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

[wilds]

That's interesting. It's not entirely the topic of this discussion, but I wonder what would happen if someone did the same thing on a 12900KS?

I bet quite well, especially paired with really fast DDR5.

 

[JZ-]

Back with some more results
5800X3d @ stock
2080 Super

corvega: 115.2
dc: 133.3


Only difference this time was my memory timings which are now 3800cl14

Think this means I take number 1 spot for corvega

 

[KompuKare]

Think this means I take number 1 spot for corvega

Yes, you've got it!

(although my sort if by the average of both and Rover896r1's impressive Diamond City score makes it hard to get the overall #1 spot).

 

[JZ-]

(although my sort if by the average of both and Rover896r1's impressive Diamond City score makes it hard to get the overall #1 spot).

Yep, no idea how to get even close to 150fps in scene 2, the difference is kinda crazy.

Also, thanks for making the leader boards, makes it nice to see the results

 

[BoredErica]

5600x 4.65ghz, 2x8gb 3600 c14 1T, 3080ti

Is there any interest for Oblivion testing? I have a save that gets me down to 63fps. Standing on top of Dragonsreach in Skyrim I can get to ~110fps. I'm trying to figure out how to get more stable FPS in those tests though. I would love to get a thread going for Oblivion but I'm worried there will be no interest. Skyrim is more often played but it already runs okay for most people.

 

[KompuKare]

@BoredErica well, yours is a better score than the other 5600X we saw earlier:

Is there any interest for Oblivion testing? I have a save that gets me down to 63fps. Standing on top of Dragonsreach in Skyrim I can get to ~110fps. I'm trying to figure out how to get more stable FPS in those tests though. I would love to get a thread going for Oblivion but I'm worried there will be no interest. Skyrim is more often played but it already runs okay for most people.

I guess the problem is having it repeatable.
Also, the ENB makes it easier to take benches.
Skyrim with a lot of mods would be nice to have datapoints on but then we'd probably have to get any participants to download a 100GB+ Wabbajack modlist.

There are two current Oblivon Wabbajack modlists too.

What might be doable is a Fallout 4 GotY save game with a large user settlement and automations to stress NPCs and drawcall more than DiamondCity does.

 

[BoredErica]

I guess the problem is having it repeatable.
Also, the ENB makes it easier to take benches.
Skyrim with a lot of mods would be nice to have datapoints on but then we'd probably have to get any participants to download a 100GB+ Wabbajack modlist.

There are two current Oblivon Wabbajack modlists too.

What might be doable is a Fallout 4 GotY save game with a large user settlement and automations to stress NPCs and drawcall more than DiamondCity does.

If you want to decrease FPS, you can set weather to raining. That drops FPS a bit and I do it for all of my personal testing. I have a FO4 save where I get 13 FPS lol. Maybe I am a pessimist, but I expect it to be hard enough to get anybody to test vanilla Oblivion, let alone download a bunch of mods. I could try to find saves with more stable FPS. I have a personal test with my mods that is more consistent, but of course it just uses my own load order.

For Oblivion, I think just having ImpeREAL City mods will tank performance enough on its own.


I'm a bit confused how JZ scored 36% higher FPS vs ZiiZoraka's 5800x3d. They even have similar ram, both being 32GB cl14 with only a small difference in 3800 vs 3600. Ditching the result with AMD GPU for a second, JZ still has 22% higher FPS than fish's result. So where I'm at is:
1. I don't understand the 22-36% performance disparity
2. The results are for FO4, and Oblivion is pretty far removed from FO4's engine.

 

[JoeRambo]

I'm a bit confused how JZ scored 36% higher FPS vs ZiiZoraka's 5800x3d.

AMD GPU + DX11 = bad if there are many drawcalls. I think they are about to release or already released the driver that will fix this problem. 5 years too late, but still progress i guess.
NV had more money and/or their GPU architecture was more friendly to draw calls submission from multiple threads inside driver and they held this advantage for years.

 

[wilds]

Just looking a few pixels to the left or right has giant variations in framerate and drawcalls. But I don't want to blame the AMD GPU on the drastically lower framerate quite yet (although it definitely cannot compete with NV here). I would love a save file where the benchmark test requires no mouse movement for easy to repeat settings.

 

[BoredErica]

AMD GPU + DX11 = bad if there are many drawcalls. I think they are about to release or already released the driver that will fix this problem. 5 years too late, but still progress i guess.
NV had more money and/or their GPU architecture was more friendly to draw calls submission from multiple threads inside driver and they held this advantage for years.

I'm aware of that, but even Fish's results was 22% off. That's an entire (pretty decent) generational difference in FPS. Do we chalk it up to ram?

In other news, I managed to get stable FPS in Oblivion. Imperial City, raining, FPS varies from 65-67. This is without mods. With ImpeREAL City - Unique Districts mod enabled, FPS drops to 48-50 with same save.

 

[JoeRambo]

I'm aware of that, but even Fish's results was 22% off. That's an entire (pretty decent) generational difference in FPS. Do we chalk it up to ram?

Fish's result is also NV, so also not comparable to that lower result. Between NV results, 64GB 3200CL16 is bottom of the barrel memory, quad dual rank DIMMs, probably secondaries/tertiaries are really relaxed as well. I'd expect X3D to help more, but it seems this benchmark continues to scale with memory speed even when 96MB of L3 are present.

