Plotting 16 in parallel, 4 threads and 3390 RAM in plotman on 4x 1TB pro 980. Staggered by 30 min. 5900x can handle them so far, but I expect some bottlenecks to happen soon (not being able to start a new plot after 30 min because 16 are still active). But 32 GB RAM seems to be on the border of what is enough for this setup. Provided you use Linux, headless and do nothing else on the machine.
I found that I could plot 4 parallel K33s on a 1.81TB NVMe. By a small margin, I should not be able fit the 4 K33s based on the “rules”, but I can. It appears that there is some leeway available.
My CPU got very hot and bothered the one time I ran 4 parallel K33s so I now stick with three parallel (where my CPU is a little hot, but not bothered, lol). 4 K33s should not have fit but it worked.
I don’t like the idea of pushing my CPU or my NVMe to their limits. I like to see happy system components.
You’re right – this is probably because they reduced plotter RAM requirements quite a bit in the early days! So there’s a lot of outdated info based on earlier versions of the plotter.
your CPU will be fine, get a better cooler.
Do you know anything about my cpu load, cooling setup now, expansion capabilities, or have any other information to back your bad advice with?
Keep giving blanket advice like that and you will kill someone’s computer.
Just so you do know, the 4xK33 plot took 33 hours with the case open and a big fan blowing in because the CPU was struggling so badly. System fans are all on full power and full time in the bios. The PC has an excellent cpu fan and no room for expansion. During this plot my Mhz dropped due to the heat and congestion.
So, with 4K33s I had constant 100% CPU usage 100% of the time and 33 hour production. 3xK33s take only 17.5 hours and my CPU is good in all respects.
My machine happily produces 4 K-33s a day plotting 3 in parallel, or struggles hard to produce 3 K-33s a day with 4 in parallel.
My system is far better suited to running 3xK33s and increasing my cooling was never an option to begin with. Sticking with 3 parallel K-33s is the answer, not a new cooling fan.
Hit Alt=f4 to reply
What cpu do you have? a jet engine?
i7 4 core 8 thread in a small form factor, refurbished Optiplex 7020 32GB RAM, Windows 10. Bought cheaply and well configured to accomplish my plot production goal on my budget.
Main plotting done on an 2TB (1.81) 970 EVO Plus NVMe, secondary on 2TB SSD. OS and paging file on their own 512GB SSD.
USB 3.0 drive arrays for storage.
My main plotting drive, the 970, lives at a happy 54C while plotting 24/7 and the CPU lives near 100% usage, but gets breaks to 85% and does not get that hot … as long as I push it no harder than the 3 K33s.
This rig was originally conceived to parallel 7 K32s, hopefully in under 24hours. After researching, tuning, test plotting, and learning more about the relative value of K33s, I eventually achieved 3 parallel K33s in 17.5 hours, more than 4 K33s a day, and over my initial goal of 7-8 K32s a day.
CPU heat was only a problem when I tested the 4 parallel K33s once and is what I was commenting on. My CPU is simply not powerful enough to do 4 parallel K33s. It slows down and gets too hot if I do.
What is the “Excellent” cooler you use on it? I am surprised a 4 core CPU causes problems with heat Is it an 4th gen i7?
No, I am genuinely interested in your problem, as I owned the same OptiPlex in the past and I like these. The discussion might also be helpful for other members of the Chia community to know that SFFs can’t be used to the max due to CPU heat problems if that is really what you have.
When I asked you about which CPU cooler you use, I am not talking about the fan. I am talking about the heatsink, which makes a huge difference in cooling performance. If you say you are using the Noctua NH-L9i or any similar cooler, and it still does thermal throttle at 100% then yeah, you are right.
BTW, your “I know it all, I’ve tested it all and the things I do are the best” attitude won’t get you that far in your endeavour.
Thanks for your polite and useful reply and sorry if I come off that way. I am an Aspy, lol!
I don’t know it all, but am a pretty good researcher and assimilator of information and PC technician.
I try to never speak without believing that I understand what I am talking about and will be the first to admit if you show me to be in error.
I can get a bit harsh sometimes … I will strive to be better!
In answer to your questions, the biggest problem is that it is the Small Form Factor when we are talking about heat.
I managed to rate the cpu fan as excellent leaving out the details.
Dell Optiplex 7020 Desktop Computer, Intel Quad-Core i7-4770-3.4GHz, 32 GB RAM, 512GB SSD HDD, DVD, USB 3.0, WiFi, HDMI, Windows 10 Pro (Renewed)
The specs say 3.4GHz but it is running at 3.8GHz through no work of mine. On second thought, increased GHz probably due to to long overdue firmware update.
It is an office quality CPU fan with good case airflow from the front case fan flowing over the cpu fan. I have it turned on full time blower mode in the BIOS. I’m getting about all the cooling the cpu could hope for.
P.S. just had my first post flagged down and your warning was prophecy, lol. Lesson learned and editing/deleting some of my posts to be more polite.
Yeah it is a nice little pc for Desktop, but the cooling system was definitely not designed for sustained high loads. Some people have changed the cooler for something better on that PC
But that requires some tinkering with the fan headers as DELL uses a proprietary cabling system. But I am not sure it is worth it in your case. As long as you are satisfied with it. But the day you get bored of it, the upgrade/tinker possibilities are endless But of course the CPU cooler will cost 1/3 of the optiplex price
Yes I agree. As the problem really seems to be the CPU doing more than it wants to. More cooling would help avoid burning the CPU but would not improve processing speed significantly.
You have swap enabled … so you dont really know if this is enough ram
disable swap for speed up and then you see if ram is sufficient.
I did that before and it was fine. But since I am going to the edge now I’d rather keep it safe, since I am away on holidays.
I think with the changes in version 1.0.4 and all the things learned in the past month, al lot of tutorials are wrong about a lot of things.
- Wrong CPU advise ( focus only on threads, disregarding single core performance. Also you can assign more threads than the CPU has to increase CPU utilization )
- Wrong RAM advise (you don’t need 3.4GB per plot when you stagger, I run 9 plots, ram usage is about 22-25GB. Also ram speeds seems to matter, at least for Ryzen systems)
- Wrong SSD advise ( focus only on TBW, forgetting to mention or emphasize sustained write speed, and for temp space the same as with the ram advise. Don’t need 256GB if you stagger the plots)
Totally agree with this. Just started optimizing this week. Last night I set up a run of plots with Swar that I thought would be reasonable but not pushing the system’s limits. This morning when I checked it, the CPU, RAM, and SSD utilization were all about 50% — which was 25-30% less than I expected.