Many overclockers say that they see lower temps with HT=Off, and so can push the OC higher, so they say.

 

 

Oh ok. I'm using my PC for bunch of stuff (video production & editing, music production, gaming, FSX, etc) so HT is, I guess, important for good rendering and general performance in Adobe and other demanding applications. I've tried 4.6 and 4.7ghz clock and literally saw no difference to FPS in FSX.

Steve,

It might be a good idea if you have time to do a guide with step by step SS were needed, it would be nice to put an end to all the misguiding posts around AM with HT, for once and all.

Thanks David. I made a bit of an attempt here, but I've been meaning to spruce it up for ages.

Back with FSX, and then P3D V1.4, I was a fan of HT off, affinity mask = 14 (use cores 1-3).

 

When I moved to P3D V2.4 I tested with HT on, and via a very strange kludge between Process Lasso and AM = 14, got all threads to work well.

 

Now that I am on P3D V2.5 (all of three days now), with HT on and no mask, it works the same as HT on and AM = 14, i.e., P3D is not using core 0 by default. Performance is good, and all threads are being utilized. The first main core is hammered into submission and is usually fully maxed at 100% busy, and its secondary thread only chugs along at 20-percent busy. All of the other cores/threads oscillate wildly but they are definitely "working".

 

And it may be good to have a refresher course on what a Hyper Thread is. They were developed to address the long amounts of time the CPU spent waiting for a memory request, on the order of a couple hundred CPU cycles. So, instead of just sitting there, the CPU developers implemented secondary instruction fetch and commit engines that could host a separate "thread". While the fetch/commit engines are separate, the secondary thread uses the same ALU (Arithmetic Logical Units), which by definition makes it a thread and not a core. Hyper Threading has been implemented on IBM Power processors, which by benchmark evidence is the best implementation, most likely because of the combination of its RISC (Reduced Instruction Set Computing) underpinnings combined with a very fast processor clock speed (which makes memory very slow), SPARC processors (also RISC), and Intel. Because Intel is a CISC (Complex Instruction Set Computing) platform with very large Last Level Caches, it has the least efficient hyper threading implementation. Multipliers are roughly 1.6 (Power), 1.5 (SPARC), and 1.4-1.5 (Intel), i.e., the processor with two threads per core cannot provide 2x performance because of the shared ALU, but can provide additive performance.

 

The shared ALU explains why the secondary thread on my primary core is lightly utilized - the main thread is hammering the core and there is no time left to share the ALU with the secondary thread. I may fiddle with disabling this thread to see if performance (smoothness) improves, but all in all, I am pretty pleased with V2.5 and bog-standard CFG tuning. 

There's been a lot of chat about affinity masks of late, and how we should set them, what values to use etc.

 

What doesn't seem to have been answered (at least not fully) is a step back from that - should we have hyperthreading turned on or off in the first instance? Can P3D make use of the additional cores, or does the overhead (which there must be, at least in terms of scheduling) mean that it's more trouble than it's worth? Does it differ between people's individual setups?

 

my sweet spot seems to be HT = ON AM=254 plus other processes (ASN, explorer etc) set affinity to core 0 

 

With (the very popular) AM = 84 i get blurries

 

Tried also AM=63,127,167, 243,244,252 etc with no luck

 

with no AM in cfg i get results similar (but imho more blurries and instable) than AM=254

 

With HT off and no AM i get more FPS but slow texture loading = blurries

 

with HT off AM=14 unplayable: blurries everywhere.

 

that is my experience iwth I7 4790K @4.7GHZ + ATI R9 270X @ 1.2GHZ, P3d V2.4

 

other tweaks:

FFTF=0.01

TBM=120

SWAP=30

Alwaysfullload=1

 

ADDON::

A2A PLANEs

ASN

REX 4TD

ORBX FTX scenery and airports

 

as for ocing i don't think HT=off could permit significantly higher frequency for my CPU because it - in any case - would require a significant increase in the voltage that would lead to overheating (beeing air-cooled). Maybe I might be able to get 100mhz more  but I do not think it would be worth. - considering blurring attitude of HT=off would force me increase FFTF -> less FPS

....with HT on and no mask, it works the same as HT on and AM = 14

Yes it may do, and on your setup it can, but this is personal experience with your particular setup and number of LPs available, and the intervention of process lassoo. As has been said already, we need to be careful about that, and this could be how a causal reader may dismiss the AM. The OS might not start all the threads in a neat row, but in the end, we use a mask to ensure threads don't go where we don't want them.

