Is there any advantage in plotting k33, k34, k35?

The simple answer to your question is yes.

One K33 provides the same chance to win as two K32s.

K32s are destined to become useless. I have read guesses of from 3 to 15 years. K33s will also become useless even further down the road if Chia grows that much.

I am plotting K33s. My 2TB NVMe plotting drive can comfortably handle 3 parallel K33 plots. Yes, I could do 7 K32s, but I do not consider this a loss.

Contrary to comments I have seen, I have found that 3 parallel K33s take 12% less time than 6 parallel K32s. This is the the second best plotting speed increase I have found. The first was NVMe plotting.

With my plotting done on a 2TB NVMe as the main and a 2TB SSD as the secondary plotting drives I now complete 3 K33s every 18 hours.

I did first performance checks before the NVMe and 3 K33s were produced in 28.5 hours. At that time 6 K32s took my machine 32 hours.

So I got a performance increase when I switched to plotting K33s and I believe they hold more long term value.

Lastly, 32 K33s fit on my 8TB (yeah right) HDDs with space left for a K32, but I don’t bother.

I would plot 2 K34s, but only 15 fit on an 8TB and I would need another NVMe to have enough plotting space.

If I am still enthused when I run out of farm space I will buy HDDs appropriate to fitting K34s and upgrade my machine to start plotting them.

There is a lot of investment in K32 thinking. In my opinion, if you are in for the long haul it makes little sense to plot K32s. One K33 is produced slightly faster than two K32s and will hold its value longer.

2 Likes

Good points here! I didn’t know plotting was marginally faster with K33s.

I’ve been worried about this recently actually: we keep hearing K32s will be good for another few years at least. But this is from the same people who estimated netspace growth! :rofl:

How mad would you be if Bram came out next week and said “uh guys, we are moving to K33 next month. Sorry!”

4 Likes

That is interesting - once i have plotted my disk collection and start to replot (poolable) i am moving towards SAS HDD plotting and retiring my SSD to an easier life (an enhanced desktop experience) - SAS HDD tend to be larger drives (600GB, 900GB). When you plot to SAS HDD you usually run 1 plot per drive - so increasing to a K33 has no additional cost

Update:

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. :smiley:
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?

Enjoy and be well!

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 don’t 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 … which it should have.

Worst thing you could do is try and fail.

Saying that there is no advantage means you do not consider being serious with Chia on the long term… 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.

A K33 has twice as much potential value as a K32.

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.

Not sure what you mean by 2 K33s having same value as one.

Anywaze, one K33 does have twice as much value as one K32 and it MAY have more longevity. K32s will one day be replaced and become worthless.

If you take in consideration new MadMax plotter I see that is probably happening sooner than later.

2 Likes

We don’t know what the next big speed shift come this winter. It’s 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 …
Can it be advantageous to have k33 or even k35 for those who have many plots?
Would fewer larger trials reduce the control time?

Good point.

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.

Naw … the access time is tiny compared to the network time.

Yes, but you would need to have A LOT of plots on the same physical system before you’d run into this problem. Like, several PiB at least.

I’m 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 didn’t keep records of the disk read/write performance before I started so haven’t 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.