I can't find this answer anywhere. I'm guessing it's because the architecture is unique but I don't know what the equivalent would be. When farming Chia you can specify the number of threads to dedicate to the process but I don't know how many threads the M1 has.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
How many threads does the M1 processor have?
- Thread starter rawdawg
- Start date
- Sort by reaction score
The gain per thread is marginal as only the first phase is multi-threaded. The remaining three phases only use one thread.
With 16GB RAM. You can only really go up to four plots at a time without over utilizing resources. Which the CPU should handle. Especially if you stagger your plotting. Which also relieves memory usage as plotting doesn't use 3.5GB all the time. Just at certain points.
So, I'd just leave it at the default of two threads. Doing at running two plots in tandem. Then waiting until those are in Phase 2/3 and start your next two plots. If you have 16GB RAM. Otherwise the most you'll want to do is two plots with 8GB RAM in a 1/1 stagger. Maybe choosing 3/4 threads to minimize Phase 1 time on 8GB RAM.
You can check the phase by looking through the logs. There'll be a break after 128 buckets and text listing the the Phase it is in.
This all assumes your plotting drive can handle the load. Very few SSD can actually take more than one plot at a time very well. Even NVME. As most just cheat with a fast cache which gets quickly overwhelmed by the sequential transfer. For consumer drives some good ones are the Samsung 970 Evo (NVMe about 1500 MB/s sustained sequential, as I recall) and Samsung 870 Evo (SATA about 500MB/s sustained sequential).
What's sad is that even a Samsung 980 balks. I'd have thought it would be a superior successor to the 970 Evo but it isn't. My 980 gets beat by dual 870 Evo (RAID 0). One plot they are the same but at two plots the 870 Evo RAID pulls way ahead. They can even handle three plots as well as the 980 can handle two.
For comparison a Crucial X8 on USB 3.1 Gen 2 gets overhelmed by one plot. As evidenced by not keeping up with the 980 or 870 evo on one plot. Meaning the SSD not thread settings is the limiting factor. The Crucial x8 is almost on par with two old 250GB 7200RPM drives in RAID 0. Even slower SSD do even worse. My 840 Evo can't hang with a single 4TB 7200RPM drive doing one plot.
With 16GB RAM. You can only really go up to four plots at a time without over utilizing resources. Which the CPU should handle. Especially if you stagger your plotting. Which also relieves memory usage as plotting doesn't use 3.5GB all the time. Just at certain points.
So, I'd just leave it at the default of two threads. Doing at running two plots in tandem. Then waiting until those are in Phase 2/3 and start your next two plots. If you have 16GB RAM. Otherwise the most you'll want to do is two plots with 8GB RAM in a 1/1 stagger. Maybe choosing 3/4 threads to minimize Phase 1 time on 8GB RAM.
You can check the phase by looking through the logs. There'll be a break after 128 buckets and text listing the the Phase it is in.
This all assumes your plotting drive can handle the load. Very few SSD can actually take more than one plot at a time very well. Even NVME. As most just cheat with a fast cache which gets quickly overwhelmed by the sequential transfer. For consumer drives some good ones are the Samsung 970 Evo (NVMe about 1500 MB/s sustained sequential, as I recall) and Samsung 870 Evo (SATA about 500MB/s sustained sequential).
What's sad is that even a Samsung 980 balks. I'd have thought it would be a superior successor to the 970 Evo but it isn't. My 980 gets beat by dual 870 Evo (RAID 0). One plot they are the same but at two plots the 870 Evo RAID pulls way ahead. They can even handle three plots as well as the 980 can handle two.
For comparison a Crucial X8 on USB 3.1 Gen 2 gets overhelmed by one plot. As evidenced by not keeping up with the 980 or 870 evo on one plot. Meaning the SSD not thread settings is the limiting factor. The Crucial x8 is almost on par with two old 250GB 7200RPM drives in RAID 0. Even slower SSD do even worse. My 840 Evo can't hang with a single 4TB 7200RPM drive doing one plot.
Hi velocityg4. Thanks you kindly for your help understanding all this! (I abbreviated the quote of your reply just to save space).