 

...A couple of minutes break, and I had tested P3D v2.5 AM=0 and no AM - produced a thread in each of the 12 LPs of my 6 core HT=On. Resulted in poor performance Core 0 was occupied by the main P3D thread on LP0, coupled with a secondary thread on LP1 contending for the throughput of Core 0.  :excl:

Right you are, Steve. 

 

To my surprise, Process Lasso intervened with the 2.5 version of Prepar3D.exe, despite being from a wholly different directory. Apparently, the process name is the key, not the item itself.

 

So, I turned it off, freed the herd to use all 8 threads and performance was horrible, similar to your brief test on your 6 core/12 thread machine. Monstrous judders on the tarmac while turning - I did not even bother flying.

 

Turned it back to using threads 2-7 using affinity mask (not process lasso), and it was better. Decided to turn off the HT on the primary core with AM = 244 and we have best, at least for now.

 

Of course, your mileage may vary!

Sounds good John.

Many overclockers say that they see lower temps with HT=Off, and so can push the OC higher, so they say.

 

Yup, very true indeed!!

I have HT on for my 5820k ( 6 physical cores ) and till now have P3D use it as it chooses, without AM.

 

When using AM some advise 1364.

That would mean 10101010100. , so core 0, off and for all other cores only the logical core is active...

 

Others like Rob have advised 4092 or 4094 and later no AM.

 

 

Steve , can you explain and advise ?

For met it works best (so far) wit HT on and no AM. AM=84 gave me blurries.

I have HT on for my 5820k ( 6 physical cores ) and till now have P3D use it as it chooses, without AM.

 

When using AM some advise 1364.

That would mean 10101010100. , so core 0, off and for all other cores only the logical core is active...

 

Others like Rob have advised 4092 or 4094 and later no AM.

 

 

Steve , can you explain and advise ?

 

First get our nomenclature correct: we have 6 physical cores with HT=Off, and with HT=On we have 12 Logical processors. So with HT=On we have pairs of LPs sharing the throughput of physical cores.

 

There's no "to AM or not to AM":

1. If you have HT=On then you need an AM.

2. If you have many cores, then you need an AM.

 

Let's look at 4094=(11,11,11,11,11,10) it may work well, but here we've got lots of threads synchronising with the main thread on LP1 which causes a problem when too many. Since each pair of LPs shares the throughput of a core with time slicing, if there is not much texture and scenery loading you can pair them up on a core to maximize the throughput of that core, like it does here. But if the texture and scenery loading is greater, then there will come a point where the sharing of the core is detrimental to the throughput of those secondary threads. So we can do 1364 which does not load so many secondary threads and does not force them to share a core. It's up to you to get your head around it and test a few to find what works with your setup, since it depends on what addons you have running, how much networking etc.

 

For met it works best (so far) wit HT on and no AM. AM=84 gave me blurries.

"blurries" means you have something else causing the issue.

 

Above we have been speaking about keeping the main thread to a core by itself, so it does not share throughput, and maximises it's potential. Also we have been talking about maximizing the potential of texture and scenery loading by ganging up threads on a core or keeping the thread to a core to itself, and making a note to not have too many.

I have no personal experience with P3D so have nothing to offer on this, I have sufficient experience with SteveW to know take his advice seriously, I have more than enough experience with overclocking and FSX tweaking to know that there are too many variables to provide a one size fits all solution.

 

IMHO If you have a hyperthreading enabled CPU you have paid over and above for this facility (i5 vs i7), disabling it makes no sense. The extra computational power of a hypertheaded CPU may mean it runs slightly hotter, but if you're overclocking then your cooler should be up to the job. If you have a hyperthreaded CPU you need to use an affinity mask, which one works best for you may vary but for a quad core i7 a mask of 84 is the best place to start.

Thanks Steve.

 

My 3 monitor setup uses photoscenery only with AG trees and some buildings on it.

Ofcoarse detailed addon airports.

 

No visual aircraft as I have a real size hardware cockpit based on the Ifly CBE.

 

Does this make any difference on the AM settings?

Best thing to do, Gerard, is to start with an AM that gives you three threads on three cores, example 2688=(10,10,10,00,00,00). Get Task manager running and display the graphs so you can see them load up when the sim starts. The first will possibly go to max or near max on LP7. The other two on LP9 and LP11 may only be at 30%, or they may be at 70% each, but depends on scenario. Have a play with that.

Steve, you said if you have "many" cores you need AM.  How many exactly is many?  I have a 2600k oc'd to 4.4 HT is off, do I need AM?