I have to believe there's a chicken bit for turning off speculative execution. But why would you want that, when the available OS level fixes probably provide an order of magnitude better performance? Helping that effort is pretty responsible IMHO.
-
-
Replying to @securelyfitz
There are no OS level fixes for the most serious problem which is P0's "variant 1", and can't be. It doesn't cross OS/cpu privilege domains.
2 replies 3 retweets 6 likes -
Replying to @RichFelker
Remember the iphone battery deal last month? Imagine the outcry if Intel pushed patches that dropped your cpu to 5 % of it's current performance. I think others might value the trade-off differently from you.
2 replies 1 retweet 15 likes -
Replying to @securelyfitz @RichFelker
I’m trying to consider how most the internet performance would have plummeted if AWS or Azure had just turned of speculative execution. Almost immediately any infrastructure within 30% of their capacity planning would have browned out. An hour later the death spiral starts.
1 reply 0 retweets 0 likes -
Replying to @JonathanMHenson @RichFelker
Entirely turning off speculative execution is not a 30% hit. More realistically it would be a 90+% hit. That's why it's not a viable solution.
2 replies 0 retweets 1 like -
Replying to @securelyfitz @JonathanMHenson
Yes, it'd likely be like the P4 disaster with deep pipelines constantly getting flushed.
1 reply 0 retweets 0 likes -
But maybe not as bad. Modern Intel chips are nowhere near as bad as P4 under fully unpredictable load (effectively random branches).
1 reply 0 retweets 0 likes -
So a typical hit of "only" 30% might be realistic.
3 replies 0 retweets 0 likes -
Replying to @RichFelker @JonathanMHenson
I like the optimism, but we've got decades of compilers optimizing to hint speculative execution instead of optimizing the code itself. It'd be nowhere as bad as P4 since the pipeline is shorter, but you're talking about a full stall, flush, & waiting for _every_single_branch_
3 replies 0 retweets 1 like -
Replying to @securelyfitz @RichFelker
My guess is that OOP may be hiding the performance hits here. And that since the world of application level engineering level programming is OOP maybe the visible impact is mitigated.
1 reply 0 retweets 1 like
Indirect calls are predicted very well on modern cpus. That's actually the vector for Spectre variant 2.
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.