Thank you for the test.

0.6J is now available. Just the same as 0.6H really...

Dear Hosters,

If any of you would like to test, here is the 0.6H11 DCon.

https://www.lfs.net/file_lfs.php?name=LFS_S2_DCON_6H11.zip

I don't think there are any notable changes and only the exe has changed. The only purpose of this is to make sure nothing is broken, ready for the full release which we plan to do on Saturday.

Updated first post with H11.

The change logs from H to H6 have been combined.

This is intended as the final test. Thank you all for the testing.

https://www.lfs.net/forum/thread/87997

Test Patch 0.6H6 is now available, with various fixes and some improvements. Thank you for the testing and feedback.

https://www.lfs.net/forum/thread/87997

Changes from 0.6H5 to 0.6H6

Misc :

Reduced min / max values for "Sound lag" setting - default now 0.08
Small error report added to deb.log if failed to start DirectX 9Ex
More efficient car distance sorting system for sound and graphics
More updated translations. Thank you translators.

Oculus Rift :

Updated ORDIRECT.dll to use 0.6.0.1 runtime instead of 0.6.0.0

Fixes :

CPU usage increased if minimised from full screen mode with vsync
Missing sound from nearby cars that were off screen in multiplayer
Frame rate limit system could cause lags when used in Windows 10
Frame info display could be clicked even when mouse was hidden
Virtual dashboard disappeared if another nearby car was drawn

Quote from Amynue :

https://developer.oculus.com/blog/upcoming-oculus-pc-sdk-0-7-compatibility-changes/

Thanks, they sent an information email about this. Our current official version will no longer work with the new Oculus runtime when it is released on 20th August. It seems to me the best plan for LFS is to finish the few remaining bugs and make the current version official as soon as possible, e.g. next week. Our Rift support in the test patch is based on Oculus 0.6, so it will work with the new runtime. Then I'll get on with the track shaders that can help Eric with the initial S3 tracks.

Quote from valiugera :

@Scawen, I found a mistake on Fern Bay. The wall is in the air, but Z is 0. Can you add ground level check the for new objects or add option for -Z?

I had to make that decision, to disallow Z values below zero, in order to reach the highest altitudes at Westhill. The autocross object data is highly compressed, to allow them to be passed quickly to joining racers. There is only one byte in the object data for altitude, so it can't go below zero. There are very few places in LFS which are below zero so this is the best decision. Maybe in future the sutocross object size will be increased, allowing more flexibility.

Quote from DANIEL-CRO :

BTW Scawen, is there any way we could see if Direct3D9Ex is used? I remember my RAM usage greatly decreasing when going from 0.6H -> 0.6H2 on Windows 7. Now on Windows 10 for some reason I don't get any decrease.

You can see near the start of deb.log

Aug 05 21:23:33 started Direct3D 9Ex

Quote from Neilser :

Quote from DANIEL-CRO :

My idea is to store two car positions for each car, one that is updated every 0.01s and used for physics calculation and other one that is used for purposes of moving view/camera/car.

Embarrassingly, I thought that was what LFS already did...

The real way to have graphical frames independent of physical updates is to use one thread for physics and another for graphics. The physics thread must store snapshots of the entire game state (car positions, wheel rotations, suspension extension, tyre deflections, smoke particle positions, etc) and the graphics thread can then read these and render frames in even steps between the last two physical frames. Very interesting but a truly massive task and for various reasons already described, not a good idea to do it now.

Quote from Flame CZE :

I'm currently having an odd problem in H5. Whenever I go to the pits (Shift + P), there is a 1 to 3 second delay before it shows the garage GUI. It just shows the garage background during the delay.

I tested it with a clean LFS install and it occurs in H5 only, it's fine up to H4.

Anyone else with the same problem by any chance?

I didn't see that in a quick test. Can you give any more info, like if you were online or in a replay, with many cars there, etc?

I recommend Test Patch 0.6H5 - it has a significant performance improvement which is helpful at Westhill.

List of changes and download: https://www.lfs.net/forum/thread/87997

There is a good case for setting the frame rate to 100 Hz. This way, there is a read from the steering wheel for every physics calculation. Some people like higher frame rates for their own reasons. Anyway, it's interesting to see in the new moving bar chart which works well.

I had hoped to release test patch H5 this weekend but that issue with the transparent autocross objects not being visible until the optimise button is clicked has taken longer than expected. Rather than a quick fix I am working on a worthwhile improvement.

