Performance Issues in PoE and some Corpus Flagships in Empyrean

[GFCat]

My Build:
Intel i7-9700F 8 cores @ 3Ghz 
32 GB DDR4 Memory
1TB SSD
NVIDIA RTX 2080 8GB VRAM
Windows 10 Home OS
1080p 75Hz gaming monitor, dual monitor setup
 

Settings:
DX Version: 11
Graphics Engine: Enhanced
Borderless Fullscreen
Scaled Mode
FPS Limit: No Limit ingame (90 fps limit with NVIDIA control panel)
Effect Intensity: 90%
Reflections: On
Volumetric Lighting: On
Glare: Off
Film Grain: Off
Ambient Occlusion: On
High Shader Quality: Off
Dynamic Resolution: Off
Geometry: High
Particle System Quality: Medium
GPU Particles: Medium
Shadows: Medium
Texture Memory: High
Anisotropic Filtering: 8x
Trilinear: Default
Anti-Aliasing: TAA
TAA Sharpening: 10 ingame (40 with NVIDIA control panel)
Sharpen VFX: On
DoF: Off
Motion Blur: Off
Distortions: On
Bloom: On
Bloom Intensity: 40
Color Correction: Off
Dynamic Lighting: On
Character Shadows: On
Contact Shadows: Off
Weapon Effects: On
Optimized Flip-Model: On

Problem:
PoE begins to have frame slowdowns, dipping from my 90 cap to sometimes as low as 65fps or so, which is below my refresh rate. Problem seems to exacerbate with repeated excursions onto the Plains. The issue makes itself especially apparent during the final phase of Plague Star. Task manager reports that no component is hitting max utilization, with the processor being taxed the hardest but only hitting a high of 83% or so. Frame dips tend to also be at their worst during sunset or night, though this may be coincidental. I also occasionally get frame dips on light intense rooms that are not in open worlds, like assembly-line rooms in Corpus Empyrean. Issues seem to be a mixture of lighting and back-end framework, if I'd wager a guess. I heard in some cases that High Quality Shaders causes a lot of FPS issues regardless of build, and disabling it has reduced the severity of the frame dips to a certain degree, but they still happen. Has anyone else had this? Thanks for reading, please let me know your thoughts or if there's some other setting that could possibly be the culprit.

[taiiat]

23 minutes ago, GFCat said:

Task manager reports that no component is hitting max utilization, with the processor being taxed the hardest

so the GPU in the scenarios that you're noticing problems, is only modest usage?
that means CPUbound of some sort. you can double check by watching the per Core Frequency in these cases, but i'm going to wager that because of a multiple Thread load, the CPU is running at about Base Clock on all Cores. which is decent but not spectacular.
or, it could be running below, if it's throttling or something.

while you can't do real Overclocking there, one thing you could try if you have an off the shelf Motherboard, is disable the Power Limit so that the Chip will try harder to maintain higher Clockspeed. this will between double and triple the Cooling necessary since this is an inefficient automatic process, rather than a manual one.

 

another thing you could try, is checking to be sure you've turned XMP on for your RAM Kit. i expect it's a very modest Kit anyways, but if XMP was off it would be quite slow.

[GFCat]

25 minutes ago, taiiat said:

so the GPU in the scenarios that you're noticing problems, is only modest usage?
that means CPUbound of some sort. you can double check by watching the per Core Frequency in these cases, but i'm going to wager that because of a multiple Thread load, the CPU is running at about Base Clock on all Cores. which is decent but not spectacular.
or, it could be running below, if it's throttling or something.

while you can't do real Overclocking there, one thing you could try if you have an off the shelf Motherboard, is disable the Power Limit so that the Chip will try harder to maintain higher Clockspeed. this will between double and triple the Cooling necessary since this is an inefficient automatic process, rather than a manual one.

 

another thing you could try, is checking to be sure you've turned XMP on for your RAM Kit. i expect it's a very modest Kit anyways, but if XMP was off it would be quite slow.

Yes, the GPU is barely breaking a sweat, so I think that can be ruled out. Base Processor Speed is 3 GHz, but Task Manager says it's running at roughly 4.4 GHz, which I think is its maximum safe speed without causing wear and tear from overclocking. I wouldn't feel comfortable clocking it any higher. As for checking XMP on RAM, is there a resource you could send me to have a look at this? I'm not sure it's necessary because I have 32 GB and it doesn't seem to be getting anywhere close to max utilization while I'm playing, even in zones like Empyrean and PoE.

