[Part 3] Measuring CPU Draw Call Performance in Fallout 4

Page 4 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

gamervivek

Senior member
Jan 17, 2011
485
29
91
Well in the Ryzen 5 3600 video I posted,there were parts on the top of buildings looking over Boston(it was a modded save too),and the FPS didn't seem to did so low as what you have seen. Here is another video from the same person at Ultra(medium shadows) using a Ryzen 5 3600 and a GTX1080TI in the built up part of central Boston:


So Swan's Pond and the Faneuil Hall area and they have mods nstalled.Then there are some videos of modded playthoughs using ENBs,higher resiolution textures,enhanced shadows,etc which show dips to the FPS you are talking about,but from what you are saying its just a normal install?? Is it with the 4K pack installed or not??

I am only asking since the videos I am watching online which seem to see those sub 40 FPS dips,tend to be all with mods like this one:

A Ryzen 5 3600 and a GTX1080TI,but with an ENB,6K shadow resolution,7200 shadow distance. The ENBs alone can destroy performance - from my testing you can see easily over 20000 draw calls,if you have an ENB,better textures,improved shadows,etc.
In that modded video, you can see the fps dip below 40 at the 15min mark. I didn't play much with 3600, but only looked at a couple of places, diamond city improved massively over the 1600(I had posted its benchmarks in the previous thread), but wasn't still 60 in the marketplace, and Faneuil hall had the worst performance below 40 when looking at the fire before its entrance. Swan's pond was below 60 as well, and dipped into 40s depending on where you looked.

As I said, it's too bad that I can't test it with 3600 anymore, and there's the caveat that I didn't reinstall or even reset windows. The other user can test with the 3700X and 1080Ti in those places with my settings(1080p ultra with DoF turned down to low), and see whether they see similar fps or not.

I do have the high resolution texture DLC installed, though I don't think it matters much in regards with performance.

Another older game, Crysis, dipped below 60fps when you get to the famous dawn part in the first level. With 9900KF it's solid 60 again.
 

MajinCry

Platinum Member
Jul 28, 2015
2,488
561
136
ENB's don't add to draw calls. At most, it implements screen space shadows, which does not affect draw calls at all. The only CPU burden is that the ENB has to do funky stuff with the game's assembly code, as there's no source code for the game.

7200 Shadow distance is low for Fallout 4. Before Bethesda released an update that just changed the threshold, the max shadow distance setting was somewhere around 15000.

The GPU driver also plays a role, as Bethesda and NVidia worked together to increase draw call performance. This can be seen in the results graph. AMD's GPU driver takes an fps hit of about 10-15fps or so in Fallout 4's draw call intensive scenes.
 

MajinCry

Platinum Member
Jul 28, 2015
2,488
561
136
What can you do for bethesda games in general to increase draw call performance?
- Run the game in exclusive full screen, this increases draw call performance by a couple percentage points

- Set the process priority to high, also increases draw call performance by a couple percentage points

- If you have a large number of cores (>2 for Oblivion, >4 for Fallout New Vegas & Skyrim LE & Skyrim SE, >8 for Fallout 4), use [NotCPUCores](https://github.com/rcmaehl/NotCPUCores/releases) to handle core affinity, so the games are getting almost exclusive usage of the set cores. This greatly reduces context switching, which wastes CPU performance.

- If you have an AMD Ryzen, make sure the game's driver thread and main threads (use Process Hacker to check per-thread CPU usage) are on the same CCX.

- For New Vegas and Skyrim, when generating LOD via the LODGen tool, make sure the atlas texture resolution is as big as possible.

Those are the current guaranteed ways to increase performance. There's the possibility that the DXVK wrapper, as it matures over time, will bring performance improvements for Bethesda's games. Unfortunately, since Bethesda's renderers are really shit and bizarrely programmed, there's a slight performance hit as DXVK has to do more and more safety checks to keep everything working.

Apparently DXVK can bring huge performance gains to Guild Wars 2, so it's not a pipe-dream.
 
