Not mine, I've got one though. Good cheap SLOG disk.
Intel Optane SSD 900P 280GB (SSDPED1D280GAX1) PCIE 3.0 x4 NVMe Add-in Card | eBay
Intel Optane SSD 900P 280GB (SSDPED1D280GAX1) PCIE 3.0 x4 NVMe Add-in Card | eBay
Intel Optane 900P 280GB HHHL AIC Gen3 x4 $110
- Thread starter ccie4526
- Start date
Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.
Can confirm works great for slog and overkill for plex transcoding volume
Ram Disk would be preferable but I didn't have that much RAM at the time, and was running a bunch of services on my truenas box. I ended up taking this AIC and splitting it into 2 partitions, one that was like 40GB for SLOG (still way too much) and the rest as a plex transcode volume. I ended up adding more ram to my Truenas box (up to128GB now) but never switched to a ram drive. I might end up adding another 128GB or so and then move to a ram drive, as transcoding multiple 4k streams can take a lot of space, but until then this works well
Plex does not need fast storage for transcoding. If on spinning rust and using it as a bitorrent target with many live torrent seeding, it COULD potentially be be a bottleneck, but any ol SSD will not be a bottleneck in ANY situation in performance.
What people fear (at times unfounded) is the life of the SSD with long terms regular writing, similar to why it's used as an SLOG drive. In other words, isolating regular/heavy writing away from regular consumer SSDs unto specialized enterprise level write heavy SSDs.
Putting the transcode directory into a RAM drive is the next step in removing all constant writing to drives and using ephemeral, unlimited writes on RAM, which in many homelabs, is so plentiful, it becomes a "why not" use case.
What people fear (at times unfounded) is the life of the SSD with long terms regular writing, similar to why it's used as an SLOG drive. In other words, isolating regular/heavy writing away from regular consumer SSDs unto specialized enterprise level write heavy SSDs.
Putting the transcode directory into a RAM drive is the next step in removing all constant writing to drives and using ephemeral, unlimited writes on RAM, which in many homelabs, is so plentiful, it becomes a "why not" use case.
what @fishtacos said. fast storage/ssd is overkill. if you're transcoding 40 videos you might need it but spinning rust is fine for 1-2. having faster storage makes things load faster though, so using a ramdrive or fast SSD for transcoding will make things snappier when you're seeking around in videos being transcoded, but its entirely overkill. i used the SSD because i didnt' have enough spare ram at the time but I could do it now, and while overkill, it will make plex slightly faster.
It's always fascinated me how the only reason we have RAM and secondary storage was because back in the early days of computing a flat storage model wouldn't work. And yet with today things 128GB of RAM being relatively cheap, we're still having to use a 'RAM + secondary storage' model versus more of the flat model offered by things like persistent memory modules. You would think in a lot of enterprise use cases a flat model would be superior.
@Samir I think the reason remains the same still. RAM is king in latency and bandwidth and cost for things like Optane DIMMs and licensing is/was expensive.
Intel's Optane in Memory Mode or VMWare's nvme memory tiering are two similar approaches to a flat memory pool with multi-tiered RAM that either exist or are in technical preview. Optane got sold out of Intel some years ago.
if I had a magic 8 ball, I'd ask it "when", not "if" KvM will introduce a similar feature in the future.
A sorta informed guess on my part would be that technology that is useful for regular use cases is either difficult to develop, not technically capable enough, not cheap enough or a combination thereof. For enterprise it's been there for a hot minute with varying levels of satisfation.
Intel's Optane in Memory Mode or VMWare's nvme memory tiering are two similar approaches to a flat memory pool with multi-tiered RAM that either exist or are in technical preview. Optane got sold out of Intel some years ago.
if I had a magic 8 ball, I'd ask it "when", not "if" KvM will introduce a similar feature in the future.
A sorta informed guess on my part would be that technology that is useful for regular use cases is either difficult to develop, not technically capable enough, not cheap enough or a combination thereof. For enterprise it's been there for a hot minute with varying levels of satisfation.