54 TBW for Chia

lihp

Active Member
Jan 2, 2021
174
43
28

at 225,94€ in Europe.
 

lihp

Active Member
Jan 2, 2021
174
43
28
not 54 tbw, but 54 pbw. the difference is 1000 times.

to my opinion you can better invest in 512GB Ram memory than in m2 nvme drive.
I disagree. Right now 50 plots parallel - walk in the park with the right tmp-drives.Unbenannt.jpg
 

Bert

Active Member
Mar 31, 2018
402
158
43
42
I use 2.5" spindles, if you raid 4 of them on a sas connector, you get more than enough iops and bw. If you can parallel process them on a server with 24 of them, that is even better and you can easily saturate 24 cores or more like that. Not the fastest option but faat enough and cheaper options as you can buy each drive from $5.
 

lihp

Active Member
Jan 2, 2021
174
43
28
I use 2.5" spindles, if you raid 4 of them on a sas connector, you get more than enough iops and bw. If you can parallel process them on a server with 24 of them, that is even better and you can easily saturate 24 cores or more like that. Not the fastest option but faat enough and cheaper options as you can buy each drive from $5.
If it works for you - congrats. I am fine with that. Just out of curiosity: how much is your plot TB/day/core?
 

Bert

Active Member
Mar 31, 2018
402
158
43
42
If it works for you - congrats. I am fine with that. Just out of curiosity: how much is your plot TB/day/core?
I was using 2650 v4 so they are not very powerful cores and I was getting 1.6- 1.8 plots a day per core. My dual 2650 v4 system was mostly bottlenecked by the output I/O as I was not able to create more than two output drives.
 

EasyRhino

Active Member
Aug 6, 2019
138
58
28
Interesting these LX3030 don't seem to be available yet.
I wonder if they're merely delayed, or if PNY is backing off the launch/
 

tinfoil3d

QSFP28
May 11, 2020
321
117
43
Japan
use "free -h" command to see the same output, cache is just the filesystem cache currently in RAM, nothing of particular interest to any user
 

RimBlock

Active Member
Sep 18, 2011
824
27
28
Singapore
to my opinion you can better invest in 512GB Ram memory than in m2 nvme drive.
I would say it depends on the volume you are looking to plot.

A decent US$200 --> 300 nvme drive, an adaptor for it if needed and 128GB ram can give a plot every 34 mins or quicker with old server hardware. For 100TB written not one NVME failure yet although. I use MadMax with tmp1 as the NVME.

If you are likely to get through a lot of NVME drives due to the amount of plots you are creating then 512 GB ram is a good option but that size of ram is pretty expensive. If Chia was still at US$1.5k pre coin then sure, but now with the current pricing.....

Small SAS drives are cheap but take more power, generate more heat, require SAS connectivity (adaptor, cables etc that home users may not have) and tend to be slower unless runing large arrays. I have 24x 10K drives and put some timings in the Chia Farm thread when MadMax came out.

I also used 4x HGST 400GB SAS SSDs raid 0 (cost around US$250 for all 4) for a while which were good speed wise but needed the drive bays for storage drives so went back to NVME on a PCIe adapter in the end.

I prefer to go with used enterprise drives as they are generally well tested and have a fair amount of knowledge floating around the internet by the time they reach ebay. Of course it is a bit pot luck on if you get a good one (lightly used) or not. I seem to have been lucky on all but a 6TB drive so far. New products, first gen, tend to have some kinks to work out especially if they are rushed to market.

If you do go for an NVME drive, check if they are using a small DRAM cache as you will get good speeds for standard use most of the time but for write intensive use like plotting, the cache will run out fast and then you drop speed quite badly. Hunt for spec sheets, try and find reviews etc and you should be ok.
 
  • Like
Reactions: Patrick