Actually 3060 is the card name:
-g or <gpu_name> GPU to use for mining, can be a number or any part of a name or any part of a serial number, optionally includes memory limit:
-g,5G 2080 -g,8G 3070 -g 4090
Multiple mining GPU can be specified.
So you should use -g,3G 3060
Instructions are below.
It worked with -g,3G 3060
But the plot creation time is 1 hour
Apparently this is due to pci-e x1 (connected via a riser)
Something really wrong I suspect. I have a ThinkStation P620 16 core 3955 with 320GB of ram and an RTX 3090. If I output to my 4tb SSD raid AOC (4x1tb) from a single instance I get a 57 second plot. If I run two instances I get 2 plots every 65 seconds. But I do neither, I skip the ssd altogether and I output directly to two 90TB raid boxes (5x18) and get 2 plots onto those HDDs every 240 seconds. I can also farm from those raid boxes without affecting the plot times by more than 5 or 6 seconds. Its faster to have nossd write directly to raid O hdd’s than copy from a very fast raid 0 SSD. My SSD AOC is 15k read/write.
This is the same base system I have, so the faster speed is likely due to the 8 channel system RAM.
No idea why you are getting faster speeds writing to raid0 HDDs vs raid0 SSDs, but this is likely due to the specs of the SSDs I would suspect. You would need a constant ~1GB/s read rate to be clearing the SSDs with how fast plots are created, which means an average of ~1GB/s write rate as well to the SSDs. I’ve found that in reality this is more like .8GB/s to 3GB/s write speeds starting after the 66% point. So the sustained write speeds of the drives becomes the limiting factor at that point.
A pcie x1 connection will be sufficient for harvesting, but not plotting. Plotting works fastest with the full x16 pcie bandwidth.
I would also recommend reading through the link above for minimum and recommended specs. Specifically:
GPU memory required for mining
Compression Level C10 C11 C12 C13 C14 C15 Min GPU memory (GiB) 0.13 0.25 0.84 1.31 2.30 4.49 Recommended GPU memory (GiB) 1.02 1.03 1.54 2.59 4.20 6.82
And
GPU plotter additionally requires from 4.26 (minimum) to 7.08GiB (recommended) video memory. “-x” option can be used to select exact GPU for plotting and, optionally, to set video memory limit for the plotter.
The 8GB of VRAM you have is not enough to simultaneously harvest and plot C15 plots.
Writing to my SSD raid 0 is much much faster than writing to the hdd raid. What I am saying is , due to efficiencies in NoSSD, writing from memory directly to the HDD raid, is faster than writing to the SSD raid and then copying it from the SSD raid to a hard drive.
I’ve found the exact opposite from my experience, which is why I suggested the 1GB/s read and write rate possibly being a factor. If there is significant slowdown on your SSD array only while trying to read as well as write, there is likely some other issue with your system. My guess would be the sustained write speed of the SSDs is to blame, but could be a number of things. I highly doubt it has to do with the software itself, however.
Good for you, I find differently. and sustained write speed has noting to do with moving plots off an ssd raid array when nothing is being written to it.
If what you are talking about is plotting to an SSD, stopping plotting, then reading off the SSD onto the final HDDs, then yes this would be much slower.
What I am describing is using a separate program to copy finished plots off the SSD destination directory into their final HDD drives while continuing to plot to the SSD, I can only assume this is what you attempted before finding writing directly to the HDD arrays was faster for you. To sustain the kinds of speeds of plotting in this way, you need to write simultaneously to about 5-12 HDDs, varying depending on how fast, full, or fragmented the drives are.
The only other thing that comes to mind is that too many simultaneous plot writings to the HDD arrays is too much for them. This wouldn’t really come down to the write speed of the HDD arrays per say, but more how much the drives have to jump around writing multiple different files at once. You would likely be able to sustain the faster plot times writing to the SSDs and copying off to many individual HDDs rather than an array of them, but its not like its worth the time to check by replotting.
My wallet address:xch1djfau3cvqku7q2usv84d6epugnjx9l9xy2kldy9h2wyfgzfkhyrsm8faaq
Does anyone know why only one of my 6 machines is displayed here? Then wait for a while and refresh it to display several other units individually.
I assume so since I am seeing it as well across my fleet.
Yes, it is completely normal.
@Anthony, any plan to try to take “profit” of CHIP-22?
Since I guess your plot solution actually requires that we farm to your “specific” account, do you see any advantage to develop & permit people to farm “NoSSD plots” with their own CNI node, or do you prefer to control the whole infrastructure?
This is not a “test if you are a scam” question : TBH, seeing the shitstrom you got since months, I could totally understand if you decide to keep the centralized approach that has been working flawlessly for many of us, for months.
I believe you are quoting the TDP of these GPUs and not the actual usage. Note that while the TDP of the Nvidia 4000 GPUs is higher than the previous 3000 generation, during normal usage the 4000 series uses a lot less power than the 3000 series.
During plotting using the Bladebit Cuda plotter with 256 GB of host ram and an Nvidia 4070 GPU, I see the GPU reports about 50W of power consumption with nvidia-smi. It is able to get a k32 plot out every 90 seconds at c7 compression, which is maybe 10-20 times faster than the CPU / SSD plotters out there - and even if you used 512GB and the CPU/RAM plotter, it’s way slower and the CPU will be using > 150W during that time.
So the conclusion of “using less power” seems well supported to me.
I can confirm this based on my usage below:
My 3090 plotters sit around 150-225W when plotting, my 4080 plotters sit between 50-150W when plotting (single plot runs per card).
New to Linux. Finally got nossd up and running by making 16 directories, and mounting a drive to each directory. It was up and going smooth. Restarted the computer, bam, everything was not mounted anymore. Any ideas? They are all external USB drives.
Can anyone help me find the problem?
You have copied the settings and cache folder among instances, they all probably has identical workerid file. Remove those files and restart clients.