  • Like
Reactions: Elfear and moinmoin

gamervivek

Senior member
Jan 17, 2011
485
29
91
Moved to Alder Lake from Zen3, decided to keep some scenarios for CPU testing.

OS: Win11 Home
RAM: 2x16GB DDR4 3600MHz 16-16-16-36
GPU: 6800XT
GPU Driver:21.11.3



5800X

First Save (Corvega)
Draw Calls:11715
FPS:69.8

Second Save (Diamond City)
Draw Calls: 8055
FPS: 81.0


12700KF

First Save (Corvega)
Draw Calls:11759
FPS:87.8

Second Save (Diamond City)
Draw Calls: 8068
FPS: 94.1



Older results with 9900KF and 1080Ti/Vega56.

 

JoeRambo

Golden Member
Jun 13, 2013
1,534
1,600
136
Looks like alder lake is beast.
I am esp impressed just how good it does in CPU limited situations like these draw calls or for example Anno 1800 late game. Getting incredible boost in performance where it matters - letting keep 60FPS + in late game for longer is game changer to me.
 
  • Like
Reactions: DAPUNISHER

Det0x

Senior member
Sep 11, 2014
621
1,010
136
Moved to Alder Lake from Zen3, decided to keep some scenarios for CPU testing.

OS: Win11 Home
RAM: 2x16GB DDR4 3600MHz 16-16-16-36
GPU: 6800XT
GPU Driver:21.11.3



5800X

First Save (Corvega)
Draw Calls:11715
FPS:69.8

Second Save (Diamond City)
Draw Calls: 8055
FPS: 81.0


12700KF

First Save (Corvega)
Draw Calls:11759
FPS:87.8

Second Save (Diamond City)
Draw Calls: 8068
FPS: 94.1



Older results with 9900KF and 1080Ti/Vega56.

Wondering if i'm doing something wrong.. why do i get so much higher fps ?
Granted i'm running a newly installed GHOSTSPECTRE win10 - SUPERLITE bench OS and a 5950x @ ~5050mhz together with tight CL13 timings pushed to the edge of stability, but still.. what gives ?

Diamond City = 113.6 fps @ 8027 drawcalls
Corvega = 83.9 fps @ 11738 drawcalls

1638891606068.png
1638891637706.png

*edit*
Have doubled checked my numbers and trippel checked that i'm using all the correct ini files following MajinCry "walkthrough how to run with benchmark"..
The only difference is that i could not download the ENB mod v0311 anymore as it have been replaced with newer version..

4. Download and install the latest ENB mod, by placing the files inside the "WrapperVersion" folder, into the same folder as your Fallout4.exe file: http://enbdev.com/mod_fallout4_v0311.htm
Could only get the newest version 0461.

But i still have replaced the ENB ini files with the ones linked
https://mega.nz/#!3kUHTY7K!GnxkhGoqX1WjnqxUsVHMbPhfmO5zFmB5Ys8dcMva048
 
Last edited:

Det0x

Senior member
Sep 11, 2014
621
1,010
136
@Det0x aren't You on your 3090? That's different GPU, not really comparable to his 6800XT.
This is purely a cpu/memory bench as far as i understand.. Its ran @ 720p settings.
My 3090 is only boosting to a whooping ~1100mhz core clock with ~40% usage in this test :p
1638902878906.png
Screenshot with settings taken right after ive closed the Diamond City scene :)
 
Last edited:

JoeRambo

Golden Member
Jun 13, 2013
1,534
1,600
136
This is purely a cpu/memory bench as far as i understand.. Its ran @ 720p settings.
My 3090 is only boosting to a whooping ~1100mhz core clock with ~40% usage in this test :p
I meant that it is not comparable between GPU vendors, i think during DX11 era, Nvidia had way more optimized drivers that were able to extract more performance by making draw call submission/processing multi threaded, while AMD suffered from single threaded driver overhead.

