9 hours ago, mSparks said:

which screenshot are you referring to?

I said that std:thread was only added in 2018. It is absolutely true that std:thread was only added in 2018, despite having been the standard that everyone was using for 25 years by then and being defined as a standard pretty much a decade earlier.

They still don't, and you've not offered any evidence that windows can manage anything like:

You really shouldn't have any problem at least matching that performance if what I say isn't true. It was on an Intel 2700k bought in 2011 and GTX1070 bought around 2015, with as you can see from system monitor 32GB of DDR3 ram.

My latest build:

And the relevance to XP12 update is XP11 has massive headroom to get better graphics out, however I keep hearing from windows people that its poorly optimised, doesn't run well and doesn't make use of the whole processor.

And my response is "they must be windows problems", because ^^^^^

Michael from LR said it himself in a video. They are only using about 2 threads even on Vulkan. It's not a windows things. It's their engine and SDK that is very single threaded in nature.

Regarding the clouds in XP12: imho they look great!!

8 hours ago, mSparks said:

which screenshot are you referring to?

The one you posted on 10/21/21 at 6:02 PM. It shows that going from 4 to 8 threads, there is a solid +98.36% boost. From 8 to 12 threads, the boost is even more than 100%! This is excellent, principally lossless scalability. You are proven wrong by your own image.

8 hours ago, mSparks said:

It is absolutely true that std:thread was only added in 2018, despite having been the standard that everyone was using for 25 years by then and being defined as a standard pretty much a decade earlier.

But you ignore, that many other, native and perfectly capable threading APIs were there already 20 years earlier.

You should not talk about these topics, if you think that std has the only relevant threading API, do not know that std::thread is not called std:thread and do not know that std::thread was introduced in 2011 so your whole claim regarding timelines is for the bin.

8 hours ago, mSparks said:

They still don't, and you've not offered any evidence that windows can manage anything like:

Windows certainly can (see once more your screenshot). But, as I explained, it is not the fault of the OS, if one particluar software does not make good use of parallelization. Other, more modern sims seem to do better.

8 hours ago, mSparks said:

And my response is "they must be windows problems", because ^^^^^

Have you thought about the possibility, that X-Plane is making poor usage of resources under Windows? Other sims seem to do better.

5 hours ago, tonywob said:

Looks very much like a certain other sim 🙂

From what I've seen of "that other sim", these clouds look a bit more detailed and less cotton candy-ish. I like them.

 

Sidney also commented on the bug tracker for the Mesa graphics library today and stated that LR will fix the Vulkan-OpenGL bridging issue plaguing AMD GPUs on Linux. Definitely for XP12, and maaaaaybe also for 11. This is huge news as it will eliminate a huge question mark that I have about 12 and its future on that GPU and OS combination.

Edited October 27, 20214 yr by Bjoern

2 hours ago, Janov said:

Yes it will.

Is it a development of that STOL quad-prop Austin was helping with some time ago?

I'm generally skeptical about this "STOL GA" craze, but this seems the one with the most reasonable configuration among the many ones I've seen so far.

 

1 hour ago, fogboundturtle said:

They are only using about 2 threads even on Vulkan. It's not a windows things. It's their engine and SDK that is very single threaded in nature.

doesn't matter, the linux kernel upscales that "completely fairly" across all cores, which is why you see all 8 HT being used to 100% during scenery load and all 4 cores being used to 100% during flight.

 

4 minutes ago, mSparks said:

doesn't matter, the linux kernel upscales that "completely fairly" across all cores, which is why you see all 8 HT being used to 100% during scenery load and all 4 cores being used to 100% during flight.

 

You don't get it whatsoever.  It's impossible to have any type of technical discussion as Linux doesn't have a "magic" scheduler. It's been documented for year that Linux had a real bad scheduler for AMD processor for years. The XP11 engine is very old. It actually perform worst when HT or SMT is enable. The whole engine was design when cpu only had 2 cores. It needs to be modernized that included the the capability in the SDK to multithread avionics or other functions. 

2 hours ago, fogboundturtle said:

It's impossible to have any type of technical discussion as Linux doesn't have a "magic" scheduler.

No it doesn't, it has a good scheduler, which distributes even single threads over multiple cores very effectively, Much like the new Mac M1 does at the hardware level rather than software.

