Screen resolution and VRAM info

[Vond]

Just ran some tests using Skyrim Performance Monitor (https://skyrim.nexusmods.com/downloads/file.php?id=6491) comparing the difference in VRAM usage between screen-resolutions.For the test I did exactly the same things with everything exactly the same except for the screen resolution. The "test" consisted of me running around in Whiterun, talking to a guard, going into 3rd person and back into 1st several times, sprinting, jumping etc til exhausted and swinging my axe, opening the map and moving around on it for a bit.

 

Results were:

1680x1050 (16:10): 1250mb VRAM max, 66min, 943avg

1440x900 (16:10): 1093mb VRAM max, 61min, 805avg

 

1680x1050 is the native for my 19". The minimum VRAM usage is from the startmenu (and I would assume loadscreens), which lowers the average quite a bit for tests of ~5min.Should also note that I did it 3 times for each resolution just to confirm my results and they were about the same as posted here, give and take a few MB.

 

 

 

Now, I am currently in process of re-optimizing my HRDLC files so they weren't active during the test which brought down the avg a bit I would suspect. Also that area isn't very demanding. Was using Vano's optimized in the meantime however (Hybrid+VN). Full SRO 1.3 installed aswell as plenty of other texture enhancers including quite heavy mods for character models.

 

That isn't really relevant however, the relevant part is that the results differed so damn much between the two resolutions, while it didn't really "feel" any different playing with 1440x900 compared to 1680x1050, and the only reason I hadn't tried this in the past is because I thought it would. The only place I really noticed it was the meminfo text in top right corner that was larger in 1440x900. Granted I have 1280mb on my GPU my choice of resolution for the future is given. :)

 

Sorry if something similar has been posted somewhere, did a search and could not find it if so.

 

Addendum: My settings are Ultra (or rather Custom, but based off of Ultra). 4xSSAA + 4x SGTRSSAA via NVIDIA Inspector. SSAO is on via ENB, same with DoF.

 

I did some quick tests running around in Whiterun aswell, and the results varied there as the 1440 config would eventually reach the "max" VRAM aswell at 1279ish, but in 1680 I got there much much faster and fps was way less stable. I'll try to do some more proper testing later when I get the time!

[frihyland]

Thanks Vond, its known that resolution and VRAM scale together but I haven't seen any numbers before, thanks for starting out the testing. We should be able to figure out the ratio with a bit more testing. I bet google might be applicable here, lol.

[MadWizard25]

Could you please post more details regarding your setup such as ultra or high settings, AF, AA, SSAO, etc. I would also suggest running around outside of whiterun, around the walls, as this seems to require much more VRAM for me.

 

A brief note is that with 1280x1024 with all hi-res STEP mods DDSopt, Skyrim eats on average up to 1.5GB vram, and 1.9GB around whiterun (not sure about inside), at about 50fps. I can post more details of my graphics settings if anybody wants.

 

Switching to 1920x1080 kills my fps. Didnt check VRAM use but assumed it went over 2GB.

[torminater]

I'll try lowering my screen resolution by a little... I already have 1280x1024 so that's going to be quite tricky :/

[Vond]

Ultra settings (or rather Custom, but based off of Ultra). 4xSSAA + 4x SGTRSSAA. SSAO is on via ENB, same with DoF.

 

I did some quick tests running around in Whiterun aswell, and the results varied there as the 1440 config would eventually reach the "max" VRAM aswell at 1279ish, but in 1680 I got there much much faster and fps was way less stable. I'll try to do some more proper testing later when I get the time!

[z929669]

A note on testing using GPU-z:

 

Suggest that you pay attention to GPU Load as an indicator for when you are in the menus versus in game. Only report averages in game using these obvious indicators. It also helps to disable all of the useless junk from the logs (keep e.g., what I have in the DDSopt gude's tables). Also suggest that you shave off the "edges". In other words, spend 60 seconds running your character through the same routine in diff worldspaces, loading several saves per session, each for a 60s run in diff places. Then in the log, you will see the ebb and flow of the pattern. Omit the loading parts and shave off 10 s (10 lines) from each 60s endpoint, such that you get an avg of 40s per run.

 

this will be pretty accurate and reproducible

[MadWizard25]

Cheers z92, this is actually a good recommendation for inclusion into the mod testing wiki, would that be alright?

[z929669]

Cheers z92, this is actually a good recommendation for inclusion into the mod testing wiki, would that be alright?

 

Fine by me

[Vond]

A note on testing using GPU-z:

 

Suggest that you pay attention to GPU Load as an indicator for when you are in the menus versus in game. Only report averages in game using these obvious indicators. It also helps to disable all of the useless junk from the logs (keep e.g., what I have in the DDSopt gude's tables). Also suggest that you shave off the "edges". In other words, spend 60 seconds running your character through the same routine in diff worldspaces, loading several saves per session, each for a 60s run in diff places. Then in the log, you will see the ebb and flow of the pattern. Omit the loading parts and shave off 10 s (10 lines) from each 60s endpoint, such that you get an avg of 40s per run.

 

this will be pretty accurate and reproducible

 

 

Sounds good, I'll do something more like that once I find the time. Just got a new HDD and some more RAM and planning on a full re-install of windows7 and everything, but once that's all done and setup I'll be happy to do some more thorough testing. I didn't use GPU-z for the testing though but Skyrim Performance Monitor, not that it matters since I have both available anyhow. :)

[MadWizard25]

okay, using skyrim performance monitor, i was able to see that my VRAM maxed out at just over 2GB in quite a few exterior places, although average was 1.6GB. still using setting i mentioned previously. This was done after a 1.5h session last night, until i crashed :) I can confirm that using all hi-res version of rextures mods at 1280x1024 screen resolution will require just over 2GB vram to be allocated. Im pretty sure that not only is it allocated, but is also used. I imagine that playing at higher screen resolutions will require up to 2.5GB VRAM, maybe even more.

[z929669]

@MW

For several reasons, I think that this is an overestimation (no offense!). We need to accurately assess VRAM in conjunction with detectable stuttering under controlled conditions before we can say exactly what is necessary.

 

My experiences are simply not consistent with needing any more than a 1.5 (dedicated) Gb VRAM allocation with default STEP using max FPS! versions of mods (but I do not have any hard evidence ... yet ;) )

[Vond]

Ran some tests last night, but sadly forgot to save the results in SPM. :( I do remember the basics though, and when playing for longer periods of time over more varied areas etc the avg/max is actually very similar with 1680 vs 1440. However, the avg fps stayed higher with 1440 compared to 1680 (pretty much never dropped below 35 at all with 1440 which is where I lock fps at, whereas 1680 would drop to 25-28 very often for prolonged periods of time in the outdoors).