To sli or not to sli that is the question

[lardyladd]

Ive noticed one or two people use sli (in their sigs) I would be grateful if someone could tell me if sli works with TS2013 and if so is it of benefit. I already own a GTX 670 and as the price of these cards is dropping I am mulling over buying another for sli thanks for your replys

[JustRight]

Hi mate,

I use a GTX690 which is a dual GPU card. I run TS2013 in SLI mode no problem. You do need to create a profile for it though and specify "Force alternate-frame rendering 2". I also run with Vertical Sync OFF.
This setup allows me to run TS2013 at 2560x1600 on my 30" Dell with TS2013 graphics set at almost maximum, except for DOF and Shadows set to 'High'. Anti-Aliasing is set to "FXAA + 2x2 SSAA" and Anisotropic Filter set to 4x.
TS2013 runs with very good frame rates except in the most complex locations which can slow it down a bit, but still very playable despite that. Both GPU's get used equally so SLI is definitely in effect.

Hope that helps.

[tads1970]

Benchmark results in SLI,

http://www.youtube.com/watch?v=8hcxvocJef4

Daz

[deniswick]

I was running two HD6870 cards in Crossfire mode (AMD's equivalent of SLI). I found that frame rates in TS2013 (as measured by Dxtory2.0) were pretty well double with Crossfire enabled compared to running with just one card.

[RichyV]

Unless or until your CPU is able to pass enough data to your single GPU to saturate it, adding another GPU will not help in TS.

The graphics engine it uses is still, very much, CPU-limited.

Above all else, I would certainly keep my money until the new TS2014 is released to see if there is any significant difference that changes this.

HTH.

[lardyladd]

Good advise Thank you........

[styckx]

SLI will give you lower frame rates. Disable one and run solely on one GPU and you will notice a big difference.

Sent from my Nexus 4

[styckx]

Here are some "numbers"

1920x1080 - Highest settings for all / 2X2 SSAA / Proc Flora on / Bloom & Dof off...

Single GPU (fullsize pic): http://i.imgur.com/9jTFr4s.jpg



Dual GPU - AFR2 (fullsize pic) : http://i.imgur.com/mkuoY2G.jpg

[RichyV]

This can be very driver/GPU/setup specific.

But your point is still a very valid one - there are no guarantees with SLI and "Game X, Y or Z" showing better performance...

[deniswick]

RichyV must be correct when he said "This can be very driver/GPU/setup specific."

I did several tests with my HD6870's when I first got them. Crossfire enabled did give lower frame rates at that time but, after I contacted AMD support, a later version of the drivers changed that. More recently I've done some more testing. I use the very beginning of the "Autumn Leaves" scenario for my "benchmark". With one card installed I was getting a steady 26 fps. With two cards and Crossfire enabled I was getting 50 fps.

[ttjph]

Unless or until your CPU is able to pass enough data to your single GPU to saturate it, adding another GPU will not help in TS.

The graphics engine it uses is still, very much, CPU-limited.

On my system (see sig), it's definitely possible to become GPU-limited when running high resolutions, downsampling and/or SSAA - and downsampling or SSAA is worth having on electrified routes IMHO.

Results from my experiment a while ago:
http://forums.uktrainsim.com/viewtopic. ... 2#p1632996

[RichyV]

Erm, maybe we are getting our wires-crossed a little. Apologies if I'm not reading it correctly...

But your table in that thread shows CPU-limitation, not GPU...

In essence, you have a GPU usage %, whatever it is, you 'up' the resolution, or shadows, or whatever and the GPU utilisation shows either the same or slightly higher but the FPS goes DOWN dramatically - this is indicative of your CPU not being able to send the GPU the data necessary for rendering the scene fast enough. So, CPU limited...

Or, at least, that's how I read it.

[ttjph]

I've shown that in several situations I can get the GPU to 100% - therefore, it's working flat-out processing frames to keep up with the CPU, and I think it's a safe assumption that in some cases the CPU has spare capacity while it waits for the GPU to finish each frame.

Looking at Chasing Yellows, with 1x2 SSAA I got 40 fps with the GPU at 100% (so the CPU was able to calculate at least 40 fps). Turning the AA up to 2x2 (which shouldn't affect the CPU), it dropped to 26 fps. The GPU was still at 100% because it was still the limiting factor; the CPU was (logically) only working (26/40=)65% as hard as at 40 fps.

If the CPU was the limiting factor I'd expect the exact opposite: assuming the GPU showed 100% at 40 fps, it would show 65% at 26 fps (spending the other 35% of the time waiting for the CPU).


You need a fast (single-thread speed) CPU to achieve high fps at max detail with TS, but you also need significant video hardware if you want to run those fps at high resolutions and/or anti-aliasing.

[andynwt]

Dual graphics cards is one of the loopiest things that the whole computer hardware industry has tricked people in to to date. What a waste of money.

[RichyV]

I've shown that in several situations I can get the GPU to 100%.

But only in 50% of your examples AND with an outdated GPU chip - I don't wish to labour the point but the sim engine, without a doubt, has been shown to be predominantly CPU-limited. 50% of your own examples show this. If you'd have been using the same CPU with a better GPU it would project your results to show that, eventually, all of your examples would have lower FPS with higher settings without your GPU ever hitting 100%.

If I put that GFX card in my setup, it would be the limiting factor on all occasions, as it does not outstrip, data-wise, the CPU I have. But, as my CPU is now several generations old, I am able to (and have with my current GFX card) hit it's ceiling, therefore if I put in a better GPU, I do not get any better performance, in fact, I can show that my current GPU can be under and overclocked and, other than at it's lowest clock-setting (650MHz instead of 950) where I can lose 1 or 2 FPS, it makes no difference to the performance whatsoever.

If you buy a cheese and onion flavoured packed of crisps and leave the shop, you have a sample where 100% of the crisps are cheese and onion flavoured, you cannot extrapolate this to mean that 100% of all crisps are cheese and onion flavoured - it's all in the sample size... [Other flavours are available ]