Will V4 benefit from the new I9 18 core processors

[AviatorMoser 353]

1 minute ago, shivers9 said:

OK...I am thinking....each HT core has multi Logical Processors (virtual cores)......still thinking....still thinking....sorry dude all I get is a headache!!

 

[SteveW 1,792]

22 minutes ago, Terrydew said:

Intel has a utility/driver Intel turbo boost max that directs app to the strongest core which the newer system identify and label in bios. You can also overclock this core more than the the others. I understand the new I9 chips will identify the top two cores.

P3D sees the number of LPs and spreads across those. With four LPs to split over the rendering process on the first LP is at its leanest (as it states in the documentation). Less than four LPs those processes don't split out and the renderer is not at its leanest. With more than four LPs the background processes split out over the extra LPs which gives the renderer more work but can load the scenario faster.

So if we start P3D without the AM it sees the full LP count, 16 LPs on an 8 core +HT. The first two will have the first process rendering and second process occupies on the next LP it 'sees' within the scope of its core affinity.

So the sim starts up split over the full LP count and then a function that reorganises those processes will not provide the desired result since all those threads of those processes split out across the LPs are re-grouped onto LPs as the jobscheduler sees fit.

The proper operation of such a function requires the sim start up with the desired affinity and then those parts are moved carefully to the desired locations by a special app that reads the Prepar3D.cfg and understands P3D modes of operation - as Gerrard correctly pointed out.

[SteveW 1,792]

5 minutes ago, shivers9 said:

OK...I am thinking....each HT core has multi Logical Processors (virtual cores)......still thinking....still thinking....sorry dude all I get is a headache!!

Hehe, the thing is an HT core on an i7 for example emulates a pair of LPs, you could have more, it is not one physical core that emulates another virtual core...  Aspirin?  

[SteveW 1,792]

For the more technically minded...the .bat (batch file) can be put in charge of starting the sim; for example  let's say on a four core +HT system we start P3D with a .bat on Hex 55 (85 decimal) =01,01,01,01 and start P3D with Affinity Mask zero (or no jobscheduler section), the sim 'sees' only those four LPs (1111 looks like a four core no HT) and operates as if started with an AM=85.

[SteveW 1,792]

What the discussion asks is will the sim benefit from more cores, answer yes, we can utilise more cores to load scenarios faster and we can put other processes like addons on cores away from those provided to the sim which is more important.

As CPUs get more cores (and maybe even more LPs per core) and as the sim in its current forms will simply split out those background tasks over more and more LPs, the requirement of the jobscheduler section in the .cfg becomes more and more important, we won't particularly want the sim to load up 36 LPs, 4, 6, or 8 will do nicely.

We should be careful to use proper nomenclature when discussing AMs because it's one of the most confused subjects. If we are talking about no HT we do this '1111' for example, and when we are talking about HT enabled with two LPs per core we do this '01,01,01,01' with comma separated pairs, and hypothetically if we have, shall we say, a four core - four LPs per core CPU we do this '0001,0001,0001,0001' to get the same result. Paracetamol? 

 

[AviatorMoser 353]

Steve, we would also have to be careful about more cores. More cores = more heat generated per chip so the clockspeeds tend to go down as number of cores go up -- which have an adverse effect on the performance -- especially for that all important one core that does A LOT in P3D.

We're seeing this in our computational science fields. These new Xeon Phi's "Knights Landing" chips have 72 cores on them, but the clock speeds are significantly reduced. As a result, the hardware is forcing us to rethink our code architectures and rewrite them to take advantage of "manycore".

[SteveW 1,792]

Absolutely. Since P3D without an AM will start up on every LP it will produce more heat especially during loading. A simple solution with 8 or less cores is to turn off HT, sort of halving the activity, but with more cores that will no longer work as a 'solution'.

[SteveW 1,792]

We hear questions like; can P3D make use of Hyperthreading? Well the Operating system makes use of HT, but due to the monolithic nature of the P3D rendering stage we keep the first two processes one to each core '01,01', we can allow two more LPs '01,01,01,01' to complete the setup. However we can gain from HT by using both LPs of the third and fourth cores '11,11,01,01' on the right. As was said those last two LPs processes will split out over those four LPs now available making direct use of HT by using up the unused cycles within the core the scenario loads faster. Mainly though, it's the operating system, I/O, networking and so on which are heavily threaded so HT enabled can make a big difference to some systems.

With P3Dv4 being 64bit it's not showing the I/O and memory shuffling that we see with v3 if we have lots of RAM, also every thread is unencumbered by the WoW64 processes we see accompanying the threads of 32bit version so performance is improved that way.

[MarkW 1,491]

I have P3D v3 running on cores 1,2,3 with core 0 excluded (4 core non HT chip).  When I was flying around in v4 I remembered to check process lasso and it was running v4 across all cores.  I forced it onto 1,2,3 and noticed a large frame rate drop....does this make sense?

[SteveW 1,792]

Yes, that's because you moved the threads of the rendering stage which started on core zero onto the remaining cores running the background tasks. As Gerrard said he uses process lasso but not for the sim. Start the sim with an AM (or any app that takes care of affinity) so that it can utilise those available LPs correctly.

[WotanUK 289]

Wow...Thank you Steve, that is VERY informative.

[anden145 2,285]

What CPU are you talking about specifically?

Is it this one?

(If so, it haven't got 18 cores. It's got 10 physical cores with 20 threads with HT on).

[TuFun 433]

19 minutes ago, Anders Bermann said:

What CPU are you talking about specifically?

Is it this one?

(If so, it haven't got 18 cores. It's got 10 physical cores with 20 threads with HT on).

http://www.intel.com/content/www/us/en/products/processors/core/x-series/i9-7980xe.html?wapkw=intel+i9+processor

[srce 57]

3 hours ago, Terrydew said:

Intel has a utility/driver Intel turbo boost max that directs app to the strongest core which the newer system identify and label in bios. You can also overclock this core more than the the others. I understand the new I9 chips will identify the top two cores.

They do, but it doesn't seem to work with Prepar3d.

[srce 57]

5 hours ago, DaveCT2003 said:

You might try using a BAT file (Google is your friend) to launch your Sim with a specific AF, or use Process Laso

Setting the AffinityMask doesn't really help, because you may still want to use Core 0, just not for the main thread that's the bottleneck.