Cracked the GPU utilisation 'problem'

[TheFlier]

OK - nothing to do with my RAM or storage, or GPU type. Simply increasing the AA from MSAA x4 to SSAA 2x2 sent my GPU utilisation from 45% to 98% :)

SSAA 2x2 also hurts my framerates quite a lot, so I notched it down to SSAA 1x2, now FPS are similar to what they were with MSAA x4, but my GPU usage is 70-80%

Appologies to RSC for thinking this problem was a coding issue.

Ude this APP (http://event.msi.com/vga/afterburner/download.htm) to display on screen your FPS and GPU usage in-game, then play with your settings to get a good balance of FPS and GPU utilitsation.

Good luck :)

[NDORFN]

What benefit does increasing GPU utilisation have if it doesn't improve framerates?

[TheFlier]

Improve quality without hurtng framerates too much?

But I must admit, improving framerates would be nice too :) maybe a future patch to increase CPU utilisation woud help?

LOL

here we go again....

[OldProf]

OK - nothing to do with my RAM or storage, or GPU type. Simply increasing the AA from MSAA x4 to SSAA 2x2 sent my GPU utilisation from 45% to 98% :)

SSAA 2x2 also hurts my framerates quite a lot, so I notched it down to SSAA 1x2, now FPS are similar to what they were with MSAA x4, but my GPU usage is 70-80%

Appologies to RSC for thinking this problem was a coding issue.

Ude this APP (http://event.msi.com/vga/afterburner/download.htm) to display on screen your FPS and GPU usage in-game, then play with your settings to get a good balance of FPS and GPU utilitsation.

Good luck :)

Okay, for the benefit of the ignoranti out here, please explain what all those abbreviations mean and how one would go about adjusting them!

[Rich_S]

Okay, for the benefit of the ignoranti out here, please explain what all those abbreviations mean and how one would go about adjusting them!

Not to be confused with the Illuminati Sorry I couldn't resist, just watched Angles and Demons.

I think I can help with some of the abbreviations:

AA - Anti-Aliasing

SSAA - Super-Sampling Anti-Aliasing. Samples are taken at several instances inside the pixel (not just at the center as default) and an average color value is calculated. This is achieved by rendering the image at a much higher resolution than the one being displayed, then downsampling (shrinking) it to the desired size, using the extra pixels for calculation. The result is smoother transitions from one line of pixels to another along the edges of objects

MSAA - Multisample anti-aliasing. Basically the same as Super Sampling but instead of using a square to grab coordinates within the pixel, MSAA uses only three points in a triangle.

GPU- Graphics Processing Unit. It's the CPU on your graphics card.

These values can be adjusted on the Train Simulator 2012 main window. They are found on the Settings tab, using the Anti-Aliasing slider. You first have to select the Custom radio button, then as you move the slider from none to Maximum you'll see a brief description below the slider of each option. Another option to consider while adjusting your AA settings are the Texture Filtering settings, which are found just above the AA slider.

Regards,
Rich S.

[TheFlier]

Okay, for the benefit of the ignoranti out here, please explain what all those abbreviations mean and how one would go about adjusting them!

Oops! sorry :)

- On the RW3 front-ent screen (you wont see this if you're bypassing it like you said in another thread) - along the top is a list of Tabs.
- Click on the SETTINGS tab.
- Half way down you'll see a slider labelled ANTI ALIASING. The further to the right you drag the slider the fewer jaggies you'll see on the screen when playing. (Jaggies are like little step-ladder effects along the edges of objects. Anti-aliasing smoothes them out to make them look crisper and clearer).
- But the further to the right you drag the slider, the more stress you put on your graphics card which CAN reduce frame-rates. The tool I mentioned above shows how stressed your graphic-card is (100% being VERY stressed), and also show the FPS you're getting. These are the two figures you need to balance. Try to get fairly high GPU usage (around 80% is good), while keeping FPS fairly high. I can live quite happily with 18fps, but its all down to taste I guess :)

Somebody else could probably explain it better than me, but this is my take on it :)

If your not sure you want to play about with trying different settings for a few hours, then I wouldn't try any of this.
Hope it helps

[PapaXpress]

I just ran into this framerate thing...

Anyway, I would like to direct you to this thread:
viewtopic.php?p=47015#p47015

I just did some benchmarks and was trying to understand the results. Snowstorm does a good job pointing me in the correct direction on why I saw a improvement in game play (framerates) but not an improvement in FPS.

The whole thread is actually a pretty good read, and if it stays on topic I think it may become extremely useful to help someone discover if they need to upgrade their video card, CPU, or both.

[TheFlier]

I just ran into this framerate thing...

Anyway, I would like to direct you to this thread:
viewtopic.php?p=47015#p47015

I just did some benchmarks and was trying to understand the results. Snowstorm does a good job pointing me in the correct direction on why I saw a improvement in game play (framerates) but not an improvement in FPS.

The whole thread is actually a pretty good read, and if it stays on topic I think it may become extremely useful to help someone discover if they need to upgrade their video card, CPU, or both.