[MillbrookWest]

7 minutes ago, GFCat said:

As for checking XMP on RAM, is there a resource you could send me to have a look at this? I'm not sure it's necessary because I have 32 GB and it doesn't seem to be getting anywhere close to max utilization while I'm playing, even in zones like Empyrean and PoE.

You can pick up zentimings for this: https://zentimings.protonrom.com/

Here's their github repo if you're so inclined, and the download includes a hash (again if you're so inclined).

This should have everything and more anyone could want about ram speed and timings.

For example:

[taiiat]

17 minutes ago, GFCat said:

Base Processor Speed is 3 GHz, but Task Manager says it's running at roughly 4.4 GHz

which I think is its maximum safe speed without causing wear and tear from overclocking.

As for checking XMP on RAM, is there a resource you could send me to have a look at this? I'm not sure it's necessary because I have 32 GB and it doesn't seem to be getting anywhere close to max utilization while I'm playing, even in zones like Empyrean and PoE.

you really need to read per Core to know what's going on, Task Manager just reports one number for all Cores so you don't know what most of them are running at.

Clockspeed has no impact on the longevity of a Microprocessor, don't worry about that.

it's not about Capacity, it's about running your Memory at the speed it said on the box rather than the base speed.
on most Motherboards there will be a toggle option for XMP on the first thing you see when you enter the BIOS, to make it easy for newbies. 

 

11 minutes ago, MillbrookWest said:

This should have everything and more anyone could want about ram speed and timings.

and more indeed, since numerous of those Stats are Ryzen specific, hehe.

[MillbrookWest]

3 minutes ago, taiiat said:

and more indeed, since numerous of those Stats are Ryzen specific, hehe.

Ah yes, whoops. I skimmed through and thought it read 5800 8-core (going blind).

For @GFCat: You can have a look at Asrock RAM configurator (or something to that tune, Google will probably set you right). My intel based systems run linux so i can't test that myself and provide links that i know work.

[GFCat]

Hmmm. I'll do some more Plague Star runs with the graphs set to all cores to see if one of them hits a ceiling. Regardless, my specs should still be more than capable of handling this, so I'm not entirely convinced this is entirely on my end. This machine is built to be capable of streaming newer games, and it has. I've done plenty of streams of things like CoD: MW and DOOM: Eternal without things like this happening. If any of my cores hit a ceiling I'll get back to you guys, but at the moment I'm still apprehensive about if this is on my end or not. As for XMP on RAM, that seems like a good thing to enable either way so I will, but beyond that I don't want to do some serious micromanaging and deeply in-depth advanced user stuff if I don't have to. I'm happy to turn some in-game settings down or off if I need to.

[GFCat]

30 minutes ago, taiiat said:

you really need to read per Core to know what's going on, Task Manager just reports one number for all Cores so you don't know what most of them are running at.

Clockspeed has no impact on the longevity of a Microprocessor, don't worry about that.

it's not about Capacity, it's about running your Memory at the speed it said on the box rather than the base speed.
on most Motherboards there will be a toggle option for XMP on the first thing you see when you enter the BIOS, to make it easy for newbies. 

 

and more indeed, since numerous of those Stats are Ryzen specific, hehe.

Okay, you were right about checking individual core processes. I'm hitting caps on some of my cores during these scenes like on PoE, so that's my bottleneck. It's likely due to running in the advanced engine with a lot of processor-intensive settings on, I suppose. There was a brief moment where I hit 100% utilization on the CPU, but that might just be from opening my capture software to snap a shot of my task manager. Any particular settings from my settings list you'd suggest turning off that would help reduce processor load while still keeping the game looking good? I'm still unconvinced I want to get into doing anything too advanced like overclocking.

[GFCat]

Apparently this issue is not exclusive to me. Others have reported having troubles in open world segments, even with powerful parts (see Reddit thread below). The OP of this thread claims the new deferred renderer is the culprit in this, so I might just play with the old renderer if I'm planning to do some open world content for a while. Some users even explicitly mention the exact problem I'm having, ie. single thread graphics processes causing cores to hit ceilings. I'm not sure if there are any other solutions, as I am unwilling to start playing around with my processor's clock just yet. Regardless, thanks for all the help, guys. I turned on XMP from my BIOS and I imagine it will make a difference, even if it's not in this game specifically. If anything else has any suggestions, I'd still love to hear it.

 

[GFCat]

After doing some research, I know for certain what the issue is. Warframe is still designed to run single-threaded processes for most of the runtime. For a CPU designed to multithread like mine, it's playing in the wrong ballpark, so when lots of processor-intense lights and shadows are on screen, the load it puts on a single core is too heavy and causes slowdowns. I may just need to switch off of the deferred renderer until it's better optimized, but does anybody know of any particularly processor-heavy settings that could be tuned down to improve CPU performance?

