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

[USER8000]

Games tend to be latency sensitive,and if you look at the difference between Ryzen 1 and Ryzen 2,the biggest one is improved latency in the caches,which games like. If you look at IPC improvements,games showed the biggest improvements due to this and DAW benchmarks showed huge improvements.

The i3-8350K beats the Ryzen 2700K in Fallout 4. That is a surprise.

What is even more of a surprise is that Ryzen beats Skylake-X.

That tells you even with a well defined core,the uncore is also important.

 

[USER8000]

Can a Ryzen 3000 owner please run this?? I have not seen any Fallout 4 benchmarks with the new CPUs.

 

[Madcap_Magician]

Can a Ryzen 3000 owner please run this?? I have not seen any Fallout 4 benchmarks with the new CPUs.

I have everything except my case. I'll be running this in the next few days or this weekend and will post performance with various configurations.

 

[IEC]

Can a Ryzen 3000 owner please run this?? I have not seen any Fallout 4 benchmarks with the new CPUs.

I've got a save game in Diamond City. What settings/where do you need me to test at?

 

[Head1985]

Can a Ryzen 3000 owner please run this?? I have not seen any Fallout 4 benchmarks with the new CPUs.

I have tested it.3700x is still slower than 6700k
https://forums.anandtech.com/threads/6700k-vs-3700x-in-23games.2567554/

 

[USER8000]

I have everything except my case. I'll be running this in the next few days or this weekend and will post performance with various configurations.

I've got a save game in Diamond City. What settings/where do you need me to test at?

View attachment 8547

There are some settings the guy who made this thread posted,which was testing highish drawcall areas of the game:
https://forums.anandtech.com/threads/part-2-measuring-cpu-draw-call-performance.2499609/

If you can run those settings then it would be really helpful!

https://forums.anandtech.com/attachments/fallout4-jpg.8461/

I have tested it.3700x is still slower than 6700k
https://forums.anandtech.com/threads/6700k-vs-3700x-in-23games.2567554/

Thank you for the test.

Not a huge percentage as the 6700K is clocked 200~300MHZ higher in your example? Core i5 8400 was faster than all Ryzen CPUs,and I think the Ryzen 5 CPUs might be competitive with the Core i5s.

 

[arandomguy]

I have tested it.3700x is still slower than 6700k
https://forums.anandtech.com/threads/6700k-vs-3700x-in-23games.2567554/

I'm curious if you can run it with SMT disabled and compare? Also perhaps with core affinity to also force the game onto one CCX?

Is your memory single or dual rank btw?

 

[Head1985]

Memory is dual rank.I will try it with smt disabled and core affinity

 

[Head1985]

Ok results:
It is with 3666Mhz cl16 ram so little bit better than in test vs 6700k with 3466mhz ram

 

[Madcap_Magician]

Ryzen 3900x stock settings and ram at 3200mt/s cas 16. 1080TI:

Corvega 11,700 draw calls: 61FPS.

Better than my overclocked 6850k.

I just reinstalled Fallout 4 specifically for this test. Fresh install of windows. Still need to tweak the processor and settings such as game mode, and other windows 10 BS, etc. I'll update sometime in the future.

I'd like more info about the two outliers in the graph. Any special tips on getting better performance? Seems like on two systems now I've had lower results than comparable processors.

 

[MajinCry]

Ryzen 3900x stock settings and ram at 3200mt/s cas 16. 1080TI:

Corvega 11,700 draw calls: 61FPS.

Better than my overclocked 6850k.

I just reinstalled Fallout 4 specifically for this test. Fresh install of windows. Still need to tweak the processor and settings such as game mode, and other windows 10 BS, etc. I'll update sometime in the future.

I'd like more info about the two outliers in the graph. Any special tips on getting better performance? Seems like on two systems now I've had lower results than comparable processors.

The GPU driver plays a big role. In the results, you see that NVidia's driver is substantially faster than AMD's for Fallout 4. The 2700X for instance, is getting 66fps at Corvega with NVidia, while with an AMD GPU the results would be a fair bit lower. Were you using an AMD GPU with your 6850k?

 

[Madcap_Magician]

Both were with a 1080ti. Different driver and windows versions though. I've been super busy, but hopefully in the next couple weeks I'll be able to do more testing and play around with it some.

 

[MajinCry]

Both were with a 1080ti. Different driver and windows versions though. I've been super busy, but hopefully in the next couple weeks I'll be able to do more testing and play around with it some.

Interesting. Windows 10 does have slightly better draw call performance in Direct3D 9, so it's possible the difference is more apparent in Direct3D 11. I wonder if there's a performance difference between the driver versions.

