1 hour ago, lownslo said:

We don't test different scenarios, but the same scenario and time how long it takes to load.  Then change the affinity mask to add another LP and time how long it takes to load.  Keep testing by adding LP's until the scenario load times stop getting shorter, then you've found how many cores your simulator can use.

Greg

Doesn't make sense to me because I only see the terrain texture loaders up at 100% when I'm in ultra complex scenery.  When I'm simple sparse scenery they're barely being used.   You're saying load times are identical in the sparse and complex scenes?  That does not fit at all w/ my experience.  It always takes way more time to load a complex scene than a super simple scene.

6 hours ago, Noel said:

But isn't this a completely dynamic situation, so if I measure time to load in one scenario, then go to one 10x more complex, what am I learning w/ re to optimal core counts and setup?  Quite often, not in this clip but most of the time in specific scenarios I might see the main thread hovering way below 100%, like around 72%, and all or most of the terrain texture loaders at 100%.  This makes me think having a few more of those simply allows more textures to be loaded in advance of being needed perhaps, and in do so textures stay nice and clear for a larger radius.   Why does this not basically describe what's happening?   My bad for displaying a clip w/ the main thread at 100% I typically avoid this by reducing scenery sliders this clip was to illustrate 100% in the texture loaders.

It would be a waste of time doing the test on a sparse scene - you would carry out such a test on your most expensive scene and stick to it. Then later when you add another more time consuming addon check it again.

Tests with a stripped down P3D are for determining where problems come in by  - elimination -

Using a logical sequence of adding or removing (eliminating) variables is the fastest way to resolve a problem.

Checking adding cores to see if they provide faster loading (at the load scenario stage everything is loading) is what you do with your most extreme setup. Park in a big bad airport facing a load of complexity and try from that.

If you add cores and you don't get that loading faster you are wasting CPU and adding Heat.

Edited August 6, 20187 yr by SteveW

4 hours ago, Noel said:

You're saying load times are identical in the sparse and complex scenes? 

No - he didn't day that. More complex scenes quite obviously take longer to load, Using the load time will ascertain how fast the sim can pull in new scenery en-route. If it takes less time to start up it also runs better through the course of the flight.

A few facts:

If the main core is at 100% then when the scene changes to a more complex scene the sim slows down because the CPU can't do 110%.

If we have HT enabled and the main core main LP is at 100% that's the entire core at 100%.

If we have HT enabled and the main core LP is at 100% and the second LP of that main core is at 20%, the main LP is actually only doing 83%.

 

3 hours ago, SteveW said:

A few facts:

If the main core is at 100% then when the scene changes to a more complex scene the sim slows down because the CPU can't do 110%.

If we have HT enabled and the main core main LP is at 100% that's the entire core at 100%.

If we have HT enabled and the main core LP is at 100% and the second LP of that main core is at 20%, the main LP is actually only doing 83%.

 

Thanks I see where you are going w/ regard to complex scenery and setup--I did not realize what the agenda was.  My situation is such that I don't believe it makes sense to calibrate for the most complex scene when I spend less than 10% of a typical flight in that complex scene, at least in terms of slider settings as I'd end up missing out on the eye candy coming w/ maxing out sliders in most of those flights, so I'll typical pause the sim and reset sliders when I see evidence I'm hitting high 90% on the main thread.  This being said yes I can go to the complex scene and test re load times.  When you say load time, you're referring to a stopwatch on starting a flight to when the plane is fully loaded, is that correct?

Now I've been under the impression I have been using an AM that puts the main thread on one LP in the pair only which is why we're seeing 100% on LP2, and 0% on LP3 as it were in the screenshot above.  Am I hampering access to any processing power for the main thread by setting it up this way?  I think in the middle sentence above you are saying, no, I'm getting all I can.  I think you suggested I do this this way so I have.

20 minutes ago, Noel said:

Thanks I see where you are going w/ regard to complex scenery and setup--I did not realize what the agenda was.  My situation is such that I don't believe it makes sense to calibrate for the most complex scene when I spend less than 10% of a typical flight in that complex scene, at least in terms of slider settings as I'd end up missing out on the eye candy coming w/ maxing out sliders in most of those flights, so I'll typical pause the sim and reset sliders when I see evidence I'm hitting high 90% on the main thread.  This being said yes I can go to the complex scene and test re load times.  When you say load time, you're referring to a stopwatch on starting a flight to when the plane is fully loaded, is that correct?

Now I've been under the impression I have been using an AM that puts the main thread on one LP in the pair only which is why we're seeing 100% on LP2, and 0% on LP3 as it were in the screenshot above.  Am I hampering access to any processing power for the main thread by setting it up this way?  I think in the middle sentence above you are saying, no, I'm getting all I can.  I think you suggested I do this this way so I have.

It's just not a question of whether you are in your most complex scenery or not Noel.

Remember you can bring things down a notch if you prefer. The described technique puts you in the position of knowing exactly what you want to know - how many cores provides maximum performance.

100% utilisation simply means you have no restriction on framerate or you are trying to render too complex a scene, too often.

Edited August 6, 20187 yr by SteveW

"When you say load time, you're referring to a stopwatch on starting a flight to when the plane is fully loaded, is that correct?"

