A lot of advice has been going around what consumer SSD/NVMe drives to use for your plotting setup, I’m here to tell you that, in my opinion, most of it has been wrong. People tend to be very focused on the TBW figures while in reality, the don’t differ as much between drives as the numbers might have lead you to believe. I’ll explain it all in my video!
I’m not sure I totally understand if I’m gettting the most out of my NVMEs or not, feeling kinda bummed right now and overwhelmed.
Using plotman on linux and newb to chia and linux, and I just feel like the machine is going slow but not sure how to determine if im getting what i paid for or not.
The difference is that userbenchmark measures peak write performance over a short duration, those numbers usually get close to advertised speeds.
The graph from Tomshardware measure sustained write speed over a longer period and that is the important speed to look at for chia plotting, as you are basically writing 24/7 so the burst/peak speed doesn’t help much
Sustained write speeds are not advertised or even mentioned in spec sheets of consumer products, because under normal use conditions, you don’t need that so much
I really like your argument that there are no “magical” nand-chips that somehow perform and outlast the others.
However, that does make me wonder, how do those enterprise drives do it then? The have both high write speeds and super long endurance.
Also it seems like all the high TBW consumer nvme’s are slow for plotting, and the fast plotters have low TBW.
Can it be that this has to do with the way they are optimized? e.g. they are slowed down, but this benefits their endurance?
Yeah, they are using double or triple the amount of chips for the endurance, since in an Enterprise product they can charge for it. And then they can also integrate en 8 channel (or more) controller, thus keeping speeds up.
Regarding the TBW to be lower on high performance drives, I believe this is artificially done so that users don’t write through their TBW values in the warranty period. If you give a drive 3600TBW but it can only do 500MB/sec, unlikely the customer is going to hit that in 5 years. If a drive can do 2000MB/sec well then they might so let’s give them a lower number to start with.
Currently my numbers are still holding strong and both the 600TBW are still telling me they’ll do at lest 2000TB actually written, so we’ll see!
Another weird thing is that there doesn’t seem to be a clear relation between:
a plot writes for example 150 MB/S
Your drive has 900MB/s speed, so you can run 6 plots. That would be easy, but does’t seem so clear cut as that. The write speeds that you see are often well below the max of the drive, but a faster drive is still faster?!
Does this have something to do with writing from multiple processes and/or to multiple files.
I just don’t know nearly enough about ssd tech to understand why there doesn’t seem to be a clear formula to calculate how much speed your ssd needs to do X plots.