1 hour ago, Schlotterknie said:

A suitable combination of HT on/off and your AM setting will most likely lower VAS consumption significantly. You can see a potential gain in FPS as a positive side effect.

That seems straight-forward enough. Thanks. Might be a couple of days before I try this. Weather is too nice to be stuck indoors.

Mike, I was confused. I do have a 1080 and the 376.33 drivers seem happy with it.

59 minutes ago, Ray Proudfoot said:

Mike, I was confused. I do have a 1080 and the 376.33 drivers seem happy with it.

Hey, that's good news. Nothing should stop you now...go Ray!

I agree, disabling CPU Hyperthreading should save you some VAS, maybe enough to meet your needs. As to whether it will improve performance and/or frame rate with your quad core, powerful though it may be, remains to be seen. Either way the outcome should be interesting. Keep us posted.

Regards,

Mike

9 hours ago, Cruachan said:

Actually, on a 6-cored CPU I think the AffinityMask 340 would appear as 00,01,01,01,01,00 leaving 2 Physical cores (4 unmasked virtual cores) free for allocation by the Windows Jobscheduler to Addons. This is in addition to the 4 unmasked virtual cores on each of the other 4 physical cores.

Reading from right to left you have the binary representation for cores 11,10,9,8,7,6,5,4,3,2,1,0. With AM=340 Prepar3D will use cores 2,4,6 and 8.

Mike

Very well explained Mike.

I did months of HT on/off - AM testing and  HT on + AM 340 was the most smooth I was able to get.

My addons I have assigned through Proces Lasso.

Sometimes a certain AM value seemed better at an airport till I was flying in rain with multiple cloud layers and then started all over again.

Steve W helped this community a lot on this subject with his knowledge. He personally advised me several times..

I've read that excellent article by Steve and for the most part he advises HT should be left on and AM set to what is best for your CPU and system. I also need a scenario to ensure the test is consistent so I can easily compare results.

But in my search I also came across this topic and wondered if you chaps have read it. It's explained really well.

https://www.avsim.com/forums/topic/444793-one-tweak-to-rule-them-all-how-to-set-affinity-mask-correctly/

 

1 hour ago, Ray Proudfoot said:

But in my search I also came across this topic and wondered if you chaps have read it. It's explained really well.

Hi Ray, just bear in mind that this post was from 2014 and the guy was using ver 2.2, I sure LM have done a lot of work on the coding since.

bob

47 minutes ago, onebob said:

Hi Ray, just bear in mind that this post was from 2014 and the guy was using ver 2.2, I sure LM have done a lot of work on the coding since.

bob

Understood Bob but same principles apply. Turn off AF and observe which cores are most used. Maximum of 4 seems to be best.

29 minutes ago, Ray Proudfoot said:

Understood Bob but sabut diinciples apply. Turn off AF and obs erve which cores are most used. Maximum of 4 seems to be best.

I have similar system to you but with a 980ti, tried all AM's but Got good results with am=85 (4 cores).

Dont use am now, use prosess lasso , cores 0, 2,4,6 and all add ons on 3 and 5.

I find that the best for a st helens lad.

Bob

 

 

 

 

 

6 hours ago, onebob said:

I have similar system to you but with a 980ti, tried all AM's but Got good results with am=85 (4 cores).

Dont use am now, use prosess lasso , cores 0, 2,4,6 and all add ons on 3 and 5.

I find that the best for a st helens lad.

Bob

 

 

 

 

 

I remeber that Steve did recommend not to use PL for P3D itself, but to use an AM for it..

1 hour ago, GSalden said:

I remeber that Steve did recommend not to use PL for P3D itself, but to use an AM for it..

I think you misunderstood Steve. He mentioned that P3D ran best on 4 physical cores, and PL was good for assigning your add-ons to certain ones at your preference that you should determine yourself on your machine. Something along these lines.

Dirk.

3 hours ago, Dirk98 said:

I think you misunderstood Steve. He mentioned that P3D ran best on 4 physical cores, and PL was good for assigning your add-ons to certain ones at your preference that you should determine yourself on your machine. Something along these lines.

Dirk.

No, I did not misunderstood Steve. He really advised to keep P3D outside of PL.

