|
intel cpus have a bad fuckup: https://www.theregister.co.uk/2018/01/02/intel_cpu_design_flaw/ it looks like the bug is in their branch predictor and allows user space programs to read kernel data and do other awful things. there's no microcode fix for it, the fix has to be applied at the OS level and it's going to have some potentially big performance impacts for some intel CPU workloads. the reductions can be as bad as 30% surprisingly, amd (says their) cpus arent't affected: quote:From: Tom Lendacky <thomas.lendacky@amd.com> amd would be in prime position to sell a bunch of new hardware except they still don't have epyc cpus available in the general supply chain. more news about the exact bug with intel cpus is expected later this week. maybe more of that news will get posted here or maybe this thread will die because yospos is dead
|
# ? Jan 3, 2018 02:46 |
|
|
# ? Jun 8, 2024 22:03 |
|
also, i managed to gently caress up more than intel and amd in the title but gently caress it
Number19 fucked around with this message at 04:33 on Jan 3, 2018 |
# ? Jan 3, 2018 02:47 |
|
Intel will probably have a boon in new CPU shipments from this as they can advertise speed improvements rebased to the neutered performance levels.
|
# ? Jan 3, 2018 02:51 |
|
It was bound to happen. When you're doing crazy optimizations in your architecture to squeeze out a few more percent it's going to open you up to new bugs
|
# ? Jan 3, 2018 02:54 |
|
intel will end up selling 30% more cpus to make up for the lost perf lol
|
# ? Jan 3, 2018 02:57 |
|
glad to see something besides iot get turbo-hosed
|
# ? Jan 3, 2018 03:07 |
|
how long does it take to do a hardware revision? is this even going to get fixed in coffee lake or whatever the next cpu is (ice lake?) it sounds like the branch predictor is busted and that might not be very easy to fix?
|
# ? Jan 3, 2018 03:11 |
|
lol. goldmine
|
# ? Jan 3, 2018 03:22 |
|
using a coffee lake cpu right now so my guess is they're going to make yet another "add an option, call it super" revision to 1166 and everyone on a different pga can get hosed in amd's position i'd give up on ryzen 2 and just shovel as many of whatever they've already drawn out the door like, now
|
# ? Jan 3, 2018 03:29 |
|
how long has NSA known about this lol
|
# ? Jan 3, 2018 03:32 |
|
Number19 posted:how long does it take to do a hardware revision? is this even going to get fixed in coffee lake or whatever the next cpu is (ice lake?) it sounds like the branch predictor is busted and that might not be very easy to fix? jawn or some other intel goon can correct me, but 'hardware revision' in this case i think means 'wait for the next uarch' mask change would be hilariously loving expensive (and a major stepping rev) and i feel like this isn't a thing you can go and fib out / cut a couple of wires bonus mega-hosed: validating the above
|
# ? Jan 3, 2018 03:35 |
|
It mostly affects synthetic benchmarks. It looks like the biggest impact are in databases. which, lol
|
# ? Jan 3, 2018 03:35 |
|
AMD is back baby. its good again. oowooao (concurrency howl)
|
# ? Jan 3, 2018 03:38 |
|
pro-click on branch prediction: https://danluu.com/branch-prediction/
|
# ? Jan 3, 2018 03:39 |
|
looks like my decision to stick with a 10+ year old amd cpu is finally paying off
|
# ? Jan 3, 2018 03:40 |
|
Elder Postsman posted:looks like my decision to stick with a ppc cpu is finally paying off
|
# ? Jan 3, 2018 03:43 |
|
let’s just speculate into kernel code, what could go wrong? and now Linux is unmapping the kernel on every context change, this is going to hurt. there is no way this is going to be fixed in hardware in a few weeks or months, it’s a major architecture change.
