Ta, I doubted pcie adapter would help.
Your talking about the sata ssd here?
If so that’s slow due to being sata, nothing to do with errors afaik.
Here the professionals dismantled these drives:
Conclusion: after sustained load, the speed is reduced to meager 128MB/S. At this rate of speed, you can’t physically reach warranty period, yes - but what’s the point?!
2 Likes
That is more or less the same thing, but I would not really bet on the difference.
When you push NVMe (e.g., Samsung 2TB), you will see about 6W usage. A 10TB HD draws about 8W. Looking at Seagate 15k SAS drives, the 300GB one draws typical 6.9W, so just .9 more than your NVMe. Pretty much the same heat output.
One more thing. My dual Xeon e5-2695 v2 draws 450W in plotting peaks, so do you think that extra Watt or two makes a dent?
I was talking about SATA SSD. Max read/writes around 550MBps. Comparing to about 3MBps for NVMe drive.
Yes, there are two types of connectors for that NVMe form factor. The only thing that I know is to stay away from the “old” one, as it potentially doesn’t have any speed advantage over the 2.5" SATA SSD (I think the physical interface may be the same - SATA).
I wanted to give you two thumbs up, but it doesn’t let me do it!!!
That is what I was trying to imply. The “pros” are running everything from memory, we “the plebs” are happy with one/two NVMes, and if we push it, those SAS3 drives are another option. So, who is the intended customer for that price range. I also compile my code on NVMe, and am really happy with that. My understanding is that for those that do heavy Video editing, they use NVMe specialized RAID0 cards, and basically saturate all PCIe lines with existing NVMes. So, again who will buy it.
I guess, they will see those NVMes flying off the shelves, and in a month or so, they will be $200-300.
I did not find XMP ( its not supported I found on different forum ) BUT I found something named D.O.C.P ( I turned on this, lets check )
D.O.C.P stands for Direct OverClock Profile; it is an overclocking profile by [Asus] developed for AMD motherboards. D.O.C.P uses XMP protocol to automatically set up data rates and timings on the AMD motherboards.
D.O.C.P VS XMP
Most latest Motherboards with Intel CPU have XMP, and AMD motherboards have D.O.C.P. Both XMP and D.O.C.P are designed for the same purpose.
I guess its the same thing. Lets see how this turns out
1 Like
Yep, same thing different name.
Just do one change at a time, and run the test, and move to the next change. You don’t want to have multiple moving parts, as it is difficult to interpret your test results.
1 Like
Okei. sorry for late reply but here it is.
This is after D.O.C.P is turned on and Working Directory 2 is added.
Before
After
We can see that we went from 744 min to 341. ( I have to say this is pretty good improvement. )
I did not try to write directly to C DIsK as people mention that this can be risky? Should I leave this or should I try?
What I am trying now is to run more scripts parallel on my other HDD. Looks like that worked as well with the same speed on all scripts.
It will be alot faster, but could kill your ssd so it’s really your decision to make.
If you do make sure to back up anything important on there.
First off, I would not use your c: drive for plotting. Not because you may ‘kill’ it (that would take considerable plotting), rather you could easily fill the drive to capacity, thereby locking up your operating system, rendering your PC inoperable.
I see you you have received many varied opinions on changing/fixing things. However, just starting out, as you are, simple, lower cost/less complex mods are much quicker/easier to implement.
To be frank, even your greatly improved 341 minute (5.68 hours!) MM time is very, terribly poor given your PC’s inherent capabilities. You are irrevocably and unnecessarily limiting results by trying to use your hard drives for plotting, no matter how they are configured.
Please, so you can get productively plotting, keep everything as it is today, buy a reasonable cost, high quality consumer SSD or perhaps two (for example, a Samsung 980 Pro 1tb) and get either a single nvme, or double nvme PCI-E adapter card to support the SSD(s). Install it and use it for -t and -2, and finally get to plotting. You should have times well under 60 minutes, and likely closer to 30 minutes with that $300-400 upgrade expense.
