Even though it's Zen2 I'd love to see some PoE benchmarking on that 32-core. Multithreading really helped PoE's performance when it was introduced, though the game still heavily relies on single-core performance. So, it would be a great comparison. It just wouldn't be fair as you're not really doing a side-by-side comparison when you can't slot a threadripper into a non-threadripper setup. When you upgraded to a 3080Ti, you must have felt like a god with that rig.
When running under high load I get about 1 FPS.
In some sitations 0 FPS, depending on the number of monsters and projectiles on-screen.
When idling in town I get 60 FPS.
When there aren't too many projectiles and monsters on screen, I get between 20-40 fps.
My videocard is under a 1-5% load when idle in POE and up to 90% during high loads.
(However its temperature generally does not go above 60c, even in those situations.)
The processor uses, at most 9 cores for POE, when under high load.
(Overall processor usage caps at 5% when under load and sits at 3-4% when idle.)
It normally only uses 3 cores when idle/sitting in town.
Please forgive my explanation below. I am most likely not using the proper terminology to describe the situation.
There are a few situations, where I get latency spikes when a feedback loop forms, from situations where the CPU tries to communicate with the GPU and isn't able to provide or receive information on time. (I suspect that this is either due to the GPU buffer getting full or not enough bandwidth on the PCI lanes between the processor and the GPU.)
This causes latency to slowly climb as the server is asking for information that my computer is not able to provide. (See the second screenshot for a great example of this.)
The most RAM I have ever seen POE request/use was ~12gigs. (Which is ALOT for a top-down RPG game. Games like Call of Duty use at most 14-16 gigs of ram.)
Hardrive usage is almost never a problem. I have never seen reads or writes go above 30%, at most. (Except when downloading and installing updates. Then the hardrive has between 80-90% usage.)
With all that said, it would be great to have an official POE benchmarking tool, but I am pretty sure that GGG would not want those kinds of results to be public.
Regarding network latency, 99% of the time my network latency sits between 15ms and 25ms. (Something is usually very wrong, if it climbs above these numbers.)
It takes alot of specialized work to build and maintain your down game engine, or how the games code interacts with the game engine.
I suspect many of the problems we are currently facing are directly due to the engine not being mature enough.
Last edited by King_Yoshi on Apr 17, 2024, 11:01:10 PM
Then I went a step further.
The nVidia shader cache is specific to PoE and it has its own stupid idiotic location in the NVIDIA/TEMP directory... so I mklink'd that to a 64MB RamDisk; that also includes the poecache being directed there as well through the GENERAL/config path.
And it worked absolutely f* great. Even out of Act1 and beyond, it didn't have many problems enough for me to even notice.
Now that's an interesting solution.
So the bottleneck for you was the rendered shade's being loaded into RAM?
Even though it's Zen2 I'd love to see some PoE benchmarking on that 32-core. Multithreading really helped PoE's performance when it was introduced, though the game still heavily relies on single-core performance. So, it would be a great comparison. It just wouldn't be fair as you're not really doing a side-by-side comparison when you can't slot a threadripper into a non-threadripper setup. When you upgraded to a 3080Ti, you must have felt like a god with that rig.
When running under high load I get about 1 FPS.
In some sitations 0 FPS, depending on the number of monsters and projectiles on-screen.
When idling in town I get 60 FPS.
When there aren't too many projectiles and monsters on screen, I get between 20-40 fps.
My videocard is under a 1-5% load when idle in POE and up to 90% during high loads.
(However its temperature generally does not go above 60c, even in those situations.)
The processor uses, at most 9 cores for POE, when under high load.
(Overall processor usage caps at 5% when under load and sits at 3-4% when idle.)
It normally only uses 3 cores when idle/sitting in town.
Please forgive my explanation below. I am most likely not using the proper terminology to describe the situation.
There are a few situations, where I get latency spikes when a feedback loop forms, from situations where the CPU tries to communicate with the GPU and isn't able to provide or receive information on time. (I suspect that this is either due to the GPU buffer getting full or not enough bandwidth on the PCI lanes between the processor and the GPU.)
This causes latency to slowly climb as the server is asking for information that my computer is not able to provide. (See the second screenshot for a great example of this.)
The most RAM I have ever seen POE request/use was ~12gigs. (Which is ALOT for a top-down RPG game. Games like Call of Duty use at most 14-16 gigs of ram.)
Hardrive usage is almost never a problem. I have never seen reads or writes go above 30%, at most. (Except when downloading and installing updates. Then the hardrive has between 80-90% usage.)
