Rob - Have you tried TrackIR in AFRFriendly? 

 

Vic

 

I'm not Rob, but I'm using AFRFriendly since 2.2 (and now 2.3) and 90% of my flight are with TrackIR. No issue reported so far, except in 2.2 the exagerad spinning cloud (not verifying yet in 2.3) not related to AFRFriendly trick.

 

NB : SLI with "only" 2 x GTX 770 ;-)

So is it the case then that if we get eventually get a P3D V2.X Profile from Nvidia with SLI support that it will deliver greater performance if all other things stay the same?

 

Reading some posts I get the impression that some hold the opinion that SLI will not really make much of difference because of the already bottlenecked CPU?

It's not that P3d is CPU bottlnecked in a classical sense. It's that P3d's main thread pins one CPU core at 100% and that limits everything. SLI will be great for keeping performance up on 1440p and 4K systems, but only a new higher end Haswell E CPU will address the main thread issue.

It's not that P3d is CPU bottlnecked in a classical sense. It's that P3d's main thread pins one CPU core at 100% and that limits everything. SLI will be great for keeping performance up on 1440p and 4K systems, but only a new higher end Haswell E CPU will address the main thread issue.

 

Not sure what you mean by "classical" sense, but if it can't load multiple core's for "main" load, then it is CPU bottlenecked. Pity, I was expecting much from compared to FSX.

 

I have 755m in SLI, a very modest mobile GPU, and all I see is that GPU loads around 50%.

 

 

So is there any point left of the AFR friendly option?

 

For me, the AFRFriendly approach simply will not work with "some" key 3rd party products and it has other issues.  So for me, the answer no further testing with AFRFriendly approach. The other approach using prepar3d.exe with those control panel and NI options will work but I'm not seeing much of an improvement ... it's going to need nVidia driver support.

 

There is always going to be some CPU bound operations ... that's just the nature of flight simulation, however, with driver support some of the CPU load could be reduced moved to the GPU.

 

I still haven't tried overriding AA settings via NI and setting AA in P3D to none ... this doesn't usually work well, but something I need to try.  My approach is basically trying to force the GPU to do more work and figure out how to have the CPU do less work ... so the basic approach is reduce settings in P3D and push setting in NI (those that are GPU based, such as filters, AA, etc.)

 

I've also bumped my res up recently to 3840 x 2160, so have re-tested again.

 

Cheers, Rob.

Ah! I see now, SLI is not doing much now.

Not sure what you mean by "classical" sense, but if it can't load multiple core's for "main" load, then it is CPU bottlenecked. Pity, I was expecting much from compared to FSX.

 

I have 755m in SLI, a very modest mobile GPU, and all I see is that GPU loads around 50%.

Before games and other 3D apps were heavily multithreaded (and I'm being very simplistic here), the CPU would draw the wireframe of objects and the GPU "painted on" the textures. It was important to balance the computational power of both your GPU and CPU, else one component would end up waiting for the other to complete it's task. Hence, the slowest component would be the bottleneck.

 

P3d is very heavily multithreaded on the CPU and also makes much better use of the GPU than FSX. Hence, for the same GPU and CPU combination a typical sim user will see much better frame rates in P3d for settings that are comparable to FSX. Or, P3d will produce much better quality than FSX for the same settings at a similar frame rate. One difficulty is that comparing display settings in the UIs of both sims is not that easy to do, since what is "High" in P3d is usually a higher setting than "high" in FSX.

 

The problem facing developers of multi threaded 3D simulation legacy apps is that no matter how much you offload from the CPU to the GPU, you are still left with a lot of work to be done on the CPU. And there has to be one main thread that in essence keeps track of everything else that is going on simultaneously. Eventually, securing more and more GPU power has diminishing returns, as the main thread running on one core of the CPU, pins that core at 100%. The result is that the user sees no increase in frame rates and both the GPU and total CPU use may neither ever approach 100%. I commonly show 50-99% GPU use and 40-60% CPU use, but one CPU core is almost always at 100%. But I have a GPU (GTX 680 4GB) that is now two generations old.

 

I would say that until someone here has purchased a system with a Intel Core i7-5960X Extreme Haswell-E CPU, I don't think that we will really be able to tell what the true advantages of SLI will be with P3d. And that's assuming nVidia gets off its duff and provides an SLI-aware profile for P3d.

Thanks!

Great explanation!

 

And I will add that I really can wait because 2.3 runs great on my system

and I couldn't be happier:-)

I can wait also. Prior to the GTX 680 I had two GTX 580s in SLI  ( I could actually run four in SLI on my setup) and that worked well for some apps and not so well for others. SLI is really designed for hard core enthusiasts who are trying to win benchmark titles.