 

[Madcap_Magician]

EDIT: How do I make a proper table? lol. Sorry for the poor formatting.

I've done more testing and am glad to report increased performance. Memory timings made a huge difference.

Note that all tests were done at Corvega and 11700 draw calls.

First things first:
Ram timings: I used the stock XMP profile for my ram. I have two sticks of dual sided Samsung B die. Each stick being 16gb for a total of 32. This is probably the worst for overclocking, but as you'll see I was able to achieve a 9% increase in performance.

The stock XMP profile's primary timings were: 15,15,15,35. With Ryzen if you have geardown mode enabled the CL will operate at an even number and will round up any odd numbers. So I was running at 16,15,15,35 @ 3200.

Updated Bios. I have an Aorus Master and updated to the latest bios. I then updated to the latest chipset drivers from AMD. This increased my performance baseline to 62. Another thing to note was that I sometimes got increased performance by closing HWinfo64. I've also seen less performance just from having Ryzen Master open.

Before changing anything I tried several settings. I disabled fullscreen optimizations - no difference. I disabled the "game bar" and other windows nonsense. This part seemed to help a hair. Maybe .5fps. Obviously not that scientific of a result, but I'll leave that garbage disabled anyway because it breaks certain games.

I then tried out process lasso and tried all sorts of combos. I disabled SMT usage. On my 6850k that helped with some games. So far I've only tested this on FO4 on my 3900x and it made no difference. I tried selecting various cores for FO4 to use and didn't find any variation that made any difference except that less than 6 cores would start to effect performance in various ways. I found the same on my 6850k. The FPS would have more dips here and there with less than 6 cores. Less than about 4 and the max FPS would start dropping rather quickly.

Anyway, onto the timings. NS was measured using AIDA 64


Test mhz timings Infinity Fab NS FPS
1 3200 16,15,15,35 1600 N/A 62
2 3200 14,15,15,34 1600 71 64
4 3200 14,15,15,34 1800 73.4 63
5 3566 16,17,17,36 1783 70.2 64.5
6 3600 16,17,17,36 1800 69.4 65
7 3600 16,17,16,34 1800 66.3 68 - Optimized sub timings.

As you can see, the subtimings and really pushing a couple of the primaries can really help. It's worth it if you have the time. Note that my setup is extremely stable. First round of ram testing was done with 1usmus's config for testmem5. Very useful program. Second round was 12 hours of HCI Memtest using memtest helper. Third round was 24 hours of Prime95.

I'll post again in the future after working on the clock rate. And perhaps again after water cooling.

Good luck peeps!

 

[Madcap_Magician]

Here is my proof. I saw that the original post showed FPS at 11,000 draw calls so I did that test as well and was getting 71fps.

Note that I opened HWinfo64 only after starting FO4. I do see regular boosting behavior under light loads up to 4.6ghz

 

[DrMrLordX]

I loaded Fallout 4, set graphics to 1080p medium, and loaded up some old saves to see how slow I could make my machine with an all-core OC to 4.4 GHz on an 3900x.

The lowest fps I observed was in the 70-72 range atop the Corvega factory. Pretty much where Madcap_Magician was during his screenshots above. Due to Fallout4's weird preoccupation with linking physics to fps and capping fps, my old .inf files had a framerate cap of 100 (for some reason), so I never got fps above that. Mostly the game just stayed capped at 100 fps everywhere. Exceptions were Corvega, looking down from a few places in Trinity Tower, and of course Diamond City.

Setup is 3900x @ 4.4 GHz w/ DDR4-3600 14-15-14-28, and Radeon VII @ 2050 MHz/1200 MHz RAM.

I might try to move the framerate cap up to 144 and see what happens. Physics will bug out but oh well!

 

[MajinCry]

I loaded Fallout 4, set graphics to 1080p medium, and loaded up some old saves to see how slow I could make my machine with an all-core OC to 4.4 GHz on an 3900x.

The lowest fps I observed was in the 70-72 range atop the Corvega factory. Pretty much where Madcap_Magician was during his screenshots above. Due to Fallout4's weird preoccupation with linking physics to fps and capping fps, my old .inf files had a framerate cap of 100 (for some reason), so I never got fps above that. Mostly the game just stayed capped at 100 fps everywhere. Exceptions were Corvega, looking down from a few places in Trinity Tower, and of course Diamond City.

Setup is 3900x @ 4.4 GHz w/ DDR4-3600 14-15-14-28, and Radeon VII @ 2050 MHz/1200 MHz RAM.

I might try to move the framerate cap up to 144 and see what happens. Physics will bug out but oh well!

You gotta max out those draw distances. Medium has very visible object fade in.

