The online racing simulator
More or Faster (LFS performance review)
Helloe

There is always people saying how high or low FPS their systems gets and other people asking how much this or that small upgrade would make things better.
So I wanted to show you guys what to upgrade and what really makes difference in LFS.

Remember these numbers are just giving you image how much the differences are in real life and they can differ with different OS's and chipsets in different systems.

Hardware I used for these tests:
Mobo: Asus Maximus II Formula P45 | Asus Rampage Formula X48
CPU: Intel E2140, E6600, E8600, QX9650
Ram: Corsair Dominator PC2-8500
GPU: ATI Radeon 4870 512MB, Nvidia 9800GX2 1GB (in single card mode it becomes 8800GTS 512MB)
Cooling: Thermalright Ultra-120 TRUE Black with single 120mm 1200rpm fan
OS: 32bit Windows XP Professional
LFS version: 0.5Z

However I used P45 mobo and 4870 in every test except tests for Nvidia's cards as I'm running that system watercooled and can't really take it apart 100 times in one day
So GPU tests with nv8800/9800 cards also runs on X48 mobo. But as I didn't compare GPU's against each other, it makes no difference

For benchmarking I used old BL1 FBM replay with 'max' cfg from http://lfsbench.iron.eu.org/ which has replaced with SO4 replay while ago.
However SO4 results coming bit later once I get it done
If you want to see some exact MhZ result with it, post here.

I split these tests to different chapters in different posts so it's easier to people link these if needed.


Here's list:
1. RAM, more or faster?
2. CPU, dual or quad?
3. GPU, does it make difference?
4. SO4 replay's results

And then to numbers





1. RAM, more or faster?





For first, I compared different latencies.

CPU used here was [email protected].



As it shows, latency makes no difference here.


Next I tested different MhZ's. For this test I had to use [email protected] to get higher RAM speeds than what E2140 supports. Will that extra bandwidth in RAM give any extra FPS?



No. Only at high speeds like 1066MhZ, there was 1-3 FPS difference.



Then the everlasting question.. Would more of RAM to get you higher with FPS?

CPU used here was again [email protected].



And answer is still no. Whatever you do with your memory, it makes no difference in LFS. At least, yet. However, I don't recommend you to run anything under 2GB nowdays
2. CPU, dual or quad?




This is the part where all the magic starts to happen :wizard:

I didn't want to go in normal routine here and benchmark only the MhZ alone.. I also cover the usually ignored and forgotten topic, size of cache.

Which is actually the test where we start.
Each CPU ran at 2.4GhZ with 300x8 settings and RAM was at 800MhZ with 5-5-5-15 timings.



As can seen here, even the bigger cache can make major impact to FPS. More.. better

Next comes the question, which one is better for LFS, dual or quad?
As LFS isn't multithreaded yet, the conculsion already at this point is.. quad makes no difference in FPS. But I give you more reasons to get quad after these charts:

Dual cores in Intel's E8000-series:



And quad cores in Intel's Q9000-series



Now you probably notice already that the Q9300 is missing from the list :magnify:
Reason behind that is.. All the Q9xxx quads have 12MB L2 (cache) except Q9300, which has only 6MB.
So I could run Q9300 test, but it wouldn't show real speed as my QX9650 has 12MB L2.
If someone has Q9300 with 4870 512MB here, yell hard and you might be able to get your score up there

Then back to reasons why to get quad over dual. I have said for a year to people to take dual over quad if it comes just to gaming.
But it finally slowly starts to show that games starts to support quads more and more. GTA 4 for example.
Also if you do rendering or photoshop stuff, quad gives big gain in matter of speed. However as long as the software supports it.
But I'm not saying that dual cores are gone already either..
They run cooler, eats less wattage and if you happen to do overclocking.. you get higher clocks way easier than with quads

So the overall conclusion in this matter is even still.. it all depends what you do with your pc.
Actually.. the numbers tells the difference between dual and quad better than my words

This is what you can expect if you rendering with dual/quad:



Even higher clocked dual can't keep up with quad when software is multithreaded
3. GPU, does it make difference?



As we already saw big FPS differences in CPU chapter.. Will we see same in GPU chapter?

Answer is cold no :geezer: LFS won't care less whatever you try to tickle it with :smash3d:

For some reason LFS has always liked ATI's cards better and so happens here again also..

All tests done with [email protected]. But Nv's cards ran on X48 mobo.

4870 results:



As might see, resolution makes no difference at all.

However slightly less FPS with Nv's cards:

8800GTS aka single card mode in 9800GX2:



And 9800GX2 results:



So conculsion, does GPU make difference? It does little bit, but don't go buy uber high-end 600$ card if you want just more FPS in LFS
Only thing what 9800GX2 could do better than 8800GTS was just better performance with AA and AF turned on
Also notice 4870 giving better FPS than 8800/9800 combo CPU limiting both cards really bad.

So what is best for LFS then after all this.. I would start from ATI's GPU and more you can get MhZ outta your CPU, better.
However if LFS one day becomes multithreaded, CPU's MhZ won't matter that much
4. SO4 replay's results



Due the hardware problems with Athlon XP system, I couldn't run tests with Radeon X800XT nor GeForce 4 MX440. :thumbsdow

But I had go with laptop and Radeon's. Basically these results just shows what hardware combo runs what FPS as nothing new found here.

First I ran tests with laptop:

Fujitsu-Siemens Amilo Xi 2550-31P
CPU: Intel T9300
GPU: ATI Mobility Radeon HD 2700 256MB
RAM: 4GB of 667Mhz 5-5-5-15
OS: Windows Vista 32bit Home Premium

Here's results with different resolutions:





And next we go for GPU benchmark.

