I remember this dude. He knows his game. And he is quite right. Windows scheduler sux for both CPU and GPU. There are registry which should be able to prioritize GPU per application type. And assign particular cores to particular application types. Neither of them work at all since that build which introduced "Game Mode". Long time before I did found tool which works as execution wrapper (configured via registry), which allows assignment particular cores to particular application. But some applications would not even start, and some were crashing. Will have to test this one, maybe it can assign threads way I want. And maybe it would do good for upcoming 3700/3800(X).

wow, sounds really good

Benchmark results will change with this utility, to bad it took them such a long time.

Windows is POS and that's me being generous with this description. Issue: i9-7900X compiling code on Windows 10 Pro 1803 vs Debian 9.6: Debian -> compile -> mouse movement smooth, no lag Windows - msvc 2017 x64 -> compile -> mouse movement LAG, sometimes hangs for a half / one sec (doesn't happen all the time depends on compile code etc., but it happens) IIRC 3-4 years ago (when using i7-4770) same things happened (actually worse) both on Windows and Debian, kernel fixes did fixed this issue, but apparently having multi billion company is not nought. M$ maybe it's better to focus on bug fixes then harvesting user data ??? (but then again how will You earn these cash ...) To me Windows day by day became more and more like a game launcher then proper OS.

Misha Engel:

Benchmark results will change with this utility, to bad it took them such a long time.

More like: Too bad MS couldn't figure this out on their own. Even AMD is somewhat responsible, but, there's only so much they can do to force Windows to adapt.

N0Name:

M$ maybe it's better to focus on bug fixes then harvesting user data ??? (but then again how will You earn these cash ...)

I'm by no means a fan of Windows (I don't really use it on any of my PCs anymroe) but that data collection involves bug info. They don't know what's causing bugs if they don't know they exist and the environment that caused them.

Looking forward to a new review on Guru3D Hilbert.

this doesnt effect 8/16 cores right only 16/32+???

devastator:

this doesnt effect 8/16 cores right only 16/32+???

It depends on the program, in some things it should have some sort of effect on 16 thread CPUs.

devastator:

this doesnt effect 8/16 cores right only 16/32+???

Neither. It only affects the 32c/64t Threadripper, though, I figure the 24c/48t would also benefit from this too, since it's the same overall structure but with some disabled cores..

schmidtbag:

Neither. It only affects the 32c/64t Threadripper, though, I figure the 24c/48t would also benefit from this too, since it's the same overall structure but with some disabled cores..

Should improve performance on any of the TR4 chips with greater than 2 NUMA nodes - so 2970w will also be helped by this. Won't affect the 2950X or previous gen threadripper since they're on 2 nodes.

Where there's a will, there's a way. A job well done. Wendell has outdone himself again.

schmidtbag:

More like: Too bad MS couldn't figure this out on their own. Even AMD is somewhat responsible, but, there's only so much they can do to force Windows to adapt.

It's likely that they could have figured it out, but chose not to devote the time. The WX CPUs are extremely niche so it wouldn't have been priority. If more of these types of CPUs are released into the mainstream then that might encourage MS to act.

D3M1G0D:

It's likely that they could have figured it out, but chose not to devote the time. The WX CPUs are extremely niche so it wouldn't have been priority. If more of these types of CPUs are released into the mainstream then that might encourage MS to act.

Well, if the Ryzen 9, 16 cores on AM4 rumor is true, MS will need to act sooner rather than later. Just typing that seems unbelievable. In 2 years AMD will have made 16/32 c/t mainstream on desktops.

MorganX:

Well, if the Ryzen 9, 16 cores on AM4 rumor is true, MS will need to act sooner rather than later.

Not necessarily. That depends on the CCX configuration and memory controller.

Just typing that seems unbelievable. In 2 years AMD will have made 16/32 c/t mainstream on desktops.

If the rumors are true, I agree. It sure is weird to think about. AMD is definitely stirring up the computer industry.

I wish he also did the testing with Windows Server 2019.

