Cockpitbuilders.com

Main => General Discussion Board. => Topic started by: FredK on April 04, 2021, 11:29:55 AM

Title: Hyperthreading
Post by: FredK on April 04, 2021, 11:29:55 AM
FYI...

I posted the following in the P3D forum...may be of interest here as well.....

I know the subject of hyperthreading affect on performance has been debated ad nauseum here and elsewhere.  However I had not personally investigated such for some years now and was curious.

My multi-channel setup provides a unique comparative testing platform.  That is, I have 4 computers for running visuals.....a server and 3 identical hardware clients (I7 8700K CPUs and 1080TI CPUs) used for left/center/right displays. By turning hyperthreading on/off on one of the client computers I can get a real-time simultaneous comparison of performance vs. the others under exactly identical conditions.

I am using P3Dv4.5 with the freeware SymSync with View Groups for multi-channel display (works better than WideView or Opus in my opinion).  I have not upgraded to P3Dv5 since all is working very well for me with P3Dv4.5 and it is a bear to reset 4 separate computers, particularly running over 100 scenery addons.  So I have been as happy as a clam of late....the setup is rock solid.

What I observed for "hyperthreading off" vs. "hyperthreading on" was the following....

- Loading pattern on Core 0 was exactly identical for both.....typically between 80-100%
- Loading on other cores was nicely distributed in each case (higher activity levels noticeable with HT off, which is logical since there are less virtual cores to absorb the work)
- Total CPU usage was consistently lower with hyperthreading on....12-17% for "on" versus 20-35% for "off"
- Total CPU package temperature was lower with hyperthreading "on".....about 10 degrees lower
- No discernable scenery loading difference
- No discernable FPS difference
- No discernable smoothness difference

So my conclusion is that hyperthreading "On" is better, but only because there appears to be less overall stress on the CPU.  I did not see that result in any performance improvement though.  My sliders are generally in the medium-max range.  I did not push sliders to their ultimate upper limits, although it appears I have room to do so.

The tests that I ran were across several of my most stress-inducing airport add-ons (including CYYZ, KEWR, KBOS, KDEN, KSFO). The pattern of performance was pretty much identical in each case.

So I have to conclude from all this that LM has done a pretty good job of off-loading CPU work to other cores...at least virtual cores within the scope of this test.  In that regard I would conclude also it would also best to not use Affinity Masking although I did not actually test such.

Fred K
Title: Re: Hyperthreading
Post by: blueskydriver on April 04, 2021, 01:14:12 PM
Nice research Fred, thank you!

IMHO, since you have the multiple machines, what would happen if you maxed out the RAM?

From what I done in the past, just going with the recommended or best amounts of RAM was not the best choice. Putting the max amount of 192GB in a 10 year old Dell T7500 has proven wonders with MSFS2020.

May I ask what are the max capable amounts of RAM that your motherboards can use?

John
Title: Re: Hyperthreading
Post by: navymustang on April 04, 2021, 01:51:39 PM
Fred - great report, thanks for giving us insight into the mysteries of P3D
Title: Re: Hyperthreading
Post by: FredK on April 04, 2021, 02:07:14 PM
Hi John

The architecture of your Dell T7500 is going to be very different from what I have, so it may be like comparing apples to oranges with what works as far as RAM is concerned.

My visual computers are three and a half years old now, but are still pretty much current regarding the mobo (LGA1151 socket/Z370 chipset). RAM is 16GB total of DDR4 3200. Max RAM possible would be 64GB.

P3D is limited to using 16GB RAM, but there may be value regarding handling background processes running simultaneously.  However from what I have gleaned from the forums is that not much will be gained by adding more.  I might just try it on one computer to see for myself though.

MSFS2020 is a whole different animal however regarding RAM usage.  My original thinking was that I would rebuild all my hardware to ready it for MSFS2020 and that the timing for that would be this summer.  I figured by that time most of the bugs would be worked out and that there would be options for avionics (SIM A or PROSIM) and multi-channel.  All that is not going to be reality any time soon it appears so that is why I am exploring taking what I have to the next level in the meanwhile.

Fred K