Sidebar Sidebar
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

Intel Chips Have Memory Access Design Flaw and Fix Could Lead to Performance Drop

So - after the 50% hit we just took with High Sierra, now we're expected to absorb ANOTHER 30%?

Thanks to PCID, at least Macs with the Magic Touch Bar will be allowed to manually set the rate of decrease - though, linked with BTFS (Bluetooth Facial Scanning), it could be lowered even further - based on initial reports, though confidence is shaky at best, given what we don't know so far.

I suspect we could end up... possibly using the Potts-Kant method... actually INCREASING speed anywhere from 3-5%, best case scenario.
 
Expect Apple to be sued again.

But on a lighter note, just think about the funny temper tantrum Steve Jobs would have had upon hearing this.

Plus, I don't think Steve Jobs would be in love with Tim Cook's new business model...

Shiny Things, Shiny Things! :D
 
Most Apple users will be unaffected as their desktop products line is a tiny fraction of their users. The overall user base will be smiling.
 
  • Like
Reactions: Stella
CrashX said:
So - after the 50% hit we just took with High Sierra, now we're expected to absorb ANOTHER 30%?

Thanks to PCID, at least Macs with the Magic Touch Bar will be allowed to manually set the rate of decrease - though, linked with BTFS (Bluetooth Facial Scanning), it could be lowered even further - based on initial reports, though confidence is shaky at best, given what we don't know so far.

I suspect we could end up... possibly using the Potts-Kant method... actually INCREASING speed anywhere from 3-5%, best case scenario.
Click to expand...

Who took a 50% performance hit with High Sierra? :confused:
 
  • Like
Reactions: meaning-matters, emmanoelle, chucker23n1 and 5 others
jerwin said:
Apparently PCID can ameliorate the performance penalty, if apple decides to go that route.
You can find out if your cpu is recent enough by using the command line.
Code: 
Typhon:~ jeremy$ sysctl machdep.cpu.brand_string
machdep.cpu.brand_string: Intel(R) Core(TM) i5-4690 CPU @ 3.50GHz
Typhon:~ jeremy$ sysctl machdep.cpu.features
machdep.cpu.features: FPU VME DE PSE TSC MSR PAE MCE CX8 APIC SEP MTRR PGE MCA CMOV PAT PSE36 CLFSH DS ACPI MMX FXSR SSE SSE2 SS HTT TM PBE SSE3 PCLMULQDQ DTES64 MON DSCPL VMX SMX EST TM2 SSSE3 FMA CX16 TPR PDCM SSE4.1 SSE4.2 x2APIC MOVBE POPCNT AES PCID XSAVE OSXSAVE SEGLIM64 TSCTMR AVX1.0 RDRAND F16C
Click to expand...
Yeah, I've been looking for PCID information, which CPU's are implementing that CPU feature. CPU's not having it need to do much more work to isolate the kernel pages, which is the source of the performance problem here. Of course, the kernel patches being worked on now, need to actually make use of the PCID feature for it to make a difference. I hear that some patches are currently even hamstringing AMD CPU's despite them not being affected at all.
 
So, this might mean foreign governments are going to go after INTEL. I’m pretty sure China is not happy with this. And if it is a much more open framework, will ARM now benefit from increased state level investment?
 
Some initial benchmarks using Linux patches are in:
https://www.phoronix.com/scan.php?page=article&item=linux-415-x86pti&num=2

tl;dr

- Synthetic benchmarks look like a terrible joke and are bound to become a laughing stock against Intel.
- A compile benchmark also took a big hit, but could be due to other recent regressions.
- Video encoding took no hit.
- Linux kernel compilation took no hit.
- The PostgreSQL database took quite a hit.
- The Redis in-memory database took quite a hit.

Games are also reported to not take much of a hit.

I think the people that ought to be sweating the most here are those running datacenters, servers, doing virtualization, not typical users importing camera photos, browsing, listening to music, and playing games. I think unfortunately for Intel, they are also guys often with plenty of cash and lawyers...
 
  • Like
Reactions: Wowereit, bwintx, leventozler and 2 others
OldSchoolMacGuy said:
Meh, that flaw is only an issue if someone has physical access to your machine and certain privileges already. That's like giving someone the key to your home and then blaming the lock company when they steal stuff.
Click to expand...
Well, it potentially nullifies the security benefits of not using an admin account as your daily user account.
 
Ah, great. Lose 30% performance across my multiple machines and get a $30 payout from a class action suit two years from now.
 
  • Like
Reactions: AxiomaticRubric
Northgrove said:
Some initial benchmarks using Linux patches are in:
https://www.phoronix.com/scan.php?page=article&item=linux-415-x86pti&num=2

tl;dr

- Synthetic benchmarks look like a terrible joke and are bound to become a laughing stock against Intel.
- A compile benchmark also took a big hit, but could be due to other recent regressions.
- Video encoding took no hit.
- Linux kernel compilation took no hit.
- The PostgreSQL database took quite a hit.
- The Redis in-memory database took quite a hit.

Games are also reported to not take much of a hit.

I think the people that ought to be sweating the most here are those running datacenters, servers, doing virtualization, not typical users importing camera photos, browsing, listening to music, and playing games. I think unfortunately for Intel, they are also guys often with plenty of cash and lawyers...
Click to expand...
Lots of Mac users run Windows in virtualization.
 
It isn't clear to me that this is a CPU "bug" so much as an exploitable design-deficiency which given the complexity of modern microprocessors is inevitable. Certainly still bad for Intel, but they aren't the only modern microprocessor affected (ARM64 is too) because the exploit subverts fundamental behaviour of virtual-memory management. I wouldn't overreact and abandon Intel over this given the information to hand.

The performance cost for Kernel Page Table Isolation is to make calling into the kernel more expensive but most GUI applications, like games, perform their work in user-land and call the kernel relatively infrequently.

If everyone's favourite Linux gaming advocate is to be believed the KPTI changes on Linux do not affect games performance which is a better indication of the affect on typical end-user software. The synthetic tests will be absolutely clobbering the I/O and other subsystems which isn’t how GUI applications usually work. Servers, databases etc. do those sorts of things and that is presumably why the kernel maintainers have jumped on this with such gusto.

On Mac it might affect high-end video/audio/etc workflow, but only if FCPX (or whatever is used) makes a *lot* more system calls than your typical game and I'm sceptical of that.
 
  • Like
Reactions: Narcaz and CaTOAGU
Intel's stock price is actually up today. Go figure. Perhaps some people are already pricing in the sale of a lot of faster upcoming chips without the bug.
 
You must log in or register to reply here.
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.
Top Bottom
Back