Best use for extra Optane and nvme SSDs

[LucasG]

Hi all, new to the forum, been watching Level 1 Tech for the last couple of months and love the channel

Recently I was super fortunate and and was able to get a couple of free SSDs and other computer gear:

x2 Intel P4800X 750GB Optane SSD
x4 Samsung 970 Pro 2 TB
Also got an old dual 8 core server with eight 4 TB drives installed

I’m trying to get the most out of them and not sure what the best configuration would be

My main system is a 5950x with a 3090, mainly use it to edit videos (Mostly in 8k pro res so pretty large files) and I also use it to game. I also use it for some machine learning, mainly with Python and Scikit learn, but I’d like to get into Tensorflow and more advance stuff in the next couple of months

Currently I have one of the Optanes in the computer and set it up with Primocache with my internal m.2 ssds (won’t cache any footage for editing, that’s on an external drive). Not sure if it would be better to just use it as a boot drive instead.

I’m considering using the four m.2’s in a pcie card to pool them all together and use them to store my most recent video projects (my camera shoots around 1 GB a second, current external drive is 1.3 GB/s bandwith, so it works, but could do with some extra head room)

Then with the extra optane card, not sure if I should use it as a cache/scratch disk for my editing applications (Davinci Resolve/Fusion), or use it in a different computer, I have a plex media server/roon music server, not sure if those applications would really get anything out of an optane card. I’ve been wanting to build a dedicated NAS for all my footage for a while now and not sure if the optane could help with that either.

I know this is a long post so sorry about that, just wondering if anyone has any advice or personal experience with getting the most out of optane

Thanks!

[LiKenun]

The P4800X’s best performance is brought out by few low-queue-depth processes working with small block sizes. Here’s what I’ve benchmarked (random read performance only):

• 2 jobs/32k gets you 77k IOps/26 µs latency for 2.4 GiB/s
• 8 jobs/4k gets you 571k IOps/14 µs latency for 2.3 GiB/s
• 16 jobs/4k gets you 569k IOps/28 µs latency for 2.3 GiB/s
• 4 jobs/4k gets you 387k IOps/10 µs latency for 1.5 GiB/s
• 1 job/32k gets you 46k IOps/226 µs latency for 1.5 GiB/s

And 512-byte reads are going to have crappy bandwidth, but they’ll be low-latency (~12 µs) with anywhere from 1 to 8 jobs hitting the drive.

If your workload happens to fit those random read access patterns, NAND SSDs can’t touch that performance. That’s where your P4800X will shine. Otherwise, it’d be a misallocation of what you have.

Can’t really speak for video editing, but Python and scikit-learn-based machine learning workloads tend to be CPU and RAM-dependent rather than storage dependent. Models are saved from memory to disk or loaded from disk to memory.

Perhaps for video editing, the niche use case is if you seek a lot. But I don’t see the P4800X having an edge over NAND with large files like the 8K ProRes videos you work with.