Atomic Systems

Thanks for the continued input here guys, really appreciate you keeping me in check and making sure I'm looking at all the angles here - often the worst thing is always having that little niggle at the back of your mind "Did I forget to do/check/install something?"

I suspect that you might be right about the stuttering thing being ram. I've been inadvertently plunged back to 4GB while I get some faulty stuff replaced - no matter, the game is a damned sight more playable now than it was yesterday lol. However, to answer your question, there wasn't much in the way of HDD activity but on checking the performance data, there are horrendous memory spikes at roughly the same times as the stuttering. I can live with that, heck, we endured this for quite some time with previous versions of the sim didn't we?

Thanks for the suggestion of using Fraps though, I'll get that installed and see what results it gives.

Martin, sadly I would put the old cards back in but I loaned them to a friend (probably a silly thing to do in hindsight - although, in my defence, I never thought for a second I'd actually have any problems with such superior hardware... lesson learned!).

As for Support not knowing... well I contacted them but they failed to grasp the concept that I'd just replaced old components with new ones and that performance was worse with such newer components than it was with old. They opted to repeatedly throw the same stupid standard email at me and then finally ended up blaming whatever they thought wasn't current tech. Despite sending a bunch of very polite questions back for them to respond to, merely as a means to help me understand why it is they arrived at the conclusion they did, I've not heard a thing from them since Monday... so, either they're too busy to respond right now or I guess I'm being ignored... either way, I'll give them some time just in case it is that they're just too busy right now.

Kr, Steve

just my 2 cent answer for the €64,000 question:

Instead of this endless ongoing guessing game between people lacking the crucial insight in the actually running code. What might actually help a lot was, if RSC were to publish settings and performance for a set of reference systems.

Make sure BIOS settings of hardware is optimale. Make sure the CPU and RAM clocking match the hardware and stay clear of the fancy stuff for now. Maybe safe default setting, with the odd couple of explicit setting you know suits yiour hardware.

Forget TSX for now and don't use leftmost position of the AA-slider.
Experiment adding anisotropic later; after finding optimal ingame setting.

Ingame settings:
begin with Graphic Detail at Master Detail Level = Customer and all settings = medium, except:
Shader Quality = High Detail
Shadow Quality = Off
Leave all check boxes unticked.

trainsmit

Thanks for your input here Trainsmit. Much of what you mentioned to check on - particularly the BIOS and the clock speeds - were checked prior to installing everything from scratch. I think your mindset was probably the same as mine here, best not rule out those things just in case. Still, I think it's great that you mentioned them as I had completely forgotten about it - so anyone following this thread, something else to try.

All the clock speeds again were at factory state when I first ran the sim - I rarely mess around with these as previously mentioned, I only did to test the theory that my CPU was the limiting factor - it wasn't. However, I haven't tried legacy mode (non-tsx) as yet - with my hardware should that really be necessary and do you get the same set of quality options in legacy mode as you do in tsx?

One thing I had tried was turning the shadow quality off entirely and the shader quality down to the lowest setting - I always leave the post processing effects off anyways (I'm really not keen on the effect) - all this had no apparent effect to performance. Mind, having said that... I haven't tried those with the clean install as yet.

Kr, Steve

Hello again:

Continuing with this thread, I post my test here to ilustrate what I said:

Configuration for both PCs:
AA 4xMSAA
Anisotropic x16 (NVidia Driver)
In Game: All to Full except Bloom and Depth of Field.
Route: Hamburg-Bremen

Main PC:
Core 2 Quad Q 9300 2.5 GHz
nVidia 9600GT 1GB
4Gb RAM
1920x1080

GPU Usage: 100%
GPU Memory usage: 100%
CPU Usage: 60-70% four cores.
RailworksProc process RAM Usage: 3.0 Gb
Fps Average: 18-20 Fps. (On detailed cities 14 Fps)

Laptop:
Intel Core i7 2.20 GHz
Radeon 6770M 2Gb
8 GB RAM
1366x720

GPU Usage: 80%
GPU Memory usage: No data
CPU Usage: 50% only 4 cores (the 4 physical cores).
RailworksProc process Usage: 3.3 Gb
Fps Average: 25-30 Fps.

As you see, only 5-10 fps more with a Pc with more RAM and CPU.

Regards.

Hi

I appreciate the amount of info' this thread contains which gives options as to what to try to get acceptable
performance in running RW3, a couple of which have been pertinent to myself.

At the moment all seems OK but I am still increasing the game settings gradually to identify the point at which it all
falls over on my specific PC. I am using both WCML & Woodhead routes as the base because if these run OK
everything else seems to fly.

Thank you for your info' that it is possible to create an independently operating offline installation, as after this last
update fiasco, my PC even after being upgraded & tweaked is probably near the limit of running RW well.

If further RW updates include machine hungry mods, I have at least an older usable version.
I am at present scanning forums to check on members experience in the best way to create this alternative.

Hi sdark2,

sorry to hear about your trouble. You mention that you have the nVidia graphics settings at their default. This is also, if I have read the various source correctly, the "official" advice from RSC.
My experience is a little different, however, as I get consistently better results with non-default nVidia specific settings.

