Windows vs Linux vs Mac: Performance Showdown

What is the FASTEST operating system available today? How much does running under Docker impact perf? What about WSL2 vs native? Dave tests Windows, Mac, and Linux to find out! Using a rigorous set of comprehensive benchmarks, he presents the data on each Operating System as run on the 16-core Ryzen 5950X, 32-core Threadripper 3970X, and 20-core Mac Ultra Studio.

For information on my book, "Secrets of the Autistic Millionaire":

Geekbench Docker: docker run -it --rm davidsarkany/geekbench && docker rmi davidsarkany/geekbench

My other channel, join now so you're there for episode 01 of my AudioBook!

Primary Equipment (Amazon Affiliate Links):

[DavesGarage]

Join me at 6AM for the Live Premiere! I set it early here (6AM in Seattle) so that it's accessible to European and (hopefully) even Australian folks.

[remrevo]

3:31 You can actually also use the -z flag from tar to decompress using gzip.
So instead of using
gunzip <File>
tar xvf <File>
You can just use
tar zxvf <File>

[GiltogCrasherDark]

Dude. I know you don't know me whatsoever, and I wholly don't expect this comment to reach your radar in a meaningful way, but in the event it might, I just really want you to know how impressive and inspiring your efforts are with this channel, both for the quality/aspects of the content, and for the sheer personal growth that I and thousands of others have witnessed over your time in developing this channel into what it is today. This isn't just a great video -- it's another marker in the genuinely fun ride of what your channel is and has been, and it reminds me so much of what YouTube used to be like, back when small communities dotted the site, and what the little tidbits of humanity that poked through the content of videos and their resulting comment sections were like. It's truly a treat to be an IT professional who can relate to the subject matter of your videos, *while* appreciating the other subtleties of your process and approach to this channel, and although I'm far from an expert on these sorts of things, I gotta say: I look forward to your newest content a lot more than I do the titans of this platform which typically dominate my feed, and I have a feeling I'm not the only one who has this perspective on the matter.

You have an excellent thing going for you, here, and I feel as though I've gained decades' worth of insight and practical knowledge just from your openness to share your own, for which I'm very grateful. I wish you the best in all your endeavors, not as a platitude, but as a fellow human being who doesn't normally take the time to write these sorts of things, and can't possibly fit the full extent of my earnestness into a simple post's worth of reading. You absolutely rock, Dave, and I look forward to learning more from you, laughing with you as a result of the stories from your good ol' days, and being present in the future to see where the channel goes from here. Sincerely: take care.

[malectric]

Hi Dave, just want to say how much I'm enjoying the rabbit hole you've drawn me into. Thanks for your work and channel!

[tmhchacham]

Nice test, Dave, I really enjoy your videos, and the comments abut what should matter along with completion notes. A true programmer.

[CoryAlbrecht]

@Dave you don't have to gunzip separately. Just 'tar -xvf' will automatically handle .gz, .bz2, .xz and more.

[dimasveliz]

I love these videos! For real, how awesome is having sth different here in YouTube.

[isoEH]

Appreciate the tons of work to present this concise shootout report. Cool! Wouldn't mind a benchmark suite vs. benchmark suite shootout. Maybe your patrons could support that? You know, just to keep those benchmark guys on their toes.

[purposepowerlove]

Docker isn't fully virtualized. That is what makes it different from a virtual machine. If it was fully virtualized, then it would be a virtual machine. Instead, it uses the host machine's kernel. Windows always gets significantly worse docker performance running Linux containers because the translation layer has more work to do. Try running a Linux docker container from a Linux (or, macOS) machine and I bet you'd see much better results. Docker is usually significantly faster than virtual machines.

[sammyfromsydney]

That's very interesting and it's reassuring that those results are as close as they are for similar hardware. In my experience What OS you use largely depends on what software you're using and which it is supported on. It doesn't matter how fast a system is if it won't run your tool.

[AGunleashed]

Cool video!, I love the raw data comparison this fantastic!!

[frankposterello]

Really interesting comparison - thanks for the tests! Would be interesting to see some traditional VM results too - but so many settings and options in those it would be tough to organize.

[kylee]

Very cool, this must have taken a lot of time! Excellent and educational!

[LMacNeill]

Honestly, this just tells me how much more optimized Windows has become over the years. Back in the late '90s / early 2000s, you *ALWAYS* ran Linux (or another flavor of Unix) if you wanted the absolute max power out of your machine. It was much closer to the hardware, much lighter, and would usually net you an easy 10% (if not 15%) gain over any version of Windows. A very noticeable difference.

It's nice to see that the software has finally caught up to the hardware -- results like these speak for themselves. Unless a very small percentage difference is that important to your project, just pick whatever OS with which you're most familiar and get to work.

[randomtuberhandle]

How have I not found this channel sooner? Ive watched a few of your videos now. Great work Dave. YouTube's matching algorithm skipped me way too long here. (YouTube, there's a flaw in your matching algorithm)

[houdwarz]

Interesting numbers ! To add to that, it would be nice to see how much further Linux could pull ahead with a kernel and Geekbench compiled for your recent CPU instruction sets.

[kmg]

Thanks for doing this, not that it really matters to me but it is good to have a sense of what the overall field looks like.

[gunnarthorburn]

It seems to me Geekbench mostly tests CPU/Memory. Another relevant thing for real world performance is I/O: reading and writing real files to a file system. I think Linux would come out more favourable for such tests. Also, what often happens in the real world is that many processes are created and then dies (when running scripts). This is something Linux is really good at, and Linux servers/systems are often built to work this way. In Windows on the other hand creating processes is expensive, thus Windows software is often larger processes with many threads.

I think the 2% difference may very well be because of Windows desktop overhead. Some processes in Windows steal a little CPU time, and pollute the cache for Geekbench while doing so.

Would be interesting to know if the Docker difference would be less on a machine with less RAM (so geekbench in docker has the same amount).

[archpenguin]

Thanks Dave! great video. Very informative.

[nielsbosmobile]

Thank you Dave, great video again! Cool to see how well Wsl2 holds up here! With the Apple Silicon being ARM, it might also be interesting to look at this from a performance-per-Watt perspective, with Linux on ARM as part of that test well. For laptops, servers, clusters and super-computers, this could give a whole new perspective on performance for those type of use-cases.