big.LITTLE impact on P3D performance - Intel 12th gen CPU?

[787flyer 633]

Hi Folks,

 

I was reading up on the details surrounding Intel's 12th generation CPU (Alder Lake) that will feature a change in CPU design, the so-called big.LITTLE design.

 

I am wondering what effect that will potentially have on P3D performance? Do you have any thoughts or predictions of whether the supposed overall performance gain will positively impact sim performance given that the high priority threads will run mainly on the big core and other threads running potentially on the "background" or LITTLE cores?

Will LM need to adapt P3D to take advantage of this thereby could we be looking at another P3D update in the not too distant future?

 

Would be interested in your thoughts on this.

 

[micstatic 4,316]

I'm also interested in this mark.  Following

[787flyer 633]

2 hours ago, micstatic said:

I'm also interested in this mark.  Following

Yes, I may consider building a new rig if its worth it, but I'd like to here from the CPU / P3D experts around here! 

[Howellerman 104]

Interesting. The various OS and application vendors have had to deal with NUMA (Non-Uniform Memory Architecture) for some time, and rather successfully. 

Now they are being offered a NUCA (Non-Uniform Core Architecture). I wonder if Winders Eleventy is going to have a fit?

[Luke 782]

The change that will be required is an update to the Windows scheduler, not P3D or any applications. Hardware abstraction has been a thing for over a quarter century now.

Cheers!

 

[fluffyflops 1,847]

Its a 20+ year old sim

It amazes me how people think its some cutting edge new platform.   god knows I wish it was.  but it isnt.  

no amount of snake oil is going to make it what you lot all want it to be.  harsh but true.

 

Edited July 24, 2021 by fluffyflops

[neilhewitt 493]

4 hours ago, fluffyflops said:

Its a 20+ year old sim

It amazes me how people think its some cutting edge new platform.   god knows I wish it was.  but it isnt.  

no amount of snake oil is going to make it what you lot all want it to be.  harsh but true.

True, but equally there's no reason to assume, nor is it likely, that a 'cutting edge new platform' would perform substantially better on the same hardware. The idea that the rendering engine of P3D is a morass of crufty old code that's somehow escaped being optimised in some simple way that would double our FPS is daft. LM has been continually developing this code for nearly 15 years. Most of the engine has been thoroughly touched since FSX in order to do the stuff that P3D now does. The port to 64-bit will have required touching quite a lot of the code throughout the sim. And the DX12 overhaul will have involved touching a lot of of it yet again. I'm sure there are ideas for optimisation on LM's backlog that haven't been prioritised and maybe never will be, but I doubt they would have deliberately ignored the opportunity for a significant performance uplift if one existed.  

Could there be different algorithms to be used in the simulation that might work better on today's multi-core processors? Maybe. X-Plane famously uses blade element theory where FSX and P3D use lookup tables, and MSFS claims to have developed a new model altogether. Clearly, MSFS can render more scenery objects than P3D for the same FPS. So there might be an opportunity to overhaul some elements of P3D similarly. But the MSFS forums have been full of complaints that the sim is getting slower, and the eye-candy levels going backwards, as it's getting updated with more stuff to occupy the main thread. When (or frankly if) they add multi-view support, people will have the same FPS-drop-per-view-opened as they do in P3D; it's just inevitable because there's only so much CPU time available on the main thread. I suppose if they wanted to be radical about it, they could basically make the sim a set of VMs where each view has its own rendering engine instance and main thread, but you'd still have to coordinate stuff between those threads and the locking required to do it would probably eliminate most of the performance gains... also, the GPU would quickly become the limiting factor, unless you had a GPU-per-view, in which case PCI bandwidth would become the limiting factor, and so on and so on. 

Based on what I know about concurrent programming and games programming (the former, a decent amount; the latter, not so much) I don't believe a fundamentally parallelised rendering engine (ie, one that could truly spread its load over multiple cores) is possible (or if possible, likely to be achievable within the development cost boundaries of a saleable product), so the problem of a 'main thread' that must run as fast as possible will persist in any future sim.

It does look like Alder Lake will deliver substantial single-thread performance uplift, though, if the leaked benchmarks are true, and so 12th gen chips may well offer an FPS boost, and potentially quite a big one. I'm sure one of the usual suspects here will volunteer to upgrade and test for us when the time comes 🙂...

[simfan1983 185]

11 hours ago, fluffyflops said:

Its a 20+ year old sim

It amazes me how people think its some cutting edge new platform.   god knows I wish it was.  but it isnt.  

no amount of snake oil is going to make it what you lot all want it to be.  harsh but true.

 

Windows is the same, original version was around 1985 and MS have been putting lipstick on a pig ever since. MSFS uses a lot of FSX code. In some large businesses, platforms have been around since the late 60s and develop through evolution not revolution. Many game engines are very old and improvements are transplanted in through new graphics engines like Vulcan. The age of the original software is not all that relevant IMO.

For the time being P3D is still the best flightsim by a country mile, all things considered. I think it's the MSFS word not allowed that are dreaming of a brighter future to be honest.

[d.tsakiris 417]

11 hours ago, fluffyflops said:

Its a 20+ year old sim

I think enough of the original code has been changed to render that statement false. Same goes for Windows.