So no wonder NV is chewing more draw calls.
 

Det0x

Senior member
Sep 11, 2014
621
1,010
136
I meant that it is not comparable between GPU vendors, i think during DX11 era, Nvidia had way more optimized drivers that were able to extract more performance by making draw call submission/processing multi threaded, while AMD suffered from single threaded driver overhead.

So no wonder NV is chewing more draw calls.
Hmm okai, i didn't know about this difference.. Do we know if its the same nowadays with newer gpu architectures such as RDNA vs Ampere ?

Btw where is yours Alder lake submission ? Wanna see how much faster AL is vs regular Zen3..
I will do a new run after i get Zen3 v-$ early next year :)
 
  • Like
Reactions: lightmanek

gamervivek

Senior member
Jan 17, 2011
485
29
91
Moved to Alder Lake from Zen3, decided to keep some scenarios for CPU testing.

OS: Win11 Home
RAM: 2x16GB DDR4 3600MHz 16-16-16-36
GPU: 6800XT
GPU Driver:21.11.3



5800X

First Save (Corvega)
Draw Calls:11715
FPS:69.8

Second Save (Diamond City)
Draw Calls: 8055
FPS: 81.0


12700KF

First Save (Corvega)
Draw Calls:11759
FPS:87.8

Second Save (Diamond City)
Draw Calls: 8068
FPS: 94.1



Older results with 9900KF and 1080Ti/Vega56.

3090 results with Alder Lake, RAM is a bit higher clocked but CR2.

OS: Win11 Home
RAM: 2x16GB DDR4 3800MHz 16-17-16-36
GPU: 3090
GPU Driver: 497.09

12700KF

First Save (Corvega)
Draw Calls:11759
FPS:97.1

Second Save (Diamond City)
Draw Calls: 8068
FPS: 120.2

Updated gallery

 

Hotrod2go

Member
Nov 17, 2021
72
25
41
3090 results with Alder Lake, RAM is a bit higher clocked but CR2.

OS: Win11 Home
RAM: 2x16GB DDR4 3800MHz 16-17-16-36
GPU: 3090
GPU Driver: 497.09

12700KF

First Save (Corvega)
Draw Calls:11759
FPS:97.1

Second Save (Diamond City)
Draw Calls: 8068
FPS: 120.2

Updated gallery

With those FPS, you have physics mod installed? over 60 & the engine goes nuts with physics.
 

gamervivek

Senior member
Jan 17, 2011
485
29
91
Not playing it right now, have few hundred hours already. I had tried the physics mod and it seemed to work well without breaking stuff, but can't maintain the fps in the boston area.

Current >60 fps is due to the config files used for the benchmark.
 

Hotrod2go

Member
Nov 17, 2021
72
25
41
Not playing it right now, have few hundred hours already. I had tried the physics mod and it seemed to work well without breaking stuff, but can't maintain the fps in the boston area.

Current >60 fps is due to the config files used for the benchmark.
I see. I play "mod light" that is about 7 mods & use vortex to manage them. At 1440p & after adjusting couple values in 3 .ini files, I got the game pretty much locked at 72 FPS with no physics mods but silky smooth gameplay atm. Even more so with my new 32GB kit ram @ 4600MHz CL19, not saying it wasn't smooth with the previous 16GB kit @ 4400Mhz CL18. On rare occasions it will drop below 72 FPS but always over 60 FPS. On a 144Hz monitor the game engine halves the FPS, if I didn't fiddle with the .ini files it locks at 48 FPS.
All this with an RDNA card.
 
  • Like
Reactions: DAPUNISHER

USER8000

Golden Member
Jun 23, 2012
1,522
753
136
Moved to Alder Lake from Zen3, decided to keep some scenarios for CPU testing.