If you ever get 1000 or 2000 plots made, and want to continue, at that point you can decide if things like enterprise SSDs or RAMdisks are necessary, but I would get the basics to start and learn how everything works together first. Good luck!
1 Like
yes this!
I can recommend WD black SN750 as a budget alternative to the Samsung drives.
Beware though that not all SSD are equal, so only buy those that are confirmed to be good for plotting Chia.
A single nvme like that + pcie-m.2 converter will set you back 150-200$ at most
Don’t go crazy on expensive enterprise drives or whatever unless you intend to plot like crazy.
As Voodoo says, get good ones. Like these. The US70 is excellent, like the 980 Pro, and others. I have the US70 as well, it’s not expensive, and practically the equal of the 980 pro.
Uhm. My understanding is that we are trying to identify the issues with your setup at this point. As such, running those full plots is really waste of time. We want to get your plots down to around 20-40 mins level with [P1] table around 10-20 sec. Therefore, running those intermediate 3h or 7h plots is just a big waste of time / electricity.
[P1] Table results are potentially the best approximation of CPU/memory speed. So, really, there is no reason even to wait for the end of Phase 1. If you check, the ratio of old P1 Table to full plot was ~2.2% and after the memory change is ~2.5% (so a good estimation of a full plot time, thus no need to do a full plots). Also, the ratio of the new P1 table to the old one is 59%, where the total plot creation is 46% respectively (again, good estimation of that change). Again, P1 table reflects mostly CPU/memory, where the total gets hit by drives performance - thus that difference, but still darn close.
Is that result good? We don’t know, as you didn’t do a Task Manager memory section snapshot. Can we push your memory further? We don’t know.
Well, the context is important. It is NOT risky at all to give it one run. It is risky (or wrong - IMO) to do it as a permanent solution.
However, you didn’t answer yet how many plots/drives you intend to do. If your goal is just to fill those two 10TB drives you have, then I would still run it either off those HDs, or off your C: drive - again, 20TB worth of plots may be OK for that C: drive using Primo Cache to 50% reduce those writes (I don’t know it TBW value, and the current level). Even with your current setup, if you run Primo Cache on your temp2, you will see big improvements, but there is no point to test it for now, as it will only cloud the data we want to see.
Again, to do that run, just seeing the P1 Table result would be a good indication. However, the whole Phase 1 would be better, as that phase is heavily temp2 bound. Again, no risk at all to make just one Phase 1 run.
So, I would run one such test without any fears.
I would also suggest, that you run one more test (Phase 1 only), where you swap temp1 and temp2 folders. The results should be exactly the same as the last test, but would confirm that you don’t have any issues with your D: drive (what is what everyone initially suspected).
1 Like
I plan to plot my 70TB disk space.
Right now I have plotted 10TB so I have 60TB left.
i have looked at the health of my disks, it should be fine but I can mix it up to try.
What I think I wanna do is to buy US70 or 980 Pro, the cost doesn’t matter for me its more that my hardware knowledge is limited. I will check tomorrow if I have any empty nvme space in my pc, if not I will get a single NVME or dobble. At this rate I get only 2 plots per day, I feel that’s to little.
Both of those NVMes are PCIe 4 drives, where your box is only PCIe 3, so speed wise you will not get much/any benefits from using those. Although, they may have higher TBW rates (last longer); however, if you want to plot that 60TB you mentioned (or double of that), Samsung 970 Evo Plus will also serve you well. Also, potentially for the price of that 980 Pro, you can almost get 1TB and 500GB 970, so you will be able to split your temp1 and temp2 folders, and have faster plots.
If you have another box that you could use that NVMe after you are done plotting, and that computer has PCIe 4, that 980 Pro is a very good choice as well.
I don’t think that your mb has NVMe slots. You will need to also purchase those PCIe to NVMe card(s) to make it work. I think that you have plenty of free PCIe slots on your mb.
Sorry for only talking about those Samsung NVMes, but I only have experience with those, and WDs (used to be very good when I purchased those, not sure right now).
1 Like