Intel Gags Customers from Publishing Performance Impact of Microcode Updates

[jigar2speed]

I would love to see TPU publish comparison results. - TPU please check if this legally doable and can you make it happen ?

 

[bug]

I would love to see TPU publish comparison results. - TPU please check if this legally doable and can you make it happen ?

It's probably legal, but it will certainly bar TPU from getting samples in a timely manner in the future

 

[R0H1T]

Phoronix benches, in some cases an unmitigated disaster

Spoiler

https://www.phoronix.com/scan.php?page=article&item=l1tf-foreshadow-xeon

 

[hyp36rmax]

Even then it will be hard(er) to enforce outside the US, at least it should be. And if it's not, then someone should just leak server benchmarks to reddit or pastebin.
There's no reason why the performance impact should not be visible to everyone, it's like Intel owns us now!

Shouldn't be as hard to enforce like we think it may be. Intel is large enough to possibly have registered offices and extensions in each major market to mitigate and control their IP's for exactly this. The company I work for has a similar approach. Our regional offices are registered separately and can be considered it's own entity.

In the whole scheme of things it's still ridiculous but probably happens all too often, we just happen to highlight and make this particular case a big deal.

 

[TheOne]

https://www.tomshardware.com/news/intel-cpu-microcode-benchmark-mitigation,37684.html

Intel is apparently updating the EULA.

 

[HD64G]

And this proves that AMD has't any big problem to patch any vulenrabilities without losing performance, while Intel is trying to keep people silent from proving that they lose heavily with each new patch. Zen 2 will have a great chance to turn tables around muchly for both server and pc market me thinks.

Phoronix benches, in some cases an unmitigated disaster

Spoiler

https://www.phoronix.com/scan.php?page=article&item=l1tf-foreshadow-xeon

WOW! We are talking about tanking performance full steam! Return to 2005 all the way! I can understand their panic watching these results. AMD stock will hit all time high next year imho, while Intel's will fall heavily after the Zen2 server contracts are being revealed.

 

[cadaveca]

WOW! We are talking about tanking performance full steam! Return to 2005 all the way! I can understand their panic watching these results. AMD stock will hit all time high next year imho, while Intel's will fall heavily after the Zen2 server contracts are being revealed.

I think you missed the part that both HT and SMT must be disabled for full mitigation of this "flaw", which means nearly half of AMD's CPUs performance will be lost as well. But good job on spinning it to match what you'd like.


Also, this is far from over. we can expect nearly 60-70% performance loss in some scenarios with all of this mitigations for these current crop of hardware flaws fully in effect. most of users here might be 5% loss though on average usage models. Most high performance losses are very specific scenarios, like older software from 5+ years ago, and things like OpenGL.

You've very easily highlighted WHY INtel has taken this stance on publishing benchmarks. Kudos to you, sir!

 

[bonehead123]

tag it, snag it, gag it.....

typical Intel yip yap......only in America

 

[HD64G]

I think you missed the part that both HT and SMT must be disabled for full mitigation of this "flaw", which means nearly half of AMD's CPUs performance will be lost as well. But good job on spinning it to match what you'd like.


Also, this is far from over. we can expect nearly 60-70% performance loss in some scenarios with all of this mitigations for these current crop of hardware flaws fully in effect. most of users here might be 5% loss though on average usage models. Most high performance losses are very specific scenarios, like older software from 5+ years ago, and things like OpenGL.

You've very easily highlighted WHY INtel has taken this stance on publishing benchmarks. Kudos to you, sir!

So, where are the tests that show AMD CPUs need to disable SMT? And since FX arch CPUs aren't vulnerable to that as they don't have HT, they will come very close to the i7s until series 5 that will disable their HT maybe? And all those who paid double or more will have to make their i7s to perform as i5s? Or allow their PC to be vulnerable just to not lose the performance they paid for? 3rd way is to go for Ryzen me thinks.

 

[Assimilator]

FFS Intel, stop being stupid and fix your 10nm node.

 

[R-T-B]

I'm not talking about big data or cloud per se, the updated ToS also seems to target individuals.

They don't. When is the last time an Intel ToS popped up on a microcode update for you?

Hint: If you didn't agree, it means nothing.

btarunr Take a look at these for instance

https://www.servethehome.com/intel-publishes-l1tf-and-foreshadow-performance-impacts/

You do realize that's only for guest os systems WITHOUT the microcode that you can't trust? Using software mitigations?

Way to cherry pick... They are benching VM performance without the microcode vs with for pete sake.