OS: Win11 Home
RAM: 2x16GB DDR4 3600MHz 16-16-16-36
GPU: 6800XT
GPU Driver:21.11.3



5800X

First Save (Corvega)
Draw Calls:11715
FPS:69.8

Second Save (Diamond City)
Draw Calls: 8055
FPS: 81.0


12700KF

First Save (Corvega)
Draw Calls:11759
FPS:87.8

Second Save (Diamond City)
Draw Calls: 8068
FPS: 94.1



Older results with 9900KF and 1080Ti/Vega56.

I remember you had a Zen2 processor - so how was Zen3 vs Zen2 in performance?
 

KompuKare

Senior member
Jul 28, 2009
720
266
136
Finally got around to trying this.
Just upgraded to a Zen2 3600 on a B550M - might see about grabbing a Zen 3 later if prices drop a bit and then I'll re-run this.
But for now, I had two GPUs to try. My 470 4GB which got I bought used as a temp card and a brand new 3050 8GB which I got from the OCUK deal at MSRP (£240).

Radeon 470 4GB run:

OS
Windows 10 x64 21H2​
CPU
Ryzen 5 3600 stock​
RAM
2 x 32 DDR4 3200 16-20-20-39-74 T1​
GPU
Radeon 470 4GB​
GPU driver
21.10.2​
Save 1 Drawcalls
11,712​
Save 1 FPS
51.2​
Save 2 Drawcalls
7,996​
Save 2 FPS
60.0​
Yes, V-sync was disabled for the above!

And for GeForce 3050 8GB run:

OS
Windows 10 x64 21H2
CPU
Ryzen 5 3600 stock
RAM
2 x 32 DDR4 3200 16-20-20-39-74 T1
GPU
GeForce 3050 8GB
GPU driver
511.32
Save 1 Drawcalls
11,730
Save 1 FPS
56.2
Save 2 Drawcalls
8,045
Save 2 FPS
66.0

Same Ryzen 5 3600 running at stock, and at 480P low it really should only be driver overhead so it looks like Nvidia's overheads are better.
 
  • Like
Reactions: USER8000

USER8000

Golden Member
Jun 23, 2012
1,522
753
136
Can someone test this with the 5800X3D`? This could be the perfect chip for modded Skyrim/FO4.
On Overclockers UK people have been running the benchmark in this thread:




One Ryzen 7 5800X3D test so far - another will be added when the forum member gets the CPU.
 

Madcap_Magician

Junior Member
Apr 26, 2018
11
31
91
I redid testing on my 3900x with a current bios, windows 10, etc.

I ended up with slightly worse performance with the 3900x than I did originally on page 2 of this thread.

My new 5800x3D performed fantastically in the test.

110fps in corvega and 125 in Diamond City with 3600mhz ram and optimized timings and subtimings.

Slightly less FPS with 3200mhz and the xmp settings for timings.

Note that the "CPU Package power" for the 3900x is 65 watts in diamond city with 77 FPS.
59 watts at 125 fps for the 5800x3D :O.

Quite the improvement! I think there's room for some efficiency tweaking. I noticed that package power is 54 watts in diamond city with the ram and infinity fabric set to lower speeds at a loss of only 2 fps.

Pretty incredible results. I'm quite happy with them!

FYI I am using zero mods enabled during testing even though I'm launching the game from mod organizer. I do have a few creation club items installed though.
 

Attachments

ZGR

Golden Member
Oct 26, 2012
1,920
419
136
Was feeling way too lazy to start a new game, so I downloaded a save from Nexus Mods (my current save cannot fast travel)

Did 2 tests. One in Diamond City and Corvega using default video settings (disabled Vsync and increased resolution to 1080p and went to borderless), and the other using much, much longer range draw distances that I use for my modded game.

Sorry if I am in wrong location for Corvega, this is fast travel position aiming at city area.