Windows Home and Pro (for processes rather than applications) has a very simple scheduler that puts multiple threads onto a single core even when it has spare cores (because they weren't next in the round robin). 

Windows Enterprise/server includes a slightly better scheduler (that starts to get close to Linux) which can get you close to Linux performance.

3 hours ago, mrueedi said:

shows that going from 4 to 8 threads, there is a solid +98.36% boost. From 8 to 12 threads, the boost is even more than 100%!

Sure, you can get that performance to, just download and buy that copy of windows here:

https://www.microsoft.com/en-us/evalcenter/evaluate-windows-server-2022

Just no piracy OK, Once you see its true and if its got to be windows stump up the $1069 for 16 core support and stop complaining its Laminars fault cheap/free windows doesn't have it.

2 hours ago, fogboundturtle said:

The XP11 engine is very old.

No it isn't, its about 6 months old, that is not "very old".

2 hours ago, fogboundturtle said:

It actually perform worst when HT or SMT is enable.

On windows yes, because windows treats those extra core threads as no different to a real core so you end up more likely to run everything on 1 core and/or 1HT instead of 1 or 2 full cores.

2 hours ago, fogboundturtle said:

The whole engine was design when cpu only had 2 cores.

Prior to XP11.30 yes, we have moved on since then.

2 hours ago, fogboundturtle said:

It needs to be modernized

Nope, microsoft windows does, its not LRs responsibility to write the operating system XP runs on to include modern capabilities used BY EVERY OTHER OPERATING SYSTEM - including "stupidly expensive windows versions".

If you are holding out for Asobo or Laminar to rewrite the windows NT kernel for home/pro versions of windows you are going to be waiting a very very long time.

Edited October 28, 20214 yr by mSparks

Firstly one should understand that Linux was never designed to be used as a desktop , hence it's power management was always on full throttle , the whole hardware was full throttle and still is and yes it performs much faster , though there is a power management one can use for linux , my tests were Blender / Gimp / XP11 , XP11 is just smooth , it's not a fps gain but it's miuch smoother and the system reposnse is much better , even blender / gimp just load faster .

  Any server will use full throttle , yeah there is a difference between windows server  / Linux server in performance , to me the usefulness of a windows server is only till AD. Windows based software developement is entirely different,

Windows Desktop OS otoh was designed with power management in mind , hence expect some difference even if used of FUll performance mode.

Edited October 28, 20214 yr by Humpty

Guys, can you please take the Windows/Linux to another topic and let everyone else talk about XP12... Thanks

11 hours ago, fogboundturtle said:

The XP11 engine is very old.............

Don't want to join this BS parade while LR are posting pics every couple of days that look better and better each time, but oh god not this again.

The "very old" stigma are graphics features (i.e. billboard clouds vs the new clouds AKA the bigger word is "rendering pipeline" features which is what Xplane12 is all about), but Xplane's flight model, scenery loading, road traffic and many other crucial aspects are fully multi-threaded under the friendly Vulkan environment. 

The problem is not "Xplane is not multithreaded", it is actually that it might be too much in that regard when plugins are involved, the exact opposite of your claim, it is an actual concern brought a couple of months ago in the slack channel.

Now please guys just leave this thread alone with your nonsense, no one cares even more when most of the claims are totally wrong.

 

Edited October 28, 20214 yr by akita

4 hours ago, akita said:

Don't want to join this BS parade while LR are posting pics every couple of days that look better and better each time, but oh god not this again.

The "very old" stigma are graphics features (i.e. billboard clouds vs the new clouds AKA the bigger word is "rendering pipeline" features which is what Xplane12 is all about), but Xplane's flight model, scenery loading, road traffic and many other crucial aspects are fully multi-threaded under the friendly Vulkan environment. 

The problem is not "Xplane is not multithreaded", it is actually that it might be too much in that regard when plugins are involved, the exact opposite of your claim, it is an actual concern brought a couple of months ago in the slack channel.

Now please guys just leave this thread alone with your nonsense, no one cares even more when most of the claims are totally wrong.

 

Too multithreaded ? come on. You can say that with a straight face. Even LR research says that their application isn't.  I have no idea about XP12 multithreading since they haven't talked about it but Xplane 11 is one of the last multithreaded flight sim on the market. The fact that if you disable HyperThreading or SMT , Xplane 11 runs better is an indication of how the engine is not design for modern CPU. This is not rocket science. If Xplane 12 resolve this issue, good for them. It will be a welcome change. 

