Discussion Thoughts on Hyper-Threading removal ?

[bit_user]

For context, on a CPU like the 14900k that has 16ecores, turning HT off lowers performance (in heavy MT scenarios that can take advantage of HT) by about 10%. It's nowhere near 20-25-30% that was mentioned in this thread. That is the case obviously because only 8 of the 24 cores have HT.

Yup, thanks for the perspective. I had said I'd have expected HT removal to coincide with additional cores, but I guess just having some IPC gains + clock speed could cancel it out.

Now I don't know how much space HT takes and what one can do with that space, but I'm pretty sure something better can be done than just increasing performance by 10%.

I dunno... microarchitecture is hard. You could bump up cache sizes a bit. If it were done in a more consequential place, like L2 cache, then maybe you could regain like 3-4% in IPC, at least?

Since I'm primarily gaming, I always turn it off because unless you run out of physical cores, HT just gimps game performance

Due to bad scheduling, as I mentioned.

and increases power draw by a ton at the same time.

Hence, why Apple & ARM haven't bothered with it.

 

[mjbn1977]

For context, on a CPU like the 14900k that has 16ecores, turning HT off lowers performance (in heavy MT scenarios that can take advantage of HT) by about 10%. It's nowhere near 20-25-30% that was mentioned in this thread. That is the case obviously because only 8 of the 24 cores have HT.

Now I don't know how much space HT takes and what one can do with that space, but I'm pretty sure something better can be done than just increasing performance by 10%. Since I'm primarily gaming, I always turn it off because unless you run out of physical cores, HT just gimps game performance and increases power draw by a ton at the same time.

Yes, but depending on the architecture hyperthreading might not be necessary. Its not like they turn it off on Raptor Lake. They will not use hyperthreading on a different architecture. I don't know where the rule is written down that every processor architecture will benefit from hyperthreading. What if they intestinally created the future architectures specifically without it in order to get highest performance. I remember when the first hyperthreaded CPUs came out and everyone complained or was worried that this will prevent the CPU to reach its full potential. It all comes down to the actually performance.

 

[bit_user]

What if they intentionally created the future architectures specifically without it in order to get highest performance.

Before any mitigations against side-channel attacks, there wasn't really a downside to a CPU having the potential for hyperthreading/SMT. CPU cores already have all the machinery to map physical registers to logical ones and track dependencies between different instruction sequences. SMT doesn't really add any big, new requirements there.

What might have changed is the mitigations & protections needed for all the side-channel attacks that have come to light since Spectre & Meltdown.

I remember when the first hyperthreaded CPUs came out and everyone complained or was worried that this will prevent the CPU to reach its full potential.

While a CPU might suffer when actually using SMT, I have yet to hear about the (unused) potential for SMT causing a performance detriment. That might be starting to happen, but I have yet to see/hear good evidence of it.