Exactly: So long as you stop the watch at the same time each trial. Choose a convenient reference.

So as you can see, in any case you can make your own test - time your preferred scenario whatever the complexity.

The aim is simply to see the same thing happen more quickly. If it does, then that is an indication the sim will perform well when new scenery comes into view.

6 hours ago, SteveW said:

"When you say load time, you're referring to a stopwatch on starting a flight to when the plane is fully loaded, is that correct?"

Exactly: So long as you stop the watch at the same time each trial. Choose a convenient reference.

So as you can see, in any case you can make your own test - time your preferred scenario whatever the complexity.

The aim is simply to see the same thing happen more quickly. If it does, then that is an indication the sim will perform well when new scenery comes into view.

Sounds good I will do some testing soon w/ this old box and see how it shakes out.   Thanks!

OK well I STARTED some testing w/ how I'm set up now w/ HT on an an AM of:  111111110100    versus HT off w/ an AM of:  111110.

Ultra complex scenery load:  26% faster w/ HT enabled

Low to moderately complex:  15% faster w/ HT enabled.

It appears I'm better off w/ HT enabled for this test.  This seems to be making it clear that HT on is a better thing than not which implies to me for some reason whatever is allocating terrain texture loading tasks seems to be finding more places to get work done, which of course was my gestalt in the first place despite others saying HT is not good thing!  Well it may not be for some but it still makes sense w/ this system.  It will not be best for getting maximum clock speed for the main thread but it seems to clearly be a better option than HT off for terrain loading.  So with no further testing which I will do just for fun to start to limit LPs, I'm still thinking an 8-core is a good thing, and HT might well be as well for it, so I'm back to a 9900K!   This doesn't surprise me too much because I've known the sim really runs well especially in terms of terrain texture updating--I never see blurries and I've always figured how I'm set up has helped some with this for sure.

Edited August 7, 20187 yr by Noel

56 minutes ago, Noel said:

OK well I STARTED some testing w/ how I'm set up now w/ HT on an an AM of:  111111110100    versus HT off w/ an AM of:  111110.

Ultra complex scenery load:  26% faster w/ HT enabled

Low to moderately complex:  15% faster w/ HT enabled.

It appears I'm better off w/ HT enabled for this test.  This seems to be making it clear that HT on is a better thing than not which implies to me for some reason whatever is allocating terrain texture loading tasks seems to be finding more places to get work done, which of course was my gestalt in the first place despite others saying HT is not good thing!  Well it may not be for some but it still makes sense w/ this system.  It will not be best for getting maximum clock speed for the main thread but it seems to clearly be a better option than HT off for terrain loading.  So with no further testing which I will do just for fun to start to limit LPs, I'm still thinking an 8-core is a good thing, and HT might well be as well for it, so I'm back to a 9900K!   This doesn't surprise me too much because I've known the sim really runs well especially in terms of terrain texture updating--I never see blurries and I've always figured how I'm set up has helped some with this for sure.

Thanks for sharing your data, Noel.  Yes, the 9900K will be a very good choice for P3Dv4.  Be careful though... plenty here will argue that your test is somehow flawed and that the 9700K is the best choice.  In the end, better to have more processing power and flexibility than too little. 😀

Greg

9 hours ago, Noel said:

It appears I'm better off w/ HT enabled for this test. 

That's correct and what I always recommended otherwise we would all be fools to pay for it.

HT maximises your parallel tasks and uses the core completely when you have two per core but you need to guard your monolithic tasks and not share cores for those.

The only reason that HT gets a bad name and not recommended is because those faulty set-ups use the AM incorrectly or not at all. That old rubbish reduces fps and increases heat.

Remember that with HT enabled and AMs doubling loading tasks per core these do more work than with HT off  - obviously as you see faster loading.

More work done = more heat made.

And so I also recommend any overclock be moderated to accommodate HT.

On ‎8‎/‎3‎/‎2018 at 2:36 PM, Noel said:

OK then it's pretty much as I guessed w/ the big question remaining unanswered:  what then is the optimum number of cores or LPs as it were to use for P3D? 

So - I claim I answered your big Q after all. 😉

No single system is the same so it is a Q that can only be answered with advice on how to find out on that system - there is no single answer to those type of questions.

Ted for one - take note! If I can't answer your Q it is because it could be irrelevant or might bring confusion and in so doing it would be stupid for me to do so - read into that what you like. Apologies for any misunderstandings, that's just lack of time to post the full gamut all the time.

Yes thanks again SteveW I appreciate it!  I've been at flight simming for a very long time and when I went to that AM w/ HT enabled I have a performance quality that I believe is pretty close to as good as it can be for my old hardware and I could tell because when I'm living w/in my means as it were its always perfectly smooth w/ clear terrain textures.

If and when I get around to building a new box w/ a 9900K or what have you what I would like to do is put the main thread on a core that is running as fast as it can, and clock the others lower to manage heat etc.  I'm guessing clocking those terrain texture loaders down won't hardly impact their ability to keep up w/ their work loads. 

You're welcome Noel. Well, it's up to you to find all that out when the time comes. Seems you have things sorted out and think you can relax and enjoy knowing that you're not blowing away some easy had performance at least.