 

[mtcn77]

That feels like a gross oversimplification...
DirectX 12 does address this a little bit by forcing vendors to move closer to the hardware earlier in development if they want the best performance, and with the removal of the monolithic thread much of the anti-AMD bias has been removed from DX. I feel that's why Mantle was released, as a clear demonstration that there was a bias in DX 9 and 11 and that the bias could be addressed without penalising nVidia. Open source versions of Gameworks will also allow more scope for closing the bias gap between nVidia and AMD, but it's wirth noting that the developer of Gameworks (nVidia) is leaving the work of closing this bias gap to the people they sell Gameworks to...

AMD is execution port discreet architecture whereas Intel is port unified. Each have benefits, but unified schedulers increase single thread throughput. Hence, AMD is better at playing along Nvidia gpus because AMD doesn't have thread optimisation while Nvidia does. This has implifications. Directx 12 despite all its glory, is still based on Mantle API and runs off 6 main recorder threads. Both AMD and Intel cpus support 6 threads, extra 'thread optimisation' threads are an Nvidia exclusive. Notice, had AMD cpus not supplied sufficient execution resources like past Ivy Bridge cpus, this would be a totally different story and constant cpu peaking would incur a bottleneck and have made thread optimisation overhead unusable, simultaneous to its weak performance without thread optimisation, leaving the cpu with no clear recommendation for fluid run time.

 

[DrMrLordX]

You gotta max out those draw distances. Medium has very visible object fade in.

I was mostly just benching against @Head1985 . Did he max out draw distances?

 

[DrMrLordX]

@MajinCry

I maxed out the draw distances on 1920x1080 Medium (fade sliders all to the right). I left distant object detail and object detail fade at medium.

Here is a short video of what I got:

https://www.youtube.com/watch?v=mcibwfBBz-8&feature=youtu.be

(I de-embedded it so you could watch it on a larger screen, just copy-paste to a new tab and maybe watch in theatre mode).

Sorry the fps counter is so small. You should be able to make out that I hit as low as 68 fps at Corvega. Game is still capped at 100 fps maximum (I haven't bothered tweaking that).

 

[TheELF]

@MajinCry

I maxed out the draw distances on 1920x1080 Medium (fade sliders all to the right). I left distant object detail and object detail fade at medium.

Here is a short video of what I got:

https://www.youtube.com/watch?v=mcibwfBBz-8&feature=youtu.be

(I de-embedded it so you could watch it on a larger screen, just copy-paste to a new tab and maybe watch in theatre mode).

Sorry the fps counter is so small. You should be able to make out that I hit as low as 68 fps at Corvega. Game is still capped at 100 fps maximum (I haven't bothered tweaking that).

You really need that ENB thingy to show you the draw calls if you want to compare to anything because that's the whole point,randomly moving the camera around might never cause you to hit a high draw call spot because often it's a matter of just a few pixels.

 

[DrMrLordX]

You really need that ENB thingy to show you the draw calls if you want to compare to anything because that's the whole point,randomly moving the camera around might never cause you to hit a high draw call spot because often it's a matter of just a few pixels.

I guess? If I can figure that doodad out maybe I'll try it again, but at that point I wonder what is the point of the draw call test, if you can't invoke the high draw call count except in one tiny viewpoint that's . . . as you say, a matter of just a few pixels.

 

[TheELF]

if you can't invoke the high draw call count except in one tiny viewpoint that's . . .

You can and you do invoke it all the time in this game,just because sometimes it's only a few pixels it doesn't mean it's always that.
The point of the thread is to have a specific draw call amount to coincide with the FPS you get for as many CPUs as possible.

 

[DrMrLordX]

You can and you do invoke it all the time in this game

In my experience, that's not the case. I could only manage lows that low from one tiny spot on top of the Corvega factory. Diamond City got close-ish and so did the exterior of Trinity Towers if I looked at exactly the right places. I got trouble exactly nowhere else.

I'll still try it out next weekend or whenever I get the time to figure out how to use ENB to dial in on that one little spot with max draw calls. I mighta already found that spot though.

 

[MajinCry]

In my experience, that's not the case. I could only manage lows that low from one tiny spot on top of the Corvega factory. Diamond City got close-ish and so did the exterior of Trinity Towers if I looked at exactly the right places. I got trouble exactly nowhere else.

I'll still try it out next weekend or whenever I get the time to figure out how to use ENB to dial in on that one little spot with max draw calls. I mighta already found that spot though.

Press Shift+Enter to bring up the UI. This spawns a second mouse for the overlay. On the bottom panel, expand the "Profiler" category to view the number of draw calls.