5 ms·
Does Geekbench test for TLB flushes at syscalls or whatever other kind of "mitigation" they introduced in 10.13.2 ? Where can we see true benchmarks of performa
by HugoDaniel 9y ago
Does Geekbench test for TLB flushes at syscalls or whatever other kind of "mitigation" they introduced in 10.13.2 ? Where can we see true benchmarks of performance comparison between previous and current version in regards to this bug ? Isn't Apple being really nasty by exploiting the lack of knowledge of the general public ?
I am no expert in any way but would like to know how did they "mitigated" this without a serious performance hit.
- brigade 9y agoMore like Phoronix doesn’t do any analysis of what they’re benchmarking and the general public quoting them can’t do that for them. Besides, syscalls have historically been slow on OS X; back when the kernel was 32-bit the kernel and user space had completely separate address spaces (unlike windows and default Linux) and I’ll bet even most power users never realized.
- Strom 9y agoYou should try defining this mystical true benchmark that you seek. Apple decided that heavy userspace execution and web performance are representative of common workloads. You are welcome to argue for a different kind of workload. As it stands now, at best you're implying that it's common for users to execute syscalls in a loop.
- HugoDaniel 9y agoCan you tell with any degree of precision what kind of benchmark the "GeekBench" is and does ? What guarantees do you have that it is testing "real world common workloads" ? Do you think common users execute guassian blurs in a loop and ray tracers before perfoming rigid body physics in their "common workloads" ? As you said, I know I am welcome to argue for a different kind of workload, specially when I am being taken for a fool.
- Strom 9y agoI'm not too familiar with GeekBench, but I'm assuming it's userspace code. Whether it traces rays or does something else isn't that important, because the slowdown is with syscalls. As for what common users execute, based on my observations over the years it's mostly web browsing. However my personal vision of common workloads is a distraction. I have made no claims to know common workloads. I merely stated that Apple has chosen a particular set to be representative.