The online racing simulator
I did it via windows, when I had win server 2012 r2 installed, and it was a royal pain in the rear. It did not seem to have any effect, all it really did was slow it down.

Will try it again soon, now that Win 10 is installed.
Changing affinity does not seem to do much, if anything for LFS.

Changing the Win 10 system to run background programs at equal speed does not improve things either.

I can currently get about 8 LFS running at about 30 FSP. But it drops to 20 or less every once in a while. I have to limit the FPS to 30 to do this, or some windows will pull frames away from the others.

If I click on one of the LFS windows, to give it focus, the LFS window with focus jumps up to like 100 FPS (if I do not have it limited to 30).

Is there a way to prevent LFS from slowing down when it does not have focus? I believe this is why it is jerky. The other windows are not running as a background window.

I have so far only been able to run in full-screen-windowed mode.
I cannot run LFS in true full screen mode, because it locks the mouse to the monitor with focus in this mode, and ALT TAB is required to change between windows... which minimizes the other LFS windows, so I cannot get more than 1 visible.
I don't believe this is a general priority but a graphic-power problem. I'm no expert but it is known LFS consumes much more resources in object intensive situations (like it is the case in every game surely).
That FPS are jumping up in full screen might be because you left the part of the course which has that many objects and/or windows indeed gives the application more resources there. But if this belongs to all instances you could be on the start again.
Some things I would try:
- Did you monitor the level of CPU and especially graphic card use?
- Trying to map graphic card resource (reservations) in graphic card driver/tool level to the game instances instead of on operation system level.
This is not due to the car's location on the track, all the cars were AI driving together in a multiplayer game anyway. I've done MANY tests, and I am SURE it is not that. In most of my tests, I put the view from cars further back in the line, so it works harder to render more cars up front, to get a worse case scenario.

Windows 10, runs the window with focus faster. I can get them all running unlimited at about 20-30 frames, but if I click on one of the LFS windows, it jumps to 120-200 fps. And the others drop to 15-25 fps.

Right now the only solution is to limit them all to 30FPS, so the one with focus does not take over. I tried 50FPS, which almost works, wish I could set it to 40FPS.


I've no idea how to allocate graphics card resources...
Not available on the stuff I am using.
Fire Pro (which is part of my problem)

I do know that only about 10-20% of the Total CPU is being used.


This whole thing is compounded by another problem I have.
I have 2 graphics cards, each with 6 Mini Display ports, so a total of 12.

In my last test, I had 6 monitors plugged into one card, and a 7th plugged into the other card on its own. The LFS window running on the 7th lone monitor, always ran at a MUCH lower FPS, like 10-25FPS, when the others are running 30-100FPS. Even if I swapped and moved a different LFS window to that monitor.
Any ideas as to why that is?!

They are NOT crossfire linked, because that will disable half of my display ports.

If this keeps up I may need to give up on the single server idea, and go with separate NUC computers. And that will just cause electric power demand issues.
It is quite possible that Windows try to assign more resources to the foreground application, sacrificing anything that runs on the background. One thing you may try is disabling desktop composition (http://www.sevenforums.com/tutorials/127411-desktop-composition-enable-disable.html), the guide I found is for Win7 but hopefully it hasn't changed in Win10. You should definitely pin each LFS instance to one CPU core. If you don't do that, LFS processes will bounce around all the cores causing CPU cache misses and inducing extra overhead for the CPU scheduler. There is also the question of how Windows manages the multiple non-crossfired GPUs in your machine. It's possible that all LFS instances are being processed on the primary GPU and the rendered image is copied to the secondary GPU's memory so that in can be displayed on a monitor hooked up to the second GPU.
..
Quote from cargame.nl :Tried VMWare yet? Or anything similar.

Yes he did and he eventually ditched it. I also don't see how you'd get a decent 3D acceleration from within VMWare guests.

I've been thinking that building this shared gaming machine on top of Linux and run LFS through WINE might be an interesting option. It would however require decent amount of experience with Linux and Xorg configuration to get it working.
..
I had VMware working when I first started, however LFS was upgraded, and the new graphics code did not work properly in VM Ware.

I just tried running it with the 2 cards crossfire linked, the performance was no better... only change was less display ports.

I think my biggest issue is using workstation graphics cards...

I wonder if 2 or 3 of these would work better:
http://www.newegg.com/Product/Product.aspx?Item=9SIA1UH4BY6751&nm_mc=KNC-GoogleMKP-PC&cm_mmc=KNC-GoogleMKP-PC-_-pla-_-Video+Cards+-+AMD%2FATI-_-9SIA1UH4BY6751&gclid=CN7Ourfu7s0CFQqoaQodDy8FKA&gclsrc=aw.ds

If Only I could get AMD to trade my two $3000 W9100 cards for three of those... and let me test it out see if it is any better.
I sort of wonder if the power is such a big deal with per-seat computer... I mean, some reasonable mini board with mobile CPU (if some chipset gfx can handle lfs well?), booting from network (having only one HDD on one stronger machine, maybe single power source would be able to power cluster of 3-4 boards.. and most of the stuff would be decentralized and easier to replace in case of failure.

