And after 6 pages, no-one actually knows the 'optimum' number of cores for P3D.

Can P3D use 10 cores ? Apparently yes, if you watch Robs' 7900x videos.

So which is better..... 10 cores running at 3.6 Ghz, or 5 running at 5 Ghz ?

 

Ultimately, 10 running at 4.8 will trump everything.

Of course the X series are all over a grand......plus the motherboard.

But are we talking FPS or smoothness ? They are arguably different things.

As Rob only runs at 30 fps, he is after smoothness.... terrain loading, Ai traffic etc., kicking in without causing pauses.

It is very difficult to make a decision. 

I would say that once a minimum of 6 cores are being used, speed then becomes important.

I bet 6x 4.5 Ghz beats 4x 5Ghz at both FPS and smoothness.

But above 6 cores, then speed is the main factor.

I can't decide whether to go for an x-series now, or wait for the first 6 ot 6.5 Ghz cpu to turn up in a few years.

 

Edited August 6, 20196 yr by Gabe777

The industry have been waiting for a 6ghz CPU for the past 5years and will be in the next 5years. 

2 hours ago, Gabe777 said:

And after 6 pages, no-one actually knows the 'optimum' number of cores for P3D.

P3D has a monolithic process and a paralleled process.

If an added core does not decrease the time to load the scenario, that's already too many cores.

Easy-peasy.,

 

Another problem of using too many cores is that since the motherboard is also a limited resource they consume the bus and operate in each other's way. More threads and more cores use more shared cache degrading the performance of the main process.

So in effect this is not a useful question "what is the optimum number of cores for P3D?"

instead

"what is the optimum number of cores my system would desire to have when running P3D?"

Not as many as folk think will help, like it would for say Paint3D, use them all there.

 

16 hours ago, SteveW said:

Another problem of using too many cores is that since the motherboard is also a limited resource they consume the bus and operate in each other's way. More threads and more cores use more shared cache degrading the performance of the main process.

So in effect this is not a useful question "what is the optimum number of cores for P3D?"

instead

"what is the optimum number of cores my system would desire to have when running P3D?"

Not as many as folk think will help, like it would for say Paint3D, use them all there.

 

But Steve this is not really addressing the issue.

With respect..... and I mean that.... rhetoric isn't helping.

If I have an i7 7700k running at 4.5 Ghz on all 8 threads, should I get a i9 9900X with 10 cores at 4.5 to 4.8, or wait for a 6 or 8 or 10 core where at least one or two run at 6 Ghz...?

There seems to be so many perms and combs..... but no 1 answer.

So for example.... which is better:

10 cores at 3.5 Ghz, 

4 cores at 5 Ghz, OR

6 at 4.2 Ghz... ?

i.e. at what point is the nos. of cores, overkill or even disadvantageous ..... 6, 8, 10 ?

Looking at a Rob video.... he had quite a long period of (what appeared) to be... 100 percent use of 10 cores in a PMDG in a busy airport during takeoff.

This seems to suggest P3D can utilise a shed load of cores.

 

#confused #frustrated

Cheers. 

Edited August 7, 20196 yr by Gabe777

7 hours ago, Gabe777 said:

But Steve this is not really addressing the issue.

With respect..... and I mean that.... rhetoric isn't helping.

If I have an i7 7700k running at 4.5 Ghz on all 8 threads, should I get a i9 9900X with 10 cores at 4.5 to 4.8, or wait for a 6 or 8 or 10 core where at least one or two run at 6 Ghz...?

There seems to be so many perms and combs..... but no 1 answer.

So for example.... which is better:

10 cores at 3.5 Ghz, 

4 cores at 5 Ghz, OR

6 at 4.2 Ghz... ?

i.e. at what point is the nos. of cores, overkill or even disadvantageous ..... 6, 8, 10 ?

Looking at a Rob video.... he had quite a long period of (what appeared) to be... 100 percent use of 10 cores in a PMDG in a busy airport during takeoff.