[taiiat]

14 hours ago, GFCat said:

As for XMP on RAM, that seems like a good thing to enable either way so I will, but beyond that I don't want to do some serious micromanaging and deeply in-depth advanced user stuff if I don't have to.

yeah, i wasn't going to ask you to do any manual Overclocking(though from my own experience i can say that would be a resolution). there's also very little of it that you can even do on this setup anyways, in order to do Memory you'd have to have a Z Series Chipset. while your CPU isn't an unlocked model so the only Overclocking you can do is as aforementioned, disabling Power Limits so it does it by itself. so just make sure XMP is on 
anyways.

 

14 hours ago, GFCat said:

Okay, you were right about checking individual core processes. I'm hitting caps on some of my cores during these scenes like on PoE, so that's my bottleneck.

hold up. i was only asking you to keep an eye on per Core Clockspeed.
as unfortunately, CPU Graphs in Video Games are totally useless. the only way for you to tell if you're CPUbound in a game is if your performance is uncapped (or below your cap), and your GPU isn't maxed out.

this is because, CPU instructions do not infinitely scale, so Graphing them is not very accurate. while GPU instructions do, so if a GPU says it is maxed out, it actually is.

 

 

13 hours ago, GFCat said:

The OP of this thread claims the new deferred renderer is the culprit in this

oh, yes. the 'Enhanced' Graphics feature, as well as Dx12, can both currently bring some negative effects on performance (and stability). they're still WIP features at this point. they usually work well enough, but may not always.

13 hours ago, GFCat said:

but does anybody know of any particularly processor-heavy settings that could be tuned down to improve CPU performance?

there's very little that you can do, Video Settings are almost exclusively GPU, in every game.
the main things in Warframe would be..... Geometry Detail, and Field of View. yeah, it's not much. some other Video Settings can have an indirect effect on the CPU but they're almost entirely GPU.

[MillbrookWest]

On 2021-09-17 at 3:54 PM, GFCat said:

After doing some research, I know for certain what the issue is. Warframe is still designed to run single-threaded processes for most of the runtime. For a CPU designed to multithread like mine, it's playing in the wrong ballpark, so when lots of processor-intense lights and shadows are on screen, the load it puts on a single core is too heavy and causes slowdowns. I may just need to switch off of the deferred renderer until it's better optimized, but does anybody know of any particularly processor-heavy settings that could be tuned down to improve CPU performance?

Most things are single threaded - sometimes almost inherently so. Threading predominantly helps where you have problems that you can inherently run in parallel (as in: each problem has no relation to any other problem). 

What you typically see "threaded" (at least "easily") is the renderer, since the gpu typically does not give one iota as to how it does its work; it's parallel by nature. Logic tends to follow on from work previously done, so "threading" in the worst case ends up adding additional overhead if the task you are performing could just as easily have been performed on a main thread - switching threads is a very slow operation.

Games will typically manage a small set of tasks which run on their own thread; Physics, networking, AI, etc. These tasks will spawn additional threads to handle whatever systems run under those main threads, but that generally relies on the engine and what the devs want done. Frostbite, for example, is capable of very pretty images, but their AI borders on brain dead. UE is extremely flexible, but you could get more performant with your own solution etc. If you look at the 'newest' methods for optimizing your game-code (Data oriented design), you have really big loops that iterate through your data - in essence, the design is to maximize the work done on a single core (cache locality to be more correct, but whatever).

That aside, you also have languages that have no real concept of "multi-threading". Warframe for example uses Lua as its scripting language; Lua has no multi-threading. But this is the general rule of thumb for any scripting language (like python for example).

On 2021-09-17 at 3:54 PM, GFCat said:

so when lots of processor-intense lights and shadows are on screen, the load it puts on a single core is too heavy and causes slowdowns. I may just need to switch off of the deferred renderer until it's better optimized

While the new renderer imho does still need some work (i still see the occasional huge frametime spikes),  unless something has gone catastrophically wrong, the new renderer is predominantly meant to be the remedy for this. Forward+/Deferred allows devs to have a huge number of lights in the scene for 'cheap'.

While the lights may be a symptom, it may not be the cause. It could be, whatever triggers the light source also triggers a script. What this script governs could be slow, and having many of these script fire off in a short period of time could be what causes the issue. But this would just be conjecture.

