This is less of an issue if you have fast RAM.

Well I'm running DDR4-3600...

I just tried this method, limiting TW3 to first 8 virtual cores (Ryzen 7 1700X has 4 physical, 8 virtual on each of two "core complexes" ). I don't see a major improvement really. Same good 40 fps on average. Limiting it to just 4 virtual cores on the first CCX as you described, degrades performance to around 25 fps. Anyway, there is no point in limiting it more than amount of virtual cores on one CCX.

performance governor does help to avoid periodic framerate drops though. I think in my case the game actually hits RX 480 GPU limit and isn't so CPU bound, so may be with Vega it would go higher.

Testing on Vega64 (core restriction):

0-3 : ~29fps (47% GPU load)
0-7 : ~25fps (40% GPU load)
All cores : ~21fps (33% GPU load)

Sitting in White Orchard 1080p. I also see a more stable frame rate when limiting to the first 4 virt cores.

Badly CPU bound here...

That 15% could be because of camera view between two tests. I tried positioning them close, but they could still be a bit different. But I do expect performance governor to improve it somewhat, since the game is CPU bound.

This should be solvable by just not moving your mouse at all after loading the save game. The camera view should stay constant when loading the same save file. That way you could eliminate the different view having a different impact on resources.

Testing on Vega64 (core restriction):

What CPU do you have? Such low GPU load surely means you are CPU bound.

UPDATE: from your profile, it looks like you are using Intel i7 2600k. That's quite old (Sandy Bridge). And I thought you are using Ryzen that's why you are limiting cores. I don't think there much point to do that for Intel.

So in your case, getting a recent CPU would help a lot.

Although not Witcher 3 related, I found that limiting cores in some games can improve (or worsen) performance, In Guild Wars 2 I run it with cores 0-7.

Only Beko can help you because him have strong single thread cpu

Without forget wine dont use correctly many cores, is most common use a few cores (real cores) but with higher frecuencies

However him have at stock speed and wine improve with higher single thread and higher frecuencies

At 5.0ghz will be interesting, in this point is around 50% more strong in single thread than ryzen overclocked

Buy ryzen cpu thinking about wine is bad choice but in other tasks case video encoding, 3d rendering, virtual machines is really good

^_^

Badly CPU bound here...

What CPU do you have? Such low GPU load surely means you are CPU bound.

As I said :)

UPDATE: from your profile, it looks like you are using Intel i7 2600k. That's quite old (Sandy Bridge). And I thought you are using Ryzen that's why you are limiting cores. I don't think there much point to do that for Intel.

So in your case, getting a recent CPU would help a lot.

Nope, this rig has a TR1950x in it.

Nope, this rig has a TR1950x in it.

So that's 16 physical cores, 32 virtual ones and using 4 core complexes. So why is limiting to 0-7 worse for you than 0-4? Something surely doesn't work right.

I guess Thread Ripper support in the kernel is still flaky.

Buy ryzen cpu thinking about wine is bad choice but in other tasks case video encoding, 3d rendering, virtual machines is really good

It's not bad if games are properly parallelized. Games that load just one core are really badly designed. You shouldn't need 5 GHz monster to have good performance.

Nope, this rig has a TR1950x in it.

So that's 16 physical cores, 32 virtual ones and using 4 core complexes. So why is limiting to 0-7 worse for you than 0-4? Something surely doesn't work right.

I guess Thread Ripper support in the kernel is still flaky.

Correct and quite possibly!!

This could be related to the performance issues the high end NVidia users are seeing...?

Buy ryzen cpu thinking about wine is bad choice but in other tasks case video encoding, 3d rendering, virtual machines is really good

It's not bad if games are properly parallelized. Games that load just one core are really badly designed. You shouldn't need 5 GHz monster to have good performance.

FWIW - it's not the system. I installed Win10, to test, and I'm seeing good core distribution and 100% GPU load. This yields (as mentioned earlier) >100fps @1080p...

It's not bad if games are properly parallelized.

Games that load just one core are really badly designed.

You shouldn't need 5 GHz monster to have good performance.

Yes if games using correctly multi core dont be appear this problem

But as youre said and is truth, most games dont use correct many cores, only few cores with higher frecuency

Almost forget wine dont use correctly multi core

Resuming: games dont use multi core properly and wine dont use multi core correctly

^_^

It's also easier said than done. It's not just about supporting multiple cores, but being able to split up the load from the highest loaded core. This is not an easy or obvious thing because some tasks simply can't be split. This is also true for OpenGL (and DX) stuff where there is an amount of work that must be done on a single thread. If that thread starts dominating the speed, that's it and only higher single core speed can help you. This is where the new API will help in the future, but it won't help in current games.

There may be another thing why parallelisation is limited to a smaller number of threads. It's fine to split completely separate tasks into separate processes, but when you have to split a single task or related tasks, you will take a performance hit. Devs may not be willing to do that.

It's also easier said than done.

It's not just about supporting multiple cores, but being able to split up the load from the highest loaded core.

This is not an easy or obvious thing because some tasks simply can't be split.

This is also true for OpenGL (and DX) stuff where there is an amount of work that must be done on a single thread.

If that thread starts dominating the speed, that's it and only higher single core speed can help you.

This is where the new API will help in the future, but it won't help in current games.

There may be another thing why parallelisation is limited to a smaller number of threads.

It's fine to split completely separate tasks into separate processes, but when you have to split a single task or related tasks, you will take a performance hit.

Devs may not be willing to do that.

Very good points and as your said is very difficult solve this situation*

*For this reason higher frecuencies are required for now

In my tests is common find wine use first core (core 0) at maximum and other with low charge**

**some exceptions case lotr war in north - medal of honor airbone when wine use both cores (core 0-1 at maximum) without csmt (with csmt appears low performance)

Personally recommends if stay using other tasks case information apps send to cores unused and leaves cores 0-1 only for wine

Example: taskset -c 1 gkrellm

However some games in wine use only core 1 at maximum case: truck racer, farcry 3 and others

^_^

No progress on the distorted monsters fix freeze so far.

It's a pity, since it seems to be the only game breaking bug for me. May be a few AMD users can sponsor recent GPU for Józef Kucia?

I haven't played a lot recently but I'm yet to see this freeze on my Vega64 system...

The freeze only happens with an experimental patch. The patch is not in any release versions.

I haven't played a lot recently but I'm yet to see this freeze on my Vega64 system...

Yes, indeed. Did you read the bug comments? Freeze happens only with the patch that fixes invisible / distorted monsters bug. And because of it, it can't be upstreamed yet.

It's a pity, since it seems to be the only game breaking bug for me. May be a few AMD users can sponsor recent GPU for Józef Kucia?

Ah.. did I miss this? Does he have a Patreon account or something?

I'd donate. Even as NVIDIA user.