On topic of scaling, that new #1 result is epic, one has to wonder how much room ADL still has got on GEAR2? Memory controller is already @ 1.75Ghz, not far from 2-2.1Ghz achievable with DDR4 @ GEAR1 mode. Very impressive, and we are still so early in DDR5 cycle,

 

[BoredErica]

In cases where 5800x3d overperforms, would we expect faster ram to also overperform with 5800x? I remember FO4 having the reputation of scaling with ram. Given that, maybe not surprising it scales well with more cache?

Because if that's true, it'd make sense if opposite is true. Maybe games that don't benefit much from faster ram won't benefit much from 5800x3d.

 

[JZ-]

After seeing some replies over on the overclockers forum, I gave this another try.

Corvega: 122.6 fps
DC: 136.1



Did some BCLK overclocking this time:
5800X3D @ 4.539Ghz (102mhz BCLK)
DDR4 32GBCL14-14-14-14-28-1T@4012Mhz
RTX 2080 Super

My corvega scores had a decent increase, however my dc scores pretty much stayed the same. I have no idea how the other X3D owner was able to hit 150 in S2, I can't even hit 140, maybe something else limiting my system?

 

[KompuKare]

My corvega scores had a decent increase, however my dc scores pretty much stayed the same. I have no idea how the other X3D owner was able to hit 150 in S2, I can't even hit 140, maybe something else limiting my system?

Hm, I initially though that since S2 has a higher NPC load where S1 is so high up that's it mainly drawcall that Intel just handled it better but then over on the OCUK thread we now have 5800x3D leading in both:

While I sort it this by Average of both, yours is the best score for S1.

 

[BoredErica]

Oh yeah. Here are my ram timings in case you want to log those...
Exciting to see how Raptor Lake and Zen4 perform.

 

[BoredErica]

With https://github.com/clamchowder/MsrUtil/releases I tried to measure L3 hit rate with my 5600x. It's reading 70% for Oblivion/FNV, 52% for SSE, 49% for FO4. This lines up with my understanding that larger L3/faster ram benefits perf more on newer Bethesda games vs old ones.

 

[Tiens]

Oky I got an urge to post my results with a R5 3600 with some decent RAM and good enough timings. I just wanted to add another R5 3600 to the mix for the sake of reference:

R5 3600 (stock)
2x8GB DDR4 3733MHZ 16-16-16-32-48
GTX 1070 Ti

1080P low settings, high shadows/ Maxed all distance sliders -
First Save (Corvega)
Draw Calls: 11725
FPS: 64.4fps

Second Save (Diamond City)
Draw Calls: 8056
FPS: 78.2fps

 

[USER8000]

Latest table from Overclockers UK forum.

 

[KompuKare]

Latest table from Overclockers UK forum.

And hint, hint, hint, beg, beg, beg:
we would really like to have Zen4, Raptor Lake and especially Raptop Lake + DDR4 results if anyone has those CPUs and want to run the bench!

 

[BoredErica]

And hint, hint, hint, beg, beg, beg:
we would really like to have Zen4, Raptor Lake and especially Raptop Lake + DDR4 results if anyone has those CPUs and want to run the bench!

I bought a 13600kf but I'll still be on ddr4 3800. It will take a few weeks.

tbh, I think it'd be better to just have a save where the camera is pointing at specific angle to hit the draw calls we want to test. It gets rid of ENB requirement + errors with moving camera around. Since it's how it was done from the start though, we're probably in too far to change things.

Less interest in Oblivion bench, since Oblivion/Fallout 3/Fallout New Vegas perf is not as relevant since SSE/FO4 are more popular. The performance situation is more dire IMO though. Take 7700x and slap on some L3 cache and frame rates will look okay even in most demanding part of Fanuel Hall. In Oblivion there is significant stutter moving around outside at high speed but testing requires setting up a script which most people can't/won't do. It will be lot harder to tame those vs getting high steady FPS in Fanuel Hall/Corvega.


Also latest version of prp now reduces draw calls in Corvega slightly.

 

[USER8000]

Latest table from Overclockers UK forum with a Core i5 13600K.

 

[BoredErica]

Finally out of bios flashback hell and started OCing. :^) So far scoring far, far higher than my 5600x. More than I expected in FO4/SSE, little less than I expected in Oblivion.

 

[USER8000]

Finally out of bios flashback hell and started OCing. :^) So far scoring far, far higher than my 5600x. More than I expected in FO4/SSE, little less than I expected in Oblivion.

Can you check the L3 cache hit rates on your new CPU?

 

[Hotrod2go]

5600x 4.65ghz, 2x8gb 3600 c14 1T, 3080ti

Is there any interest for Oblivion testing? I have a save that gets me down to 63fps. Standing on top of Dragonsreach in Skyrim I can get to ~110fps. I'm trying to figure out how to get more stable FPS in those tests though. I would love to get a thread going for Oblivion but I'm worried there will be no interest. Skyrim is more often played but it already runs okay for most people.

I realise this post is over 6 months old but I'm up for the Oblivion testing. What do I need to get going here if your still interested?
Btw, what is that overlay software in your screenshot? apologise for the nOOb question!

 

[USER8000]

Another update.

I realise this post is over 6 months old but I'm up for the Oblivion testing. What do I need to get going here if your still interested?
Btw, what is that overlay software in your screenshot? apologise for the nOOb question!

There is a thread here:

Info - Oblivion FPS Test

The Fallout 4 Draw Call thread is successful so I was wondering if anyone was interested in an Oblivion FPS test. The reason why I want an Oblivion test is because the engine in Oblivion is very different from FO4. Not sure if anyone is interested but I figure it was worth a shot at least. Here...

forums.anandtech.com