Evening all. Just wondered if anyone has any 1st hand experience with compressed CPU plotting and farming. I’m not too interested at buying a GPU just yet as I don’t have a huge farm (just 80TB). Electricity costs are not a huge concern as I have a fairly large solar array. I can create a C7 plot in 43 mins ( I know this is not quick by any means) with a dual Xeon setup and would like to pursue this method. I’m not really too sure about how the decompression side works when farming and if the Xeons would be suitable so if anyone else is doing this please feel free to chirp in
My 2c is that c7 is too much for cpu.
However, I am on ancient i7 and pentium
@BadgerStork, is this based on C7 CPU plots that you have plotted / are farming or what you’ve read? With the dual Xeons I have 20 cores / 40 threads so was hoping it’s possible.
Is that c7 on gigahorse or bladebit?
Because c4 gh = c7 on bb
Max has a table on his gifahorse github page with how much you can farm.
Xeon e5-2695v2 can farm .21 PB @c7
No I have not tried c7
I have 11TB of c6. With 20 cores I think you are fine
For those wondering…
FF C6 takes 10-14seconds on a pentium
Core(s) per socket: 2
NUMA node(s): 1
Vendor ID: GenuineIntel
CPU family: 6
Model name: Intel(R) Pentium(R) CPU G645 @ 2.90GHz
[2023-03-04 04:36:19] INFO pool: Partial accepted ch=742e3983e717 compressed=true diff=1 elapsed=13.446146578s gpu=false new-diff=1 plot=c102662294fd size=32 [2023-03-04 05:18:02] INFO pool: Partial accepted ch=957ca43e285d compressed=true diff=1 elapsed=11.856205514s gpu=false new-diff=1 plot=3e6b1f08f8a4 size=32 [2023-03-04 05:26:59] INFO pool: Partial accepted ch=52deae86fdf5 compressed=true diff=1 elapsed=10.7895441s gpu=false new-diff=1 plot=41c4b03b2bbc size=32 [2023-03-04 05:33:22] INFO pool: Partial accepted ch=11bacc957ce2 compressed=true diff=1 elapsed=13.495006105s gpu=false new-diff=1 plot=39a791005448 size=32
It’s on GH. Nice one, I’ll have a look at that table.
Depending how many more plots you want to add, you may want to check out nvidia k2200 card - Dell Nvidia Quadro K2200 4GB GDDR5 PCI-E DVI/DP Video Card 0GMNNC | eBay. You can get it for ~$45 or so, and it draws max 40W. With the number of plots you have, most likely the average power draw from that card will be around 5W or so (not too many plots to process). Your Pentium’s TDP is 65W, so even if you reduce it by 10W, you will already have about 5W saving. Although, I am not sure how that would affect your lookups (my guess it would shorten them, but again, it is only my guess).
As a reference, this is my lookup distribution for ~200 TB of C7 plots (maybe there are some C8 there, but not that many) using that card:
If the processing power doubles with every C-level increase, scaling the y-axis by 0.5 would roughly approximate your potential results.
This chart below shows GPU utilization, what correlates with power draw (that square wave). When it is hitting the max, power goes up to about 25W or so, in those dips goes down to ~5W (you can see that 5W on the second line). Although, I am not too sure how reliable is that power reporting.
Not sure what are your Xeons, but from what you provided, looks like e5-2600 v2 with a TDP of 130W. I assume when your are plotting, your box pulls ~400W. Not sure what would be your CPUs utilization during harvesting, or rather if you are using that box as a harvester (I would not). It is really hard to beat the price of those low-end GPUs, both as the initial cost, and as recurring (electricity). My farmer / node runs on e5-1230 V2 (69 TDP) running close to idle.
By the way, when Max will release the decompression module, I intend to move the farmer / decompression module to an AMD APU drawing about 10-20 W at the wall, and adding whatever GPU needed to pull the decompression job (basically making harvester(s) also light weight).
Useful info but this is the CPU thread
The recycled pentium above can plot 6 c6 in about a day and a bit
I just hit two partials and the time went up to 20 seconds, so it really is tight
@Jacek , I’m using 2 x E5 2690 v2. I would like to give the CPU method a go and see how I get on. If I find it doesn’t work out and I decide to go with the GPU method, would I have to replot again with the GPU or could I farm the compressed CPU plots with a GPU?
No, all plots (from the same manufacturer) have the same format. So, you can plot with either CPU or GPU and also farm with either one. The plotting doesn’t have to match harvesting.
I would also think that harvesting all plot formats could be done potentially by someone not related to either Max, NoSSD or Chia. Once the plot is on the disk, most likely the plot structure can be reverse-engineered, and plot decompression may not be as challenging as compression, so even using chia’s harvester (when it will be available) could be a jumpstart for such project. So, it may be that at some point we may see such harvester (e.g., arc_plot guy could do it, as he already has good enough experience).
So, that is ~450 W at the wall when plotting. Also, depending how much RAM you have, you should be getting 2 plots in that 40 mins (each CPU producing one plot per 40 mins).
Potentially, that CPU beats k2200 GPU as a plotter, as I even not know if that GPU can pull a plotting job - at least I have not tested that.
The GPU plotter and the CPU plotter make the same type of plots, so you are good to go
Just out of interest, what pools are you guys using? I’m currently using Space pool, but I don’t think I can use compressed plots with this pool. I guess I may have to start looking for a new pool!
Hi Gareth, Why don’t you go for it and join Flexpool and try it out? If you get concerned about your night time electricity usage buy a cheap GPU. The Tesla P4 is a tiny 8GB card with the same processor as the GTX 1080 ti and is rated at 75W. It is vGPU compatible too which means you can split its resources should you want to try more than one pool.
Why do you think that?
You can use compressed plots with Space pool, I have already tested with Gigahorse Cuda_plot and farmer.