This seems to suggest P3D can utilise a shed load of cores.

 

#confused #frustrated

Cheers. 

my experiance with this , and its my thoughts can be subjective but its my option today.

the one i have tested in P3D V4  and all cpus OC to max stable FRQ the same with mems.

7700k at 5.3ghz , 8700k at 5.2ghz or 5.4ghz , 7920x 5.0ghz (ht off ) , 7980XE 5,0ghz ( htoff) and 9900k 5.5ghz ( ht off)

7700k need ht on , 8700k its hard if you want to have 200mhz more with htoff ( loadtimes suffer little with ht off ) dont use them anymore.

the difference 7920x and 7980xe at same frq 5ghz 12 vs 18 cores nothing the one that i think worked best was the 7980XE with HT off and 8 cores disabled in bios get more cashe per core.

but for me its crystal clear the winner is the 9900k at 5.5ghz with 4266mhz 2x8gb sticks 16gb, thats my personal thougts .in other words FRQ is King 

I dont even try to run less then 4,8ghz in P3D with settings i like, it begin to stutter with less then 4.6ghz even with 18cores / 32 treads

the 6ghz on 1-2 cores with normal cooling , its not clear if thats possible yet look at the AMD rumor 5ghz all core boost ended with 4.2ghz , with 2 more cores 8-10 more heat i would be happy with 5.6ghz as max OC. its more likely that the frq is in line with the 9900k and a IPC gain of 15-20% or more ( ipc mesured at same FRQ) but that is in the future , time will tell.

Edited August 7, 20196 yr by westman

7 hours ago, Gabe777 said:

But Steve this is not really addressing the issue.

Yes, it is - and it precisely addresses the issue.

! repeat.

If you do not see a decrease in time when loading your test scenario with another core, then don't use it as it can only impact the PC with extra heat.

And your test had better be good.

With P3D we have one core that provides the performance, and any number of further cores to construct the scene.

If the scene cannot be constructed any faster with more cores, then that is quite simply a waste of power and adds heat.

We will see smaller and smaller decreases in loading time as we add each core.

Remember to test a repeatable scenario with no randoms, guaranteed repeatable, no flatlines, and that it is measured for well cached and non cached speeds.

At some point one curve crosses the other, it is a law of physics, and from then on more cores consume their power and do not provide useful benefit at the same time.

None of that, adding cores, can improve the throughput of the main core, you have to manage software and Affinity for that. How's that done in those tests? 

 

Edited August 7, 20196 yr by SteveW

When I discuss performance with P3D users most are familiar with the idea that it is does not cap our settings. It tries to do what we want. Not like a game which would often cap the scene for performance. Also most users expect to enter a scene that is a little too much for the system. Most users have a usual haunting area that has more typical performance requirements.

Choose a scenario that fits with those desires. We could choose the worst possible scenario or a more typical one.

A problem with most tests that is not understood, is that the simulator reaches flatlines that affect the way further reaching performance characteristics behave. Producing irregular results.

A symptom of overreaching settings, can cause any kind of behaviour, including INCREASED fps.

 

Edited August 7, 20196 yr by SteveW

The trend with more cores is a more conservative clock because they all add heat. Increased clock frequency gets harder to do. An optimised P3D system would avoid too much heat. We would want an optimum core count, not an 'any number we can get' count.

The number of cores that can transact concurrently might only be 6 on an 8 or 10 core. Using more cores will chop up the transactions disturbing the main process of P3D. Won't hurt Paint3D.

With a typical 8 core, check to see if 8 cores really helps more than just 6 of those 8, AM=126=01111110 (HT enabled use 5460). Try the same with a ten.

 

OK. Let's set up an appropriate test. Mainly we look across the controls out over a scene, an approaching airport say on final. We usually want that to be good at  least.

Start a flight, in the VC looking across the controls and out over the airport buildings including the tower. Make sure the system can repeat the scenario the same way each time, save the flight.