With all that said, it would be great to have an official POE benchmarking tool, but I am pretty sure that GGG would not want those kinds of results to be public.
Regarding network latency, 99% of the time my network latency sits between 15ms and 25ms. (Something is usually very wrong, if it climbs above these numbers.)
It takes alot of specialized work to build and maintain your down game engine, or how the games code interacts with the game engine.
I suspect many of the problems we are currently facing are directly due to the engine not being mature enough.
Even though it's Zen2 I'd love to see some PoE benchmarking on that 32-core. Multithreading really helped PoE's performance when it was introduced, though the game still heavily relies on single-core performance. So, it would be a great comparison. It just wouldn't be fair as you're not really doing a side-by-side comparison when you can't slot a threadripper into a non-threadripper setup. When you upgraded to a 3080Ti, you must have felt like a god with that rig.
When running under high load I get about 1 FPS.
In some sitations 0 FPS, depending on the number of monsters and projectiles on-screen.
When idling in town I get 60 FPS.
When there aren't too many projectiles and monsters on screen, I get between 20-40 fps.
My videocard is under a 1-5% load when idle in POE and up to 90% during high loads.
(However its temperature generally does not go above 60c, even in those situations.)
The processor uses, at most 9 cores for POE, when under high load.
(Overall processor usage caps at 5% when under load and sits at 3-4% when idle.)
It normally only uses 3 cores when idle/sitting in town.
Please forgive my explanation below. I am most likely not using the proper terminology to describe the situation.
There are a few situations, where I get latency spikes when a feedback loop forms, from situations where the CPU tries to communicate with the GPU and isn't able to provide or receive information on time. (I suspect that this is either due to the GPU buffer getting full or not enough bandwidth on the PCI lanes between the processor and the GPU.)
This causes latency to slowly climb as the server is asking for information that my computer is not able to provide. (See the second screenshot for a great example of this.)
The most RAM I have ever seen POE request/use was ~12gigs. (Which is ALOT for a top-down RPG game. Games like Call of Duty use at most 14-16 gigs of ram.)
Hardrive usage is almost never a problem. I have never seen reads or writes go above 30%, at most. (Except when downloading and installing updates. Then the hardrive has between 80-90% usage.)
With all that said, it would be great to have an official POE benchmarking tool, but I am pretty sure that GGG would not want those kinds of results to be public.
Regarding network latency, 99% of the time my network latency sits between 15ms and 25ms. (Something is usually very wrong, if it climbs above these numbers.)
It takes alot of specialized work to build and maintain your down game engine, or how the games code interacts with the game engine.
I suspect many of the problems we are currently facing are directly due to the engine not being mature enough.
TLDR: Single-core CPU performance matters.
Threadripper not being optimal for gaming, who would have thought?
Threadripper not being optimal for gaming, who would have thought?
Um.. no?
Did either of you even read what I wrote?
The processor is not the bottleneck..
It doesn't even hit 5% overall usage.
The single cores that are used for POE don't even hit 80% usage.. so it clearly isn't the processor...
I believe that the bottleneck is either the GPU's cache or the communication between the processor and the GPU. (aka PCI lanes)
On another note, my previous post was @hmcg020.
He asked for some performance specs, so I gave them.
Also, as a FYI, what I gave WAS A quick and dirty TLDR version.
Most normal benchmarking writeups are about 10-15 pages long...
Last edited by King_Yoshi on Apr 18, 2024, 8:12:18 PM
The processor is not the bottleneck..
It doesn't even hit 5% overall usage.
The single cores that are used for POE don't even hit 80% usage.. so it clearly isn't the processor..
Something wrong on your end, mate. I play a super spammy build, maybe worse than yours, and my CPU and GPU spikes are not anywhere close to as crazy as yours. The latency spiking, though, that's got to do with engine limitations.
Original creator of the "Poor Man's Ward Loop" build: https://www.pathofexile.com/forum/view-thread/3480922
Windows 11 Enterprise 64-bit, i7-13700K 5.30GHz
PNY RTX 4080 16GB GDDR6X, 32GB DDR5-6000 CL36
Samsung 980 Pro, Seasonic Prime GX 850W Gold
Let's make it simple: Certain spell/MTX effects are unoptimised for CPU/GPU usage.
If you're having issues try to post your build/spells (and any MTX) as well. In addition to hardware specs (lol)
These fps issues are a case of blind men and a donkey. Everybody is having their own unique experiences.
Example: With Last Epoch 1.0 relase necor Cthonic Fissure spell, when specced into cold damage, was reducing fps to 1 for some players. A temporary workaround was not to take the cold node. This was fixed with 1.02 patch.
