Hi folks,


I'm having some performance issues with LFS on my pc.

It's an Athlon 3200+ with Asus A7N8X-E deluxe motherboard (nforce 2 chipset with soundstorm), 512mb pc3200 ram (timings 2.5-3-3-6) and a Radeon 9800 non-pro 128mb


I'm using the latest drivers, but for a long time I've having low performance, since minimum sleep was forced to 1ms in LFS (don't remember exactly witch patch forced that).

Even before, setting minimum sleep to 1ms instead 0ms, although solved some controller lag issues, dropped the performance significantly (although some users didn't have any performance drop doing this).


I run LFS on 1280x960 at 2x Anti Aliasing and 4x Anisotripic filtering, and driving alone on Blackwood (hotlapping) with all gfx options on max, I get a performance that is between 58 (lowest) and 83fps (highest). I remember that with minimum sleep to 0ms I could do a whole lap using v-sync and sticking to 85fps (I'm not using v-sync nor frame rate limiter right now).


I think that this performance is pretty low on this system, specially if you consider the conditions and the system that is running lfs. On on-line races, the performance is significantly lower than off line (of course), what is not good specially on starts with full grid.



Just to be sure, I formatted my pc installed the latest drivers (even updated the motherboard bios) and windows xp updates, and still got the same performance, and I already tested another driver version for the video card, without results.



Don't know if it's a bug with my setup, but I think that I had to get a better performance on my pc.

I did some benckmarking using Sisoftware Sandra and 3dMark2005, and the numbers that I got were normal for my system, so I believe that it's something only related to LFS.


Anyway, Merry Christmas for all !

I suppose that the framerate that you posted is when you are alone. I got similar problem caused not by min sleep but EAX HW "acceleration". When I've installed non-EAX driver for my soundcard, I can do 102 fps when I am alone on BW GP(P4 3.2E, 512MB DDR2 3-3-3-10, GF6600GT)

The min sleep might somehgow affect soundcard driver and do this mess.... disable sound by SHIFT+N and check what MINIMAL fps you get..

Merry Xmas to everyone...

Quote from Mogar :

It's an Athlon 3200+ with Asus A7N8X-E deluxe motherboard (nforce 2 chipset with soundstorm), 512mb pc3200 ram (timings 2.5-3-3-6) and a Radeon 9800 non-pro 128mb

Hi Mogar Its just occured to me that with a bios update your memory may be default set to asynchronous ,your XP3200 barton cpu could be a 333Mhz or 400Mhz version and if its the 333Mhz (FSB166 in bios) then you will get better performance if your memory runs at the same (100%) speed the SPD will want to run the memory at 400Mhz (200) so this may explain some drop in performance , I think you may have already checked this but you never know . ATI Catylist versions 3.11 appear to work best for Radeon cards and LFS however I have similar setup to you and mine seems fine with the 3.12 ver however I do not use the CCC I have not yet been able to get this working well enough to make me want to use it .

Good luck ..and Merry Christmas

SD.

Quote from SparkyDave :

ATI Catylist versions 3.11 appear to work best for Radeon cards and LFS however I have similar setup to you and mine seems fine with the 3.12 ver however I do not use the CCC I have not yet been able to get this working well enough to make me want to use it .

Good luck ..and Merry Christmas

SD.

I think you mean this issue with the 5.12/5.11 drivers, not the 3.xx

Well spotted

SD.

well, timings and clocks for memory were checked, and the benchmark results that I got were pretty consistent with the reference numbers, so it doesn't look like I have a performance problem on my pc.


about eax drivers, that I simply don't know because I use the on-board sound (my motherboard has soundstorm), and I'm not familiar with tweaks for it. Anyway, I've double checked and I have no dsp effects being applied, and it's set as 2 channels only (I use headphones). However I will check if without sound LFS runs faster.


about video drivers, I'm using catalyst 5.13 right now, and since ATI released those drivers with control center, I've been using it. I will do some testing, but I remember that my pc has always been affected with minimum sleep settings in all versions of S2 (except for patch P and Q that I didn't test anything higher than 2ms).


