Stuff being anything but P3D. The thing was to keep anything clear of that main task on the main core allocated (right end).

So with 6 cores +HT  340=00,01,01,01,01,00 the 00 on the right is core zero (LP0 and LP1), the 01 left of that is core 1 with one LP (LP2). That is the main P3D task.

But if you have only four cores and want to use HT then I would use AM 85=01,01,01,01 which gives P3D all four cores same as with HT disabled the natural AM is 15=1111 all four ones.

Then if you can move addons onto AM12=1100 they occupy two cores away from the right hand. With HT enabled AM80=01,01,00,00

With the use of process lasso you don't assert that on P3D (or FSX or any app that has AM control of its own) as the outcome is not what is intended. Also moving stuff around after it started by selecting checkboxes in Task Manager Set Affinity does not do the same thing.  We always only use the given method (of assigning an AM in the JOBSCHEDULER section of the cfg) for those apps (P3D FSX) that have the control..

 

 

 

Edited November 7, 20196 yr by SteveW

So in testing on Monday a ten core with HT and without HT:

HT enabled I used AM340=01,01,01,01,00 four cores one HT LP of each.

HT disabled I used AM30 11110 four cores one LP of each (only one available per core).

I put an aggressive addon (deliberate high CPU load and fully configured simconnect client on AM10240=10,10,00,00,00,00,00 which uses cores away from those four.

The results were almost identical bar for a few percent CPU overhead increased with HT enabled.

Why? With HT enabled the use of double the level one cache per core and halving the thread switching time per core gives that little lift. However without care HT enabled will perform badly.

...additionally. In those tests I increased GP load for the simulator (increased settings) to become GPU bound. In that case very little difference was seen in the CPU configuration because it is waiting on the GPU.

Edited November 7, 20196 yr by SteveW

3 hours ago, HUSSAR said:

That didn't work, sound "skipping" then Prepar3D CTD.  I also have Riva Tuner Statistics Server running at Scanline sync x/2 at 60 which locks frames to 30 FPS, I think it is interfering with Process Lasso.

Something else is wrong.

Changing AMs should never reveal any differences in operation of anything.

To check that try HT enabled, and two LPs of one core with P3D AM=3=11 and you should be able to run the whole shebang on that one core and see no differences in operation.

Edited November 7, 20196 yr by SteveW

4 hours ago, SteveW said:

Why? With HT enabled the use of double the level one cache per core and halving the thread switching time per core gives that little lift. However without care HT enabled will perform badly.

Does the above apply to when mimicing non-HT mode as well with HT on? I.e. 01 01 01 01 01 01 vs say 11 11 11 11 11 01 AM for P3D ? Still better cache use and switching time per core? How come?

Thanks

Edited November 7, 20196 yr by Dirk98

Yes, because even though the P3D tasks are using one LP per core, anything that does happen on the sister LPs will not detract so much from the sim task. If it were only one LP per core, we get one lot of cache and anything else happening on that core disrupts the task. That's why HT is more expensive. But HT creates problems by causing some apps to double their task count and that will burden the CPU anyway.

If we can't see better performance with a game, better to turn off HT. With many cores 6+ we might want to limit a game to using only four and if possible they can be started with a batch file.

Many games are handled by another process, like with some steam games, so are difficult to administer. Pure games machines stick to no more than 6 cores no HT.

Edited November 7, 20196 yr by SteveW