Yes, I only have 8GB RAM so I'll take your advice on my limitations. I have an OWC Envoy Pro EX (thunderbolt 3) SSD.
It's weird though, whenever I try to do parallel plots even at the suggested RAM with 2 threads my system bogs down. I've experimented with delays around 300-400 minutes but the progress is really slow.
This is using the GUI and I don't see the option to set a number of plots to run in parallel. I'm guessing to do that I need to use the CLI.
Set queues not parallel. Set one queue. Check the log then set the other queue at the right time. Phase 1 should end at around 31%. I try to set it in the 35-65% range.
Now that I've had some time to ponder. I'll have to just give you some observations from my plotting experiences. My setup was much different. With three Windows computers plotting two sending finished plots over Ethernet to my Farmer/Plotter. I never added my Macbook as I was already fully utilizing all my available plotting drives. I know macOS is more efficient at plotting than Windows but all my CPU horsepower is in Windows and I hate fiddling with Linux.
Plotting Speed
Plotting is slow. Expect a minimum of 6 hours per plot. I came across a few posts giving a measure of 6-7 plots a day on M1. So, at two plots. That's going to take about 8 hours to complete a plot on the M1. Another post shows 6 hours with just one plot using two threads. So, 8 hours per plot in parallel is more efficient. I'm guessing this is an SSD issue. Although it could be an issue of insufficient memory dragging it to 8 hours.
This can be longer or shorter depending on SSD performance. I have no idea what the OWC Envoy will do. I only saw some garbage benchmarks that don't hit an SSD with an extended sequential transfer. Which will eat through the cache and give the real reading.
With nothing else going on. You can try something like move 50GB of video or other large file from your internal SSD to your OWC. Then watch how low the data written/sec goes in the Disk tab of Activity Monitor. Towards the end of the transfer. Which should fully saturate the cache. Really this should happen much earlier than 50GB.
Queues
This gives you more control over the plotting than parallel. You just set multiple queues for parallel plotting. The pro articles I came across recommended this. So, that's what I did.
Memory
You only have 8GB RAM. That's shared by the GPU too. How much varies depending on your screen and what you're doing. But with two plots running. That's probably about it for your M1. It should have all other software closed. Any cloud syncs (iCloud, &c) should be disabled. Any network services should be off. With two plots running. I expect about all it's good for is one tab open for light web browsing, e-mail and writing a document in AbiWord.
You can try fiddling with the memory settings for the plots. Try to get RAM usage down some. If this is your daily driver. Plotting will impact you with so little RAM. I was fine on my main system as I have 32GB but there were times where I felt a small slowdown. My older media computer with 16GB was impacted significantly and I had to limit to two plots. That was a CPU issue.
Even with one plot. After shared memory, background tasks and plotting. You are probably left with 2-3 GB RAM for everything else. Plus about half your high performance cores' resources remaining during Phase 1.
CPU
I don't think this is an issue. Unless you are doing a CPU intensive tasks or multi-tasking. But I expect RAM is affecting you before the CPU plays a role.
Cost (TLDR, you won't make much, It'll take days or weeks depending on your farm size and it'll burn through your SSD)
What do you expect to get out of this? When I started plotting. My expected Chia win was 30 days with the storage I dedicated and Chia was worth a lot more. By the time I finished (a couple weeks) it was at three or four months and now it is at nine months (8 months yesterday, 7 Months a few days ago). I was aware of the network growth and figured a break even by September. But then official pools got delayed. I also didn't realize how high the TBW per plot was. As I was going to plot and hope for the best then replot for pools on the 17th of May.
I bothered because I had a bunch of plotting drives, RAM and CPU resources available. I've been wanting to upgrade my Media Server storage a long time now. So, I just figured. I don't need the storage right away. I may as well let the drives pay for themselves. Maybe make a little extra if I won a reward before having to switch to a pool.
At this point I have to either. Replot to join a pool, use my current plots with HPool or hope to get lucky. I haven't looked much so I don't know if official pools are available or if there is another delay. As Google's results are mostly more than a few days old. I won't be doing HPool though. They are getting close to 50% network capacity which could put the blockchain in jeopardy. Plus many figured out how to cheat HPool and double dip their plots.