And this performance that I said that I'm having is on hotlapping, so I'm alone on the track and not connected with LFSworld.

tried disabling sound, but without results.... and disabling all filtering (AA and AF) resulted on a very small gain in performance.. so it doesn't look like a video card problem.

there is a own thread about the poor performance of the 5.12/5.13 catalyst drivers, you should downgrade them to the 5.11 or (the better way) the omega drivers

thread: http://www.lfsforum.net/showthread.php?t=3639

Quote from herki :

there is a own thread about the poor performance of the 5.12/5.13 catalyst drivers, you should downgrade them to the 5.11 or (the better way) the omega drivers

thread: http://www.lfsforum.net/showthread.php?t=3639

If you have read my original post, you will see that I was using the 5.11 drivers before I decided to format my hd.


Anyway, I'll give a try, but as far as I know, those issues with catalyst 5.12 and 5.13 are far different from mine. Ppl having issue with latest drivers don't use the same video card than me (this make a difference although they are on a single driver pack), and they are having performance like 15-20 fps, and not like me.

So, my issue is [r]NOT[/r] solved. I will try to switch drivers to 5.11, but since it WASN'T working before, i doubt that it will work this time.


Please be careful before editing the thread's title.

Installed the 5.11 drivers, and got the same performance, so again, it WASN'T a issue with 5.12 and 5.13 drivers.


I will PM a moderator to change the title of the topic



Just tried a lap on patch L, and the performance was 78fps minimum and 113fps maximum (versus 58fps min and 83 max that I get on patch P) on same conditions (same settings inside lfs, same settings for video card, a lap on BL with XRT on hotlap mode). But I did a test with minimum sleep (on patch L, testing 0ms and 1ms), and while it affected the performance, the performance drop was very small (like 3-5 fps). So it looks like it's not a issue with minimum sleep, but a performance drop on patch P (and Q also, since I didn't noted a performance difference between them).

I've removed the "solved" from the title. I can't see where you said in the original post that you had been using the 5.11 drivers - i thought when you said that you were using the latest drivers, that must have been the problem. We've got an ongoing problem with ATI at the moment, people are reporting that ATI are simply denying that any game can run slower with the new drivers, but not actually testing LFS to prove their point.

I often change the thread titles, when the solution appears to be found, to save other people the trouble of reading about problems that are already solved. But you are welcome to mail us of course if somehow we have read it wrong.

By the way, i can't think of any reason why P and Q would be slower than L. As far as i know, more things are just done in more efficient ways and all known bugs were fixed, so really, P should be a bit faster and Q a bit faster still, given the same track and car combination. Sorry i can't help more at the moment.

Now I think I've solved


I noted that on patch L I was using the "clr+sky" screen clear method (don't know if it was the default method on that patch) and on patch P and Q I was using "ellipsoid". When I changed on patch Q it for clr+sky, I got a very nice performance boost (more than 10 fps instantly, and on some sections of the track, got more than 25fps more).


I tried to look if there are image quality differences between those two methods, but I didn't noticed immediately. I can't tell if there is a difference at all, but if there is a difference it's minimal.


So, for now, I think that I solved my issue changing the screen clear method. I will some more testing just to be sure that it was it.



[editing before posting]

Well, just did a test on KY1 with FO8, watching a replay from my pb hotlap. With ellipsoid method, I got 70-78 fps, with clr+sky method, got 92-116 fps (what is a huge difference). So I can consider my issue solved.


BTW, what are the differences between those two methods that could lead to a such performance drop without an apparent better image quality ?


And thanks LFS devs for the lightning fast answer for my private message

Quote from Mogar :

about video drivers, I'm using catalyst 5.13 right now, and since ATI released those drivers with control center.

just re reading this can I ask do you mean CCC Catalyst Control Center? or CP Control Panel ? as it has been commented on in the ATI forums that the CCC can be a real resources hogger .

as to the ongoing ATI drivers thing they are aware of the problem with LFS even if they deny it so we can hope for somthing in a future set of drivers that restores the performance .

SD.

EDIT oops posted while I was writing

Ok, i've added "solved" to the title again.

"Clear + sky" clears the entire screen to a background colour, then draws the sky dome.

"Ellipsoid" does not do a clear, but simply draws a sky dome and a grey underworld dome (so it's a complete ellipsoid).

"Clear + sky" is probably faster on most modern cards, even though every pixel in the sky area is drawn over once with the background colour then again with the actual sky texture. I don't know why that's quicker but i guess it's to do with how the card's hardware works. On older cards, the ellipsoid option can be quicker because it doesn't waste its time drawing all those pixels twice.