Quote from Neilser :

Holy crap. H4 is now installed. After a race a moment ago, I got this weird music in my ears for the first time ever!
Whatever you did Scawen, it made the music work on my PC and I don't ever remember hearing it before...
(I turned it off again mind you, cos it's, em, awful )

The entry screen music option is enabled when you first run this version because the audio options are stored in a more compact way and it doesn't read the old setting when it starts. Normally that music is switched off when you unlock.

Quote from nacim :

I'm back, with some benchmarks !

Sadly, I was expecting better results.

Most of the benefit of the H4 restructure is seen in stereoscopic 3D modes, where around 5% to 10% of CPU time is saved in the render, due to sharing calculations between the two eye views. I think the main benefit from H3 to H4 in non-stereo views was the bug fix - those few buildings that were drawn a slow way. That may help a little with the frame rate drop in the 3 places, but not much as it was only a few of the buildings.

But still, so much is visible at those locations that it does still slow down. I have one more thing to try. I think for the objects with a low triangle count, the CPU is still wasting too much time deciding if those objects should be drawn, and sending a request to the card for their specific vertices and triangles to be drawn. I think it is worth trying a test of just grouping these lower detail objects into a mesh that is drawn complete every frame. This would certainly help CPU usage. I will need to experiment with the number of vertices that is the threshold between these smaller objects, and the larger objects that it is worth not drawing when they are not visible.

Quote from nacim :

No, I used a single lap of 5 IAs with FBM @WE1 with replay camera on first IA.
They have done the same time, and I don't think replay is accurate for a benchmark because it doesn't compute physics.

Physics is computed in single player replays.

Test Patch 0.6H4 is now available : https://www.lfs.net/forum/thread/87997

Misc :

New Audio Option "Sound when window is inactive" (off / on)

Oculus Rift :

Reduced CPU / GPU usage by sharing scene preparation for both eyes

Fixes :

Some buildings at Westhill track were drawn using a slow method
LFS failed to resave custom textures found to be in wrong format
Using mouse wheel to change gear did not work properly at high fps

Thanks for that info. I just did a quick test adding that flag in the places where a DSBUFFERDESC is filled in. For some reason the music did keep playing but there was no car sound. But I think they use a similar system so I'm not sure why that is. I'll look into this.

Sorry that this has taken so long to fix. It must have been frustrating.

I have now looked into this and fixed it in my version. I hope to release another test patch this week with some more performance improvements.

Quote from cargame.nl :

Quote from Scawen :

In your case, is there any noticeable change from H to H2 / H2 to H3 or does it all seem much the same?

Wanted to do some screen recordings of the screen(s) here to show the differences but for some reason its not so easy reproducable at the moment... But, this machine has a bit more power. The streaming computer I cannot use anymore because of screen lag (whats the proper name for this?) .. That machine has an i7 720qm ... Its a bit weird to buy a different system just for LFS, then to see that single thread performance of much newer processors are also questionable ... Its also the wrong approach of "fixing" a problem.

I will do some random streaming to show what issues Im talking about. Will link later to video's.

I also wonder, it affects the screen, it's hard to drive.. But will it affect physics/steering input as well? I guess so. Will it affect the Rift experience? I think too, because that device is even more hardware hungry? ...

I'm confused by your text. Not really sure what you mean. It's not clear to me if you are saying there is a difference between H / H2 / H3 on your computer.

Anyway, when you say "screen lag" maybe you are talking about the effect normally called "input lag" which is when there is a noticeable delay between (e.g.) moving the controller or mouse and the on-screen steering wheel moving. This can happen when the CPU is not loaded but the GPU is drawing quite a lot. Then the GPU can get several frames behind the CPU (while trying to work through its list of things to draw and never catching up to real time) causing this strange effect. This effect can be quite extreme in some cases.

LFS has a prevention method for this (checks that the card is finished before sending more instructions to the card) so most people would never have seen that. It should be better in H3 than it was in previous versions, because it now uses an event query (DX9 function) instead of a lock (DX8 style). Though on most systems this shouldn't make a noticeable difference.

One test result I heard suggests that the input lag prevention mechanism may not have worked properly on SLI setups before H3 but should be OK now. Though LFS doesn't get any benefit from SLI anyway and may run faster with SLI disabled.

Last edited by Scawen, Fri 3 Jul 2015, 12:26.