cpu Skylake CPUs in some cases more than 50% slower than the ealier generations

[Anghammarad]

6 minutes ago, VegetableStu said:

 

I dunno... one mio for me is 1.000.000, then the next 1.000.000.000 I know as one mrd, 1.000.000.000.000 I learned as billion... well this is in good old germansky with metric counting... the us counts everything different... imperial... perhaps this is why there are naming discrepancies

 

 

[TrigrH]

 

non issue.

[Arika]

1 hour ago, VegetableStu said:

-video-

this makes a whole lot of sense. my IT teacher was British

[Sakkura]

On 18/6/2018 at 2:53 PM, Bananasplit_00 said:

Pretty interesting issue, wonder what made Intel increase the pause delay

Look at the OP, Intel explains it on the slide. It improves performance in some situations, while reducing performance in others.

 

Overall they must have found the positives outweigh the negatives.

22 hours ago, Syntaxvgm said:

I think we're looking at non gaming aspects here. They literally multiplied the pause time by more than 14x times the cycles and that certainly does something. I have no idea why they did this, there's certainly a good reason, but nice of them to break existing code, which imo should never be done unless absolutely necessary.

The slide explains why they did it. And it's not like it breaks existing code, just some of it might run a little slower while other code would run a little faster.

 

14x slower only applies to this particular latency, and code doesn't just consist of PAUSE. The performance impact on real code would always be considerably less than a 14x slowdown.

19 hours ago, huilun02 said:

They made architecture changes knowing it will negatively impact existing code.

Ah, Intel, explains

They explain how it would positively impact existing code too...

[leadeater]

3 minutes ago, Sakkura said:

code doesn't just consist of PAUSE

But that is the only code I know how to write. App release plan foiled, darn.

[Syntaxvgm]

16 minutes ago, Sakkura said:

Look at the OP, Intel explains it on the slide. It improves performance in some situations, while reducing performance in others.

 

Overall they must have found the positives outweigh the negatives.

The slide explains why they did it. And it's not like it breaks existing code, just some of it might run a little slower while other code would run a little faster.

 

14x slower only applies to this particular latency, and code doesn't just consist of PAUSE. The performance impact on real code would always be considerably less than a 14x slowdown.

They explain how it would positively impact existing code too...

It's not just single pausing, its really a problem with spinlocking, and I think it's a bigger problem in some cases than a simple slowdown if it's not patched. 

I think what the blog post shows is broken code to me performance wise, though I expect code written in an inefficient way with this to be effected most. 

Btw for anyone wondering I do believe gen 1 ryzen pause is 8 cycles. 

Relevant 

 

[Mira Yurizaki]

47 minutes ago, Syntaxvgm said:

It's not just single pausing, its really a problem with spinlocking, and I think it's a bigger problem in some cases than a simple slowdown if it's not patched. 

I think what the blog post shows is broken code to me performance wise, though I expect code written in an inefficient way with this to be effected most.

PAUSE only affects spinlocking. If you're not spinlocking, then this shouldn't affect you.

[HunterAP]

16 hours ago, leadeater said:

Any news? No issue, has the issue?

The test app uses .NET Framework, and the patch required for it is not out yet, so for now I can't test it.

.NET Core should be fixed with version 2.0.3, although I haven't tested it.

 

I'm going to keep checking on the .NET Framework updates to see if they do backport a fix for this, but for now I guess I can't test for a solution.