In this blog post, I try to guide a person to figure out how many plots they can make a day. This question is by far the most asked question. Feedback welcomed.
Overall looks very good, one quibble
Using 3400 RAM for 2 threads in a plotter setting is close to ideal. Setting more RAM will NOT be used by the plotter. It will NOT make the plot faster.
It’s been a while, but slightly more RAM did make plotting a tiny bit faster as of 1.0.4. I’m talking like two percent.
I’d also add that after a certain point, even staggering, you are going to overwhelm the NVMe interface if you are plotting too many simultaneous things to the same drive. I would say a good rule of thumb is no more than 5 plots to the same NVMe drive if you want to be safe? And probably no more than 3 simultaneous plots to a SATA SSD as a similar safety rule, even with staggering.
I think some oversimplification is good here just to keep people out of the weeds…
Good point. I think its phase 2 where the Extra RAM gets taken advantage of. That’s the only one that doesn’t list any buckets.
I think sustained speed, TBW and price per TB should be considered together.
The comparison should consider between
- added benefit of “plotting faster”
- added cost of “equipment dead faster”
2 Likes
The only resource I could find on sustained speed is the toms hardware page. Is this metric featured on the SSD Spec sheet? I see that max throughput is usually listed, but I haven’t notices sustained write speed.
2 Likes
Very informative post! I have a question about your cores + threads calculation though: shouldn’t it just be num_threads/2? A 12-core, 24-thread CPU has 12 cores, each of which can run 2 threads at a time, meaning it can run 24 things at once, not 36. Or, are you using the 1.5x multiplier on the number of threads as a proxy for the plotter only using multiple threads for stage 1?
I’m interested how you get the cores + threads / 2. If I have 16 threads, I figure I can run 4 plots at 4 threads each concurrently. What am I missing?
Yep, this is how we ended up with the cores+threads method.
1 Like
Threads only count in phase 1 of plotting, everything else is single threaded.
1 Like
Ok, so I guess if they are staggered to where not all are in stage 1 at once, they would probably work. Gotcha.
Manufacturers sadly don’t like sharing it because then their “8000MB/sec+” claims on the box are getting doubted and granted, for desktop workloads, generally those are perfectly fine, it’s just this “not so normal” workload which exposes the true underlying speeds.
Tomshardware and Techpowerup have been testing it since TLC SSDs starting introducing these caching mechanisms, MLC didn’t suffer from this as much as long as they used controllers with enough memory channels. And well, that’s still true on the TLC side really. Some SSDs use 4 channel controllers, others use 8. In steady state that can make the difference between 500MB/sec and 1000MB/sec sustained even though both SSDs use the same TLC chips and the same amount of flash.
3 Likes
Thank you for your explinations Quindor. I wouldn’t have known about this limitation if you had not mentioned it in the other thread.
I wonder if Chia will end up changing how SSDs are marketed.
3 Likes
Certainly don’t start with more than 4-5.
Edit: Assuming your disks are connected with PCIe 3.0 x4. Some (generally larger) disks use more lanes and some newer disks use PCIe 4.0 (if your motherboard/chipset/CPU do). In those cases, you can use more bandwidth for sure.
Then once your last job (started when all the previous ones are already running) finishes for the first time, record the total time, then you can try adding one or two more (with stagger) and see if there’s a slowdown and whether it’s worth it.
1 Like
Note I added this blog entry to the category description, because I tire of twenty “how many plots for my rig” posts every day…
@Blueoxx let’s be sure to keep that blog entry up to date so it’s a key reference for everyone!
2 Likes
I’m going to need to update the formula to 250 for temp space.
2 Likes
Thank you! This will help a lot. I feel like if I tried to find the best settings for plotting by trial and error, by the time I will have found the best settings, there will be no more plots left to make because my HDDs will be full.
Not sure if you mentioned this, but I read somewhere that the real advantage of using 6750 GB of RAM is not speed but fewer writes on the temp disk (preserving the SSD). Is this true?