Thanks for that :)

My system specs:
Mobo - GA MA770-UD3 (v1.1 FK)
RAM - 3.3gb DDR2 @ 800 (corsair 4x cm2x1024-6400C4)
CPU - AM3 athlon II x3 (450)
GFX - radeon HD4870 (512mb) CCC 11.9 driver
2 x 500gb Hdds
WIn XP sp3

Here's a few screenies

[Griphos]

Super sampling renders your screen in double height or width or both, and then scales that down to your resolution, so it's rendering twice as many pixels in one or both directions in order to increase the quality of the AA. It works, but as you see, it is also not a terribly efficient way to do things. That's why it's using your GPU more. It's not achieving a better "balance." It's just doing more work. That work comes at a cost in performance. Not so much if you only use SSAA 1x2 or 2x1 which doubles the resolution in one direction only. There's no advantage other than getting somewhat better AA by means of a not terribly good approach to AA.

The recent update seems to have accomplished some better "balancing" of CPU and GPU itself, at least on some routes, although I have no idea how. I can monitor that usage, but not see how it is being accomplished, of course, since nothing I can use to monitor breaks things down by task. It is clear that not all routes are created equal. Inevitable, I guess, given the model of contracting out development and publishing for third-party developers.

But, as to the topic, I wouldn't use SSAA just to achieve better "balance." Use it if you can't live with the AA you get from Multisampling. I prefer Multisampling. Better balance of performance and appearance on my system.

[PapaXpress]

I am trying to study this whole thing, because I just purchased a 560Ti 448 and I am trying not to return it. I came across an article at Tom's. Here is a link to the last page which I think has the most meaningful information (though the benchmarks are good to look through).
http://www.tomshardware.com/reviews/ant ... 65-19.html

This is the bit that really hit me:

Supersampling Anti-Aliasing

The highest-quality AA mode, supersampling essentially renders the frame at a higher resolution and downsamples the result. This causes a performance hit so large that Nvidia straight-up removed this mode from its GeForce drivers some time ago.
...

Does this mean what I think it does? If I set SSAA in RW my NVidia card is really giving me something else? And if so, what is it giving me? MSAA?

[TheFlier]

yes - but before I was only getting 50% gpu usage with AA at MSAA x4, which to me seemed like a waste of potential card.
By increasing the AA to SSAA 2x1 I get more use of my card (80%) for better quality images, with only a very very small hit on FPS (about 1 overall), but if I use 2x2 (instead of 2x1) gpu usage hits 100% and my FPS drop by about half.

I'm pleased anyway :)

[PapaXpress]

OK, so I poked around some more, and I found that Tom's was a bit off. Nvidia does not offer SSAA from their control panel, but third party tools, and games (if offered) can turn it on. I am also reading that SSAA is a bit of hit and miss for most people. Some like it for when then walk around, but others prefer MSAA+8 and never look back. There is also a consensus that SSAA can render a bit blurry, but that is probably game specific as I have not noticed it (yet) in RW.

[TheFlier]

If I use MSAA x8 my fPs drop 25% lower than SSAA 2x1, but my GPU usage drops too to about 50%. Now what does that say? That my card doesn't do MSAA very well? (who knows) Whatever - The quality results between MSAA x8 and SSAA 2x1 dont look much different so I'll stick with SSAA for now.

Which kinda brings me back to what my obsession with GPU usage was all about in the first place.

WHY WAS BOTH MY CPU AND GPU NOT BEING UTILISED MORE THAN 50%

OK - now I'm getting 80% GPU usage, but still my CPU never gets more than 50%. If nothing else, I'm now sure that it's not my GPU that's the bottleneck. And it isn't RAM as ppl with >8gb are having the same issue. So too are ppl with >1gb ram on their gfx cards. So what the hell is it about this game that's holding-up the CPU?

[PapaXpress]

First from what I am reading I think ATI will handle SSAA better than Nvida.

As to why the CPU is getting hammered there are a couple reasons. First it has to swap out scenery at an amazing pace. Then there is the argument that PhysX is forced on the CPU instead of the GPU (for Nvidia cards).

If I get time tonight I will try a few things.

1) Get that benchmark with No AA for Snowstorm
2) Get a bench for MSAA with CPU utilization
3) Get a bench for SSAA with CPU utilization

If you can think of something else I can try please through it out there.

[TheFlier]

First from what I am reading I think ATI will handle SSAA better than Nvida.

As to why the CPU is getting hammered there are a couple reasons. First it has to swap out scenery at an amazing pace. Then there is the argument that PhysX is forced on the CPU instead of the GPU (for Nvidia cards).

If I get time tonight I will try a few things.

1) Get that benchmark with No AA for Snowstorm
2) Get a bench for MSAA with CPU utilization
3) Get a bench for SSAA with CPU utilization

If you can think of something else I can try please through it out there.

Well, It's not that the CPu is getting hammered, the problem is it ISN'T getting hammered!
In a game like this either the CPU or the GPU (preferably BOTH) should be running at or close to 100%. If the GPU isn't at 100% it because its waiting for something to do, and only the CPU can give it tasks. If the CPU isn't 100% then it because its waiting on something. If the cpu isn't Isn't at 100% but the GPU is 100%, then the CPU is waiting on the GPU. In the case of RW3 it is NOT waiting on the GPU.

I did a disk-drive test earlier and saw minimal disk access during a scenario, except for the occassional burst as it loads in a Tile. So I don't think the CPU is waiting on the disk either.

PPl with 1600mhz RAM also have GPU and CPU running at less than peak (I've seen one with GPU=70% and CPU=30%). So it's NOT memory either.

If you are able Papa, I'd love to see a WAIT-STATE report for the CPU, to see exactly what it's waiting on.

Best of luck :)