[fluffyflops 1,847]

10 hours ago, neilhewitt said:

True, but equally there's no reason to assume, nor is it likely, that a 'cutting edge new platform' would perform substantially better on the same hardware. The idea that the rendering engine of P3D is a morass of crufty old code that's somehow escaped being optimised in some simple way that would double our FPS is daft. LM has been continually developing this code for nearly 15 years. Most of the engine has been thoroughly touched since FSX in order to do the stuff that P3D now does. The port to 64-bit will have required touching quite a lot of the code throughout the sim. And the DX12 overhaul will have involved touching a lot of of it yet again. I'm sure there are ideas for optimisation on LM's backlog that haven't been prioritised and maybe never will be, but I doubt they would have deliberately ignored the opportunity for a significant performance uplift if one existed.  

Could there be different algorithms to be used in the simulation that might work better on today's multi-core processors? Maybe. X-Plane famously uses blade element theory where FSX and P3D use lookup tables, and MSFS claims to have developed a new model altogether. Clearly, MSFS can render more scenery objects than P3D for the same FPS. So there might be an opportunity to overhaul some elements of P3D similarly. But the MSFS forums have been full of complaints that the sim is getting slower, and the eye-candy levels going backwards, as it's getting updated with more stuff to occupy the main thread. When (or frankly if) they add multi-view support, people will have the same FPS-drop-per-view-opened as they do in P3D; it's just inevitable because there's only so much CPU time available on the main thread. I suppose if they wanted to be radical about it, they could basically make the sim a set of VMs where each view has its own rendering engine instance and main thread, but you'd still have to coordinate stuff between those threads and the locking required to do it would probably eliminate most of the performance gains... also, the GPU would quickly become the limiting factor, unless you had a GPU-per-view, in which case PCI bandwidth would become the limiting factor, and so on and so on. 

Based on what I know about concurrent programming and games programming (the former, a decent amount; the latter, not so much) I don't believe a fundamentally parallelised rendering engine (ie, one that could truly spread its load over multiple cores) is possible (or if possible, likely to be achievable within the development cost boundaries of a saleable product), so the problem of a 'main thread' that must run as fast as possible will persist in any future sim.

It does look like Alder Lake will deliver substantial single-thread performance uplift, though, if the leaked benchmarks are true, and so 12th gen chips may well offer an FPS boost, and potentially quite a big one. I'm sure one of the usual suspects here will volunteer to upgrade and test for us when the time comes 🙂...

A lovely long nicely written post. 

You missed out the bit saying that all the technology isn't going to fix dynamic lighten killing the sim 

Its not going to fix that if you put a ton of animated burger King workers in KORD restaurants to get kids to buy it (FS dreamteam)  by adding eye candy to get sales that kills the sim

It's not going to fix when adding a load of effects like flytampa does, once again to make sales which kills the sim

No amount of super cpus or 3090ti is going to fix a 20 plus year old game. 

You guys are living in dream world,  I applaud you for positive thinking, but it's 20 years old sim

Meanwhile at 60 gigs I can play red dead redemption 2 on a huge map with an massive amount of extras including, animals, ai people, flowers that grow, all in amazing detail. And as smooth as hell. 

Why? Because its brand new, using new technology and an army of people that developed in costing millions and millions and millions.   Its breathtaking. 

It's a world away from a few people at LM trying to make good a 20 year old botched attempt by ACES 

 

Edited July 25, 2021 by fluffyflops

[Ray Proudfoot 8,930]

@fluffyflops, I've approved your post even though it's inflammatory in the hope that sensible dialogue will follow. If P3D appalls you so much why use it?

If you continue to hammer LM's product and 3rd party addons future posts may not be approved. Last chance from me.

[Luke 782]

6 hours ago, simfan1983 said:

Windows is the same, original version was around 1985 and MS have been putting lipstick on a pig ever since.

Having used every single version (some a lot less than others) since Windows 1.0 on a 286 in the late 1980s, this is categorically false. The NT-based kernels most folks have been using since 2001 (and some of us from even earlier) are a completely different beast than original windows. All they share is the name.

 

[simfan1983 185]

35 minutes ago, Luke said:

Having used every single version (some a lot less than others) since Windows 1.0 on a 286 in the late 1980s, this is categorically false. The NT-based kernels most folks have been using since 2001 (and some of us from even earlier) are a completely different beast than original windows. All they share is the name.

 

My point was that it makes no difference what the age of the original code was as most software continues to develop in an evolutionary fashion. Some of the steps are bigger than others (such as NT). I doubt that P3D shares a lot of code with FSX (apart from the bits that work well enough and don't need re-writing).

I've also used Windows since v1.0 and MS flightsim since v2.0. Windows has wasted more of my hours fixing issues than I could count. I think lipstick on a pig is a perfectly fair description.

[jabloomf1230 2,445]

2 hours ago, Ray Proudfoot said:

@fluffyflops, I've approved your post even though it's inflammatory in the hope that sensible dialogue will follow. If P3D appalls you so much why use it?

If you continue to hammer LM's product and 3rd party addons future posts may not be approved. Last chance from me.

You should have cut this off with his first post. Using the same logic, Tesla should stop building cars because their products share so much in common with a Model T Ford.

As you know, I'm a proponent of MSFS, but it's not THAT much better than P3d5 and XP11. These threads produce nothing of value except for people that like to point out that their one is bigger.

[simfan1983 185]

To get back on topic, any processor that helps with single thread CPU performance will be worth looking at for P3D. I'm still running v4.5 with a 8700k at 4.5 GHz and with the graphics settings reasonably high I get no stutters. A 5+ GHz CPU would probably enable me to run higher settings for even more immersion but it wouldn't make a huge difference.

Every Twitch stream I see of MSFS is a stutter fest when using a reasonably complex aircraft so I'm happy with things as they are. Maybe after the next update MSFS will be competitive with P3D but at the moment it's just not.