Beginner Rig Setup : Feedback

HP Z840 Workstation 44Cores/88Threads, 2x Xeon E5-2699CV4, 128GB RAM, HP Zdrive

Processors: Dual, 2x Intel Xeon E5-2699CV4 (22C/44T, 2.2/2.4GHz)
Memory: 128GB RAM DDR4 ECC
Graphics: NO GPU (Add own GPU)
Storage: 512GB HP Z Turbo drive PCIe M.2
4x Caddy Trays for additional HDDs
Opitical Drive: Super Slimline DVDRW+/-
Power Supply: HP 1125W 90% High Efficient PSU
Integrated: SAS/SATA RAID On-Board 0,1,5 & 10
Integrated: Dual Gigabit Ethernet
Integrated: Audio
OS: Windows 10 Pro

Corsair MP400 4TB NVMe PCIe M.2 SSD

28TB storage : WD 14 TB Elements Desktop External Hard Drive - USB 3.0, Black. Soon to be 280TB.

I’m planning to run 16 plots at a time due to the 4TB SSD. I may add another 4TB in time.

As I am new to this, any further advice on the above? Anything I have missed?

Check out the madmax plotter.

This may work better than parallel plotting on your system

1 Like

Agreed, with madmax you could do without the nvme’s. Just use the turbo drive as temp1 and 110GB ramdisk as temp2. Not sure how good that turbodrive is, but in any case with the new plotter you only need 256 total temp space

OK, I am running 14 plots in parallel using GUI (still learning).

I have 4TB NVMe SSD and it is 100% maxed out with average response time of 2623ms and 164MB/s write speed.

CPU is 4% and RAM is 18%. SSD is bottleneck. Am I right to be worried about the SSD performance and should I drop the number of plots in parallel?

Looks like your Corsair has bottlenecked the whole affair. When u see the SSD is 100% utilized and has long response times, it’s being overloaded with requests. So it’s not the size that’s the problem, but the nvme controller and multiple requests.

Sorry to tell you if you’d chosen 4x 1tb Corsairs or whatever, you’d be better off, as u would have 4 nvme controllers working at only 4 jobs each, not 1x nvme controller working at 16 jobs. With what u have I’d try 8 jobs (maybe less), see how it goes. May find completion times are high enough to get 16 plots as before, just split up 8+8 or 6+6+6, etc… Try various combos until you Corsair doesn’t cry for mercy as shown my response times and faster output. Or send it back and get other (nvme) drives, smaller size.

I will continue my testing. I have tried 1, 4, 8 and 14 in parallel and response times and throughput are similar.

What plot parameters do you use? With 16c/32t Ryzen CPU, 128GB, but using 5x1tb nvmes, I easily ran 15-16 plots at the exact same time, 6min delay. 4098 memory 6-8 threads each. For all to complete, the session took about 11.5-13 hrs.

W/ur two CPUs and 44 cores & 110MB cache between them, you should be doing great, quite the mystery. GUI only, BTW.

I think I have the answer : QLC cache. Unsuitable for high throughput workloads.

The SSD will be returned and replaced with two of these:-

Samsung 970 EVO Plus V-NAND M.2 2TB SSD

still curious why you won’t try madmax plotter.

Make a 110GB ramdisk as temp2 and use one nvme as temp1. With the rig you have that should do very well, save 75% of the wear on the nvme and also much easier to work with.

On my 2x E5-2680v2 I get 50+ plots per day that way

I have 2 x 1TB Evo Plus on the TR. They work great, good choice.

I am sending back the Corsair today so will try MadMax before I order the Samsungs. Of course, it will be first time using MadMax and I will have to follow the tutorial above.

Switching from 2Evo Plus 1TB to 1Intel P4800X 375GB made my plotting in MadMAx over 600sec faster per plot.

created 110GB RAMdisk.

Running Mad Max.

-t = 2TB NVMe SSD.
-2 = 110GB RAMdisk.
-d = 28TB External HDD

I’ll post up time when it finishes.

Why would you not go for a Threadripper (Pro) but a Xeon?

C:\Users\Z840\Desktop\MadMax>chia_plot.exe -n 1 -r 32 -u 256 -t F:\ -2 R:\ -d G:\ -p **** -f ****

CPU 30-40%
MEM 95-98%
SSDs are fine, slightly peaky.

Phase 1 took 996s
Phase 2 took 410s
Phase 3 took 671s
Phase 4 took 207s
Total : 2282s

@ChiaMax MadMax is working well. Any further adjustments I should make to increase performance? Many thanks.

@Voodoo thanks for all your help above.
@Chia.Switzerland this is the kit I have now. No more changes!!

110GB RAMDisk
-u = 256 and -v = 128. I have tried different combinations.
Cores are 22 but threads at 32 provide best results.

Averaging 2600s/43m/33plotsPD

chia_plot.exe -n 1 -r 32 -u 256 -v 128 -t F:\ -2 R:\ -d G:\ -p **** -f ****

I don’t understand why your plotting using ramdisk is 700sec slower than mine using a single nvme on a single processor 14 Core System, there must be a huge bottleneck somewhere.

you are using the ramdisk as temp2 (-2) right?

You should be much faster. -r is threads, you should try 80