Cache-Drive: Block-Size recommendation?

Hi there!

I´m on DR17.2 and i´m going to install a PCIe-based RAID into my MacPro (2019).

Setup:
• Mac Pro 2019 , 28 Core, 256 GB RAM, Dual Radeon VEGA II 64GB, RADEON RX6900XT 16GB
• SONNET m2. 4x4 PCIe
• 4 x Samsung EVO 970+ 1 TB
configured as a RAID 0.
This Volume is ONLY used as a Cache-Drive for DR (Render-Cache Format: ProRes422HQ)

Now I'm wondering what's the best setting for Block Size when initializing the RAID-Volume?
I´ll use SOFTRAID to create and handle the Volume.

Any recommendations?

THAAAANKS!!

If you work with big files (eg. 1GB smallest) then higher should be better. For hardware RAID I set 1MB or more.
Issue is that smallest taken disk space will be at least 1 block (at least this is how I understand it). If you have 1MB block size and 1000x 1KB text files then they will take 1000x 1MB on the disk (wasting some space). In todays disk and file sizes this is small issue though.
Another point is that if you have 2MB file in size and want to write it to 4 disks in RAID 0 then with block size 1MB you will be able to write only to 2 disk at the same time, not 4. This will limit speed. With typical video files sizes this agains should be not and issues.

I’d happily waste some space if I therefore get the best performance from the drive.
I should have been more precise; I’m looking for the best performance for this cache-drive.

and I’m not really sure if we are talking about big or small files.
When examining the cache-folder i see some subfolders with a lot (!) of small files ( some kb in size)
and some folders with larger ProRes Files ...
So I’m not sure where to go AND if the Block Size even does matter in this scenario...

Your video cache files definitely won't be small
Bigger block size = better read speed (at least in theory).

4x SSD should give you about 2GB/sec so this is plenty for ProRes. Border line should be 10bit DPX cache for UHD at 60fps.

Best to contact the programmer of the sonnet card but probably 64k. Bigger hasn't been better for years, and can block the oh i dunno i stopped listening when it was explained to me properly. The default for the card that are attached to. It's academic anyway as the bottleneck will be elsewhere in that Cupertino bucket anyway.