Threadripper performance is almost halved when running Windows 10

>Threadripper performance is almost halved when running Windows 10

phoronix.com/scan.php?page=article&item=2990wx-linux-windows

Why is Windows 10 so much slower than other operating systems?

Attached: threadripper2990WX_linux_vs_windows.png (710x417, 38K)

Other urls found in this thread:

youtu.be/mK9bTzIWIQA?t=30
twitter.com/SFWRedditGifs

>wangblows 10 literally sucks
duh

Windows sucks since it's first public version bro

Looks like classy MS scheduler issues
The task question it's, why is Antergos so behind Ubuntu LTS?

>Why is Windows 10 so much slower than other operating systems?
poor NUMA awareness

needs to copy and encrypt all the data to send to the NSA before 7-zip does.

Attached: 1533995222068.png (849x749, 358K)

This is why Windows 10 isn't the goto server.

>The task question it's, why is Antergos so behind Ubuntu LTS?
Likely differences in kernel config

This is Win10 Pro. Perhaps they lock NUMA awareness to Windows Server

so is a pirated w7 ultimate still the best choice if i need to play games and can't deal with the bullshit linux needs to do to play games on it ( while also losing 2% performance just because)

>muh games
Why are you even in this thread discussing a high performance workstation processor?

Attached: 1532924504132.jpg (400x285, 33K)

Reading more about it, the WX series of Threadripper are locked to be in NUMA-mode. While the regular X series Threadripper have both NUMA and UMA mode(default). While *linx systems have NUMA awareness automatically, the Windows 10 probably require bios change from UMA to NUMA.

Well it's just one specific benchmark but I wonder that too. I really don't have much myself but my transfer speed with rsync more than doubled going from Ubuntu to Arch. Like said it's most likely a differnece in kernel config or maybe other system setting. Might be some compile-time flag for a library as well. ClearLinux is optimized for Intel so interesting it's the one where AMD performs best, guess some optimizations might benefit x86_64 in general.

>almost halved
That's LESS THAN HALF.

If you had two machines with threadrippers running Windows, they would still be slower than just a single machine running Linux.

Jesus. It's like just throwing $1700 right in the garbage. Why do people even use Windows again?

Normal thread ripper doesnt have this issue because it doesnt have 2 gimped dies that cant directly access memory

The botnet uses at least 16 threads to send all your pictures to the HQ

Oy vey goyim, why use amd on linux when you can use intel on windows?

Attached: ayymdpoorfag.png (633x900, 112K)

dipshit

2 gimped dies for an extra 950. nice product you are shilling pajeet

>Windows 10 PRO
>PRO
That's the problem. The test should use the "Workstation" version because the "PRO" version limit the core that you can use. upgrade to "workstation" now goy.

>processor works as expected on one operating system but not another
>BUH MUST BE DA GIMPED DIES LOLOL
Neck youreself.

I know youre a brainlet but look into it. It actually has 2 dies that cant directly access memory. 2 gimped dies

So mad he has to defend this.

They even give you a special mode to turn those 2 gimped cores off for workloads that use memory

Use the stripped version of windows 10. It disables updates but less than 1 gb is run when first installed and it barely uses the cpu. Windows 10 features ruin the os

>Buy 32 core CPU
>Why you use 32 cores? Shut it down!
>Shuts off gimped cores.
>Now uses 16 core CPU

God's we were strong then.

Its not an update/bloat issue. Its a core problem with windows 10(and probaby older windows). The system is limited to 32 threads. This is why the if you disabled half the cores from Threadripper, it suddenly "works" proper again. Windows is shit.

In theory, they support higher threads, but since no single desktop variant CPU had 32 cores 64 threads before, this wasn't a tested "feature" of Windows.

Its not a windows problem the 2990wx has 2 gimped dies. stop lying to yourself. It handles 64 threads just fine in cinebench on windows because it isnt memory intensive

Attached: supported_features.png (324x326, 18K)

>schedule memory intensive tasks on IMC cores
>schedule other less latency intensive tasks on IF cores
>suddenly performance goes up
WOW

Attached: 1534039763939.jpg (226x223, 19K)

If its 2 gimped dies, then disabling half would lower the speed. Instead it doubles the speed.

Retard.

>not benchmarking on gentoo
you had one job phronix

This is completely disregarding the results which literally show that it's fine, just windows blows at being NUMA aware.

>flash
literally no one should be using that

The scheduler is garbage.

If it weren't a windows problem, it wouldn't be performing so much better on Linux. Clearly, linux somehow makes better use of the "gimped" dies. There's no excuse for any part working so much worse on an OS people actually pay money for.

Oof, that didn't age well.

Windows defense force is here.

Stop using logic against braindead subhuman kike shills.

2 gimped dies for an extra 950 and drones defend it this hard. Just save yourself the pain and buy the 16core tr that isnt crippled by gimped memory

*nix was always made for a server operating system

Learn to code, Wangblows pajeet shill.