56 minutes ago, fogboundturtle said:

The fact that if you disable HyperThreading or SMT , Xplane 11 runs better is an indication of how the engine is not design for modern CPU.

In my opinion this only reflects somethings else where at least I'd agree with @mSparks here... from my experience coding my own implementation of a task scheduler a while ago.

In short: although on the hardware level a HT "core" is not a full core by itself, the simpler Windows API used to enumerate the CPU topology is not distinguishing the two. In other words, a CPU with 8C/16T is seen as 16C from the API standpoint. Under these conditions, if you're building multi-threaded app and if you pin threads on cores (like you'd do most likely for a task scheduler), you'll end up pinning 2 threads on the same physical CPU core: the main one, and its HT sibling.

This is the reason most often than not you get better perfs with Windows games when disabling HT in the BIOS, or when setting the affinity mask to mask off every other CPU (i.e. the HT siblings).

edit: about the Windows API in more details

There are 2 principle ways that I know of to get the topology: the simpler way, and the complex way. Most projects I've seen are using the simpler way GetProcessAffinityMask but this one doesn't distinguishes HT cores from real cores. The "complex way" instead is GetLogicalProcessorInformation which is more demanding but complete.

However: https://docs.microsoft.com/en-us/windows/win32/api/winnt/ns-winnt-system_logical_processor_information

Windows Server 2003 and Windows XP Professional x64 Edition:  This member is also 1 for cores that share a physical package. Therefore, to determine whether the processor supports multiple cores or hyperthreading on systems prior to Windows Vista, use the CPUID instruction.

 

Edited October 28, 20214 yr by RXP

2 minutes ago, RXP said:

In my opinion this only reflects somethings else where at least I'd agree with @mSparks here... from my experience coding my own implementation of a task scheduler a while ago.

In short: although on the hardware level a HT "core" is not a full core by itself, the Windows API used to enumerate the CPU topology is not distinguishing the two. In other words, a CPU with 8C/16T is seen as 16C from the API standpoint. Under these conditions, if you're building multi-threaded app and if you pin threads on cores (like you'd do most likely for a task scheduler), you'll end up pinning 2 threads on the same physical CPU core: the main one, and its HT sibling.

This is the reason most often than not you get better perfs with Windows games when disabling HT in the BIOS, or when setting the affinity mask to mask off every other CPU (i.e. the HT siblings).

 

You can control CPU affinity if you wanted too but this is does not hide the fact that when running Xplane 11, It only uses about 30% of my GPU (3080Ti). This clearly indicated a core bottleneck. Now they talked about moving operation that were performed by the CPU to the GPU to try to relieved that core bottleneck. I am curious to see how this will perform. I want Xplane to take fully advantage of my PC to improved visuals and performance. When I see an aircraft like the IniBuild A300 barely running at 30fps on a system like mine, it's a shame. You can clearly see that IniBuild is confined with the current limitation of the engine. FPS does matter in Xplane as when you get below 20fps, it start messing with the flight dynamics which is why VATSIM will kick you off their network if you go below 20fps. Theses are all things that requires modernization of the Xplane engine. I've seen the latest screenshot from a visual perspective , I think it looks great so I am looking forward to the public beta (one day)

9 minutes ago, fogboundturtle said:

You can control CPU affinity if you wanted too but this is does not hide the fact that when running Xplane 11, It only uses about 30% of my GPU (3080Ti). This clearly indicated a core bottleneck. Now they talked about moving operation that were performed by the CPU to the GPU to try to relieved that core bottleneck. I am curious to see how this will perform. I want Xplane to take fully advantage of my PC to improved visuals and performance. When I see an aircraft like the IniBuild A300 barely running at 30fps on a system like mine, it's a shame. You can clearly see that IniBuild is confined with the current limitation of the engine.

Look at the bright side: plenty of unused power under the hood to make XP12 running way better than XP11 !

1 minute ago, RXP said:

Look at the bright side: plenty of unused power under the hood to make XP12 running way better than XP11 !

100% agree with you on that. Looking forward to XP12 and see if I can get these numbers up.