Intel releases their internal performance analysis after Spectre/Meltdown fixes

[WYP]

How hard has Intel's performance been hit?



Read more about Intel's internal performance impact assessment of Spectre/Meltdown fixes.

 

[hellboy99]

And what about Gens before Gen 6? AMD from now on with me and sooner rather than later, not looking forward to see what happens my i7 4790K

 

[NeverBackDown]

And what about Gens before Gen 6? AMD from now on with me and sooner rather than later, not looking forward to see what happens my i7 4790K

Your fine.

 

[AlienALX]

I doubt any one will actually notice it, but it will be there (the deficit). However, one thing I do know is that Ryzen plus is going to look an awful lot better than it was about to. Especially if people bench it against new results after the Meltdown patch.

That is the one that really causes issues. And of course, AMD are not susceptible to that in any shape or form. That is the one that literally needs a new kernel to address the CPU in a completely different way. And it causes latency and other issues which is why there is degradation in certain things.

And if you were very conservative and said that overall you have lost around, say, 8% performance? then wow, every single last % of that can, and will, be exploited by Ryzen +.

In fact, it almost closes the deficit between Ryzen and Coffeelake to absolutely nothing. Instead of it being "sort of" on par with Broadwell E.

 

[Darkdayzzz]

Your fine.

Consumers will be fine honestly, not like anyone trying to use Spectre will be pursuing a normal everyday person...not like Spectre is even easy to attempt to use to nefarious purposes. The person or group would really have to know their stuff so I wouldn't even sweat it.

I've got a 4790k myself and I in no way give a toss if I update or not. I won't be affected and I know it. Why? cuz no one cares about my downloads / porn collection / or achievements.

Yes that is silly logic and I know that but the point is - most normal everyday people will NEVER be targeted by cyber related attacks. Old people and the very young or the rich are who are targeted 99% of the time lol.

 

[barnsley]

Consumers will be fine honestly, not like anyone trying to use Spectre will be pursuing a normal everyday person...not like Spectre is even easy to attempt to use to nefarious purposes. The person or group would really have to know their stuff so I wouldn't even sweat it.

People said that about the tools leaked from that NSA hack and that didn't end well, did it?

Someone will work out how to create a module that exploits it, charge x% of a bitcoin/litecoin/whatever and start selling it to people who distribute malware. As with everything IT Security related, its only a matter of time before something happens.

 

[hellboy99]

Your fine.

I hope you're right, my rig is used for music production so I just have to wait and see what effect it will have on me.

 

[NeverBackDown]

I hope you're right, my rig is used for music production so I just have to wait and see what effect it will have on me.

It effects mostly VM's/DataCenters/Servers/etc. Consumers don't really have that kind of environment. It has to do with CPU branch prediction and when it mispredicts the correct next instruction, consumers don't really ever get this problem as you don't do complex enough tasks to warrant a failed prediction. Most CPUs have a 99% branch prediction success rate and I am sure it's even higher. CPUs that were amongst the first to get this would be more affected as the algorithms are less efficient.

 

[SPS]

It has to do with CPU branch prediction and when it mispredicts the correct next instruction, consumers don't really ever get this problem as you don't do complex enough tasks to warrant a failed prediction. Most CPUs have a 99% branch prediction success rate and I am sure it's even higher. CPUs that were amongst the first to get this would be more affected as the algorithms are less efficient.

Misprediction happens all the time.

 

[hellboy99]

I asked Intel today about Gens prior to their 6th Gen CPU's and this is the reply I got from them:

Hi, rest assure we are working diligently to address all concerns and will address yours as soon as possible. Thank you for your patience. You can find all the information regarding this topic in the following links: http://intel.ly/2Fua12f

 

[NeverBackDown]

Misprediction happens all the time.

They have an extremely high accuracy and if it does occur they have methods to make it so the already executed instruction isn't entirely wasted. Which is one of the issues with the malware in question.

 

[SPS]

They have an extremely high accuracy and if it does occur they have methods to make it so the already executed instruction isn't entirely wasted. Which is one of the issues with the malware in question.

All they can do is use the history from the previous execution of that code. When a program first boots the branch prediction will have no accuracy. And your statement isn't really correct. If the prediction is incorrect they must go back and take the other branch, simple as that. The spectre bug is about forcing a misprediction so that it speculatively executes malicious code reading outside of the program's own memory that would have normally crashed if the prediction was correct. When it does this the unauthorised memory read will pull that data into cache, however, when the CPU rolls back as it predicted wrong, it doesn't clear the cache. This means they can then read back a bit they wrote to determine if it is in the cache or not, if it is they can read a byte successfully from another program's address space. If you keep repeating this procedure you can effectively read another program's memory for passwords etc or a ROP style attack. Luckily this is quite hard to do out of a testable environment where all variables are known. I'd highly recommend reading the spectre paper.

 

[looz]

https://www.epicgames.com/fortnite/forums/news/announcements/132642-epic-services-stability-update