Using the saved flight, time the loading stage to the same point each time. Add a core and repeat the test. Until the additional core does not appreciably decrease the time to load up to that point. Don't touch the controls.

At that stage you have enough cores.

Try other saved locations. The aircraft turning onto final with AP, is a fairly good repeatable test and changes the view which is more GPU specific.

That's probably a very good hour or two spent with a new system, knowing that those other lovely cores are not wasted.

Alright. Let's for argument's sake we find 6 of our 8 core are starting the sim as quickly as it can. No amount of extra cores can help the fps.

How do we improve fps?

We can drop a setting or lower the screen resolution, but we don't want that.

Instead we want to be sure we are getting the best throughput on the cores that do the critical tasks. That way, whatever we decide to set we know the thing is going to be doing the best it can do, more or less.

Look at the cores doing the work and keep other programs, away. Even small exe's running a switch panel can alter the behaviour.

With all the extra cores, consider to use an optimum number of cores for the simulator and the others for other things.

17 hours ago, SteveW said:

Yes, it is - and it precisely addresses the issue.

! repeat.

If you do not see a decrease in time when loading your test scenario with another core, then don't use it as it can only impact the PC with extra heat.

And your test had better be good.

Ah I understand now. Sorry.

So an indirect test of cpu core usage is the scenario loading time.

Got it ! Clever. #impressed.

Thnx

21 hours ago, westman said:

my experiance with this , and its my thoughts can be subjective but its my option today.

the one i have tested in P3D V4  and all cpus OC to max stable FRQ the same with mems.

7700k at 5.3ghz , 8700k at 5.2ghz or 5.4ghz , 7920x 5.0ghz (ht off ) , 7980XE 5,0ghz ( htoff) and 9900k 5.5ghz ( ht off)

7700k need ht on , 8700k its hard if you want to have 200mhz more with htoff ( loadtimes suffer little with ht off ) dont use them anymore.

the difference 7920x and 7980xe at same frq 5ghz 12 vs 18 cores nothing the one that i think worked best was the 7980XE with HT off and 8 cores disabled in bios get more cashe per core.

but for me its crystal clear the winner is the 9900k at 5.5ghz with 4266mhz 2x8gb sticks 16gb, thats my personal thougts .in other words FRQ is King 

I dont even try to run less then 4,8ghz in P3D with settings i like, it begin to stutter with less then 4.6ghz even with 18cores / 32 treads

the 6ghz on 1-2 cores with normal cooling , its not clear if thats possible yet look at the AMD rumor 5ghz all core boost ended with 4.2ghz , with 2 more cores 8-10 more heat i would be happy with 5.6ghz as max OC. its more likely that the frq is in line with the 9900k and a IPC gain of 15-20% or more ( ipc mesured at same FRQ) but that is in the future , time will tell.

Thanks for the detailed info Westman.

As a matter of interest, you say the i9 9900k with HT set to OFF, is your best option. What estimate would you place on FPS and Smoothness gains, compared to to the i7 7700k. (I am guessing FPS would only be around 20 to 25 percent better ?)

I currently run the i7 7700k at auto turbo..... all 8 threads at 4.5. But as I run at 30 fps fixed, wonder if there is a benefit to upgrading to the i9 ?

I really want 60 fps at least for low n slow vfr, but if I try that currently,  I can get 60 for a lot of the time, but tend to get a lot of blurry scenery coming down to land and autogen that has disappeared. 

Fixing FPS at 30, cures all that.

Maybe I am better waiting for a large increase in power, from the next or one after.... the 11 or 12000k ?!

6 hours ago, Gabe777 said:

Ah I understand now. Sorry.

So an indirect test of cpu core usage is the scenario loading time.

Got it ! Clever. #impressed.

Thnx

It's not going to be precise, gives you more than you need probably. We move from scene to scene and look in different directions. But a general Scenario well chosen takes as long as it takes to load up. At the start, that's the most significant run of throughput we can see as the background pulls in everything. From there, those background threads just chirrup along pulling in data. 

Edited August 8, 20196 yr by SteveW