Well, I won't switch back to 5.13 right now since 5.11 is working just fine, but anyway, people that had performance issues are using a newer generation video card (I believe that although on the same package, each card family has your own driver).

But anyway, switching back to 5.11 didn't solved my issue immediately, the real change was changing the LFS settings.


And I'm using the CCC version of all drivers since it was released, but again, this wasn't the issue, since I've been using them for a long time, even before I noted the performance drop.

Quote from Scawen :

Ok, i've added "solved" to the title again.

"Clear + sky" clears the entire screen to a background colour, then draws the sky dome.

"Ellipsoid" does not do a clear, but simply draws a sky dome and a grey underworld dome (so it's a complete ellipsoid).

"Clear + sky" is probably faster on most modern cards, even though every pixel in the sky area is drawn over once with the background colour then again with the actual sky texture. I don't know why that's quicker but i guess it's to do with how the card's hardware works. On older cards, the ellipsoid option can be quicker because it doesn't waste its time drawing all those pixels twice.

Humm.... Interesting.....

Anyway, I think that this information could be useful on a troubleshooting guide, or on a recommendation list for better performance (since I believe that there is a lot of users unaware about how it can change the LFS performance, in my case a huge difference)


And I hope that this "investigation" can be useful for other LFS players and to LFS progress

VGA: Asus Radeon 9600XT
Drivers: ATI original 5.11 + Asus Enhanced Driver 1.22 + Asus SmartDoctor 4.80 + ATI Tray Tools v1.0.4.780
Replay: LFS Benchmark
LFS Settings: Custom, most of them max

I've heard that few people use ATI Tray Tools instead of CCC, and now I use the ATT too. 1MB download, quite easy to configure.
Going from 5.6 to 5.11 I got extra 2 to 5 FPS.

I wrote ATI original 5.11 because, I have previously used ATI drivers from ASUS homepage, but the latest drivers available from ASUS, are 5.10, I think. I have no info about what does ASUS do to the drivers, maybe only the install backrounds are changed.

I have no reason to change to a newer version, maybe if 5.14 come out with an considerable FPS increase in LFS, I'll probably try them.

Edit: It was on clr + sky setting, if patch Q didn't change it, I've always had clr + sky in S2. Just checked: in S1 0.3H I have clr + sky too.

well, I will post my results with the lfs benchmark

CPU: Athlon XP mobile 2500+ @ 2.2ghz (aka 3200+ ), 512mb ddr 400mhz latencies 2.5-3-3-6, Asus A7N8X-E dlx motherboard, Nforce drivers 5.10

Video card: ATI Radeon 9800 non-pro 128mb (standard clock) at 1280x960@85Hz , 2x AA, 4x AF

Custom lfs settings (everything in max)


Frames: 4787 - Time: 103406ms - Avg: 46.293 - Min: 26 - Max: 73 (clr+sky)

Frames: 4801 - Time: 103438ms - Avg: 46.414 - Min: 29 - Max: 73 (plain sky)

Frames: 4752 - Time: 103516ms - Avg: 45.905 - Min: 29 - Max: 68 (ellipsoid)

Well, that replay from benchmark is cpu limited on my system, so there is no significant difference between those 3 modes. And did a test switching the resolution to 800x600, and got a very small performance increase, but on hotlapping (and maybe on-line races) it does make a difference.

Hey, that one did make quite a difference

Athlon64 3200+, 1GB Ram, 6600GT (PCIe)
Benchmark run @ max.cfg, 1600x1200, 8xAA, 16xAF

Ellipsoid
2006-01-05 21:51:21 - LFS
Frames: 3798 - Time: 103922ms - Avg: 36.546 - Min: 29 - Max: 48

Clr + Sky
2006-01-05 21:54:17 - LFS
Frames: 4599 - Time: 103500ms - Avg: 44.434 - Min: 29 - Max: 59

None
2006-01-05 21:56:27 - LFS
Frames: 4423 - Time: 103390ms - Avg: 42.779 - Min: 30 - Max: 59

PS: I let the benchmark run once before testing, so there are definitely no texture-load lags on the first run.

The max bechmark defaults to ellipsoid and maybe that isn't such a good idea because it doesnt affect the quality really. I might change that the next time the package gets updated because of a physics change in lfs.