Perhaps he is around here and can react on this.

The discussion was about 4 core, 6 and 8 core processors. Steve and other in the thread suggested that P3D runs better on 4 cores, not more and not more, so without HT on a 4-core processor 85 was the best choice or no affinity at all. However if add-ons were used as Steve mentioned  in combination with HT and/or more >4 core processor, PL was one of the good tools to redistribute the load on cpu cores. Steve will chime in here I hope. Nothing negative whatsoever was said about using PL.

Dirk.

 

Edited May 11, 20179 yr by Dirk98

Hi guys, Just quickly: When P3D v3 is started with no AM and HT enabled it will make use of a four core CPU as if it had 8 cores - they are 8 logical processors (LPs) over four cores. That means there will be two processes per core rather than one, the sim splits out over eight main processes rather than four. With the overhead of the sim and the system managing the extra processes we have more VAS consumed and also cores unnecessarily sharing compute throughput with other processes that would work better with a core to themselves. And so we use AM=85 with the four core+HT and AM=340 with the six core+HT to restrict the sim to four processes. Now also, each thread of a 32bit process consumes an extra little bit of VAS due to the WOW64 processes, it's not a lot, a few hundred Mb at most. In other words with or without an AM the VAS is similar within a few hundred Mb.

When we use Process Lasso with P3D and we don't specify an Affinity Mask (AM) for the sim it starts up on all eight LPs (on a four core), then PL asserts an affinity and those eight processes simply move and gang up onto the fewer logical processors we assigned - it's not a good idea to do that because the sim has configured for the CPU it 'sees' when it starts. *Any app that handles its own affinity - let it do it.*

VAS issues may instead be caused by modules (dll) or gauges (gau) incorrectly programmed that do not release used memory allocations, memory space is gradually consumed and never released. Exe apps won't cause any issues like that. Pare the sim back to stock and see what the VAS does when you add certain addons.

Those with an 8 core CPU can turn off HT and they then still have a super-four-core-with-HT-on-steroids. An 8 core CPU and HT on will make 16 processes without an AM showing vastly more performance with HT off. It's not a solution, an AM and HT on would produce better results in the long run, but with 16LPs available any kind of software will be fooled into making too many processes making HT ON/OFF a bigger difference than with a four core CPU. When you see an 8 core with HT off that's a sign there's some processes stretching out over too many LPs on that system with HT on.

 

 

.

Many thanks for your explenation Steve.

My AM=340 is still working great .

Regards

Gerard

There was a mention of randomizing assigning of primary task to a core on every p3d start. I use AM=254 so as to give main task a whole physical core. Can this be randomized ensuring the main thread has a separate core? as currently the task always gets put on core0.

3 hours ago, him225 said:

There was a mention of randomizing assigning of primary task to a core on every p3d start. I use AM=254 so as to give main task a whole physical core. Can this be randomized ensuring the main thread has a separate core? as currently the task always gets put on core0.

Hi, 

I think you may be referring to a statement made by me. Mine has proved to be a somewhat unique situation and I cannot recommend it to those of you who may be using quad or even six cored cpu's. The 'discovery' was purely accidental as I had forgotten to remove the 21760 Affinity Mask (AM) when I moved from HT=ON to OFF. I was struck how much smoother the sim was performing which lead me to monitor what was happening. There will certainly be a logical explanation and it is likely that this effect is not truly 'random' in nature. There are not many of us currently using i7-5960X processors with 8 physical cores so this is a hard one to test. I did persuade Rob Ainscough to give it a try, but his objective results using his similar setup did not support the 'theory'.

In the end this may turn out to be System specific. Even if it is a valid observation, which can be applied in other situations involving fewer physical cores, then a) I could not guess at what AM values should be tried, and b) trying to randomise core assignment for the main P3D could prove counterproductive as it raises the chances of contention with another 3rd Party threaded process. Perhaps the 8 core situation allows more flexibility in this regard. I really don't know. 

What I do know, however, is that it continues to work for me.  The Prepar3D AffinityMask=21760 / Hyperthreading=OFF in the system BIOS combination remains my preferred configuration. 

Regards,

Mike