Just for testing if you are up to it, try setting the "maximum # of pre-rendered frames" = 0 in the nVidia settings. And then max all the RW3 graphic settings except shadow detail. What do you get then? What is the GPU load then? Frame rate?

MP

Hi MP, thanks for the tip - I'll give that a try.

Since I last posted on here, I've been testing a variety of different games and, up to now, I've had no problems with any in terms of graphics performance. I have had a little problem with Mass Effect 3 in that it seems to stutter and freeze occasionally (quite similar to railworks when it loads in the tiles), but generally the frame rate is a flat 60fps with all settings maxed. Certainly was interesting to play some games such as DCS A-10C Warthog and Need for Speed Shift 2 at 5760x1080 resolution, with all settings maxed out, and still get reasonable (30+fps) frame rates.

Also ran a few benchmarks via 3D Mark 11 and Vantage (on Extreme settings) - the graphic quality is absolutely stunning! Frame rates are quite high considering the intensive nature of the tests but has highlighted that my cpu really isn't up to the job of handling the throughput requirements at maximum load from the hefty gpus I've got. If railworks delivered the same degree of stunning graphic quality that I'm seeing on 3D Mark 11 with the same frame rate, we'd all think we'd died and gone to heaven!

Kr, Steve

OK tried the pre-rendered frames tweak - gpu and cpu loads are the same. Frame rate is unaffected.

Kr, Steve

Alright. I have a single card setup with an oc'ed GTX570 and a quad core CPU @2.8Ghz, and setting pre-rendered frames to zero changes the processing scheme so that the GPU does substantially more work - with the moderate CPU load more or less unaffected. With much better frame rates as a result. Maybe it is SLI and DX9, being single threaded, that doesn't play well together with an application that is not specifically (RW3 is not, I believe) designed to utilize SLI.

MP

Hi All, been doing some more testing today. Thought I'd prepare a graph to show just how supposedly "taxed" my system is.

The graph shown here (apologies for the lack of clarity, flickr seems to downscale everything I upload) is recorded with nVidia's System Monitoring Tool (bizarrely, it doesn't record framerate). Scenario is Chasing Yellows on Oxford to Paddington route. Frame rate never climbs above 14fps and is mostly planted at around 9fps. Game settings are > TSX Enabled, AA Off, Texture Filtering on Aniso x8, All in game settings on lowest. The points where the gpu load drops to zero, is where the game freezes. Disk activity is < 20% throughout.


Click the image for a larger view

Does my system appear to be pushed to the extreme here? Does the CPU appear to be overloaded and responsible for the frankly dire performance?

Kr, Steve

EDIT: For those interested, just calculated the averages for the entire 15min test. CPU = 37%, RAM = 77%, GPU = 21%

sdark2 wrote:Does the CPU appear to be overloaded and responsible for the frankly dire performance?

Looking at the graph, the CPU seems to hover around just below 20% a lot of the time, which is about what you would expect if it is CPU limited and most of the processing is on one core (you have a 6 core machine I think, and 100%/6 is about 17%). So I would guess yes, you are CPU limited.

This doesn't explain why things worked better before you replaced the graphics card though (which if I recall was the original issue?).

Thanks for posting nigeltouatievans. For reference, the average individual core loads are as follows:

Core1 = 18%, Core2 = 49%, Core3 = 27%, Core4 = 53%, Core5 = 2%, Core6 = 5%

At no point during the entirety of the test, with the exception of loading the application at the start, does any of the cores exceed 62% - the raw data can be provided for analysis should anyone wish it.

Kr, Steve

sdark2 wrote:Thanks for posting nigeltouatievans. For reference, the average individual core loads are as follows:

Core1 = 18%, Core2 = 49%, Core3 = 27%, Core4 = 53%, Core5 = 2%, Core6 = 5%

At no point during the entirety of the test, with the exception of loading the application at the start, does any of the cores exceed 62% - the raw data can be provided for analysis should anyone wish it.

Kr, Steve

I'm pretty sure Windows balances the load between the cores to some extent, so a single threaded program using 100% CPU will show up as less than 100% on several cores - at any one time the core usage should add up to 100% (in actual fact each core works at 100% for a very short time, but this is below the resolution of the monitoring program).

EDIT: Railworks is not single threaded, so you would expect more than 100% of one core to be used most of the time. The point though is that you can be CPU limited without any particular core showing 100% as long as the total is at least 100% of one core.

I'm not sure I understand what you mean here... could you elaborate please?

Kr, Steve

sdark2 wrote:I'm not sure I understand what you mean here... could you elaborate please?

Kr, Steve

I may be wrong (others are welcome to correct me), but as I understand it the monitoring program will take samples which give average loads over say a second. The operating system (Windows) may move a single thread from running on one core to another very quickly, say it runs on core 1 for .3s, core 2 for .3s and core 3 for .4s. Then you see (in the average over 1s, or whatever the minimum resolution of the monitoring program is) usage of 30%, 30%, 40% for cores 1, 2, and 3, even though the process is CPU limited and single threaded.

The timings are probably completely wrong (in particular I imaging Windows switches processes around much more often than that), but hopefully this illustrates the general idea.