I have been quietly farming for a while and came back here and the forums are alight with GPU plotting, compressed plots and Dev fees. Not to mention people are speaking in acronyms.
I have been reading a lot. But am still stuck as to what to invest time/money into. I think I want to keep my current rig if I can.
I previously was plotting with MadMax on Debian ~55mins.
I have a dell R620 with 120GB of ram (2), and an 8 disk RAID0 (t). I have a second farming node server/PC with a Raid card with a bunch of random disks hanging in single disk mode and about 5 USB3 externals full of uncompressed plots ~35TiB on SpacePool.
I want to plot some more as I have some spare 2TB drives. Am I fine to keep doing what I’m doing? Anything else I need to know? - Did I read compressed plots will be native (and free) soon? Can I create compressed plots with CPU plotting? Any shortcomings with staying on CPU/RAM/RAID0?
2TB drives likely wont be cost effective unless you have very cheap electricity. You can certainly keep your uncompressed plots, they’ll still be working, but compressed plots will give you more effective farming space.
Compressed plots are native to create on the latest chia client, but only the high RAM (256 GB) + GPU option I believe. In the latest bladebit beta build the lower RAM + GPU options are available.
I haven’t been following CPU plotting for a while, so Im not sure on the status of creating compressed plots on CPU. The downsides to CPU plotting is it is slower and uses more power per plot to make than GPU plotting.
Depending on electricity costs it may well be cheaper to shut the farm down and just buy XCH. For efficient farming you need high capacity disks - smallest I’m running is 16TB, with some 18TB drives and even 20TB drives. 8TB drives stopped be efficient a long time ago in the UK.
You can carry on as you are, or replot compressed plots, I’m presuming you mean 128GB of ram, if you download the latest Bladebit you can plot with 128GB of ram but will need a suitable GPU. Plotting now takes minutes.
But, yes you can continue farming your current plots, but you should update Chia to the latest version.
Dev fee’s only apply to NoSSD or Gigahorse plots, not Chia plots.
NoSSD have the highest compression, plots a 50% smaller, which means you have twice as much effective space, but Gigahorse (Madmax) will be releasing a new version soon, and I suspect CNI (Chia Inc) will be many months away.
Edit to add. Compression is really the opposite, the plots are not complete, and part of it is calculated at the time its required. Not all plot sizes need a GPU to farm. No SSD has C1 to C15, Gigahorse is C1 to C9 currently, and CNI (Chia official) is C1 to C7. You can mostly farm up to CNI C5 on a CPU.
I came back here looking for similar information and this post is helpful, thank you.
I have also been similarly farming away on OG Chia Plots (solo - no pools) since Chia inception back around April 2021. It seems I have some catching up to do, and I want to replot and join a pool as my win rate solo with OG plots has declined significantly.
There are a few questions I have not managed to find the answer for, as I have been out of this for a while and have never pooled (ever). I also have limited time to work on replotting so only want to do it once - right first time. My questions are (and there may be more to come…);
I will be replotting one plot at a time on one system most likely. Can I continue to farm my OG plots solo whilst gradually replacing them one at a time as a replotted compress pool plot is complete ? i.e. Can the same single full node farm OG solo and compressed pool plots together, or does this need a different full node ?
The system I am hoping to use is my current full node. It has always been sufficient for OG solo purposes. It is a 16-core i7-10770 with 128GB DDR4 3200, a couple of NVMe SSD’s, 1x 3060 and 1x 3060Ti.
Can the latest versions of Chia 2.x use say my 3060 for farming the compressed plots and my 3060Ti for plotting at the same time ? Would it be best this way or the other way round (plot on 3060 and farm on 3060Ti ?)
I read the latest bladebit beta has 128GB mode, but is this any good on a system with only 128GB of RAM or does it need more RAM on the system to be able to plot and farm at the same time (Windows 11) ?
Which are considered the best pools to join ? I will be sticking with official Chia software only.
Anything else I should know ?
Would appreciate anyones input and apologies if there are a lot of these questions already asked and answered but as I say I am a bit limited with time for the foreseeable.
