HT and AMs are difficult to understand - let me try to explain!

AM 15=(1111) on a four core is equivalent to no jobscheduler section, or AM=0.

An LP is a Logical Processor, we have 2 LPs for each HT enabled core, or 1 LP for each HT disabled core.

The "Best" AM for FSX/P3D is to allow FOUR LPs, so that's 4 cores on a 4 core AM=0=15, or 4LPs of a 4 core with HT enabled, AM=170=(10,10,10,10) or AM=85=(01,01,01,01). P3D and FSX work best with 4 main threads on 4 LPs.

Notice I separate HT enabled bit pairs with commas. That's not just to make it look neat, that's to show we have an HT pair for each core.

However, if the first LP, or core, *on the right*, that is core zero, or LP0 or LP1 with HT enabled, has an "aggressive" addon running, then the sim main thread performance will suffer. So that's why on a 4 core an AM or 14=(1110) is often recommended since core zero is "masked". The equivalent on an HT enabled system is 84=(01,01,01,00) or 168=(10,10,10,00), these mask core 0 by masking LP0 and LP1.

So if our addon ruins on LP0 or core 0, that's why the AM is introduced to help the sim avoid that addon running on core zero.

Since HT disabled 14, or HT enabled 84 or 168 only present 3 lPs, the sim will split/group it's main threads over those three LPs.

P3D and FSX run better with 4, so with HT enabled we can make an AM that provides 4 LPS for the sim.

P3D and FSX know how many processors are available and split across them, that's why it is often said "avoid an AM and let P3D/FSX handle it" - but they don't know about HT enabled and is a mistake to suggest AMs are not needed.
 
When the sim splits and groups it's main threads over 4 LPs it can run better if it is helped onto LPs that do not group two main threads on the same core, so that when those main threads want maximum throughput, they are not sharing with another. That's why I suggest an AM equivalent to 116=(01,11,01,00) - the middle two, although both o a core together, never call for max throughput together. But an AM like 212=(11,01,01,00) groups two LPs that do max out together during terrain loading and should be avoided to provide smoother operation.

Remember the fps is hardly affected by changes to the AM, only the smoothness of the sim in operation with addons is improvable.

Here's a screenshot of AM=116=(01,11,01,00) in operation on a four core HT enabled with an aggressive addon simulated on LPs 0 and 1, see LP2, LP5, and LP 6 (on separate cores maxed out during terrain loading but not affecting each other. And below LP2 maxed out with no contention from LP3 when the sim is in full swing:

Hmmmm...

 

I've avoided affinity masks to date, but this thread has made me think it's worth a try

 

I use a lot of high res photoscenery e.g. 25cm/pixel, sometimes with complex airports and/or autogen trees. Mainly flying the NGX.

 

Using an overclocked 5820K (six core) CPU, with HT on. P3D v3 only.

 

Am I right in thinking:

Leave HT on

Apply an affinity mask of 340

 

Or do I need an affinity mask with more cores active for photoscenery purposes?

Good question. The different scenery types do present different problems.

 

Any PC I try turning off HT means it becomes a bit sluggish, not much, but turning back on HT the PC becomes obviously more zesty, so I would leave HT enabled if possible.

 

With the 6 core HT enabled and an AM of 340=(00,01,01,01,01,00) this gives good performance and leaves two cores for the jobscheduler to utilise for other processes.

 

Adding another core means the sim splits out into five cores and I've only seen that result in less smoothness with autogen. But with the photoscenery why not try 1364=(01,01,01,01,01,00) and see what it produces. It's possible the load is different and even though more threads are used (using up more VAS, which isn't a real isssue), it may be a better trade off with heavy load scenery.

What are your experiences ?

 

Well you already know my experience, but to clarify my experience, this was with my 5960X and ONLY my 5960X which is 8/16 logical ... a "real" core and an HT core are both called a "logical core", the single CPU die is referenced as one Physical CPU ... to further confuse matters if you aren't already.