5775C @ 4.2 GHz | 3.8 GHz cache | 1.9 GHz L4 | 32GB 4x8GB 2400 MHz CL10 | 1080 ti

Corvega Default:
corvega2.jpg
95.5 fps - 3549 draw calls

Corvega Custom Preferences:
corvega.jpg
73 fps - 7381 draw calls

Diamond City Default:
dcity2.jpg
83.4 FPS - 3562 draw calls

Diamond City Custom Preferences:
dcity.jpg
60.2 FPS - 8557 Draw Calls

When my game has all the mods running, I only get around 20-30FPS in heavy combat (hundreds of AI!). The real goal for my next build is to get near 60 fps in these 'old' games.

Will be doing exact same save and settings for 5800X3D in a couple days I hope.
 

Attachments

Last edited:

KompuKare

Senior member
Jul 28, 2009
720
266
136
Was feeling way too lazy to start a new game, so I downloaded a save from Nexus Mods (my current save cannot fast travel)

Did 2 tests. One in Diamond City and Corvega using default video settings (disabled Vsync and increased resolution to 1080p and went to borderless), and the other using much, much longer range draw distances that I use for my modded game.

Sorry if I am in wrong location for Corvega, this is fast travel position aiming at city area.

5775C @ 4.2 GHz | 3.8 GHz cache | 1.9 GHz L4 | 32GB 4x8GB 2400 MHz CL10 | 1080 ti

Corvega Default:
View attachment 60617
95.5 fps - 3549 draw calls

Corvega Custom Preferences:
View attachment 60618
73 fps - 7381 draw calls

Diamond City Default:
View attachment 60620
83.4 FPS - 3562 draw calls

Diamond City Custom Preferences:
View attachment 60621
60.2 FPS - 8557 Draw Calls

When my game has all the mods running, I only get around 20-30FPS in heavy combat (hundreds of AI!). The real goal for my next build is to get near 60 fps in these 'old' games.

Will be doing exact same save and settings for 5800X3D in a couple days I hope.
Was feeling way too lazy to start a new game, so I downloaded a save from Nexus Mods (my current save cannot fast travel)

Did 2 tests. One in Diamond City and Corvega using default video settings (disabled Vsync and increased resolution to 1080p and went to borderless), and the other using much, much longer range draw distances that I use for my modded game.

Sorry if I am in wrong location for Corvega, this is fast travel position aiming at city area.

5775C @ 4.2 GHz | 3.8 GHz cache | 1.9 GHz L4 | 32GB 4x8GB 2400 MHz CL10 | 1080 ti

Corvega Default:
View attachment 60617
95.5 fps - 3549 draw calls

Corvega Custom Preferences:
View attachment 60618
73 fps - 7381 draw calls

Diamond City Default:
View attachment 60620
83.4 FPS - 3562 draw calls

Diamond City Custom Preferences:
View attachment 60621
60.2 FPS - 8557 Draw Calls

When my game has all the mods running, I only get around 20-30FPS in heavy combat (hundreds of AI!). The real goal for my next build is to get near 60 fps in these 'old' games.

Will be doing exact same save and settings for 5800X3D in a couple days I hope.
Thanks for that, but somewhere in the OP or a few post later, there should be the save games to use.

Actually, it is not that easy to find, as there have been various parts to this, but this is where the original instruction were:


So basically download two saves, backup your INIs, install provided INIs and edit them, install the latest EBN, disable vsync, etc.

As it happens, your Diamond City screenshot looks close to where the benching save is, but the Corvega save for the benching is way up in tower overlooking everything, hence why its draw call numbers are so high.
 

ZGR

Golden Member
Oct 26, 2012
1,920
419
136
Thanks for that, but somewhere in the OP or a few post later, there should be the save games to use.

Actually, it is not that easy to find, as there have been various parts to this, but this is where the original instruction were:


So basically download two saves, backup your INIs, install provided INIs and edit them, install the latest EBN, disable vsync, etc.

As it happens, your Diamond City screenshot looks close to where the benching save is, but the Corvega save for the benching is way up in tower overlooking everything, hence why its draw call numbers are so high.
Thanks, will re run in a bit! definitely felt a bit lost hopping through the threads!
 

ASK THE COMMUNITY