OK, it says CUSTOMERS...

What about tech Journalists and websites publishing reviews or opinion pieces? Is that covered by this, or is this next on the list?

And what is TechPowerUp's response to this?

Only enterprise.

 

[HTC]

Dunno about you dudes but i'd like to see a set of benches that showed no mitigations 1st, then mitigations for those security issues that affect both Intel an AMD, and finally all mitigations enabled.

Just to see the cumulative effect these security issues have caused.

 

[R-T-B]

You've very easily highlighted WHY INtel has taken this stance on publishing benchmarks. Kudos to you, sir!

I said it earlier... The social web in general has begun to scare me. People care not to learn the truth, only who to blame.

Just to see the cumulative effect these security issues have caused.

Outside of a VM this latest one has no impact.

 

[mtcn77]

I think you missed the part that both HT and SMT must be disabled for full mitigation of this "flaw", which means nearly half of AMD's CPUs performance will be lost as well. But good job on spinning it to match what you'd like.


Also, this is far from over. we can expect nearly 60-70% performance loss in some scenarios with all of this mitigations for these current crop of hardware flaws fully in effect. most of users here might be 5% loss though on average usage models. Most high performance losses are very specific scenarios, like older software from 5+ years ago, and things like OpenGL.

You've very easily highlighted WHY INtel has taken this stance on publishing benchmarks. Kudos to you, sir!

HT and SMT don't do similar things. If you look over at "LanOCLanOC.org.org";

what Intel gains from L1-HT is mainly bandwidth, a.k.a. execution speed whereas AMD's L2 - which is what physical cores get - is not slower than L1-SMT apart from latency, 2.8ns>1.1ns. With the distinction of i9's AMD's physical cores are faster and more patent than Intel.

 

[HTC]

I said it earlier... The social web in general has begun to scare me. People care not to learn the truth, only who to blame.



Outside of a VM this latest one has no impact.

Then do them benches with that one both inside and outside of VMs, and the same for other (if any) such applicable instances.

 

[hat]

It's a lot to keep track of. There's been a lot of holes to patch, and those patches have come in varying quality.

 

[dorsetknob]

And what is TechPowerUp's response to this?

They can write any shit in their TOU's, doesn't mean it's enforceable.

Intel is apparently updating the EULA.

https://www.theregister.co.uk/2018/08/23/intel_microcode_license/

Intel HAS BACKED DOWN

 

[R-T-B]

Then do them benches with that one both inside and outside of VMs, and the same for other (if any) such applicable instances.

Who does? The benches I quoted from intel don't.

EDIT: my bad, was on a phone and misread "then" as "they."

 

[Dr_b_]

But guys Intel is going to make a dGPU in 2020 that will be so awesome. And you won't be able to benchmark it either, under penalty of death.

 

[eidairaman1]

As far as I recall non-disclosure agreements we're only for when people were testing products before they released this what they are trying to do now we should have a right to be able to see how much of a impact any patch has on any piece of hardware Intel is obviously trying to hide.

Watch as their user base starts dwindling because of such underhanded tactics.

 

[GoldenX]

Looks like they are rolling back.

 

[R0H1T]

They don't. When is the last time an Intel ToS popped up on a microcode update for you?

Hint: If you didn't agree, it means nothing.


You do realize that's only for guest os systems WITHOUT the microcode that you can't trust? Using software mitigations?

Way to cherry pick... They are benching VM performance without the microcode vs with for pete sake.



Only enterprise.

They're Intel's own performance numbers, for servers - hence servethehome link.
So it is possible that the terms above could be used to restrict internal benchmarks being showcased against the competition - for servers & even if it is just enterprise - that's still wrong.

Just to be clear, the language is broad enough to target anyone downstream, so it's not as if only enterprsies are at risk.
This at a time when the competition is gaining ground on them seems less of a coincidence to me.

 

[cyneater]

This is a new low.
I hope someone sues them.
This is disgusting.

Yup a new fixed cpu and mobo for every use effected
Then I would forget..
ill buy the new ram sounds fair to me.

 

[R-T-B]

for servers & even if it is just enterprise - that's still wrong.

No disagreement. Just clarifying.

Just to be clear, the language is broad enough to target anyone downstream

No it's not. Language is irrelevant if you never digitally agree to it.

 

[$ReaPeR$]

Yes, that why we have so many strong economies developed in the absence of a healthy stock market

do you remember the 2008 crash? i do. maybe you were born yesterday tho.. so yeah.. "healthy"..