SLI is really designed for hard core enthusiasts who are trying to win benchmark titles.

 

Or people that want to get closer to "as real as it gets" -- I certainly have no interest in benchmarks, I'm the crazy guy that's good with 20 fps 

 

Heck I don't even have a full cockpit setup ... however that is next on my list.  I'd never discourage folks like Pete Dowson and others pushing the simulation envelope to get that much closer to reality ... it's that desire that has helped and driven others.

 

Cheers, Rob.

It's not like I haven't messed around with SLI either and I applaud your efforts at pushing the limits of the P3d graphics engine.

 

It's just that nVidia has always treated SLI like an afterthought that is more of a marketing ploy than a mainstream technology. I'm surprised that they don't see the PR value in P3d. People always point to Crysis as the torture test for gaming systems. P3d is much tougher. Maybe nVidia just sees P3d as some legacy app which isn't completely based on GPU computing. I hope that they wake up and change their minds.

Before games and other 3D apps were heavily multithreaded (and I'm being very simplistic here), the CPU would draw the wireframe of objects and the GPU "painted on" the textures. It was important to balance the computational power of both your GPU and CPU, else one component would end up waiting for the other to complete it's task. Hence, the slowest component would be the bottleneck.

 

P3d is very heavily multithreaded on the CPU and also makes much better use of the GPU than FSX. Hence, for the same GPU and CPU combination a typical sim user will see much better frame rates in P3d for settings that are comparable to FSX. Or, P3d will produce much better quality than FSX for the same settings at a similar frame rate. One difficulty is that comparing display settings in the UIs of both sims is not that easy to do, since what is "High" in P3d is usually a higher setting than "high" in FSX.

 

The problem facing developers of multi threaded 3D simulation legacy apps is that no matter how much you offload from the CPU to the GPU, you are still left with a lot of work to be done on the CPU. And there has to be one main thread that in essence keeps track of everything else that is going on simultaneously. Eventually, securing more and more GPU power has diminishing returns, as the main thread running on one core of the CPU, pins that core at 100%. The result is that the user sees no increase in frame rates and both the GPU and total CPU use may neither ever approach 100%. I commonly show 50-99% GPU use and 40-60% CPU use, but one CPU core is almost always at 100%. But I have a GPU (GTX 680 4GB) that is now two generations old.

 

I would say that until someone here has purchased a system with a Intel Core i7-5960X Extreme Haswell-E CPU, I don't think that we will really be able to tell what the true advantages of SLI will be with P3d. And that's assuming nVidia gets off its duff and provides an SLI-aware profile for P3d.

I cannot seem to find any reference about this but No one has addressed the use of a duel CPU system, my guess it has been tried and not worthy of the results, has it been tried?

I cannot seem to find any reference about this but No one has addressed the use of a duel CPU system, my guess it has been tried and not worthy of the results, has it been tried?

Dual cpus dont really help much but I could be wrong - I have dual xeon machine at work not impressed

People always point to Crysis as the torture test for gaming systems. P3d is much tougher.

 

Agree ... DX12 should help give developers more control over Multi-GPU processing and leverage.

 

 

I cannot seem to find any reference about this but No one has addressed the use of a duel CPU system, my guess it has been tried and not worthy of the results, has it been tried?

 

More CPUs will only help if your main synchronizing thread (single CPU) isn't already maxed out.  100% CPU utilization across all cores is never going to happen in a flight simulator ... I've explained why many times.  Now multiple PC's is viable, because each PC effectively has it's own "main" synchronization process, however they are bound by the efficiency of communication across those PCs.

 

However, if the bus isn't saturated, more CPUs could be used to do other things like TrackIR, ASN, EFB, FRAPS, etc. etc.  In these situations you probably want to limit what P3D uses and control what CPUs the other products use.  So more cores isn't a complete waste if you manage who's on first, 2nd, 3rd, etc. 

 

Cheers, Rob.

In the FWIW department - I've noticed something different between 2.2 and 2.3 using both AFR-Friendly and the NI SLI.

 

Using the EVGA or the NI utiolity to monitor GPU usage - in 2.2 I would get a distributed load across the two cards - about 50% and 50%.

 

Using 2.3 in similar heavy cloud etc - I am getting in the 90% - 100% usage for EACH card.

 

I've noticed that if it stays at 100% for long - FPS start to degrade.

 

I turned HT ON and using all cores there seems to be a subtle improvement in the pesky micro stutters.

 

IMHO, the overall improvement with this SLI would not be worth the $$ of adding a second card - that is until NV gets around to providing a profile.

 

Vic