 

HT ON does improve memory bandwidth and as Steve suggest "smooth" data flow.  However, there is something not working as expected when high numbers of "logical" cores are present ... I don't know if this is an Intel problem, something deep within C++ threading model, an OS issue, or a code issue in P3D.  But I have gained considerable VAS (not just a few MB) with HT OFF and I'm able to sustain my Vsync 30Hz (30 FPS) -- keep in mind I didn't "discover" this VAS savings, it was information passed along to me from a reliable source using the same CPU.  When I tested claim, I expect it to NOT be valid or be marginal (within acceptable error limits).   When I saw the results 500MB VAS savings, I re-tested several times thinking I must have done something different ... every test I came out about 400-500MB ahead in VAS usage with HT OFF.

 

I just finished doing a video from Drzewiecki Design - New York V2 using these setttings:  http://www.robainscough.com/P3DV3_Settings_Medium.html and my usual set of add-ons  ... I used the R66 and PMDG 737.  I was at 2.7GB-2.9GB (display VAS usage in the video) with the R66 and about 3.1-3.2GB with the PMDG 737 ... HT OFF ... I turned HT ON drop OC frequency down to 4.5 (stable for HT ON) -- I couldn't load the saved flight, got an OOM.

 

5:21 min mark I display VAS ... also again 9:18 mark.  Couple of spots the PMDG 737 gets below 30 fps (27-29) but mostly stutter free.

 

Video is available in 4K res, but you'll need to give YouTube about 30-60 mins to finish re-compressing it.

 

 

Like I said, I couldn't do this test with HT ON - immediate OOM.  Now, before you go buy a 5960X, let me re-iterate ... you can't extrapolate that to ALL CPUs will have a VAS savings with HT OFF.  Steve's done a fantastic job presenting his information from his testing ... he might be a little dubious of my results and that's expected, because so was I.  But HT is staying OFF for now.

 

Cheers, Rob.

What AM did you use with HT enabled Rob? Can you try one that provides just 4 LPs with HT enabled? e.g. 340=(00,00,00,01,01,01,01,00).

 

..if there's no AM provided the sim will use up a shedload of VAS and run poorly.

 

 

 

Going back to the discussion title: "HT off saves VAS and increases fps ...".

 

1. HT off saves VAS. Correct. That is if you've not included an AM.

 

2. HT off increases fps. Correct. That is if you've not included an AM. Since the jobscheduler moves a process along by one, that's just another LP of the same core with HT enabled, but when HT is disabled, it's a fresh core.

I am gonna try P3Dv3 this weekend.  Lets just make this conversation easier

 

For HT off - are you using AffintyMask to select the cores only while HT is enabled via bios (but not using the HT cores)

 

Or best solution is to disable HT in the Bios and then use Affintymask or leave it as.  I have 6 physical Core Intel (5820K).

On the 6 core, with HT disabled use an AM to select 4 cores, e.g. 30=(011110).

 

With HT enabled use 340=(00,01,01,01,01,00).

It's not called HYPE-rthreading for nothing.

4 core no HT... AM = 15

Thanks, Manny, and with HT on?

 

Jim

It's binary Jim, so we lay out a string of ones and zeros to define which LPs are available from right (LP0) to left (LPN-1) so a 4 core with HT on is 11111111 bits all LPs set on. I separate them with commas to denote that HT is on 11,11,11,11. The same applies here with all LPs set on, the AM=0=255=(11,11,11,11)=no jobsheduler section.

It's binary Jim, so we lay out a string of ones and zeros to define which LPs are available from right (LP0) to left (LPN-1) so a 4 core with HT on is 11111111 bits all LPs set on. I separate them with commas to denote that HT is on 11,11,11,11. The same applies here with all LPs set on, the AM=0=255=(11,11,11,11)=no jobsheduler section.

 

Well I'm here with Steve HT on with AM, off makes no sense here local ;-)

Yep, so if we use all cores on a 4 core with HT off, that's AM=15=(1111), it's equivalent to no jobscheduler section. But if we enable HT then we must use an AM=85=(01,01,01,01) or AM=170=(10,10,10,10) since otherwise the sim make eight jobs, two to a core, instead of four. That could be why many report extra heat with HT on, reducing their overclock. Since the extra jobs would place twice as much work on the CPU without an AM to keep them to four. In testing there's less true performance (fps-delta) with three or five jobs, rather than four. A lot depends on how things are tested.

 

What of the extra jobs beyond four? Mostly they are repeats of what is already taking place, but they can get utilised if they finish first when others are interrupted and are discarded as they become surplus. So in a sense a few more jobs running may prove less interrupted overall by other processes since another doing the same job might still finish in the usual time. However that's more heat and more work for the pooling job.

