ZOS_JessicaFolsom wrote: »We understand and respect the frustration in regards to the performance issues. We have our Engineering and QA teams currently working on tracking down and testing several things that we believe are contributing to the problem. This is a bit of a nebulous issue, as it has several contributing factors. Each time we nail down a contributing factor and fix it, we push that fix live. We promise to continue to keep you updated, and we do apologize that the issue is still persisting.
@ZOS_JessicaFolsom
There is nothing nebulous (or vague) about it, the problem is the change in how shadows on objects are rendered after the graphics update in 1.2.3.
For whatever reason, the engine is telling the GPU to enter a perpetual idle state while the CPU processes object and shadow rendering. This takes a lot of time internally, and after the process begins the GPU is never ordered to come out of idle mode (until such time as one changes map states, such as traveling to a totally different zone or restarting the client)
Your engineering team needs to look into those idle states. Then you will resolve the issue within hours instead of days/weeks.
To put this into perspective.
A typical mid range GPU will have anywhere from 96 - 1500 processing "cores" dedicated purely to graphics rendering. They are all much weaker individually than a CPU core, but they are much more numerous. Parallel multiple effects processing is by this virtue much more efficient on a GPU.
When you order them into the idle state, the 2-6 (sometimes 8 ) cores a high end machine has, is tasked with doing the work thousands of smaller graphics cores normally do. This is highly inefficient, and even on extreme systems will cause a marked drop in performance. The better the processor the less you see, but eventually you can and will be bogged down.
This is your bottleneck. Release rendering of shadows back to the GPU processing cores and you will solve your lockup problem.
This is made plain by the fact that leaving the map/transitioning to another location/restarting the game, removed the idle flag. The cores process normally until such time as you enter a scenario where the client orders the GPU back into the idle state and the CPU takes over (once again crashing down to under 5 FPS and locking it there, never to recover because the flag is never removed)
To put this into perspective.
A typical mid range GPU will have anywhere from 96 - 1500 processing "cores" dedicated purely to graphics rendering. They are all much weaker individually than a CPU core, but they are much more numerous. Parallel multiple effects processing is by this virtue much more efficient on a GPU.
When you order them into the idle state, the 2-6 (sometimes 8 ) cores a high end machine has, is tasked with doing the work thousands of smaller graphics cores normally do. This is highly inefficient, and even on extreme systems will cause a marked drop in performance. The better the processor the less you see, but eventually you can and will be bogged down.
This is your bottleneck. Release rendering of shadows back to the GPU processing cores and you will solve your lockup problem.
This is made plain by the fact that leaving the map/transitioning to another location/restarting the game, removed the idle flag. The cores process normally until such time as you enter a scenario where the client orders the GPU back into the idle state and the CPU takes over (once again crashing down to under 5 FPS and locking it there, never to recover because the flag is never removed)
To put this into perspective.
A typical mid range GPU will have anywhere from 96 - 1500 processing "cores" dedicated purely to graphics rendering. They are all much weaker individually than a CPU core, but they are much more numerous. Parallel multiple effects processing is by this virtue much more efficient on a GPU.
When you order them into the idle state, the 2-6 (sometimes 8 ) cores a high end machine has, is tasked with doing the work thousands of smaller graphics cores normally do. This is highly inefficient, and even on extreme systems will cause a marked drop in performance. The better the processor the less you see, but eventually you can and will be bogged down.
This is your bottleneck. Release rendering of shadows back to the GPU processing cores and you will solve your lockup problem.
This is made plain by the fact that leaving the map/transitioning to another location/restarting the game, removed the idle flag. The cores process normally until such time as you enter a scenario where the client orders the GPU back into the idle state and the CPU takes over (once again crashing down to under 5 FPS and locking it there, never to recover because the flag is never removed)
If you ever make a game... I'm buying it!
ferzalrwb17_ESO wrote: »To put this into perspective.
A typical mid range GPU will have anywhere from 96 - 1500 processing "cores" dedicated purely to graphics rendering. They are all much weaker individually than a CPU core, but they are much more numerous. Parallel multiple effects processing is by this virtue much more efficient on a GPU.