Isn't this very similar to what AMD has already built into the newer Ryzen Master software? https://community.amd.com/community/gaming/blog/2018/10/05/previewing-dynamic-local-mode-for-the-amd-ryzen-threadripper-wx-series-processors

schmidtbag:

Not necessarily. That depends on the CCX configuration and memory controller. If the rumors are true, I agree. It sure is weird to think about. AMD is definitely stirring up the computer industry.

Correct as it was shown this issue only occurs in Windows 10 when you have more than two CCX's. The desktop parts I would have to assume will be 2 chiplets so this issue would still only effect threadripper and EPYC. With that said it may not be a huge issue as Windows Server 2019 may not have this performance regression since I have not seen it tested.

N0Name:

Windows is POS and that's me being generous with this description. Issue: i9-7900X compiling code on Windows 10 Pro 1803 vs Debian 9.6: Debian -> compile -> mouse movement smooth, no lag Windows - msvc 2017 x64 -> compile -> mouse movement LAG, sometimes hangs for a half / one sec (doesn't happen all the time depends on compile code etc., but it happens) IIRC 3-4 years ago (when using i7-4770) same things happened (actually worse) both on Windows and Debian, kernel fixes did fixed this issue, but apparently having multi billion company is not nought. M$ maybe it's better to focus on bug fixes then harvesting user data ??? (but then again how will You earn these cash ...) To me Windows day by day became more and more like a game launcher then proper OS.

For compilation, you should either disable Windows defender entirely, or you should add all compiler, IDE, project folders into exceptions. I had like 3 times speedup doing so. Best is to have linux based virtual machine as there you set up everything and can create full backups before changing components easily. That's if type of project you are working on can be done on linux.

schmidtbag:

... I'm by no means a fan of Windows (I don't really use it on any of my PCs anymroe) but that data collection involves bug info. They don't know what's causing bugs if they don't know they exist and the environment that caused them.

I understand how all of this work and I do agree in general, but... we are talking here about, like what, 750 billion dollar company (it's forth or fifth largest, and riches, company on surface of this planet)? Do they give ME Windows 10 Pro for free? No, I had to pay for it. I don't have any obligation to M$, especially seeing their shady practices. I did purchased product and I expect it to work without my input whatsoever. They can purchase i9-7900X, heck half Intel CPU line-up for that matter, and test it out. To Me they are corporate rats driven by greed. No wonder that they use Linux kernel for IoT, but I do suspect that there is another reason for this move.

Fox2232:

For compilation, you should either disable Windows defender entirely, or you should add all compiler, IDE, project folders into exceptions. I had like 3 times speedup doing so. Best is to have linux based virtual machine as there you set up everything and can create full backups before changing components easily. That's if type of project you are working on can be done on linux.

It has nothing to do with defender, it's kernel / scheduler that it's broken (and I have defender / update / many other, useless to me, services disabled / so no auto updates, "virus scanners" or search speed up's for me 🙂 ). I use Debian for compile and of course Windows and it's just easier to compile and test right away on Windows then cross-compile for win and then run VM to test this out (not to mention issues that comes sometimes with VM itself). I could point few issues with my HW, with is Asus Prime Deluxe, GSKill RAM, Samsung SSD / WD HDD and i9-7900X, but I will skip it as mostly it's now resolved, by me - asked hw question on M$/Asus forums / contacted with support and no one had a clue what the heck was going on so ... that's the level of support that You get purchasing a rig for over 3k USD. Irony here is that free (as in free beer) software works correctly @@. Unfortunately if You need video NLE then there is no other option then either Mac or Windows and from this both I prefer lesser evil (M$). Not to mention games and other software, but things slowly are moving forward on Linux side, maybe someday we will see quality software like Adobe / 3dsMax / Vegas / Premiere natively running on linux box .

JamesSneed:

I wish he also did the testing with Windows Server 2019.

Yes, separately Wendell confirmed Windows Server uses the same flawed kernel.