You can submit your EE.log to warframe support and they'll take a look at it and maybe something comes of it if the issue is something that can be quickly actioned - if you get a period of poor performance, you can post an in game message in your squad chat, and this will show up in the logs as a marker.

[sam686]

On 2021-09-16 at 7:42 PM, GFCat said:

DX Version: 11
Graphics Engine: Enhanced

Enhanced and DX11 is slow in solo free roam when both used at the same time. This drops my FPS down to 70 FPS. AMD Ryzen 9500x 3900x 12 core 24 threads, AMD 5500 XT, 144Hz monitor.

DX12 + enhanced give much higher FPS up to about 120 FPS in solo mode, other then some DX12 FPS stuttering or micro-freeze issues. Edit: I am wrong on this part. Enhanced render and high shader quality slow down both DX11 and DX12 when CPU limited.

Or just go with DX11 and classic render for 144 FPS in solo mode.

Multiplayer 4 player squad in free roam often drops my FPS down to as low as 60 FPS. All this slow downs and my 12 core CPU are barely at 20% CPU usage, only 1 or 2 CPU cores are fully used with about 6 more partially used CPU cores.

[taiiat]

7 hours ago, sam686 said:

Multiplayer 4 player squad in free roam often drops my FPS down to as low as 60 FPS. All this slow downs and my 12 core CPU are barely at 20% CPU usage, only 1 or 2 CPU cores are fully used with about 6 more partially used CPU cores.

off topic but, if your GPU isn't maxed out in those scenarios, that's something you could address with Overclocking both the CPU and RAM :)

[sam686]

7 hours ago, taiiat said:

off topic but, if your GPU isn't maxed out in those scenarios, that's something you could address with Overclocking both the CPU and RAM :)

I did the opposite, slow down CPU speed using Ryzen Master to test things. I wonder why enhanced renderer with high shadow quality massively drops frame rate when CPU limited?

I was kind of wrong with it affecting only DX11, this slow down FPS on both DX11 and DX12 due to slow CPU.

Plains of Eidolon Caputra
2560x1440, CPU 4.0 GHz, Shadow quality high
Dx11 Enhanced: 102 FPS
Dx12 Enhanced: 102 FPS
Dx11 Classic:  134 FPS
Dx12 Classic:  135 FPS

1280x720, CPU 4.0 GHz, Shadow quality high
Dx11 Enhanced: 147 FPS
Dx12 Enhanced: 152 FPS
Dx11 Classic: 240 FPS
Dx12 Classic: 253 FPS

1280x720, CPU 0.6 GHz
Shadow Quality Low, Med, High
Dx11 Enhanced: 46, 35, 28 FPS
Dx12 Enhanced: 47, 35, 27 FPS
Dx11 Classic: 49 FPS (Shadow quality high)
Dx12 Classic: 57 FPS

[taiiat]

19 hours ago, sam686 said:

I did the opposite, slow down CPU speed using Ryzen Master to test things. I wonder why enhanced renderer with high shadow quality massively drops frame rate when CPU limited?

Ryzen is a weird animal
inter Core Latency is enormous on multi Chiplet Ryzen Chips, on single Chiplet models it's better but still kinda slow
single Chiplet models have halved Memory Write Bandwidth for some reason (i don't even know why, it doesn't make much sense compared to how CPU's are otherwise designed)
RAM access is disasterously slow

with things like that in mind, whenever the system needs to stream some data, the Core that's requesting it could potentially end up waiting a long time for it.

these quirks get better or worse depending on which generation of Ryzen, 5000 compensating for it the best, 3000 somewhat compensating, and the two before that faring far worse overall.

 

buuuuut it's not necessarily that to blame, they're just likely candidates. 

depending on how Shadows are handled in Enhanced, that could also explain differences. such as if they aren't as pre-computed. or also potentially if they are but are significantly higher quality/larger? or if there are just a looot more of them?

as i can only speak from the angle of the Computer Hardware, i can't tell you exactly what the game is doing.

[MillbrookWest]

Unless i've missed something; one (enhanced) is drawing [dynamic] shadows for everything in the scene, while the other one (classic) excludes shadows outright using only a baked solution.

The enhanced engine is even doing grass, which traditionally is always cut for the cost associated with them.

 

[taiiat]

4 hours ago, MillbrookWest said:

Unless i've missed something; one (enhanced) is drawing [dynamic] shadows for everything in the scene
even doing grass

well that would qualify for the lots more Shadows, and higher quality ones.
so my hunch could very well be it, but i don't have a Ryzen system to fully Overclock and test with to confirm

i probably should have switched and looked myself, but i'm GPUbound enough already on my Hardware so i'd been avoiding it.