cross-posted from: https://kbin.social/m/linux/t/646160
With currently reviewing the HP Z6 G5 A workstation powered by the new 96-core AMD Ryzen Threadripper PRO 7995WX Zen 4 processor, one of the areas I was curious about was how well HP’s tuned Microsoft Windows 11 compares to that of Linux.
While this is cool, but I am interested in a comparison with a fresh windows install. This article says it’s out of the box from HP, I wouldn’t be surprised if they have some dumb processes running, chunking performance… I’m confident linux would still outperform but this is quite an insane gap on display.
That’s a fair comment. But on the other hand, if you are spending a fortune on a CPU the size of your hand (look at that thing in the article!) then there’s a good chance you’re using it for business purposes, and either you or your IT department will be very keen to have a completely vender-supported stack. Enthusiasts with fresh OS installs will not be representative of users of this tech - AMD haven’t really been targetting it at gamer desktops.
Of course, comparing both would be even better, see whether it is an HP crapware issue…
Don’t most businesses cut the bloat out and put their own builds on it? Sure they put their own software on that will hurt performance but it seems fresh vs fresh would be give better metrics.
Totally agree, it’s two different tests and use cases. Most people will run it how it comes out of the box and that’s probably more representative of the real world.
I just think it’s not entirely fair to say “windows is 20% slower” when we have no idea which trash HP loaded it up with. If I managed an IT Dept and learned my $$$$ hardware lost 1/5 of it’s performance I’d certainly be pushing HP for solutions. Or maybe they’d prefer to take 20% off the price?
20% is a LOT. That’s probably because of the random shit that nobody ever asked for but windows is always doing in the background anyway. Building a search index, windows update (which consumes an insane amount of CPU for a completely unreasonable amount of time sometimes), other individual updater services (because there can’t be one program that updates everything because every vendor does their own proprietary bullshit to handle updates), compressing and sending all you personal data to microsoft and of course the pre-installed McAffee (on trial license) that works hard to make your system less secure (that HP probably installed for you because apperently you haven’t paid enough money for the computer, so you must pay with your patience and your privacy as well). Depending on the benchmark, the pathetic legacy file system windows uses might also play a role.
No, it’s because the windows scheduler literally cannot handle that many cores. it simply does not know how to allocate work effectively.
The Windows scheduler is so stupid chip manufacturers manipulate the BIOS/ACPI tables to force it to make better decisions (particularly with SMT) rather than wait on MS to fix it.
Linux just shrugs, figures out the thread topology anyway and makes the right decisions regardless.
This is the best summary I could come up with:
Going back to the original AMD Ryzen Threadripper processors, Linux has long possessed a performance lead over Microsoft Windows.
With Linux typically being the dominant OS of HPC systems and other large core count servers, the Linux kernel scheduler has coped better than various flavors of Windows when dealing with high core count processors.
Ubuntu 23.10 was run for providing a clean, out-of-the-box look at this common desktop/workstation Linux distribution.
The HP Z6 G5 A for all testing was configured with the Ryzen Threadripper PRO 7995WX at default frequencies, 8 x 16GB DDR5-5200 Hynix RDIMMs, Samsung MZVL21T0HCLR-00BH1 NVMe SSD, NVIDIA GeForce RTX A4000 16GB graphics.
A full review on the HP Z6 G5 A Threadripper workstation will be published in a separate article on Phoronix in early December.
From there the up-to-date Windows 11 Pro Build 22631 (H2’23) was tested against Ubuntu 23.10 with its stable release updates.
The original article contains 436 words, the summary contains 148 words. Saved 66%. I’m a bot and I’m open source!
and makes it 100% slower with Snaps.
