Try my string, just to test. But change your drive letters etc…
What does that mean? cuda_plot_k32_v3**-2**
Try -S 3, 4 but not 8. I know u have more VRAM, but 8 is not a good choice IME.
However, I don’t have x2 GPU to test… maybe it needs 3+3 = 6 or like that??
Max released another version of the cuda_plot_k32_v3.exe
and when you run
It shows you 3.0.2-979b1a0
1 Like
I’ve tried -S 4, but it was the same result
1 Like
Its the cuda_plot_k32_v3.exe ( but it is the second release of it so I named it cuda_plot_k32_v3-2.exe to keep track)
2 Likes
Yes, that’s understood.
To expand on that, pool has some time period and resistance / throttling built in when processing partials. What it means is that assuming the farmer sent only 1 proof with X difficulty, due to the throttling, the pool will not recognize it as the full farm size, but let’s say only .25%. As no new proofs are sent, after that predetermined time period expires, the farm size will drop to 0. Conversely, assuming that the farmer constantly and often sends 1 proof and the pool will have the proper farm size, if 1 proof will be missing, the same time period / throttling will be used to lower the farm size down by 0.25.
So, depending on what is that time period used, with low proof count per day, the pool may actually never show the full farm size. (I didn’t look at the code, don’t know how long is that time period.)
If we add to that randomness of proofs, the same thing will apply, most likely resulting in proper farm size if averaged over longer time periods.
So, focusing on a single point pool report doesn’t really make sense, we need to compare longer trends. Therefore, I said that so far the data I have is kind of garbage (only about 2 days).
Now, the payments are not calculated based on farm size, but shares * diff submitted. Therefore, if there is no payment throttling / time periods used, it should not matter whether the farmer supplies on average 1 or 1k proofs per day. The payments are related to “pool shares” (proofs * diff), and at least Foxy pool reports proofs in 15 mins slots (for 1 day reports). With higher diffs, there will be less proofs, but the diff will compensate for lower proof counts. So, switching from 1 day to 30 days report should smooth those daily results, and show no difference in shares count. Although, it may as well be that the charting part is doing some extra work to do smoothness and higher diffs can be shown as overall lower shares.
Hope that it makes sense.
Sorry, forgot to mention the context of those exchanges. The latest c3x GH compression recommends partial diff starting at 20k and going up to 200k for c33 (when the GPU is close to be maxed out). That means smaller farms with 3060 ti or lower performing GPUs will have a handful of proofs per day with 20k diff.
Can you try single GPU with -g 1 to use second GPU only? Also -S 8 is too much for 3060 unless they are 12G model.
Multi-GPU is not supported in Windows usually, at least nobody has succeeded in that yet.
1 Like
Looks good. There are 2 places where the GPU spilled over 3 challenge cycles, so not that bad (I mean, it is bad to have 3 full cycles in the row, but you have only 2 of those). As we have 3 challenge cycles to submit partials to the pool, those are potentially close calls (if partials were found in the first cycle). So, based on that, you don’t have that much room to expand your farm. Actually, your current partials rate is about 1.5 partials / 15 mins, so if all is “normal” those 2 spill-over places were where partials were found (plus a higher eligible plot count). Also, we need to remember that those 2-cycle spill-overs show that there are more places that may run into the trouble.
You mentioned that you have around 800 c31 plots and roughly 30k plots total, so those c31 plots represent about 2% of your farm. Maybe those plots were included in the eligible plots calculations during those spill-over cycles. Again, the GPU cycles needed are no-of-eligible-plots + no-of-partials.
Your 3060 should hold up to around 500-600 TB raw, and you are rather far away from that. Maybe those c31 are the main culprit? Maybe if you want to push your 3060 closer to the max (and not add another GPU) you would need to kill those c31 ones?
Also, I think that your 500 diff was kind of close to the sweat spot (not too many 15 mins gaps). Looks like you pushed it to 1k now. I think that efficiency gain for that is rather minimal (no-of-eligible-plots will not change, just no-of-partials will drop by half, so maybe at best you will have just one such 3 challenge cycles / 30 mins; not that much GPU cycles savings). Anyway, if you could tolerate it for several hours, that would be interesting to see what you will get. (I set my diff at 10k now for the day.) Although, I think that some of those charts are not really calculated properly (using the accurate diff levels, if those change over time), what may be really confusing. Also, the thing we really want to monitor is daily payments, and that is the most tricky one, as those vary with no of proofs submitted, and pool earnings, making it really hard to normalize.
Evening folks. Having an error that my old tired brain can’t figure out.
Specs e5-2699 v4 x2 , 512 gigs ddr4-2400 .
error=./cuda_plot_k32_v3: line 1: payload:allShortcutsEnabled:false: command not found
Command string-sudo ./cuda_plot_k32_v3 -n 740 -C 30 -M 256 -t ‘/run/media/gm/Temp1/’ -t ‘/run/media/gm/Temp2/’ -d ‘/run/media/gm/1/’ -d ‘/run/media/gm/2/’ -d ‘/run/media/gm/3/’ -d ‘/run/media/gm/4/’ -c
It is not safe to run programs with sudo. So, instead of doing that, you should change the ownership of cuda_plot… to your local user (sudo chown) and change properties to 0500 (chmod) (can read and can execute, but cannot write for the user (to turn it into a rogue program), forbidden to touch by any other (potentially rogue) accounts.
I have the following, the c31 plots will need to go come the filter change. I’ve upped the diff to 1000. I’ll probably start replotting the c19’s Saturday.
C19 3864
C30 1889
C31 829
1 Like
Yup, I have the extra zero there, plus didn’t account for compression. 
Those c19 and c30 are wash as far as GPU cycles. On the other hand, c31 are 2x heavier. I bet that once you get rid of those c31, you will not see those GPU spills anymore.
I would really replot those c31 first and after that watch lookup times (to eventually see whether you could squeeze few c31 at the end of the process).
1 Like
Need to download actual binary, not the HTML page. See Raw button: https://github.com/madMAx43v3r/chia-gigahorse/raw/master/cuda-plotter/linux/x86_64/cuda_plot_k32_v3
Yeah his GPU is almost 50% idle. It’s ok to get 1-2% stale partials, if it means you can farm twice the space 
1 Like
I’m ret@3ded.Thanks Max
It’s not directly, but df -l and look at the file sizes should let you count them.
May be I should try Linux… Not familiar with it.
Try Linux Mint, thats what I found the easiest, although I did have some major issues when I moved to Cuda plotting but a fresh install solved that.
Once it’s setup and working it’s very straightforward.