P3DV5 HT ON or OFF with 10700K

[DaedalusX 11]

Just wondering if there is a new community consensus on the use of HT in P3D. It used to be strongly discouraged but maybe it's changed now ?

I run a 10700k with all 8 cores at 5.0 ghz with idle temps at 33-35. If I turn HT ON my temps are pretty much the same. So do you have to set a specific affinity mask or does P3D now recognize HT properly ?

[Bob Scott 7,716]

Your idle temps aren't really a measure of much...the trick is to keep temps within bounds and the CPU stable with it under load, as it puts more stress on your overclock...higher temps, more voltage droop etc.

P3Dv5 appears to spread out the workload of the simulation "core" (not to be confused with a hardware CPU core) workload to LPs 0, 2, and 4, as opposed to all of it on core 0 in v4.x and earlier versions.  I run HT off on a 10-core 10900K, but if I were to run with it on, I would probably try it first using the CPU's ability to do per-core HT and only enable it on the cores beyond core 4 (in your case cores 5-7) and see if it makes any observable difference.

[mrlip 56]

On a 8 core Ryzen 3700X I run with SMT off, no affinity mask other than the profile specific one FSlabs sets 

[DaedalusX 11]

It does seems that at 5.1ghz under heavy load it is more stable with HT OFF than ON - With HT ON, even though the temps under load are similar, I do get the odd CTD under prime95.

[Bob Scott 7,716]

13 minutes ago, DaedalusX said:

It does seems that at 5.1ghz under heavy load it is more stable with HT OFF than ON - With HT ON, even though the temps under load are similar, I do get the odd CTD under prime95.

Typically, HT off will allow you to get a 100-200 MHz bump in best overclock at a given set of voltage and LLC settings, at least if your CPU cooler is keeping up.  And conversely, if you did your OC with it off and then turn it on, then you may have to nudge your OC down a notch or two.

I also stopped using P95 as my go-to stress test...I think it's much too aggressive, especially if you run it with AVX instructions enabled.  I use RealBench now for a more realistic stress test.

[DaedalusX 11]

27 minutes ago, w6kd said:

Typically, HT off will allow you to get a 100-200 MHz bump in best overclock at a given set of voltage and LLC settings, at least if your CPU cooler is keeping up.  And conversely, if you did your OC with it off and then turn it on, then you may have to nudge your OC down a notch or two.

I also stopped using P95 as my go-to stress test...I think it's much too aggressive, especially if you run it with AVX instructions enabled.  I use RealBench now for a more realistic stress test.

Thanks. I do prime95 with all AVX turned off. My cpu seems to max out at 72deg so I think I should be fine with a RTX3080 FE even if it'll add a few degs. Cooler is N-D15

[Kilo60 660]

So, is it best to still go for INtel chipsets for the single core (Core 1) performance over AMD Ryzen's which are better at multi-core applications?

[Afterburner 374]

I have a six core AMD Ryzen CPU with HT on, and when I fly over heavy scenery, I see core/thread 0 being at max, but all of the other threads (out of 12) also work hard. If you overfly sparse scenery, most of the other cores run at low %. In FS2020, I see four threads running high, while the rest are almost idling.

[G-RFRY 1,708]

I am bedding in my upgrade at the moment with HT on i want to run the sims and burn-in with AI learning before i OC but i will test with it off as well. i changed my case as well to fit 360 cooler and fitted Noctua industrial 140mm case fans, runs cooler than i expected.  

Edited September 15, 2020 by G-RFRY

[downscc 1,327]

I built an i7-10700K workstation about three weeks ago, with Noctura NH-U12A air cpu cooler and have run numerous stress and performance tests.  I noticed that P3Dv5 loads core 0/thread 0 fully during normal use with HT enabled and I recall a discussion in LM P3D forum a year or two ago by a LM developer.  I disabled HT on my AS Rock Z490 Extreme4 board and noted that now P3D loaded core 0 fully but note this is double what was observed on half of the core.  I keep HT off.  CPU temps run in the low 60's C during heavy loads, normally closer to 40 C.

LM comments then still apply:

The default behavior is to use all cores. The only purpose of applying an affinity mask would be to disable one or more cores. We don't recommend this, but we recognize that it may be beneficial in some cases. For example, there could be an add-on that communicates with another application. There could also be other unrelated windows applications running, that users don't want to close when they run Prepar3D. Higher fidelity training devices often rely on additional applications to perform simulation work. In that case, the system might be configured to allocate specific cores to specific applications. This can be done via OS settings, but the AM setting in the cfg makes it a bit easier.

