Actually, no need to go that fast. With 1Gbps network (wired) I was getting ~125MB/s transfer speed.
Upgrading to 2.5 Gbps maxed out the xfer to USB 3.0 disk to about 250 MB/s. No need to spend to get to 10Gb, when 2.5Gb adapters are super cheap and run on any old CAT5e cable. Only exception if you want to fill two drives at the same time, you’ll be back to ~125MB/s for each.
3 Likes
This is phenomenal advice ! I didn’t even know there was a 2.5Gbps. I’m definitely grabbing some of those for the home network. Thank you !
2 Likes
I invested in 3x 2.5 Gb switches and adapters. Thanks again for the tip ! I’ve been avoiding this as I thought a major require was required. This will make a huge difference for my setup.
No problem. Sometimes good things are easy and useful.
Has anybody derived major performance gains by playing around with -k parameter?
i have same processor and have same 32g ram use ssd 3000/3000 speed and make plot for 2h one where is problem
I ended up adding 64 GB of RAM. 4 - 16 GB modules to match my other 16 GB ones. Two modules per each CPU to balance them. Set RAM disk to 260 GB for both temp and temp2 directories. Now my Dell Poweredge R820 creates plots in memory. It takes between 68 and 72 minutes to create a plot with MadMax, and it is very stable. I use visual interface of MadMax to create 147 plots and usb 3 card to transfers plots to Exos 16TB hard drives. BTW, I found my first block yesterday with spacepool.
Hey thanks again for the advice. I not have two plotters running around the clock feeding one common farmer across a 2.5GBps network. Can transfer a single plot in 7 mins, which means no network bottleneck. Overall investment was only $600 to serve my chia and home Plex/unraid and office backup needs. Money well spent !
1 Like
Hello,
With this setup:
Ryzen 5 3600
DDR4 16GB 3200MHz (OC the memory to 3600)
M2 500 Go 970 EVO Plus
SSD 1T with Windows 10
AsRock B550 gaming 4/ac
(running Chia farm on the same machine)
Using Madmax (NTF plots): -r 10 -u 256 -v 128
Config1:
M2 500 Go 970 EVO Plus as tmp1 & tmp2
=> Plot time: 75 min
Config2:
Then I put another NMVE dive (WD Blue SN550 1 TB, on a pcie slot):
One drive on tmp1 & other on tmp2
=> plotting time: 60 min (achieved after a pc restart for 1 plot but never reached again for some reason)
Config3:
Enabled raid0 and put the 2 drives in raid0:
=> plotting time: 81 min (not amazing)
Config4:
I have then used again the 2 drives for tmp1 & tmp2.
/!\ I forgot to remove raid0 settings in bios
=> plotting time: 82+ min (82 min best)
Then I disabled Raid0 settings and used Config2 (2 drives) and I’m getting 75 min a plot…
It’s curious that crystaldiskinfo is not showing the two NVME drives anymore…
Questions:
- Can anyone explain why I cannot get the previous performance with tmp1 and tmp2 in sperate drives after disabling raid0?
- (not so related to the issue) I’m thinking of installing 110 GB of ramdisk for tmp2. Comparing extremely fast and low latency SSDs with RAM, they tend to be cheaper per GB than RAM but have a shorter lifetime (not so important for high end models).
Can anyone guess what would be the impact on the plotting time?
I think it would be considerable since I’m using low end SDDs but not sure if I would better invest on a better CPU
Thanks!
Just changed my 3600 by a 3900x
just after changing the CPU the plotting times were roughly the same (72 min) and the CPU was not reaching more than 60% which made me suspicious about the health of my NVME…
So I changed it by a 980 PRO
The result is now 55 min per plot
Not amazing looking at other guys here
the CPU is running at 4 GHz and memory is OC to 3600 MHz (from 3200; 16GB of memory)
I’m using the same 980 pro as tmp1 and tmp2
I’m using swar with madmax to fill different drives
I’m also running full node on the same machine
any tips to improve my performance?
The speed boost would be the RAM disk option, that is only available with 128 GB Ram and above. The speed of the RAM is not as important. I´ve got a 5950x (16 cores) and 128 GB RAM with just 3200 MHz, Ramdisk is the second temp, a Sabrent Rocket 4 (PCI 4.0) SSD the first temp for Madmax. With that, it takes 28 Minutes per plot. Turning off the Ramdisk and just using the SSD, i drop massively in speed.
1 Like
that is soooooo tempting
I’m a bit in doubt since some people say that the difference is not so amazing
besides, I would have to put 500 $ more to the machine and I have only 60TB left to finish my farm (I could put 24 TB and some patience instead of the RAM)
I’m plotting 38 a day so 2 weeks left to be out of space. So I’ll skip the RAM for now.
26 plots/day (above specs) + 12 with with my 10 year old i72600k using an overused and about to burn out WD blue
I would keep it that way. I´m at 48-50 plots per day, so not to far away. At the end you want to be cost effective. I happened to have that machine due to 3d artist work, otherwise, i had gone much cheaper.
1 Like
I did another test for fun: i changed the first temp from my PCI Gen 4 SSD to a USB connected SSD. In essence, this brought down the datatransfer between my ramdisk and the first temp to 400 MB/sec. Clearly, the processor was idle a lot more then before.
The result: 48 min (USB SSD & ramdisk) compared to 29 min (PCI Gen4 ssd & ramdisk). So again, the CPU wouldn´t help here. At the end, the perfect thing would be a 256 GB ramdisk, but that would enforce a threadripper setup (5950x is maxed out at 128 MB) or a similar Intel Setup. PCI Gen 4 comes close enough, maybe a raid-0 SSD setup on PCI Gen 4 would increase things a bit more.
Finally, all this would cost a lot compared to the percentage of win it gives 
Exactly that is the key thing
Just in case you aren’t aware, like I wasn’t, enabling CPU affinity while plotting improves plotting and farming:
I was getting stale partials while plotting and farming on the same machine. When I used 22 threads out of my 24, the plotting speed went from 55-59 min/plot to 48-49 and I got no stale partials anymore 
1 Like
Also note that stale partials are for pools, if you solo mine you can go with higher response time and not lose profitability