for Meltdown
https://i.imgur.com/D6s5ltF.png
https://www.epicgames.com/fortnite/f...ability-update

Guess, there might be little reason for a gaming server to be super secure so they might not need the patch. Eventually anyway, once they separate their concerns so that another server handles authentication etc. and then the server serving only games can be left less secure. But that kind of work is not instant.

Quote:

---

Seems one one of the methods the Linux kernel guys want to use is pretty useless on Skylake+ (Skylake, Kabylake, Coffeelake etc.) as it gets optimised?
Retpoline as a mitigation strategy swaps indirect branches for returns,
to avoid using predictions which come from the BTB, as they can be
poisoned by an attacker.

The problem with Skylake+ is that an RSB underflow falls back to using a
BTB prediction, which allows the attacker to take control of speculation.

---

https://lkml.org/lkml/2018/1/4/724

So, 'minimal' real world impact as long as you're not running any of those loads.

The thing is it seems some benchmark software will need to be rewritten because of the way these vulnerabilities are being mitigated, currently some (most?) benchmarks hammer the very thing that was exploitable and the fixes have hobbled that, most everyday workloads don't hammer a processors local memory so when a context switch does happen the penalty is relatively small, however if, as it seems benchmarks are doing, you perform lots of context switching those small delays add up.

Quote:

---

Originally Posted by outwar6010

I thought this was the emergency fix and not the full one which is expected to cause a bigger difference... Also Hardware unboxed noticed fairly big differences in their m.2 speeds post patch.

---

full patch is out, you are using it

I'm running an I5 processor are there any benchmarks showing performance impact while streaming pron? It's the biggest workload for my PC and I'm sure a real world task many users would be interested to see...

So this is this decade's Y2K.

Quote:

---

Originally Posted by the_avatar

So this is this decade's Y2K.

---

I was thinking the same thing myself earlier today.n lots of speculation about performance hits, but no real world evidence.

Quote:

---

Originally Posted by peterb

I was thinking the same thing myself earlier today.n lots of speculation about performance hits, but no real world evidence.

---

I'd be careful saying that mate, an engineer will pop out of a socket and knife you.

The only reason Y2K didn't come to be a problem was because of the millions of hours of work that went on, around the clock, by dedicated engineers, coders, programmers, and various other folks. It was an absolute massive effort that was pretty much a complete success

This is nothing like Y2K: Y2K was a very real issue, that could be fixed. This is being mitigated. Only way it'd be like Y2K would be if at the stroke of midnight all clocks started running slightly slower and there'd be feck all we could do about it :P

Quote:

---

Originally Posted by Tunnah

I'd be careful saying that mate, an engineer will pop out of a socket and knife you.

The only reason Y2K didn't come to be a problem was because of the millions of hours of work that went on, around the clock, by dedicated engineers, coders, programmers, and various other folks. It was an absolute massive effort that was pretty much a complete success

This is nothing like Y2K: Y2K was a very real issue, that could be fixed. This is being mitigated. Only way it'd be like Y2K would be if at the stroke of midnight all clocks started running slightly slower and there'd be feck all we could do about it :P

---

Fair enough <removes tongue from cheek> :)

Quote:

---

Originally Posted by philehidiot

I honestly expect that ~90% of their customers will notice no difference whatsoever. Those running specific workloads will have problems but hopefully once the hole is sealed there will be optimisations released which will mitigate the impact. At the moment everyone is panicking for no justifiable reason. Wait and see what the real world impact actually is. At the moment we've got Which magazine putting crap all over Facebook implying that people will all experience around a 30% performance hit rather than explaining this was a synthetic benchmark on an OS that hardly any of their readers will use and employing a workload in a setting they're even less likely to use. Once the hole is sealed, they can work on optimisations for those people who are really impacted. For gamers, this just appears to be a non issue for the time being. We just need to be less dramatic about this - the security hole is clearly a major issue. The fallout should be manageable although there are some people who are going to have to use the opportunity to switch to Ryzen. In that case, I would say they should be reimbursed by a class action or something similar.

