Skyrim Multithreading tests on a 3960X Hexacore @ 4.65Ghz

[SMB92]

Test 4 I reverted back to default INI settings and the core load changed. One thing that I have taken from this is that, if iNumHWThreads is the only setting that works, it might be the case that setting an even or odd number actually does change the cores. The other thing is, in both variations of the INI settings tested, the second busiest core jumped from core 2 to core 3 (3rd and 4th cores). Notice that iRenderingThread2HWThread is set to 2 in the first set of settings, which would correlate to Core 2 as one might expect, however in the default test (test 4), it is set to 1 and no setting is at 2, but iAIThread1HWThread is set to 3, which would correlate to Core 4, which saw minimal usage in test 4. So obviously this determines that these settings do absolutely nothing apart from that first setting. I will run a test with those setting mixed up and see if the cores shift again. I expect not.

[jones177]

I decided to have some fun last night & run SMB2's inis.

 

I first did a test run with my inis at 1080. I run Skyrim at 1440. I got 59-60fps until I reached Whiterun then 50-54fps.

 

When I changed over to SMB2's inis I got a lower frame rate. Basically the difference between uGridsToLoad=7 & uGridsToLoad=5. The game ran smooth except for minor stutter on turning. The fun part was that a lot of things were slightly different from camera view to player speed.

 

I then upped the resolution to 1440. My save would not load. SPM told my that ram went up & up then crashed. I decided to start a new save & that worked. I then got carried away checking out the differences in the game I forgot I was testing. The last thing I did was try to load my test save that crashed on load. It crashed again but I got a better view with SPM. RAM went up to 3.160. Vram dropped as it does in the load screen. There was no I/o peak. I don't read a lot in this because with my inis I have gone up to 3.8 before a crash.

 

I then went back to my inis. The new save I made for SMB2's inis loaded & ran fine. It took about 2 minutes for the camera view & player speed to adjust to my inis.

 

I then loaded my test save that would not load a 1440, but ran at 1080. Skyrim was totally trashed. 80% of the textures were black & 20% of meshes were missing. I did a fast travel to Whiteruns interior & things looked worse.

 

If I had not done a pretest save my game would have been trashed. My rule is to not overwrite saves while testing. all auto & quick saves were backed up before testing. The saves made in testing were treated like toxic waste & disposed of.

 

Now I have a new mystery. Why would a resolution change trash a save? I think this will become an issue as people go from 1080 to 1440 or 4k.

 

Thanks to SMB2 for shearing the inis

 

Later

[SMB92]

It's very apparent that every setup is different. My game consumes around 7.3GB at 1080P, about 9.5GB at 4K (which is only 30FPS and diving anyway) and I can't say I have ever played with the speed or camera settings, as of yet at least. I get the same rate on 5 and 7 grids as yourself, just that 5 for me is almost unbearable. HT exacerbates the stutter a fair bit, not tragically but not great. I'm a little baffled over the crash you got though, I'm not sure what ENBlocal settings you have but mine is set to 10GB. 6 on card and one ENBhost @ 4GB. That interests me the most. 1440P for me is the same as ugrids 7 so thats my trade off there. I just can't stand the pop-in XD

 

I sometimes get crashes using Nvidias DSR, the ones you get from C++ or something or rather. I think it is definitely of concern if it is trashing saves, this should be looked at further.

 

Also Test 5 is up.

[FiftyTifty]

Something that may be of interest, is that you can open TESV.exe with a text editor, and find all sorts of settings not in the .ini files by default. Try doing a ctrl+f search for "thread", and see what comes up.

[SMB92]

Was planning a search with ESS, didnt know this. Thx.

[jones177]

Hi

 

I still use 8 mb ram so that makes sense that I am running out. The new save I made has less overhead but would probably start to crash after a few levels.

 

I am going to go back to ugrids 7. I used 5 to get good frame rate with a GTX 980 at 1440. This test showed that the 980 ti was fine with 7.

