Intel Hyper Threading Performance With A Core i7 On Ubuntu 18.04 LTS

[Weasel]

So Linux prefers clearly performance over security. OpenBSD on the other hand prefers clearly security over performance. I'm with OpenBSD on this. You seems to be with different OS probably. What's the problem? And, btw, OpenBSD is not niche in BSD. :-)

You're not going deep enough with the security first principle in my opinion. You should take out your CPU completely from the motherboard. Then with zero processing power, no security threat will ever exist or run and you can claim perfect safety. Win!

In all seriousness, security is great and all, until it impedes functionality or performance, which are more important, and the fact that you're not running a single-threaded in-order CPU proves it (in other words, a god awfully slow CPU like the original Atom). We make compromises between security and functionality/speed, because the latter are way more important for most people.

[starshipeleven]

In all seriousness, security is great and all, until it impedes functionality or performance, which are more important, and the fact that you're not running a single-threaded in-order CPU proves it (in other words, a god awfully slow CPU like the original Atom). We make compromises between security and functionality/speed, because the latter are way more important for most people.

HT does not provide a so large performance boost, and it has some drawbacks too (apart from safety) so it all depends from what the user values more.

I've been disabling HT in Windows gaming systems since a while back, and it actually was better for performance (because the process scheduler apparently is too stupid to actually fill first one thread of each core and then start filling the other.

[austin754]

Not completely surprising.

First, I think the reason graphics-magick and vpxenc see little or no gains is that they are naturally limited by parallelism and throwing even more virtual cores at the situation isn't going to help. For example on my older i7-4770 system with 4 physical cores and 8 hyper-threads, I see the cores only scheduled 42% for vpxenc and 30% for graphics-magick. This is total user+system time divided by elapsed time. So while these benchmarks have multiple threads, I wouldn't expect them to scale with more physical cores either. Your other benchmarks are mostly close to 100% scheduled on the CPU cores. Similarly, I wouldn't expect single-threaded benchmarks to be affected at all (perhaps be just the tinyest bit faster with hyper-threading off if you haven't pinned them to a single core and they jump between cores and caches).

The i7-8700 is 5-wide decoder and has matching functional units for a lot of processing in parallel. Your test results suggest to me that these benchmarks can't always take advantage of that full parallelism and for a parallel program at least, it is better to share two threads worth of this decoder/functional units/etc because the algorithm benefits more from the extra thread than the incremental slowdown of sharing. Where I do wonder is about newer cores with fairly wide issue and without hyper-threading.

I could hypothesize of programs that theoretically down because they are either thrashing some shared resource e.g. cache, power; or because nominally the program speeds up only slightly with more threads but otherwise thrashes slightly more. For example, I'd expect stream to not speed up and perhaps even slightly slow down. I don't expect most Phoronix programs (often smaller) to fit that case.

--mev

[creative]

Ubuntu Studio 16.04 4.15 low latency rt kernel: i7 7700 65watt

Shadow of Mordor: 95fps HT off. 94fps-HT on.
HT on showed 5fps higher minimum frames per second. <<-Deal Breaker

Tomb Raider: Showed no difference as for avg fps.

I will leave on HT due to higher lower minimum FPS.

As for Windows 10 I can not say. I rarely boot into that mess of a 25GB base install after updates. Right now its only redeeming factor is Skyrim SE and a few other games.

Between Linux and windows, Linux has the highest amount of installed games per two distros on station. Really kind of ready to get rid of Windows again, its novelty is seriously starting to wear off. Years and years of Linux only tends to do that to me.

[creative]

As for HT for overall system performance being on, the amount of applications/process's your system can run at the same time? Answer? HT on for me. Even though I don't like to run a ton of things at once its nice to know some headroom is there.

Why buy a pricey hyper-threaded processor to gimp it later. Its like buying a shiny new wheel barrel to warp one of the handles out of shape and let some of the air out of one of the tires and warp a lower truss.

That three legged dog will have some difficulty if it goes on a run with you.

And if Intel proves to get stranger and stranger, its back to AMD for me, I think...... Seems I remember more segmentation faults with all the AMD chips I had or maybe oh maybe.. I almost feel like going to my gaming consoles again, and recording using a Tascam DP-008EX for recording and doing much more drawing and painting.

[gamerk2]

Ubuntu Studio 16.04 4.15 low latency rt kernel: i7 7700 65watt

Shadow of Mordor: 95fps HT off. 94fps-HT on.
HT on showed 5fps higher minimum frames per second. <<-Deal Breaker

Tomb Raider: Showed no difference as for avg fps.

I will leave on HT due to higher lower minimum FPS.

As for Windows 10 I can not say. I rarely boot into that mess of a 25GB base install after updates. Right now its only redeeming factor is Skyrim SE and a few other games.

Between Linux and windows, Linux has the highest amount of installed games per two distros on station. Really kind of ready to get rid of Windows again, its novelty is seriously starting to wear off. Years and years of Linux only tends to do that to me.

Games really won't benefit in terms of max FPS, but you will get a boost to minimum. This makes sense when you consider that games still use only a handful of heavy threads; the i7 has enough cores where HTT doesn't make much of an impact. Where it does help is preventing lower priority threads (system stuff) from temporarily bumping one of your game threads to get work done (say, updating the system clock). The processing time is small, but does add latency to the program you are trying to run. That's where HTT helps out a ton, even if you can't measure it on the FPS.

[F.Ultra]

"Tests were done on Ubuntu 18.04 LTS x86_64 with its default Linux 4.15 kernel."

It would be interesting to see if there is any actual difference between the kernel disabling HT and the bios disabling HT. Difference being, that HT in software has overhead, so if you don't compile the HT software in the kernel, perhaps the kernel can work slightly better since the scheduler has different expectations?


CONFIG_SCHED_SMT:

│ SMT scheduler support improves the CPU scheduler's decision making
│ when dealing with Intel Pentium 4 chips with HyperThreading at a
│ cost of slightly increased overhead in some places. If unsure say
│ N here.

│ Symbol: SCHED_SMT [=n]
│ Type : bool
│ Prompt: SMT (Hyperthreading) scheduler support
│ Location:
│ -> Processor type and features
│ Defined at arch/x86/Kconfig:1009
│ Depends on: SMP [=y]

There is no "HT software" if you disable it in BIOS or if you disable it in the kernel you still disable HT from happening. The kernel config that you have found there is if the kernel scheduler should be HT-aware or not (i.e if it when deciding which core to put a thread on should e.g not schedule it to another virtual core of the same physical core and so on).

[Weasel]

There is no "HT software" if you disable it in BIOS or if you disable it in the kernel you still disable HT from happening. The kernel config that you have found there is if the kernel scheduler should be HT-aware or not (i.e if it when deciding which core to put a thread on should e.g not schedule it to another virtual core of the same physical core and so on).

Also, the BIOS *is* software. Software that runs before the OS, but it's still software, so if it can disable HT, the OS can as well (unless the CPU prevents it later down the line only).

[bridgman]

almost feel like going to my gaming consoles again, and recording using a Tascam DP-008EX for recording and doing much more drawing and painting.

Neat... I didn't know affordable portable digital studios like that existed. When I first read your post I imagined something like the old Teac I always wanted back when I had time for playing & recording:

http://www.thevintageknob.org/teac-A-3340S.html

[coder]

A few weeks ago I did Geekbench benchmarks on Windows with HT enabled and disabled; multi-threaded performance was slightly higher with HT on, but single-threaded performance was slightly higher with HT off. The numbers weren't anything extremely different though.

Certain benchmarks seemed to show this when Skylake was launched. I'm a little surprised not to have seen more about it.