EDIT: What is going to happen to Intel is that people will probably stop buying their CPUs until new architecture comes out fixing the flaw. By the sounds of it, AMD won't have such issues.

EDIT EDIT: I do wonder if the CEO will be investigated for selling his shares with insider knowledge??

---

1. Most of the Internet runs on LAMP stacks. That's Linux (The OS Devs that found this), Apache, MySQL, PHP. All of those need to do two things: Read and Write to Memory and Hard Drive. When you take all of the little slow downs, they stack up.

2. The fix has been found to kill scientific/mathematical calculation speed. While the calculations still occur, they are slow, and if software was specifically designed or engineered to get a response in a certain amount of time, then you need to completely redesign the software, and any time dependent observations may be affected up to 30% or more (especially if virtualized).

3) The fields most affected by math and science calculations off the top of my head are most publicly funded science research, medical imaging, financial databases, etc.

Fortunately, you will only experience a partial slowdown if you visit websites that have AMD hardware. Unfortunately, most of the enterprise level world runs on Intel.

Couple this with the insider trading and the known security flaws of Intel, and you have Intel getting BTFO'd, if you will.

If Intel rallies from this breech of trust, I will be astounded.

I haven't chimed in on this yet though I've been following it closely.

I would just warn you that there are certain performance regressions from the OS patches, but there are some other (possibly larger) performance regressions for 'Intel' platforms from the also required firmware patches, most of which are missing right now, except if you have certain Asus boards as far as I know. So I would expect the real impact of this to come through only as the firmware / BIOS patches appear alongside the OS patches, whether that be Linux Windows or Mac/others.

Keep in mind the change numbers you see now may not be the actual outcome once vendor bios fixes hit.

edit: it is still likely to mainly hit certain server workloads though

Quote:

---

Originally Posted by Tunnah

I'd be careful saying that mate, an engineer will pop out of a socket and knife you.

The only reason Y2K didn't come to be a problem was because of the millions of hours of work that went on, around the clock, by dedicated engineers, coders, programmers, and various other folks. It was an absolute massive effort that was pretty much a complete success

This is nothing like Y2K: Y2K was a very real issue, that could be fixed. This is being mitigated. Only way it'd be like Y2K would be if at the stroke of midnight all clocks started running slightly slower and there'd be feck all we could do about it :P

---

I meant in terms of the sensationalized predictions of 50% performance hits and the like. I'm guessing future CPU's will have to be redesigned to prevent this on a hardware level.

Oh no, the world will end!!! Much like Y2K was over blown. And no, that's not to suggest this was/is a non issue, just that it got over blow a bit, and allowed all the Intel HATERS another reason to freak out over. Now, the nit picking begins.

Quote:

---

Originally Posted by the_avatar

I meant in terms of the sensationalized predictions of 50% performance hits and the like. I'm guessing future CPU's will have to be redesigned to prevent this on a hardware level.

---

The 30% performance hit figure was not a prediction, it was a measurement of a real workload. Compiling code in Linux matters to some of us as that is basically our job.

Quote:

---

Originally Posted by ETR316

Oh no, the world will end!!! Much like Y2K was over blown. And no, that's not to suggest this was/is a non issue, just that it got over blow a bit, and allowed all the Intel HATERS another reason to freak out over. Now, the nit picking begins.

---

I know quite a few people who are running around dealing with this. Just because it doesn't impact gamers (much) doesn't make it unimportant.

Quote:

---

Originally Posted by the_avatar

I'm guessing future CPU's will have to be redesigned to prevent this on a hardware level.

---

They've been designed to prevent, or rather mitigate against it, on a hardware level for over six years, unfortunately because the original core design has been around a lot longer than that it contains legacy "hardware", that's to say Intel introduced x86 instructions and the ability for the hardware to process those instructions many years ago but because there wasn't a known way to exfiltrate data from the processors local memory during speculative execution, and because writing extra code that excluded certain types of data from speculative execution takes time, i suspect, most developers didn't bother using the new instructions.

MFENCE, LFENCE, INVPCID and the associated PCID (Process-Context Identifiers) have been around for ages.