Hi,

 

Ugh! This just gets worse. The more I read the more I understand. Yet confusion remains as my, admittedly crude, testing (see below) does not appear to be bearing out the expectations. Steve's contributions, in particular post #16, have helped immeasurably towards my understanding of what is going on:

http://www.avsim.com/topic/478995-ht-off-saves-vas-and-increases-fps/?p=3334243

...and I thank him for that. It is clear he has spent a great deal of time and effort trying to enlighten dullards like myself and for that he has my sincere thanks.

 

I too have an 8 core water-cooled i7-5960X running, for the time being, at default clock (optimised defaults on my Asus Rampage V Extreme mainboard). Performance is very good and, I have to confess, sometimes I wonder whether I really need to overclock. Doubtless I will at some stage since that has always been the intention behind this build. My goal was to first ensure data integrity by installing everything I need first, including P3DV3, ORBX, REX,etc,etc. I have 2xAsus Strix GTX980s running quite well in SLI at 2560x1440 outputting to an ROG-PG278Q G-Sync enabled monitor. I do not use VSync in any game except a highly modded Skyrim which will only run correctly with the monitor's refresh set at 120Hz and Adaptive (half refresh) selected in NCP. G-Sync is used for Prepar3D and ensures generally very fluid performance until the frame rate drops below 30 when it can hitch occasionally with slight monitor flicker.

 

My settings are similar to Rob's Medium settings for Prepar3D V3, except for water which I have on Ultra.

 

The plethora of suggested AM settings tried are as listed below:

 

HT Disabled in BIOS = 8 core = 8 LP

 

AffinityMask=224

1,1,1,0,0,0,0,0

AffinityMask=240

1,1,1,1,0,0,0,0

AffinityMask=248

1,1,1,1,1,0,0,0

AffinityMask=252

1,1,1,1,1,1,0,0

AffinityMask=254

1,1,1,1,1,1,1,0

 

HT Enabled in BIOS = 8 core HT = 16 LP

 

AffinityMask=21504

01,01,01,00,00,00,00,00

AffinityMask=21844

01,10,10,10,10,10,10,00

AffinityMask=21824

01,01,01,01,01,00,00,00

AffinityMask=43008

10,10,10,00,00,00,00,00

AffinityMask=43648

10,10,10,10,10,00,00,00

Affinity Mask=54272

11,01,01,00,00,00,00,00

Affinity Mask=62720

11,11,01,01,00,00,00,00

Affinity Mask=64832

11,11,11,01,01,00,00,00

 

Yes, I've tried 'em all yet, like Rob, I found the best performance with minimal/no stuttering appears to be achieved while using No Hyperthreading (OFF in the System BIOS) and NO Affinity Mask in P3DV3.

 

I test at the default Elgin AFB (KVPS) in the Raptor. Graphic drivers version 358.91

After initial load I cycle round the cockpit a few times, ditto outside to ensure full texture loading. Frame rates settle around 45 (Hyperthreading - OFF), lower with HT enabled. I then take-off over the sea, perform a climbing turn back towards land at around 3-4000ft and settle straight and level at 250-300knots. I view out the cockpit at 45% observing the land texture update. I then roll the Raptor several times in either direction and look for any hesitancies during the manoeuvre.

 

As I said, fairly crude testing, but used consistently each and every time for every test. When flying away from heavily populated areas the frame rates rise dramatically and the sim remains butter smooth courtesy of G-Sync. I was flying last night in ORBX FTX: NZ South Island and the experience was simply jaw dropping!

 

While Prepar3D V3 is running:

CPU temp does not rise above 35-36degrees C

GPU temps rarely rise above 60degrees C and are often lower.

GPU usage is around 40-50+ % for both.

 

Regards,

Mike

Steve I am using HT on and 340 on my six core but for some reason on an hour fight the ground textures turn to blur and mud and I loose all definition. Is this to do with my core setting?

 

This is with Global and Pago Pago and no photo scenery that I am aware of. 

SteveW's summary brings clarity to this HT/AM discussion that's been going on for weeks now. Thanks so much.

 

But for me a few questions remain. What are some examples of an "aggressive addon"? And how do we determine if it's running on Core 0?