I wasn't too concerned about SSD wear the first pass. As my farm is too small to really hurt all those SSD. I also know my regular TBW. Even with the plotting. They've probably got ten years before coming close wear limit. If I re-plot that may go down to 8 years. I'd just hate to dump another 70 TBW on my Samsung 980. Could really care less about the POS Crucial or the 870 Evo's. As the Evo's were originally destined for some old computers and got repurposed for this temporarily. The Crucial has just been sitting, unused, for over a year.
Just keep in mind. You are going to use about 1.6TBW per plot with Bifield enabled. More without Bitfield but why would you turn it off? On an 8TB HDD that's going to be around 115 TBW of SSD wear for 73 plots (remember GB to GiB). I can't find what the Envoy is rated for. If it's using TLC memory, it'll take a decent chunk. If it's QLC memory, I'd be very worried. If it's MLC (I doubt) 115TBW is nothing.
It's also going to take you 10 to 13 days to fill an 8TB drive. If the plotter never crashes and you never let it sleep or reboot. If anything interrupts the plotting. You need to delete the partially completed plot files. Then start the plot all over. Real world may take longer. The CLI is probably more stable as I had the GUI crash a few times. But I just stuck with the GUI as I was already losing interest and didn't want to look up the CLI stuff.
I mean if I was going to fill up a hundred 16TB drives with ten plotters. I'd be all over CLI, but just a couple 8TB drives wasn't worth the bother. Also the GUI crashes was more likely my fault. As I was learning the plotting ropes and figuring out the optimum settings. I was fiddling a lot with it and force quitting plots. Once I got everything tuned in and set with a queue depth of ten. Things ran smoothly.
Just an FYI. I also GPU mined a few years ago. At the time it seemed like a complete waste. But I sat on the Bitcoins and hardware after I gave up. A few years later. Those old GPU hit prices to break even on the hardware. While the Bitcoin shot through the roof and made a very nice profit. I guess the point is. Right now you're not going to make anything. You have to believe in Chia or convert your Chia earned to a crypto you do believe in. You might break even on a hard drive by January if you're plotting now. In reality you have to just sit on the coins you earn or convert to another crypto. Hoping for a decent return in a few years. Just like any other investment. As the get huge profits fast time passed in March/April for Chia.
Now that I've had some time to ponder. I'll have to just give you some observations from my plotting experiences. My setup was much different. With three Windows computers plotting two sending finished plots over Ethernet to my Farmer/Plotter. I never added my Macbook as I was already fully utilizing all my available plotting drives. I know macOS is more efficient at plotting than Windows but all my CPU horsepower is in Windows and I hate fiddling with Linux.
Plotting Speed
Plotting is slow. Expect a minimum of 6 hours per plot. I came across a few posts giving a measure of 6-7 plots a day on M1. So, at two plots. That's going to take about 8 hours to complete a plot on the M1. Another post shows 6 hours with just one plot using two threads. So, 8 hours per plot in parallel is more efficient. I'm guessing this is an SSD issue. Although it could be an issue of insufficient memory dragging it to 8 hours.
This can be longer or shorter depending on SSD performance. I have no idea what the OWC Envoy will do. I only saw some garbage benchmarks that don't hit an SSD with an extended sequential transfer. Which will eat through the cache and give the real reading.
With nothing else going on. You can try something like move 50GB of video or other large file from your internal SSD to your OWC. Then watch how low the data written/sec goes in the Disk tab of Activity Monitor. Towards the end of the transfer. Which should fully saturate the cache. Really this should happen much earlier than 50GB.
Queues
This gives you more control over the plotting than parallel. You just set multiple queues for parallel plotting. The pro articles I came across recommended this. So, that's what I did.
Memory
You only have 8GB RAM. That's shared by the GPU too. How much varies depending on your screen and what you're doing. But with two plots running. That's probably about it for your M1. It should have all other software closed. Any cloud syncs (iCloud, &c) should be disabled. Any network services should be off. With two plots running. I expect about all it's good for is one tab open for light web browsing, e-mail and writing a document in AbiWord.
