...ask yourself why it seems best to turn HT off MikeT707, what is your application?

 

Some say to turn HT off save a little heat and they can get another 300MHz out of their overclock. But since the results clearly show HT enabled better for the PC, and with an *appropriate Affinity Mask* P3D/FSX also perform better. The issue with VAS and OOMs is purely because the wrong, or no AM was applied. So get an i5 if you want to experiment with the last few MHz of the overclock. If you got the money the i7 is the better experience.

Would 01010101 for P3D and 10101000 for addons be an acceptable solution, subject to testing? Am looking for a logical start point. I realize 01010101 for P3D has been suggested earlier. Just wondering about both being used at the same time.

 

Admittedly, I do not understand affinity that well. It does appear the above will keep P3D and its addons on separate LP. This of course assumes that segregating processes is the major consideration.

 

Thanks

I'll venture to say my performance improvement with HT OFF (on my 5960X which has a 20MB L3 Cache) was likely due to more L3 Cache hits as their are 1/2 as many threads filling it up.  L3 Cache is shared by all cores be it "real core" or "HT core" for Intel i7 CPUs, I don't believe it is shared on i5 or i3 CPUs.  L1 Caches are 8 X 32KB - per "real core" not shared, and L2 is 8 X 256KB per "real core" not shared.

 

I don't now what happens with "HT cores" when it comes to L1 and L2 cache usage ... there are only 8 of them not shared.  Maybe Steve can fill in his thoughts on L1, L2, L3 with HT.

 

But like Steve suggests, for flight simulators I would focus on only i7 processors.

 

Cheers, Rob.

...ask yourself why it seems best to turn HT off MikeT707, what is your application?

 

Some say to turn HT off save a little heat and they can get another 300MHz out of their overclock. But since the results clearly show HT enabled better for the PC, and with an *appropriate Affinity Mask* P3D/FSX also perform better. The issue with VAS and OOMs is purely because the wrong, or no AM was applied. So get an i5 if you want to experiment with the last few MHz of the overclock. If you got the money the i7 is the better experience.

 

That is a helpful explanation, Steve. Thanks for that. So HT turned on with the correct Affinity Mask provides the best configuration. I will stay with my i7

 

I'll venture to say my performance improvement with HT OFF (on my 5960X which has a 20MB L3 Cache) was likely due to more L3 Cache hits as their are 1/2 as many threads filling it up.  L3 Cache is shared by all cores be it "real core" or "HT core" for Intel i7 CPUs, I don't believe it is shared on i5 or i3 CPUs.  L1 Caches are 8 X 32KB - per "real core" not shared, and L2 is 8 X 256KB per "real core" not shared.

 

I don't now what happens with "HT cores" when it comes to L1 and L2 cache usage ... there are only 8 of them not shared.  Maybe Steve can fill in the his thoughts on L1, L2, L3 with HT.

 

But like Steve suggests, for flight simulators I would focus on only i7 processors.

 

Cheers, Rob.

i7 it is.

 

Great thread.

I'll venture to say my performance improvement with HT OFF (on my 5960X which has a 20MB L3 Cache) was likely due to more L3 Cache hits as their are 1/2 as many threads filling it up. L3 Cache is shared by all cores be it "real core" or "HT core" for Intel i7 CPUs, I don't believe it is shared on i5 or i3 CPUs. L1 Caches are 8 X 32KB - per "real core" not shared, and L2 is 8 X 256KB per "real core" not shared.

 

I don't now what happens with "HT cores" when it comes to L1 and L2 cache usage ... there are only 8 of them not shared. Maybe Steve can fill in his thoughts on L1, L2, L3 with HT.

 

But like Steve suggests, for flight simulators I would focus on only i7 processors.

 

Cheers, Rob.

Today I will also try with HT off. My 5820 has 15 mb L3 cache.

Do you currently use a specific AM Rob or is it still no AM ?

Say you got a four core with HT enabled and an AM=85=(01,01,01,01) provides the same 4 core scenario the sim "sees" with HT disabled, except that other processes cause the same thread swapping that HT improves! Cache is no better or worse off. My tests show the HT enabled with AM 85 better performs than HT disabled no AM, which I find concurs with the HT ideas Intel provided to help improve performance of these processors.

 

With the addons, the jobscheduler will generally arrange them and avoid the most used cores, but it can use some help if there's some to give. With a 6 core or more we still only need provide 4 cores to the sim and the others will get targeted better by the jobscheduler. If the addons can be kept to certain cores, make sure they are provided with at minimum 2 LPs or they can be delayed. With the four core we can use three cores and leave one for the jobscheduler, but with that I recommend HT on and a spread of sim threads that the four LPs AM 116 provides as it keeps all maxed processes to separate cores, even though LPs exist on the same core.