/thread

Wow I was thinking about that too.
Normies btfo, enjoy your nocores

Attached: 869715642700.png (918x720, 520K)

>REEEEEEEEEEEEEE 2990WX IS SHIT AND SHOULDNT EVEN EXIST BECAUSE REASONS
I don't really get you guys? 2990WX fills in the market segment between the 16c threadrippers and 1P epycs, its extremely good value for those that want that extra computational power but their use case doesn't make use of the extra memory bandwidth or other features of epycs.

...

Been doing it on a living for years. Has nothing to do with the drone defense force doing damage control for this aweful quad channel memory config on the 2990wx. Its fucking gimped, steve from hardware unboxed explained this very clearly and even he isnt using it because it chokes in memory heavy workloads

>posting dumb slut
>the post itself is just as dumb
checks out, huh

And yet Steve literally said it's actually wangblous being shit, pajeet programmer.

No he didnt lmao

Braindead wangblows pajeet programmers on suicide watch: youtu.be/mK9bTzIWIQA?t=30

Attached: 1495158670874.jpg (433x419, 29K)

Just disabled the gimped cores and its fine lmao

Source?

>stop ruining my shit flinging with reasonable arguments

Using logic is anti-semitic

What? Using logic isn't anti-semantic

>server cpu
>windows 10
this is what happens when you are an AyyMD fanboy and spend a shitton of money on something you don't need.

Attached: 1.png (276x230, 115K)

Because intel on windows is still both more expensive and slower than amd

>TR
>server CPU

ok, linux that's cute. What about bsd performance?

>I'll take shit no one uses for 400 Alex!

>bsd performance
noone who uses bsd can afford anything beyond core2duo.

He didn't ask how many people you think use it.

>bsd performance
macOS only runs on Intel

ok, now how about BSD performance

> Why is arch and anything arch based so shit?

>pay more to unlock the full potential of your hardware
wow

wow, surprised to see so many people ignorant of bsd.

>bsd

>NUMA awareness

MAYA HEEE

MAYA HOOO

Attached: numa-maya.png (637x432, 226K)

Fucking kek

Haven't seen all the benchmarks yet but are they so bad that people are already looking for excuses?

Yeah its pretty bad, windows babbys are trying super hard to shift the blame.

Just click the link. Windows gets absolutely destroyed in every benchmark except Solitare.

What do you mean excuses?
Windows scheduler is known to be a fucking pain, indiscriminately to every new CPU configuration.
When Intel reintroduced HyperThreading with Nehalem, the Windows scheduler was so broken it was recommended to disable HT in BIOS for better performance. And it took no less than two years of begging from Intel and top server vendors for Microsoft to fix that shit.
When Ryzen 1 hit the shelves, it was Windows scheduler that was broken again, but now it took "only" a year to fix.
Fast forward to present, and Threadripper 2 is wrestling with the same fucking scheduler. Wonder when the patch comes out this time.
At this point I'm surprised people DON'T know what a piece of shit Windows is. The year of the Linux desktop can't some soon enough.

On windows, the 32 core gets raped by the 16 core variant.

OP's link @ Phoronix does test for both Windows and Linux and we see a huge disparity. On linux, the 32 core functions like its supposed to. On Windows, the speed is half of that.

If you've read both comparison reviews, the blame is on Windows implementation of NUMA and the scheduling.

If you've only read the Windows gaming reviews, its the "memory bandwidth" issue.

>still no bsd
Jow Forums has become /v/eddit

Holy shit, the link to the test suite is right at the bottom of the article. Do it yourself, we're not here to spoonfeed you.

>we

>Using pajeetOS

Attached: 1530301954199.jpg (690x460, 186K)

BSD is slower than Linux for years, what performance?

*BSD's scheduling varies a lot depending on what *BSD are you running, and the ones that aren't muh secure or muh old (ie. OpenBSD and NetBSD) are guaranteed to be comparable to Linux or faster

Of what?

Show me proof mate. Baseless claims won't do shit. I wanna see TR2 running on it.

Nobody cares because Theo won't touch ryzen.

Go to phoronix and just look for any Linux vs BSD comparison, m8. In more than 90% of tests BSD is slower

>Baseless
Are you retarded? The author of ULE from FreeBSD studied CFQ to make it, and LKWT from DragonflyBSD was made with the sole purpose of SMP which obviously requires NUMA awareness

Don't forget windows 7 not supporting bulldozer for a year and then making the fix as difficult to get as possible.

I'd like to see the DragonflyBSD on that one to be honest.

your mom didn't age well

>The system is limited to 32 threads.
Clearly not, since all 64 logical CPUs are actually being used.

>*nix was always made for a server operating system
It was literally made for interactive timesharing use.

It's not like the OS knows whether a certain thread require low memory latency.

Now THAT would be interesting
there's a real genius behind DragonflyBSD and he doesn't have any sex

This but unironically

Different cpufreq settings than the other distros, general Arch shittery

Applications that don't (yet) work on Linux. All we need is Adobe and ONE major CAD vendor to port.

Intel compiler.

Linux does. Linux is NUMA aware and thus beats the bejeezus out of Windows on NUMA systems.

They actually claim that their NUMA implementation is better than Linux' due to it being later and having learned the lessons from Linux.

>SMP which obviously requires NUMA awareness
There's nothing about SMP that "requires" NUMA awareness.