You can try fiddling with the memory settings for the plots. Try to get RAM usage down some. If this is your daily driver. Plotting will impact you with so little RAM. I was fine on my main system as I have 32GB but there were times where I felt a small slowdown. My older media computer with 16GB was impacted significantly and I had to limit to two plots. That was a CPU issue.
Even with one plot. After shared memory, background tasks and plotting. You are probably left with 2-3 GB RAM for everything else. Plus about half your high performance cores' resources remaining during Phase 1.
CPU
I don't think this is an issue. Unless you are doing a CPU intensive tasks or multi-tasking. But I expect RAM is affecting you before the CPU plays a role.
Cost (TLDR, you won't make much, It'll take days or weeks depending on your farm size and it'll burn through your SSD)
What do you expect to get out of this? When I started plotting. My expected Chia win was 30 days with the storage I dedicated and Chia was worth a lot more. By the time I finished (a couple weeks) it was at three or four months and now it is at nine months (8 months yesterday, 7 Months a few days ago). I was aware of the network growth and figured a break even by September. But then official pools got delayed. I also didn't realize how high the TBW per plot was. As I was going to plot and hope for the best then replot for pools on the 17th of May.
I bothered because I had a bunch of plotting drives, RAM and CPU resources available. I've been wanting to upgrade my Media Server storage a long time now. So, I just figured. I don't need the storage right away. I may as well let the drives pay for themselves. Maybe make a little extra if I won a reward before having to switch to a pool.
At this point I have to either. Replot to join a pool, use my current plots with HPool or hope to get lucky. I haven't looked much so I don't know if official pools are available or if there is another delay. As Google's results are mostly more than a few days old. I won't be doing HPool though. They are getting close to 50% network capacity which could put the blockchain in jeopardy. Plus many figured out how to cheat HPool and double dip their plots.
I wasn't too concerned about SSD wear the first pass. As my farm is too small to really hurt all those SSD. I also know my regular TBW. Even with the plotting. They've probably got ten years before coming close wear limit. If I re-plot that may go down to 8 years. I'd just hate to dump another 70 TBW on my Samsung 980. Could really care less about the POS Crucial or the 870 Evo's. As the Evo's were originally destined for some old computers and got repurposed for this temporarily. The Crucial has just been sitting, unused, for over a year.
Just keep in mind. You are going to use about 1.6TBW per plot with Bifield enabled. More without Bitfield but why would you turn it off? On an 8TB HDD that's going to be around 115 TBW of SSD wear for 73 plots (remember GB to GiB). I can't find what the Envoy is rated for. If it's using TLC memory, it'll take a decent chunk. If it's QLC memory, I'd be very worried. If it's MLC (I doubt) 115TBW is nothing.
It's also going to take you 10 to 13 days to fill an 8TB drive. If the plotter never crashes and you never let it sleep or reboot. If anything interrupts the plotting. You need to delete the partially completed plot files. Then start the plot all over. Real world may take longer. The CLI is probably more stable as I had the GUI crash a few times. But I just stuck with the GUI as I was already losing interest and didn't want to look up the CLI stuff.
I mean if I was going to fill up a hundred 16TB drives with ten plotters. I'd be all over CLI, but just a couple 8TB drives wasn't worth the bother. Also the GUI crashes was more likely my fault. As I was learning the plotting ropes and figuring out the optimum settings. I was fiddling a lot with it and force quitting plots. Once I got everything tuned in and set with a queue depth of ten. Things ran smoothly.
Just an FYI. I also GPU mined a few years ago. At the time it seemed like a complete waste. But I sat on the Bitcoins and hardware after I gave up. A few years later. Those old GPU hit prices to break even on the hardware. While the Bitcoin shot through the roof and made a very nice profit. I guess the point is. Right now you're not going to make anything. You have to believe in Chia or convert your Chia earned to a crypto you do believe in. You might break even on a hard drive by January if you're plotting now. In reality you have to just sit on the coins you earn or convert to another crypto. Hoping for a decent return in a few years. Just like any other investment. As the get huge profits fast time passed in March/April for Chia.