I have two windows 11 pc with same motherboard and same ram (128 GB) with 230 TB space (about 3.000 C7 plots each). Two boxes are on two different internet servers. One pc has intel i7, 10 th generation and the second has i5 10th gen. The first one has found 4 blocks since replotting (4 months ewt: 22 days) and the other with i5 has found no blocks since 4 months! Any idea? Missing signage points are similar about 30-60 per day. Should I try to overclock the second box? Or is this just bad luck?
1 Like
If everything is running OK, then its likely just bad luck, at 4 months and with a 22 days ETW, that’s about 546% effort, so getting less common, but there was someone posting recently who was at 1000% effort (10 x ETW), then they won.
The pool I’m on, also got up to 730% effort recently, until I found a block, so it’s not uncommon.
3 Likes
re-plot them if no more HDD you need to fill in
1 Like
How’s the CPU usage, is it close to 100% on the 2nd one?
1 Like
I recently took note of the intro of compressed plots in 2021 and have been replacing my 400T of uncompressed with compressed. During this work, I noticed that the proofs were taking 10x the time and while my uncompressed plots were consistently under 1 second, the compressed plots were in the 10-20 seconds range.
Since I needed a GPU for plotting anyway, I read here that you can enable GPU for proofs in config.yaml. Once I did and restarted, my proofs on compressed plots were again under 1 second.
It appears to me that the new requirement for a harvester is a GPU!
4 Likes
A fact I’ve been dragging my feet on. I totally regret buying a slim case PC at this point lol.
1 Like
Very likely between the compression and number of plots the cpu can’t keep up. As others have said it sounds like it’s time for a gpu harvester.
2 Likes
What version of Chia software are you running? Is it the same on both harvesters?
Another few performance tidbits:
-
the host I run the full node on: a Threadripper 32 core with 128 GB RAM and an Nvidia 3060GPU - the full node is currently using about 1.2 cores continuously just to service incoming queries
-
the host I used to replot the 400TB is a 64 core Threadripper with 256 GB RAM and a 4070 GPU. I now use it for all harvester duty on 400TB of k32/c7 plots and it uses about 0.2 cores continuously to service the load
The average response time for the 4070 based harvester is half that of the 3060 harvester - about 0.6 seconds average. Also - it seems the full node is much less loaded when there is a remote harvester than when the harvester is running on the full node.
These machines are a far cry from a Raspberry pi - and the increase in harvester work with c7 compression is enormous. While the $600 I spent on the 4070 seems warranted by the 30% improvement in earnings with compressed plots, the machine requirements for being a full node and harvester are pretty $$$.
1 Like
Version 2.1.3 on both, which is latest I believe.
1 Like
Dunno why you are replying to me . My comment was to the OP .Nice systems btw . I farm and node on a old 3950x @4 Ghz,128gb of ddr4 3600 and a 3080ti with a couple of netapps .I plot on an old z840 2 xeon e5-2699 v4,512 gb ddr4 2400 with a asus tuf 3060ti and several u.2 ssds. Usually just under 2 minute plot times.
Biggest workload I have seen is decompressing 15 plots of c20 .Causes seek time to dip into the 2 second range . Just got serious about chia around 2 months ago .
C20 sounds like the nossd stack - the formal chia stack only goes to c9 I think…
Nope gigahorse. I got tired of those errors which were bladebit based .Series 3 cards can experience the grr result error. At first it was me fighting with bladebit and the 3950x plotting . Plotter crashed and crashed read that the bladebit plotter and AMD architecture didn’t “like” each other . After alot of hair pilling and cursing I used MADMAX’s plotter that just worked .Welp I held out from gpu compression , but I finally gave in. Used Bladebit, replotted and began getting that grr error . Swapped to gigahorse which farmed my bladebit plots (Thanks Ronski) and have never had 1 single error. So I decided to play around and compress the hell out of everything . With the upcoming halving that everyone is so worried about I will just split my drives to two harvesters instead of replotting to lower compression.
Ahh - ok! I read nossd’s pitch which is big on how they can get 2x compression with their plotter at something like c20 - is there an open GPU plotter that gets that (2x) and is compatible with the v2.1.3 chia distro full node?
Nah. It’s nossd or gigahorse. I’d rather pay Max a small fee and not have to continually babysit my farm with better performance and compression than bladebit . Nossd scares many , myself included for having a backdoor in their software so they can upgrade their software. I just do not agree with that. Someone figures that out or a bad actor at their pool and boom .All your xch is gone or all of a sudden you are part of a bot net and find you are part of a cyber war with some group or country LOL.
5 Likes