But I'm lazy bastard, not going to pick real parts and calculating real power consumption, so you have better idea. Smile I was just since your first post worried about your centralized way, I somehow don't like these things, where single failure takes down everything.
I currently get about +10% more FPS running LFS with Linux using wine on thu same hardware. It could be worth a try, with less overhead and more configurabilty.
With his workstation gfx card, 12x wine and FFB wheels he may run into huge (missing good drivers) problems with linux.. then again, what's the deal, just to put some decent recent distribution on the usb stick, and boot from it for test. If you would be lucky to have working drivers for all the HW, the rest will be working better than win.
That'd depend a lot on the hardware. Recent Linux distro and nVidia card would most likely work very well with Logitech wheels.
For the time being I am not really in the mood to mess with *nix OS.

I just bought a RX 480 card. If this $240 card gives better frame rates then the $6,000 worth of firepro cards I have, then it is time to put the fire pro's on ebay, and go with non-workstation cards.

While Cross linked the two FirePro W9100 cards, are in the mid to high yellow on the Steam Vive VR Ready Test.
Why would FirePros be faster in real time 3D rendering ? They are not made for that. And I have a RX 480, and I think it could handle easisly multiple instances of LFS with solid framerate. BTW, don't forget to undervolt it though, stock voltages are too high for that card.
I just ran the test for comparison.

As I said, the firepro was a mistake.
When I started this, no one had made a single machine that would run 8 to 12 players before, so mistakes where expected.

I thought I was erring on the side of caution by getting the extra gpu and ram for each instance of LFS. I thought it would run better if each monitor had its own display port from the graphics card, instead of using an MST hub. I did not think the difference between cad card and game card would be that bad, and I wanted a card for CAD so I bought one while they were on sale.

I may still keep one for doing CAD since I've been intending to install Maya.

I am in a similar boat with the Opteron Processors/Server Motherboard, however, every time I price separate PC's it winds up costing me more in cash, or power. However I do not know just how small I can go with separate PC's and the small ones are getting more powerful, eventually, if not already, LFS will be running well on a HDMI PC stick...


The setup I am using now, with the server, 8 monitors, 8 G29 wheels, is still under 15Amps, but not by much. which is the limit of most standard outlets that I may be given for power at most venue locations. If I split up the server into separate computers with their own power supplies, I believe it will put me far above 15Amps.



Nacim, can you run LFS 8 to 10 times on he RX 480, each running on blackwood with 10 AI? They can all be over lapped on the same monitor, as the frame rate does not change much when moved to separate monitors.
..
Just redid the test again... it runs just fine.
Attached images
SteamVR_2xW9100.png
And it actually does better with only 1 card enabled.

I tried to run Race Room with the two cards linked together, and as the menu and simulator ran, it kept going forward and backward... like it was redoing everything a second time. It was unplayable...

I disabled one of them, and it ran just fine.

LFS feels a LOT better then RaceRoom.
Attached images
SteamVR_1xW9100.png
Quote from SimulatorRental.com :Nacim, can you run LFS 8 to 10 times on he RX 480, each running on blackwood with 10 AI? They can all be over lapped on the same monitor, as the frame rate does not change much when moved to separate monitors.

Just did it with 8 instances, VSync turned on, post process turned off (keep in mind that I have some big pow() in my shaders and post process, nothing good for performance, so if it run like that, and vanilla game will definitively work!)

It seems to be running at 60 FPS on every instance with my i5-3570K @ 4.3 Ghz (65% usage max), and the RX 480 is running from 95% to 100%. I guess it will be fine Smile

Quick screenshots of the experiment:


(see virtual desktops on the bottom)


(don't mind the dark game, it's just because post-process is off and shaders are converting pixel values to linear color space)
Attached images
Capture d’écran (2).png
Capture d’écran (3).png
Thanks,
Is LFS itself reporting 60 FPS?

When I test it I have to have them all open on screen while in cockpit view because it takes a few FPS more. I tile or spread them across monitor in order to see the proper FPS from LFS itself at the same time. All while none of the LFS windows have window focus, because the active instance, gets a big boost, while it pulls the others down.

Still waiting on RX 480 order, I better get a revised card after waiting this long.
Quote from SimulatorRental.com :Thanks,
Is LFS itself reporting 60 FPS?

I think so, inactive window sometimes drops to 55 FPS, but most of the time it's 60.

Quote from SimulatorRental.com :When I test it I have to have them all open on screen while in cockpit view because it takes a few FPS more. I tile or spread them across monitor in order to see the proper FPS from LFS itself at the same time. All while none of the LFS windows have window focus, because the active instance, gets a big boost, while it pulls the others down.

I did try with cockpit view too, same result Wink

Quote from SimulatorRental.com :Still waiting on RX 480 order, I better get a revised card after waiting this long.

Yeah, I really recommend you to buy a card with a non-reference cooler, stock cooler is loud and GPU is hot. With a good cooling solution, overclocking should give you a nice performance boost!
Like which card? all the 480's seem the same to me... I would like to have a remote fan/water sink like the pro duo.
SAPPHIRE NITRO+ Radeon RX 480 looks like a stock RX 480 killer, with better cooling, overclocking, and way less noise.

Or if you want watercooling, you can also buy a reference card and then buy either a GPU watercooling kit, or do a full watercooling loop on the PC (CPU + GPU). Wink
Card looks decent, but not available yet.

FGED GREDG RDFGDR GSFDG