 

My camera view is farther to the right. I set it back in 2012 I just forgot about it. The player speed is a mystery.

 

In ENBlocal I use the setting generated by the RealVision installer, 12288. I know it is higher than the max but I can see why the installer was coded that way.

 

I use EnableUnsafeMemoryHacks=true & ReduceSystemMemoryUsage=true & along with EnableAmbientOcclusion=true in the enbseries.ini to help remove all stutter at ugrids 5. It will be interesting to see if I can remove all stutter at 7.

The Skyrim inis I used were made for my old 680 4gb. They have no enb tweaks in them. bFloatPointRenderTarget was already set to 1.

 

In a few months I will need to make a decision on my next build. Whether to go with a i7 six core or a i7 4 core. The only 6 core within my budget is the i7 5820k. The i7 5820k has about the same core strength as my old 2600k. I would get a performance boost in modern games but none in skyrim. The verdict is still out on the i7 6700k when it comes to skyrim & the 4790k is now a generation down. Getting more out of 6 core would be a big help in making a decision.

 

Thanks for the effort.

 

Later

[SMB92]

I know a guy with a 5820 that is getting the same performance as me. I gave him all the tweaks i had and with that and his 980ti it was the same. I seen another user with 6600k that wasnt getting as good performance as myself, but i am still inclined to say the 6700k would be the better bet than a Haswell. I intend to upgrade to Skylake-E whenever that is out personally as I prefer workhorse CPUs but from a gaming perspective any increase in overall MIPS as where the bonus is at, even if that is 5-10% per gen. I would hope that Skylake-E provides me with near 50% increase but I am no dreamer there, i would probably realistically like 30%, maybe a bit more. This game absolutely prefers grunt on a single core level, if a 6700ks single core outweighs the 5820s then thats what i would recommend above all else. Having said that i would not recommend anything less than a quad core of course. Also you will probably save money, well perhaps if you didnt need to buy the new board for Skylake.

 

Also my RAM is particularly rare, it is a low latency 2400 kit, ost me 600 back in '12. Havent found many 16GB kits of it since. I deliberately run it at 2133 (2150 in my case) as it achieves the same data rate if not better in most applications than at 2400 due to the lower latency. I havent really played with DDR4 yet but i am very skeptical about any gains to be found here as far as gaming, apart from maybe better latency with loading to vram, but then the higher the frequency the higher the latency, not limited to DDR4. Also i can achieve better OCs on the CPU at 2133. I can push to 5ghz comfortably but for the daily overclock 4.65ghz provides a good balance and my Kraken x61 can keep it in the low 60s in Skyrim even in the tropical weather i live in. Just for fun i put one on the 980ti as well and that has been unprecedently good, keeping the gpu to just 45c in Skyrim on quiet mode and 28 at idle (almost ambient)

[SMB92]

I've added a rather large note to the end of the description, posting my final thoughts and observations. I will continue to add a few more tests to the list also, I have just turned off HT so I will start doing those soon. I done a few different ones already, I can tell you now that what I wrote in my notes down bottom is pretty much it, every test just about proves the same, although there has been some unexplained core activity with different settings, none of it actually affects performance. Maybe so on low end, but that's not what we are here to talk about or test. Enjoy.

[FiftyTifty]

Concerning the last couple lines, the low level APIs make draw calls pretty much a non-issue, as far as performance is concerned. Mantle demonstrated that rather well, and we now have Direct3D 12.

 

Hopefully Fallout 4 will get a D3D 12 patch; having a draw call performance limit of ~400,000 is a huge increase over the ~10,000 (best case) we have with D3D 9. D3D 11 is definitely better than D3D 9 in this regard, but it's just a double digit % performance increase over D3D 9, rather than the triple digits we see with D3D 12.

[SMB92]

One of few people i've heard mention mantle and d3d12 together ;) Well at least from my discussions on Hardforum. Some ppl outright denied it (fanbois)

 

I've also got to add a recommendation to that note for people to read the DynDoLOD documentation if they want a better understanding of the settings.