As mentioned in plotting stuck at 100%:
I’ve heard this apocryphally, but has anyone recently benchmarked it on the same exact hardware? Also trying to figure out why that would be… is the chia.exe running through any kind of emulation layer on Windows or something? 
Also, no penalty on OSX plotting, either, maybe because it is Unix underneath, somehow?
The explanation I was given is that Visual Studio doesn’t support the 128-bit integer value for the Windows version of the plotter. Only Linux based systems do.
But there are some optimizations you can do on Windows:
Interesting! Could you recall more details?
It was the case for a long time, but with most recent VS no longer is! Python on the other hand… I don’t know about, but looking for 128-bit problems commented on in Chia repo I found just this in the changelog, nothing in issues or the source
Assistance would be more than appreciated if you have experience building binary python wheels for Windows. We are pushing some limits like uint-128, avx-2, avx-512, and AES-NI so it’s not as easy as it looks…
XFS file system makes the biggest difference
Are there benchmarks backing this up, or just claims?
I have no hard fast proof just the Keybase forums say XFS is faster
Yeah I’d love to see apples-to-apples benchmark here on the same machine if anyone has the inclination! Just run a single plain default plot params under vanilla Windows, then run a single plain default plot params under vanilla Linux / Ubuntu or what have you.
I’ve heard from storage_jm that btrfs is the fastest when comparing file systems.
Yeah but by how much? This could be like increasing memory for plotting, such an inconsequential 1 or 2 percent it’s hardly worth doing… we just don’t know without the data & science 
Still waiting for someone to run this test on the same hardware … I would but man, Linux is just not my thing. I’m too lazy for that much command line machination
I feel like you’re giving me a hint. It wouldn’t be too hard to pull off since I already have a load of windows test data. Probably the best would be a test of 8 plotters in parallel.
Well no! I think anyone could do it, if they had the spare hardware. I did my testing on the Surface Pro 7 with the various params (more memory, more threads, etc)… I’m just not very good with the Linuxes, to be honest.
Yes. I moved from windows to Ubuntu and went from 2-2.5TB/day to 3-3.5TB/day. Use Plotman.
Are you attributing your significant increase due to running on Linux or utilizing Plotman? That seems like a massive increase just from the OS.
Yeah I’m gonna say … I don’t believe that.
I want to see a single plot on the same machine, with vanilla installs of Windows 10 and, say, latest Ubuntu.
@storage_jm mentioned doing some filesystem benchmarking in yesterday’s Chia Decentral video (I thought the comments about btrfs were interesting): https://youtu.be/n-QrzdXEtzE?t=2523
Also I’m not sure how well this will integrate with Chia plotting or farming but it sounds like Windows might be getting support for one or more Linux filesystems via WSL (the feature’s available in insider builds but I’ve had some issues with those in the past so I’m not going to try it just yet).
I’ve tried WSL for plotting…Definitely slower than native windows. Surprisingly good for TimeLords though.
I imagine there’s of course a performance cost to using WSL but perhaps if WSL is able to act as an interface between the Windows Chia plotter and say a btrfs, xfs, or ext4 filesystem we’d see a net gain in performance over the current Windows performance? Of course I don’t imagine that such a setup would be better than plotting natively on a (non-WSL) Linux system, but it might at least give folks another option.
Sounds like a good test! Would you like to try it?