I only had 2 slower cards after problems with AGP system and were almost dropping this test out already, but decided to run it after all.
Results doesn't show much difference between cards except, X1950 Pro was only card to start showing slight slowing down when resolution was increased. :magnify:
CPU used in tests was [email protected] and ram was running at 1066Mhz.

Here's results:





Now after all tests are done, I hope this thread is informative for many people who are looking for information what to upgrade to get better FPS in LFS.
When we get major update to 3D engine, I will make another big test then. Hopefully I have much more hardware to get my hands on then to get wider results

Thanks to http://lfsbench.iron.eu.org for making benchmarking real with LFS
Nice.
Excellent and very thorough analysis of CPU performance and GPU, Maybe a few more additions with stats perhaps testing an Older card with the mix ? x800 or 1950pro or smth (As i love stats/benchmarking)

Great work dude
diggin this review
#8 - vari
Well done

Nice to know the facts
Welldone! Finally a proper performance review.
Sticky?
This is a very nice analysis of the effects that different upgrades have on fps in LFS. It should definitely be Stickied.
Excellent, very informative and interesting. Stickied it should be
#13 - FexM
Er, I seem to notice an inconsistency: In your cpu section, in the cache chart, the e8600 gets 113 average, and then in the next graph, e8600 gets 160 average. Whats up with that?
edit: didnt see ypu gave them equal ghz. sorry

Other than that, i'd wish you had compared the GPU's to lower end ones to see the difference. And maybe a single core proc?

p.s. good to know how much of an increase would be expected from an overclock. Might push me to overclock my amd 3700 since i got a san diego core.
Quote from MijnWraak :Er, I seem to notice an inconsistency: In your cpu section, in the cache chart, the e8600 gets 113 average, and then in the next graph, e8600 gets 160 average. Whats up with that?

Other than that, i'd wish you had compared the GPU's to lower end ones to see the difference. And maybe a single core proc?

In cache run, all CPU's were working only at 2.4GhZ while in CPU chart E8600 is working at 3.33GhZ (stock speed)

about GPU part, I will run SO4 tests with hardware listed here:

AMD Athlon XP 2800+ with ATI Radeon 800XT 256MB and Nvidia GeForce 4 MX 440
E8600 at stock gets benched with:
ATI Radeon X1950 Pro 256MB
ATI Radeon X1950 XT 256MB

And to see how slow GPU can be before it starts to affect into FPS, we underclock 4870 as low as it takes before FPS decrease.
Dunno how much that helps with anything, but it at least will show how much LFS really depends on GPU

Also testing laptop with 32bit Vista Home Premium with specs like:
Intel T9300 2.5GhZ 6MB L2
4096MB RAM
ATI Mobility Radeon HD 2700 512MB


I will run tests after this Xmas madness is over so stay tuned!
It just goes to show that unless you're planning on running the absolute cutting edge of games, top-end kit just isn't necessary.
Quote from (FIN)Eza :In cache run, all CPU's were working only at 2.4GhZ while in CPU chart E8600 is working at 3.33GhZ (stock speed)

about GPU part, I will run SO4 tests with hardware listed here:

AMD Athlon XP 2800+ with ATI Radeon 800XT 256MB and Nvidia GeForce 4 MX 440
E8600 at stock gets benched with:
ATI Radeon X1950 Pro 256MB
ATI Radeon X1950 XT 256MB

And to see how slow GPU can be before it starts to affect into FPS, we underclock 4870 as low as it takes before FPS decrease.
Dunno how much that helps with anything, but it at least will show how much LFS really depends on GPU

Also testing laptop with 32bit Vista Home Premium with specs like:
Intel T9300 2.5GhZ 6MB L2
4096MB RAM
ATI Mobility Radeon HD 2700 512MB


I will run tests after this Xmas madness is over so stay tuned!

Ah okay, thanks. S04 test computer is similar to mine, so that'll be helpful.
SO4 post updated
Unlucky with testing the AGP stuff Eza, Can always try again perhaps

I decided to run the latest LFS Benchmark if anything its an Indicator that Single threaded perfomance is the key with LFS on my phenom. I only ran with the affinity masked to core 0, and ran 1FPS higher.
You can see the results here in the maxlist
I may (probably will at some point) try again and use BIOS to "uncore" the CPU this way the OS only sees the amount selected in BIOS.
Is that LFS Bench site being maintained by someone? It's a bit outdated: no Core I7 and Phenom II. I'd put my entry, but it looks stupid to put Unknown as the model.
Quote from AutoPilot :Is that LFS Bench site being maintained by someone? It's a bit outdated: no Core I7 and Phenom II. I'd put my entry, but it looks stupid to put Unknown as the model.

a phenom II and i7 for LFS?... you have got to be kidding me.
It's not (just) for LFS, and it may be interesting to see how they compare to others.
Quote from AutoPilot :Is that LFS Bench site being maintained by someone? It's a bit outdated: no Core I7 and Phenom II. I'd put my entry, but it looks stupid to put Unknown as the model.

I let vari know on irc
I've run the benchmark. With mininum graphics, I almost have the same avg fps as Eza, but on max settings I get 100 fps avg compared to Eza's 146. I think I can only blame Vista64.
This makes me curious how GTA4 would run under XPSP3
Quote from RevengeR :
This makes me curious how GTA4 would run under XPSP3

[email protected] + oc'ed 9800GX2 gave average ~62fps in benchmark and when cpu was running 3.6ghz, fps jumps between 30-60 in game.
With E8600 minimum fps was around 20 and that's only difference comparing to quad. Just average fps was 5-10 lower.

I play with:
1920x1080
texture detail: medium
render quality: highest
view distance: 40
detail distance: 100
vehicle density: 100
shadow density: 16
1

FGED GREDG RDFGDR GSFDG