|
# ? Jan 3, 2018 03:50 |
|
movax posted:jawn or some other intel goon can correct me, but 'hardware revision' in this case i think means 'wait for the next uarch' new enough intel CPUs have a feature called PCID that allows them to swap between sets of TLB entries or something (haven't read much about it yet), so if the hardware and OS both support PCID you can mitigate the flaw with a much more minor performance impact
|
# ? Jan 3, 2018 03:59 |
|
lol
|
# ? Jan 3, 2018 04:09 |
|
can't they just market their insecure cpus as now being 30% faster
|
# ? Jan 3, 2018 04:10 |
|
this is way beyond me
|
# ? Jan 3, 2018 04:14 |
|
movax posted:pro-click on branch prediction: https://danluu.com/branch-prediction/ pro click
|
# ? Jan 3, 2018 04:23 |
|
Number19 posted:there's no microcode fix for it lmao rip my existing processors, I guess but also MALE SHOEGAZE posted:this is way beyond me
|
# ? Jan 3, 2018 04:25 |
|
movax posted:jawn or some other intel goon can correct me, but 'hardware revision' in this case i think means 'wait for the next uarch' the next uarch might be hosed too. it depends on how long intel has known about this and when they started designing a fix. also if the fix is easy or something deep in their architecture that’s going to take a lot of work to correct
|
# ? Jan 3, 2018 04:38 |
|
man I hope lowtax has sa running on a bunch of loving raspberry pi s
|
# ? Jan 3, 2018 04:48 |
|
Best Bi Geek Squid posted:man I hope lowtax has sa running on a bunch of loving raspberry pi s bring back mips good enough for your router good enough for the cloud
|
# ? Jan 3, 2018 04:56 |
|
im betting on risc v
|
# ? Jan 3, 2018 05:01 |
|
more like Branch Davidian
|
# ? Jan 3, 2018 05:12 |
|
hale nop
|
# ? Jan 3, 2018 05:16 |
|
flakeloaf posted:hale nop
|
# ? Jan 3, 2018 05:26 |
|
Raere posted:im betting on risc v lol
|
# ? Jan 3, 2018 06:04 |
|
how much will this affect my FPS in shitteo shames?
|
# ? Jan 3, 2018 06:08 |
|
DuckConference posted:new enough intel CPUs have a feature called PCID that allows them to swap between sets of TLB entries or something (haven't read much about it yet), so if the hardware and OS both support PCID you can mitigate the flaw with a much more minor performance impact I forgot about those; need an effort post on that stat
|
# ? Jan 3, 2018 06:42 |
|
movax posted:I forgot about those; need an effort post on that stat it’s just ASID, right?
|
# ? Jan 3, 2018 07:03 |
|
Beast of Bourbon posted:how much will this affect my FPS in shitteo shames? lol. many, many people will never install a security patch ever again
|
# ? Jan 3, 2018 07:28 |
|
so this is probably exploitable if you get execution inside a browser sandbox process, yeah?
|
# ? Jan 3, 2018 07:30 |
|
Beast of Bourbon posted:how much will this affect my FPS in shitteo shames? not too much apparently: https://www.phoronix.com/scan.php?page=news_item&px=x86-PTI-Initial-Gaming-Tests it seems like its IO heavy stuff https://www.phoronix.com/scan.php?page=article&item=linux-415-x86pti&num=1
|
# ? Jan 3, 2018 07:42 |
|
Trimson Grondag 3 posted:not too much apparently: Oh, good, so just literally everything that I work on.
|
# ? Jan 3, 2018 07:48 |
|
yeah its going to be great. next week every cloud provider gets to tell their customers they are getting 20% less boom than they were the week before for the same money.
|
# ? Jan 3, 2018 07:51 |
|
|
# ? Jun 8, 2024 22:03 |
|
Trimson Grondag 3 posted:yeah its going to be great. next week every cloud provider gets to tell their customers they are getting 20% less boom than they were the week before for the same money. there is no accountability in the world, Intel will continue to sell things and profit
|
# ? Jan 3, 2018 07:53 |