The other use case that is often mentioned is to disable the hyper-threaded core that shares a physical processor with the primary thread. Sometimes doing a full workload on the HT core can slow down the primary. This is really chip and workload specific. We updated our job scheduler in v2 or v3 such that it typically uses core 1 without overloading it. In our tests, even with HT on, using all cores was typically better than masking off core 1 after the scheduling changes were made. Since then, we have not recommended custom AM settings.

One last thing I should note is that add-on dlls are considered to be part of the Prepar3D process. The AM settings will be applied to those dlls as well. Add-on developers may be creating their own threads and assigning heavy work loads to them.

Beau Hollis P3D Software Architect
23 Apr 2018

[PurdueKev 159]

I cannot run HT On with P3Dv5 on my 10900k, which is the same as when I was running v4 on my 8700k.  The stutters are horrible with HT On, and I was never able to find an Affinity Mask setting that would make them go away without introducing quite pronounced blurries.  If I run P3D, I have to turn HT off first.  With HT off, no stutters at all.

P3D is the only application where I have to do this - flight sim or otherwise.

(My 10900k I currently run stock, by the way - I haven't taken the time yet to try to overclock it.)

[Bob Scott 7,716]

3 hours ago, PurdueKev said:

I cannot run HT On with P3Dv5 on my 10900k, which is the same as when I was running v4 on my 8700k.  The stutters are horrible with HT On, and I was never able to find an Affinity Mask setting that would make them go away without introducing quite pronounced blurries.  If I run P3D, I have to turn HT off first.  With HT off, no stutters at all.

P3D is the only application where I have to do this - flight sim or otherwise.

(My 10900k I currently run stock, by the way - I haven't taken the time yet to try to overclock it.)

Keep in mind that you have another, possibly better option for HT, and that's to use the 10th gen chipset's capability to do per-core HT.  You might try it with HT turned on for cores 5-9, leaving cores 0-4 with HT off so that the simulation's primary threads on cores 0, 2, and 4 don't compete with other process threads on LPs that would be paired on those cores if HT were enabled on them.  No need to mess with HT in that case...you'd have 15 LPs in use, the first 5 are dedicated physical cores with primary process loads on three of them, and the last 10 are LPs paired-up across the remaining 5 cores.  I keep threatening to try that experiment, but I have a to-do list a mile long before I get around to that kind of tinkering.

[G-RFRY 1,708]

It was on but i am now trying with off in variable boost to test my cooling i have fitted Noctua 140mm case fans in my new case for better cooling, and have found it hitting high core clock. periodically on all cores when in the sim.

  https://www.flickr.com/photos/154944674@N05/50374316433/in/dateposted-public/

 

[PurdueKev 159]

8 hours ago, w6kd said:

Keep in mind that you have another, possibly better option for HT, and that's to use the 10th gen chipset's capability to do per-core HT.  You might try it with HT turned on for cores 5-9, leaving cores 0-4 with HT off so that the simulation's primary threads on cores 0, 2, and 4 don't compete with other process threads on LPs that would be paired on those cores if HT were enabled on them.  No need to mess with HT in that case...you'd have 15 LPs in use, the first 5 are dedicated physical cores with primary process loads on three of them, and the last 10 are LPs paired-up across the remaining 5 cores.  I keep threatening to try that experiment, but I have a to-do list a mile long before I get around to that kind of tinkering.

Yes, I have thought about that, but I just don't want to spend a lot of time customizing things to work with one application. I am not sure how much performance benefit I would realistically notice for P3D from the extra virtual cores anyway (I mean, I have 10 physical cores as it is!).  That's why I haven't bothered to even spend time overclocking the 10900k - when I measure per-core clock speed, the cores run at 4.8 anyway just on Turbo with stock BIOS settings. All I have done is enable XMP for my memory.  It is on my eventual to-do list to see what I could get from an OC (mine is a de-lidded k-processor, after all), but for now I am fine with the performance and I don't have to worry about stability.

The only nuisance is I have to reboot and disable HT before I run P3D (I have BIOS profiles set where only thing different is HT On or Off).

[kevinfirth 1,392]

12 hours ago, w6kd said:

Keep in mind that you have another, possibly better option for HT, and that's to use the 10th gen chipset's capability to do per-core HT.  You might try it with HT turned on for cores 5-9, leaving cores 0-4 with HT off so that the simulation's primary threads on cores 0, 2, and 4 don't compete with other process threads on LPs that would be paired on those cores if HT were enabled on them.  No need to mess with HT in that case...you'd have 15 LPs in use, the first 5 are dedicated physical cores with primary process loads on three of them, and the last 10 are LPs paired-up across the remaining 5 cores.  I keep threatening to try that experiment, but I have a to-do list a mile long before I get around to that kind of tinkering.

or just dont faff with the bios and just apply an AM that replicates that core use? 😉

although with 8+ cores, it's probably more than enough to just disable HT anyway with no performance loss..?