Yes to all questions, farm with 3060, plot with 3060ti, although you don’t say how many TiB of plots you have, but 3060 will cope with 2 PiB (1PiB after filter change) of C7 plots IIRC. 128GB of ram is plenty, but you do need a good SSD with plenty of sustained write speed.
I have 12PB so it looks like I will need to be careful and setup multiple harvesters again.
I have other systems I can use as harvesters and bring them back online as I did initially if needed and as I reach higher numbers of compressed plots.
When I started out, they were all plotting in parallel and all harvesting to the full node detailed above. Some of these systems have RX5700’s though and I understand AMD cards do not work for plotting/farming compress plots (yet ?).
I have another identical system to the one I detailed above that is currently used as a harvester for my full node, only difference being it only has 32GB RAM.
So at present, 2x systems farming. Of my remaining systems (currently switched off) there are a total of 4x more 3060Ti’s. So in total, I have 6x 3060Ti’s and 2x 3060’s (plus the 3x RX5700’s which I will ignore for now).
Next obvious question - can a 3060Ti farm more C7 plots than a 3060 ? Is farming performance based on the VRAM of the card, Ti’s only having 8GB of course rather than the non-Ti’s 12GB ?
A 3060ti will farm more, but I don’t know how much, and you can farm with less than 8GB vram. If using separate harvesters, each harvester will need its own card currently.
It might be worth selling some cards and getting a 3090 or 4090, but the card may be limited if not running PCIe4.
You can also run off some plots at different compression levels and then use Bladebit simulate to test how much you can farm, and probably best to check with the plot filter set to 256.
You may well find you can farm C5 without any problems, each step up gets a lot harder.
You may also want to consider MadMaxes Gigahorse software, which can use remote compute, so one GPU can be shared amongst harvesters.
This site has a GPU calculator, that takes in to account the several scheduled plot filter halvings, each of which will halve the number of plots a GPU can handle.
So if, for example, you max out a 3060 card now, then when the plot filter halving date arrives, you will have a headache.
I have a few questions about the charts:
I do not understand why the 3060 (non Ti) card is not included.
Depending on which GPU you choose from the drop-down menu, the output jumps back and forth between the Gigahorse table and the Bladebit table.
Are the GPUs Gigahorse or Bladebit specific? Meaning, if a table is presented on the Gigahorse side only, then that GPU is not compatible with Bladebit’s compression?
Thanks for that - interesting. Looks like I have enough systems with 3060Ti’s to last me until June 2030 with C7 plots, assuming the hardware lasts that long of course… These GPU’s have all been retired from Ethereum mining over the last 2 years buy they’ve all been well looked after and kept cool.
Thanks again, useful information. Didn’t know about the remote compute feature of Gigahorse but I think I have enough plot space to warrant multiple cards (in time) with several harvester systems.
I am unlikely to be going down the route of selling GPU’s and buying 3090’s just yet - spent enough already!
Power costs are obviously a factor but a nice by-product is the farm keeps my office at decent temperatures in winter, as it has no other heating at all.
It’s just bad data or a bug on that website, any Nvidia card 20 series and newer with 8GB ram can be used for plotting, less ram is required for harvesting.
10 series cards can also be used, but they do seem a little temperamental.
The 3060 GPUs are available in both 8 GB and 12 GB versions.
The 3060 Ti GPUs are available in only 8 GB versions.
However:
The 3060 GPUs have 3584 CUDA cores, and
the 3060 Ti GPUs have 4864 CUDA cores.
The more CUDA cores probably translates to having a faster GPU?
But what about having more VRAM? Is there any benefit that Chia can reap from 12 GB over 8 GB?
Thanks, I presume you’re just using stock clock speeds for core and mem ?
I’m used to mining Ethereum on GPU’s and it’s surprising how tuning can reduce power consumption without affecting performance. I’ll be trying this when I get some plots to farm.
Yes, just running it at stock. When I was running my 3080 on Nicehash I did tune that to control temperature and power use, but its not something I’ve done with the 3060.
I know quite a few on Discord have tuned various cards to reduce power use, its something I should do but just have gotten around to doing.