I am about to give up this plotting caper and refractor the hardware for something productive. I recently acquired what I thought was a reasonably powerful server for plotting with 2 * 6 cores/12threads / (Dual Xeon Sixcore 2 * L5640 - 2.26Ghz ) an 24gb ram
I have been doing some test runs but I am looking at +12 hours a plot and even with doing 3 plots simultaneously with 30 minute separation but that still took ~16 hours…
I am awaiting a M.2 adapter card to add 2 * 2TB NVMe M.2 SSD but not even sure it is worth it.
Can someone please put me out of my misery… please
This is all before I try to understand what is the best way to plot / pool / madmax etc… It is all feels like it is heading into deep technical territory.
Thanks for completing my otherwise lazy Sunday.
HP Workstation Z600 (KK755EA)
Dual Xeon Sixcore 2 * L5640 - 2.26Ghz (12C/24T)
24GB RAM (6 x 4GB) DDR3 PC3-10600R 1.333Mhz
1 x SSD 1 TB - Samsung 870 SATA
1 x HDD 1 TB SATA 7200rpm WD Black ed.
Nvidia Quadro FX-1800
PCIe USB 3x card
2 * 6GB External 6gb SATA drives
Awaiting **ASUS HYPER M.2 X16 CARD V2 adapter for>
2 * Corsair MP600 PRO Gen4 PCIe x4 NVMe M.2 SSD - TLC NAND
I have a dual X5670 rig (same generation), was also running with 24GB RAM, using the MadMax plotter I can get < 1hr per plot.
- I use an NVMe disk in a riser (gen 3), I’ve not measured it but I assume it’s faster than plotting to a SATA SSD.
- Make sure hyperthreading is enabled, for some reason it wasn’t the default in my BIOS, and enabling this made it faster
- Make sure you’re not getting thermally throttled, I was running in a 4U chassis with passive cooling on the CPUs and they did not like it, moving to a 1U chassis made a big difference because the chassis acts like a duct, and has high power fans at one end - very noisy though.
- Use MadMax rather than the stock plotter, you don’t need a huge ramdisk to get a speed boost, and whilst you wont be able to parallel plot, you’ll likely still be able to get more plots-per-day.
Whilst this setup works fine, it’s not very power efficient per plot, so I don’t run it anymore, but it was possible to go from 12hr plots to <1hr plots with a relatively small amount of effort, and when this was running, it was my cheapest plotter in terms of money spent on parts.
I agree with the previous post. Given your processor, high thread count, and choice of nvme, you should get much better results with madmax plotter over the stock chia software. Get the MP600 in your rig first, as that would help alot.
thanks guys - i will see how the MP600 goes and decide from there.
Let me open another bottle of Rose to beat the heat…
More to follow once I do another test however the low price of Chia worries me.
That’s normal (9-12hrs) for plotting with Chia GUI or CLI. Try madmax.
move to mad max. We started with a very powerful machine like 5800x with 64 gb ram and we were able to get one plot in 6 hours with chia gui.
Move to mad max plotter and the time was reduce to 50 minutes each plot. so I would suggest you to try with mad max before.
Although your hardware seems to be bit low specially in terms of ram mhz speed and processor but if we are considering threads here then it should cover it.
Thanks also - what is your recommendation for RAM?
I will look at madmax but should I be considering a pool? Of course this means deleting previous plots and starting out new but that should not be a problem if the speed is good. Just need to find a good source of well prices drives
Thanks again all - keep the guidance coming
I received the new memory (96gb / 1333) - installed ok
ASUS HYPER M.2 X16 Card V2
2 * MP600 Pro - CORSAIR Gen4 PCIe x4 NVMe SSD 2TB
Once installed only 1 of the 2TB cards is recognised.
I tried repositioning them on the board but always 2 installed and only 1 recognised.
My friend tells me :
so your systems does not support onboard pci bifurbication (splitting the pci express channels per card).
You’ll need a raid card that does that - that’s a € 160+ expense
Please advise as I thought I had a good rig for farming with 4TB of fast SSD - not just 2.
Your friend is correct. The Asus card does requires bifurcation by the MB as the board does not have the necessary switch on board to do that for just any MB. Those type boards are much more expensive.
FYI, honestly you only need one or two good SSDs (1tb is fine) on your system (no need for the fancy Asus board even) and use MadMax with 16-32GB memory and you’ll be plotting faster (almost) than you can store them away. MM speed depends on you SSD and CPU, but will do wonders with minimal hardware.
I have 2x 1tb PCIe 4.0 SSDs & 2x 1tb PCIe 3.0 SSDs and use two MM threads with 20-25 min plot times (2 MMax instances in parallel). The two SSDs/instances are for -d & -2, with only two SSDs for both threads the times expand a bit, but not by much on my 3955RX ThreadRipper Pro.
I’m rather tired, so haven’t read the entire thread, but you now have 128GB ram I believe, so you can use a ram disk and a single NVME drive, or pick up a couple of single NVME PCIe adapters and presuming you have two PCIe 4 lane slots available you can stick one NVME in each slot, then use the OS to mount them as a striped array, and use in conjunction with the ram disk.
Thank guys - very much appreciated.
I only (relative I guess) have 96 Ram
I have 1 x SSD 1 TB - Samsung 870 SATA but this I was using before and did not feel much speed.
the only thing I have not tried yet is MadMax. I will install today and do some tests with the installed MP600 Pro - CORSAIR Gen4 PCIe x4 NVMe SSD 2TB
With regards my previous modest faming of 90 plots - do i delete them, join a pool and farm on with Madmax + pool?
BTW - i do link the Z600 - remarkably quiet machine and well structured.
I might be wrong, but does this Xeon generation only support PCI-E 2.0? Which would result in slow SSD Speed of max 1.5GB/s
This would also explain why a Workstation does not support bifurcation, which was also introduced with PCI-E 3.0
Yes you are correct.
The specs are here https://www.dectrader.com/docs/set06/499679/Z600.html
The above document states 48GB of ram as the maximum (6 X 8GB), perhaps the OP got lucky and 96GB is working, but it doesn’t gain them anything.
This really is a poor choice as a plotting machine, but it will still plot, and if he sets up the two NVME drives as a striped array, and the other SSD’s as another striped array he could use them as the two plotting drives, with Mad Max.
Edited to add, it may be worth setting up all the SSDs in a single striped raid array, this could maximize the storage speed available, he’d need to test to confirm which setup creates plots the quickest though.
Ideally he needs to join a pool and plot NFT plots, I still have OG (original plots) and use Foxy OG as my pool.
I also have a the 3955RX 16 core with 64GB RAM running 1 instance of MM on windows. My plot times are approx 29 mins. Are you running Linux?
No Windows 10, I’m way more familiar with it and seem to be able to accomplish xcellent times w/thoughtful* setup. Got myself one of these two days ago and it’s really sweet & cheap for our TR in slot 1 on the MB > Gen4 AIC Adapter
Best times are 2 instances of MM. I’ve tried two ways as follows: First 3 all Gen4 SSDs… -t was 1tb each separate 1x 980 Pro and SG US70; -2 for both was Aorus Gen4 7000s 2tb > 20.5min/plot. Other way I tried, almost as good, but not quite… -t using two individual Gen3 970 Evo Plus; -2 the above 1tb SSDs separately. 22.5 min/plot.
I think the Gen3 SSDs, although worthy, simply can’t keep up with the all Gen4 setup.
*good times secret sauce is the MM command line, small changes make big plot time difs.