Page 5 of 9

Re: Benchmarks

Posted: 26 May 2016, 06:26
by Tipped
mkborregaard wrote:I don't have other OS's on the machine, unfortunately. It does sound unlikely that this is causing trouble, though, right?
Do you have Compiz? That is (or was, last I checked) known to cause significant performance degradations.
For reference, on the current version of OpenMW on Github, with a Nvidia 550 Ti, and openbox as my window manager, I typically get ~30-50 FPS in the exterior of Balmora. The draw time ranges from around 4 to 12.

Re: Benchmarks

Posted: 26 May 2016, 13:46
by raven
mkborregaard wrote:...
Nvidia GT 730 2GB DDR3
Is that 64bit or 128bit DDR3? You might be bottlenecked by vram bandwidth.

Re: Benchmarks

Posted: 26 May 2016, 17:04
by Capostrophic
raven wrote:
mkborregaard wrote:...
Nvidia GT 730 2GB DDR3
Is that 64bit or 128bit DDR3? You might be bottlenecked by vram bandwidth.
Truly, 64bit bus width sucks. :x

Re: Benchmarks

Posted: 30 May 2016, 21:38
by mkborregaard
Yup, it is in fact a 64-bit interface. F*ck, I did not even know this as an issue, I just bought the card last month (it was the only one I could fit in my cabinet).
I do not have compiz.

Re: Benchmarks

Posted: 02 Jun 2016, 20:48
by PrimordialBronze
Here's more data.


Processor: Intel(R) Core(TM) i5-5200U CPU @ 2.2GHz 2.2GHz
Disk drives: WDC WD5000LPVX-75V0T SCSI Disk Device
Display Adapters: Intel(R) HD Graphics 5500 , NVIDIA GeForce 820M
Windows: 7 Pro

1366 x 768
FOV: 50
VSync: off
Trilinear
Anisotrophy: 16
Water Shader: on
Refraction: on
Texture Q: high
Reflect Actors: on

Running around Balmora:

Balmora exterior: 20-40 fps
Balmora interior: 60-160 fps for various interiors

Script: 2.0-2.7
Mechanics: 1-2 up to 5 when looking at sky
Physics: 8-22
Event: 0.03-0.04
Update: 3-6
Cull: 3-11
Draw: 10-45
GPU: 12-37

Re: Benchmarks

Posted: 15 Jul 2016, 20:04
by BlueFootedBooby
Capostrophic wrote:
raven wrote:
mkborregaard wrote:...
Nvidia GT 730 2GB DDR3
Is that 64bit or 128bit DDR3? You might be bottlenecked by vram bandwidth.
Truly, 64bit bus width sucks. :x
I looked up some stats for context: 64bit is the same as the PS2, half as much as the original Xbox. :P

Re: Benchmarks

Posted: 15 Jul 2016, 23:22
by Ferik
Could've sworn the PS2 had a 256 bit RAM bus width, and the Xbox had a 128 bit bus.

EDIT: Ah, the Graphics Interface that draws things on screen is 64-bit. Huh, I never knew that. But the article I'm looking at says the GPU has a DRAM bandwidth of 2560-bit from 3 different buses: one read only, one write only, and one read/write. That's quite strange and I'm not entirely sure I understand all of it. So, the speed something can be moved into video memory at is significantly faster than it can be drawn on screen at? That doesn't sound like I have it right.

Re: Benchmarks

Posted: 19 Jul 2016, 08:00
by Greendogo
I have this machine at work:
Ubuntu 16.04
Intel i7 6700K (Skylake)
Nvidia GTX 1080 - 8GB (drivers: 367.35)
16GB RAM
500GB Samsung SSD

If anyone is interested in a benchmark of OpenMW on this Linux machine with the latest from Nvidia, let me know. Also let me know how exactly we benchmark OpenMW. Also, give me time, this is a work computer I built for machine learning research at my job, so it's in use doing Neural Network training most of the time, but I do have several hours some days where it won't be doing anything. (Also, I don't have Morrowind with me at the moment, so if anyone wants to send me an extra steam key for this, my Steam username is pcm1123 - but don't buy one just for me, I can just buy my own come pay-day).

Re: Benchmarks

Posted: 19 Jul 2016, 17:01
by SquireNed
Greendogo wrote:I have this machine at work:
Ubuntu 16.04
Intel i7 6700K (Skylake)
Nvidia GTX 1080 - 8GB (drivers: 367.35)
16GB RAM
500GB Samsung SSD

If anyone is interested in a benchmark of OpenMW on this Linux machine with the latest from Nvidia, let me know. Also let me know how exactly we benchmark OpenMW. Also, give me time, this is a work computer I built for machine learning research at my job, so it's in use doing Neural Network training most of the time, but I do have several hours some days where it won't be doing anything. (Also, I don't have Morrowind with me at the moment, so if anyone wants to send me an extra steam key for this, my Steam username is pcm1123 - but don't buy one just for me, I can just buy my own come pay-day).
I thought the 10-series was moving away from the compute cores? Or does machine learning not care about that.

Also, damn, that looks suspiciously like my specs for the next rig I was going to build.

Re: Benchmarks

Posted: 19 Jul 2016, 19:40
by Greendogo
"Compute Core" is the name AMD gives to its cores/threads (I think). Nvidia is the big player in the Neural Network space right now and they're absolutely dominating (mostly because of the quality of their chips and their ample support of research and development). Nvidia calls their cores "cuda cores" and their 10 series, specifically the 1080 and probably the new Pascal Titan being announced in August have >3000 of them. They're blazingly fast. Another important attribute for neural networks is the size of the RAM on the cards. More memory means bigger models can be trained (more classes more details taken into account, etc).