When you order them into the idle state, the 2-6 (sometimes 8 ) cores a high end machine has, is tasked with doing the work thousands of smaller graphics cores normally do. This is highly inefficient, and even on extreme systems will cause a marked drop in performance. The better the processor the less you see, but eventually you can and will be bogged down.
This is your bottleneck. Release rendering of shadows back to the GPU processing cores and you will solve your lockup problem.
This is made plain by the fact that leaving the map/transitioning to another location/restarting the game, removed the idle flag. The cores process normally until such time as you enter a scenario where the client orders the GPU back into the idle state and the CPU takes over (once again crashing down to under 5 FPS and locking it there, never to recover because the flag is never removed)
If you ever make a game... I'm buying it!
If you buy that guesswork then good luck. It doesn't explain the scaling FPS reduction I see. Occasionally, randomly, mine gets locked to 30FPS (after going down as low as 14) and needs a relog.
What I see in Cyrodiil is an FPS reduction that scales the closer I get to a large number of players. I don't even have to see the zerg so there's nothing to render as yet. If it was a simple case of a problem with shadow rendering then disabling shadow rendering would be a fix. It's not.
If I was to make a wild guess I'd be looking further back than the rendering point because it's well before (and after) then that the issues start. I have also noted that when I'm standing outside a zerg with FPS of 40 and look at the zerg it can seem as if many of them are moving at 10FPS yet the scenery scrolls more smoothly than that and the people behind me animate smoothly.
There are many times where the game looks like it's rendering a 10FPS scene at a much, much higher FPS. (A bit like convering a standard movie to 60FPS without any frame blending).
But my guess is as good as anyone else's. All we see are the symptoms with zero access to code. ZOS are entirely to blame and the only ones who can fix it.
@ZOS_JessicaFolsom
There is nothing nebulous (or vague) about it, the problem is the change in how shadows on objects are rendered after the graphics update in 1.2.3.
For whatever reason, the engine is telling the GPU to enter a perpetual idle state while the CPU processes object and shadow rendering. This takes a lot of time internally, and after the process begins the GPU is never ordered to come out of idle mode (until such time as one changes map states, such as traveling to a totally different zone or restarting the client)
Your engineering team needs to look into those idle states. Then you will resolve the issue within hours instead of days/weeks.
ferzalrwb17_ESO wrote: »
There are many times where the game looks like it's rendering a 10FPS scene at a much, much higher FPS. (A bit like convering a standard movie to 60FPS without any frame blending).
To put this into perspective.
A typical mid range GPU will have anywhere from 96 - 1500 processing "cores" dedicated purely to graphics rendering. They are all much weaker individually than a CPU core, but they are much more numerous. Parallel multiple effects processing is by this virtue much more efficient on a GPU.
When you order them into the idle state, the 2-6 (sometimes 8 ) cores a high end machine has, is tasked with doing the work thousands of smaller graphics cores normally do. This is highly inefficient, and even on extreme systems will cause a marked drop in performance. The better the processor the less you see, but eventually you can and will be bogged down.
This is your bottleneck. Release rendering of shadows back to the GPU processing cores and you will solve your lockup problem.
This is made plain by the fact that leaving the map/transitioning to another location/restarting the game, removed the idle flag. The cores process normally until such time as you enter a scenario where the client orders the GPU back into the idle state and the CPU takes over (once again crashing down to under 5 FPS and locking it there, never to recover because the flag is never removed)
All well and good - but the guys making thousands of dollars per week can't seem to find the code that does what you're talking about.
So...here we are. And...there I go..
@ZOS_JessicaFolsom
There is nothing nebulous (or vague) about it, the problem is the change in how shadows on objects are rendered after the graphics update in 1.2.3.
For whatever reason, the engine is telling the GPU to enter a perpetual idle state while the CPU processes object and shadow rendering. This takes a lot of time internally, and after the process begins the GPU is never ordered to come out of idle mode (until such time as one changes map states, such as traveling to a totally different zone or restarting the client)
Your engineering team needs to look into those idle states. Then you will resolve the issue within hours instead of days/weeks.
Greetings folks!
A hot-fix for the performance issues (FPS drop among other things) is currently being tested. We will keep you updated as soon as we are ready to implement it.
thank you!
IcyDeadPeople wrote: »