I upgraded my NVMe to a Samsung 2TB (1.81 for real) 970 EVO Plus. Using an SSD as the secondary plotting drive I now produce 3 K33s every 17.5 hours or just over 4 K33s per day.
This on a cheap refurbished, small form factor Optiplex 7020.
Producing 3 parallel K33s uses close to 100% of the cpu and 80% of my 32GB memory. I have all of the system fans on full time in the bios or I would probably start to burn my cpu. After careful monitoring, I now feel comfortable that it is not burning itself up, even working 24/7. Within reason, I have maxed out my hardware.
I have not run new K32 tests, but I was only able to produce 5.5 K32s/day before my upgrade. I could probably produce about 7 K32s a day now with the upgrade so I do believe I have achieved at least a 10% plotting speed increase plotting K33s vs K32s.
Plotting K33s on a budget has become a fun hobby. As the economics of Chia farming (total storage growth rate in particular) have become tougher, I am even happier with my choice to go K33 as it removes my major mid-term concern. If and when will my plots suddenly lose their value?
Has anyone tested if a 600GB 15K SAS drive is sufficient for plotting k=33 plots in terms of temp space requirements?
Assuming a k=33 plot requires a maximum of 521 GiB (or 559.4 GB) of free temporary space, and a 600 GB hard drive has only about 558.79 GB of real empty space, I donāt see the requirements met.
I have plotted 4 K33s on 1.81TB. I should not be able to memory wise and it makes my pc work too hard but it never actually ran out of space ā¦ which it should have.
Saying that there is no advantage means you do not consider being serious with Chia on the long termā¦ So what do you do in 5 years, possibly 10? You delete all the Tera/Petabytes you collected and start all over again? Replotting all these Exabytes will be very wasteful on energy also.
I have watched this video and he has an error in his math. In probability you can not add two independent events together. a 1/3 chance and a 1/3 chance is still a 1/3 chance with each event, the same that you can not add a 2/3 chance and a 2/3 chance ( 4/3 chance ?? ) together.
He makes no mention of 1/3rd at all. He does speak of two x2 cancelling (not adding) each other out. His explanation and diagrams were clear and to the point.
A K33 has the same chance of passing the filter but twice as much chance of winning the challenge.
Thanks. Like I said, something did not make sense when I watched it. I have had many argue, there was no point in making K33 plots claiming 2 k33 plots have the same chance as one k33 plot.
We donāt know what the next big speed shift come this winter. Itās possible that K-32s will be doomed in the next year or two. Just saying. Nothing is written in stone!
I take up this question again trying to see it differently ā¦
Can it be advantageous to have k33 or even k35 for those who have many plots?
Would fewer larger trials reduce the control time?
Two K32s will be challenged twice as often as one K33. This does not change the odds of winning at all, but if you have a large farm, K33s should considerably reduce your response time vs the same TB storage of K32s.
Challenged twice as often vs having to search 2x the data, I think we need real figures to compare and will no doubt differ system to system dependant on bottlenecks.
Yes, but you would need to have A LOT of plots on the same physical system before youād run into this problem. Like, several PiB at least.
Iām curious, has anyone actually hit this issue (high response times due to too many lookups)? How many plots do you have, and can you rule out a network latency issue, which would be a different problem entirely but would manifest in a similar way?
I am still plotting and farming with the same machine. I am almost finished plotting 96TB. My first plots took 17.5 hours for 3 parallel K33s. As my farm has grown my plot time has gone up to 18.5 hours.
Since all the machine does is plot and farm, the time increase in plotting would seem to be caused by the increased time doing lookups. It is not scientific, but it points at the the notion that lookups take more system capacity than most think about.
I am on a refurb Dell Optiplex. People with real machines would probably not see the impact as easily, lolz!
Have you checked to see if your plotting disks perform as well as they did when you started? I have noticed a progressive (albeit slight) degradation over time on my dedicated farmers - I didnāt keep records of the disk read/write performance before I started so havenāt been able to determine exactly where it comes from, but can certainly rule out farming being an influence, since all those machines do is plot.
No, I have not looked at plotting drive speed at all. Since I never ran a baseline Iāll never know for sure, but I have a Samsung 980 plus bought new for the farmer and it is still well under TBW and tests out well. I donāt think it has slowed, but do not really know.
Iām ok with it ā¦ I was starting plots in the middle of the night to keep pumping 4 K33s a day, but have become less obsessive now. I just start new plots when I wake up every day and get full sleep and three a day instead of four, lolz!
If MM plotted K33 or K34 then people would do so. I imagine there are quite a few people who have never āadded a plotā using the chia gui or CLI interface at all.