I’m not too sure on the 16GB mode atm, think there were some problems with that in the beta.
Best ask in the discord channel
Hi. Im waiting after plot finished in bb around 2-3 min for copy to finish directory (before creating new plot) But I have 2 HDD. I use 2 finish directory. How to say bb, dont need to wait copy process and start make plot to another hdd (which is not busy)
Right now it a bit stupid - finished plot and then copy to disc A, next plot to disc B and then again disc A
I think it can economy 1-2min because for copying need only reading ssd.
At the moment you can’t (this feature will apparently be added later), you should be using a fast enterprise SSD for a staging drive.
I use ssd. Just dont want to wait copy process from ssd to HDD. It would be good if it start to make new plot on another final directory.
I think that means which discord? 
Hello. Help me troubleshoot.
Bladebit_cuda 128gb utilize only 1 cpu thread / 128 available threads.
See pinned screenshot for more info.
Threadripper 3990x.
Nvidia CMP90HX (lite 3080 for mining without screen ports)
ram 128gb
raid0 x2 nvme ssd samsung 980pro.
ubuntu 22.04.3 LTS.
Madmax plotter is working fine utilizing all cpu threads.
It’s a cuda plotter…it uses the cuda cores on your GPU, no CPU cores needed apart from some memory and copying tasks.
The memory allocation always takes a long time, but you only need to do it once per run.
So make more than 1 plot at a time, otherwise you waste a lot of time
Do you think that 35mins per v7 plot on 128gb mode is not suspicious result for my gpu? It must be 10 times less.
Does nvtop show proper pcie use?
What format is the raid 0 in?
yes, should be much much faster indeed.
Leave the “-t 128” out, shouldn’t matter, but just to rule it it doesn’t cause any odd behaviour.
Other possibilities I can think of:
- something with the raid setup? try running with just a single nvme see how that goes.
- driver update
weird though, it so slow
Probably not worth bothering anyway since it produces obsolete plots
I tbought that since it is a mining card it might have reduced pcie lanes
Will check it later, so far I can say that the device usage is 100%.
Created with mdadm, ext4 file system
- Will try intel optane ssd on Monday. Incorrectly configured raid0 is one of my suspicions about the cause of the problem, but that’s unlikely.
- Also will try borrowed rtx 4090 instead of CMP90HX. So far I’m most inclined to think that the problem is with the GPU.
- Also problem may be with cudaplot code itself since I am also forced to use a second old gpu amd 550 4gb to connect the monitor, because CMP90HX has no video outputs. As a result, 2 devices are connected at the same time, but only one is visible by the cudaplot as a cuda device. Perhaps there is some confusion in the code
Well that seems like a thing. Dump that crap and ssh into the box instead
This GPU using only 4xpcie gen1 during all the process. It is a shame. Never buy a mining GPU for plotting. Going to get refund.
Are you sure its not a BIOS setting, or something else limiting it to gen 1 x 4?
Some slots on motherboard operate at a lower speed, or maybe because all resource are in use, check you motherboard manual.
According to this the card uses a PCI-Express 4.0 x16 interface.
I tried auto and x16 mode for pcie slot in bios. All the same - pcie gen1x4 and does not increase under load. Some search shows that the developers deliberately limited everything. On older versions this was solved by soldering additional components to activate x16 lines. Turns out this is not a problem.
Tested 4090 and produced z7 plot in 2.26 mins on pcie gen4x16.
1 Like
At least you know the problem now. 4090 is not much quicker than my 3080 which takes 159 seconds.