Thanks. When I get back to town, I'll give that a try.

Do you currently use a specific AM Rob or is it still no AM ?

 

HT OFF, no AM.

 

Cheers, Rob.

HT OFF, no AM.

 

Cheers, Rob.

Would people contemplating a new system be better off looking at an i5 6600k rather than an i7 6700k do we think then? Theres about £150 difference in the uk between the cpu's right now, which is wasted money if advice at the moment is to disable AM! K

But like Steve suggests, for flight simulators I would focus on only i7 processors.

 

Kevin, see above.  For my specific processor and OC level, I get better VAS/FPS from HT OFF no AM ... I would NOT extrapolate that to "all" processors, see all Steve's info.

 

Cheers, Rob.

Tested it again yesterday.

 

I have small to none effects in turning HT on or off.

Also setting AF to different numbers does not do much here...

 

At the moment I am using HT on and Cores 2-7, I run ASN on Cores 0+1.

If I switch the Affinity in Sysinternals Process Explorer, I see that the load switches to the other cores but I do not get anything "special" in P3D. No stutters or whatever.

 

As I do not experience any OOMs (I pulled the sliders far to the right some days ago and do not experience any problems) and my FPS are good I stick with that.

 

Processor: i7 6700k @4,2 GHz

If we look at the graphs in post 6 we see there are only small differences unless extreme AMs are used with extreme changes of internal simulator requirements. The improvements to look for are rate of change of fps, or in other words smoothness. Like car mpg figures, they'll refine the way the engine works to get the best out of it, and you'll see the test results. It'll be different when you get the car home, but you'll want to be sure it's running sweet. That's why I'm using an AM of 340 on my 6 core HT enabled.

Hey guys,

 

after having a scenario where I get an OOM instantly after taking off (Aerosoft EGLL, FTX Global + Vector + OpenLC, ASN weather, 4k resolution, all sliders far to the right) I thought of this thread, deactivated HT, turned up the speed from 4,2 GHz to 4,4 GHz and tried it again:

 

No OOM and far less stutters!

I am using NO AF.

 

Seems to be the best thing for me like it is for Rob!

Thanks again mate for the good work in this thread.

Yep, doubling the LPs is your problem.

 

Is anyone getting this? 6+ core CPUs and no AM, that's splitting the sim up too much already. Turn HT on doubles the LPs, doubles the VAS allocated to threading, (worse on a 64bit O/S), and way overburdens the threadpooling. You get less heat with HT off mainly only because you're running half the jobs. I would say the main problem with overclocking heat is the choice of AM for the sim. P3D works differently to FSX and FSX-SE in the AM department, where FSX and SE have the edge. So I'm waiting to see how 3.1 goes...

I've read all the posts to date, tested all the suggestions exhaustively and now feel I should assume the role of Devil's Advocate by stating that the best performance subjectively, in terms of smooth frame updates, seems only possible on my rig with Hyperthreading ON and NO Affinity Mask.

 

For a time I had convinced myself that Hyperthreading OFF and restricting P3D to using 4 physical cores was the way to go, but further testing revealed I was mistaken. Flying straight and level with occasional gentle turns produces butter smooth fluidity most of the time in scenery areas of varying density and complexity. However, as soon as frame rates dropped below 30 and steeper turns were attempted the micro-stutters began to reappear.

 

This is by no means clear cut because other factors are also involved. I have confirmed that VSync with Triple buffering is preferable to using G-Sync when frame rates are in the sub 30s. G-Sync is great when frame rates are high, but less impressive when frame rates drop. VSync with triple buffering, along with an unlimited frame rate setting in the sim, produces excellent fluidity with barely noticeable rare micro-stuttering no matter the frame rate, i.e. anything above, say, 24fps.

 

i7-5960X (still at standard clock), GTX980 x2 in SLI, 16GB DDR4 3000, P3D V3 installed with ORBX, etc, etc on a Samsung 850 EVO 1TB SSD, Windows7 64bit

 

NCP Settings:

Monitor Technology: Fixed refresh. (G-Sync switched off manually before starting P3D - unfortunately the P3D profile does not do this automatically as it does, e.g. when starting and stopping Skyrim)

Power Management Mode: Prefer maximum performance

Preferred refresh rate: Application controlled

SLI rendering mode: NVIDIA recommended (SLI)

Triple buffering: On

Vertical Sync: Use the 3D application setting.

 

I reckon I have tried every appropriate Affinity Mask setting as has been suggested for an 8 cored CPU with 16 LPs. None has produced the all round fluidity of performance I am now experiencing without the application of an Affinity Mask and with all LPs working.

 

Mike

 

Edit: I have not experienced any OOMs .....as yet!