2 x e5-2697v2; 384GB DDR3; TEMP DIR RAMDISK =>slow plotting; one plot more than 12h

Yes, I definitely did! Using numactl and stepping up to 2 instances each with their own 110 GiB ramdisk and dedicated SSD increased my output dramatically. I went from something like 54 plots per day to 72 plots per day. Single threaded MM instance was yeilding me roughly 26 minute plot times, and 2 instances of madmax yields around 30 minutes per plot. I have just ordered my DDR3 1866 for Bladebit. I’ll report what kind of output I’m getting with Bladebit also when it comes in and I have time to test it. Running dual e5 2690 v2 and 256GB of RAM currently.

1 Like

I’m in UK. Here is how much I paid for it in June without NVMe.

Hi,

Yesterday I tried two jobs:

1.) ./chia_plot -n 1 -r 20 -u 256 -t {1st 600GB SAS15k} -2 {110GB RAMDISK} -d {5TB usb external HDD}
1.)with delay 20 minutes ./chia_plot -n 1 -r 20 -u 256 -t {2nd 600GB SAS15k} -2 {110GB RAMDISK} -d {5TB usb external HDD}

VM have 40 threads {8 threads using by hypervisor} 360GB of DDR3 1600MHz

But creation time was terrible, each of them take about 90minutes. With 20 minuts delay second job finished after 110minutes. In that time I can make 4 plots if I will use just one job with 40threads and one temp dir 256GB RAMDISK.

Maybe I did not correctly start parallel jobs.

Thank you

I think to get max performance out of your system you have to ‘split up’ your system in 2 subsystems where cpu A has 192GB of RAM in its dedicated memory slots and cpu B also also has 192 GB of RAM in its dedicated memory slots. Then assign one madmax instance to cpu a and one to cpu b.

Could well be something to set up in bios. What’s the brand and model of your system if I may ask?

Don’t know how to do this exactly but there must be guidance somewhere on the internet/youtube. Others on this forum have talked about this from experience. Maybe you can ask @RobbieL811 directly for some pointers, he is just a few posts up;-)

I don’t know if you can get this working inside a KVM virtual machine though…

Start your second instance as soon as phase 1 completes. Can you share with me why you are doing this from a VM instead of a bare metal instance? I’m not sure how KVM or whatever virtualization platform you’re using is setup and/if it affects how madmax runs. I would advise trying from a bare metal system if possible. Also, I’m not sure if numa can be used in a VM. On my lunch break, I’ll try to hop on my laptop and grab all of the commands for mounting the ramdisks and running mm so that it uses a specific CPU and the ram that goes along with that CPU. On my phone now.

My understanding is you use Fedora 35 as your base OS and ran into a bug preventing the installation of the inbuild madmax plotter for chia v1.2.11.
We talked about installing madmax standalone from its github install instructions for CentOS because Fedora and CentOS share the same upstream distribution RHEL.
Guessing you as a Fedora user feel comfortable with CentOS because of it’s similar cli and tools but not wanting to risc wrecking your Fedora installation I can understand your idea to install CentOS as a VM inside Fedora/KVM and install and use madmax from there.
But that may be in the way of using the system to its potential with dual processors.

What you can do as a step in between is install Fedora 35 as a VM inside Fedora 35/KVM and test if madmax’s CentOS install instructions work out there. And run it once or twice.
Then you know it’s feasible to run madmax on your base OS Fedora 35 and you can try with @RobbieL811 's setup


Can I start a gofund-me Chia page???

10% discount, a nicely boxed wireless keyboard set and on top of all that … free shipping.
Who could resist?

1 Like

We need more than just a couple of stim checks…

Windows is shit fir Ramdisk but shouldnt actually be THAT bad. I was getting around 50 mins IIRC plotting on windows with dual 2650 v1’s. Change to ubuntu is a good start though. I went from 50 mins to 35 mins with that setup just by changing to ubuntu.