NVIDIA SLi and Grimrock 2 numbers

[hansmuff]

I created this thread viewtopic.php?f=18&t=7528&p=77043&hilit=sli#p77043

But that was in the wrong forum. I have done a little more work to quantify my findings.
Sorry about the linked vs. inline images. Deal with it

Here's my NVIDIA setup for working SLI with performance improvements over single card use:
http://screencast.com/t/HxsjnatcZKwd
http://screencast.com/t/T5uIC6Z6u

Other setup notes:
1980x1200 fullscreen, all graphics options on 'High'.
debug=true
maxFrameRate = 500
v-sync off

I haven't taken GPU-Z or in-game screen shots. However, I have a save game I loaded that starts in a dungeon (15 26 0 23), and I want up to the outside in a very busy area graphics wise (22 13 3 22).

SLI set to NVIDIA default (because there's no profile, that means "off"):
Dungeon spot 124fps (range 119 - 160)
Outside spot: 55fps

SLI set to Forced Rendering 1: total crap, slower than default, half the fps!

SLI set to Forced Rendering 2:
Dongeon spot : 193fps (range 190 - 300)
Outside spot: 67fps

So what this essentially means for me is that with V-Sync on, SLI keeps me above 60fps and therefore fluid, where the single card causes a drop to 30fps during difficult renders outside due to V-Sync halving it.

It really should be noted that SLI for the outside rendering doesn't do all that much effectively. It just happens to keep me above 60fps.

As far as CPU load in the SLI Outside scenario, I have this to show:
http://screencast.com/t/HSBYoHyBVF3W

Evidently, there is some usage of multiple cores that goes up during gameplay. It is not caused by the "Threaded optimization" switch in the NVIDIA panel; I disabled that to test, and there was no difference in CPU load patterns.

Also, it looks like the CPU could be a bottleneck here because the main thread on core 1 does peg the CPU quite a bit.

As a last experiment, I'll do a low resolution, low options, with SLI to see how much I can get out of that and what the CPU does:
Dungeon: 300fps
Outside spot: 85fps

CPU: http://screencast.com/t/8poe3E1JL

So there was some improvement by lowering everything drastically.
Inside the dungeon, that was a 50% improvement.
Outside, it was about a 27% improvement.
CPU was almost completely pegged on high details/res, but then completely loaded down at low details/res.

As a VERY last experiment, I turned off NVIDIA threaded optimization in the driver.
High details/res numbers:
Dungeon spot: 195fps
Outside spot: 75fps

I can't really draw sound conclusions from this. I just figured the numbers may be interesting, and the gain from SLI is measurable.
And NVIDIA threaded optimization, if anything, seems to be a hindrance to this game.

Last is system configuration:
i7-2600k @4.4GHz, 16GB RAM, 2x Asus 660Ti TOP at PCIe 2.0 x8 each. Win 8.1 latest patches.

[Dr.Disaster]

That's some pretty good info!

Leaving aside the SLI part which i can't compare to (only 1 GPU) i get the same behavior on the CPU part. LoG2's main process keeps 1 core really busy while several childs use up to 3 more cores on a low to medium rate. The overall performance of the game might be improveable by off-loading some of the main processes tasks to it's child(s), especialy with slower CPUs.

EDIT
It would be interesting to see how FPS behave in the sewers with it's water und reflection stuff.

[hansmuff]

I'm playing this game very slowly for a number of reasons, mostly a lack of time. I'm not in the sewers yet, but once I get there I'll post some data from there.

A benchmark mode for this game would be pretty nice, dear Almost Human Just something that flies through a couple of different areas, with low and high complexity. But until then, I might post some data, maybe in a better format if I get some time to look at how to do that.

Is there a way to take in-game screenshots without installing 3rd party tools?

[Dr.Disaster]

A benchmark mode for this game would be pretty nice, dear Almost Human

The pre-release versions 2.0.x had that but it got removed with the 2.1.x release candidates.

Is there a way to take in-game screenshots without installing 3rd party tools?

If you got the Steam version just press F12 or whatever hotkey you have configured here.

[badhabit]

So there was some improvement by lowering everything drastically.
Inside the dungeon, that was a 50% improvement.
Outside, it was about a 27% improvement.
CPU was almost completely pegged on high details/res, but then completely loaded down at low details/res.

Thanks indeed for taking the time fiddeling with the Nvidia settings and identifying best working combinations & sharing with us.

About conclusion, from your numbers (as my one) I interprete that on one hand the Grimrock engine is CPU bound for low settings (while not being multithreaded) and GPU bound for high settings, leading to the weak overall scaling between low and high settings of only 25-50%. That the scaling is especially bad for the most demanding outside areas is also unfortunate.

Yes, I agree, and it would be helpful if AH would make the benchmark/profiler available for everyone!

PS: as you have a very multicore system, could you please make a FPS analysis vs the number of available CPU cores ? 1 to 8 via CPU affinity to the Log2 process via taskmanager. I'm a little bit baffled why AH has quadcore as recommended system while it can't fully utilize my dual core... is there some FPS gain after making more cores than 2 available to LoG2?