I see that the requirements are just a tad over this 2TB limit. I have a 2TB SSD and would like to see if it will work. Will the plotting automatically stop or will it just slow down and pull resources from system ram etc?
Those high quality, high tbw 4TB SSDs are very expensive…
Pretty sure it will stop and abort with an error-message.
You could get another (preferably similar) 2TB and put both in RAID-0 configuration. Double the size and increase read/write speed also.
Try turning on filesystem compression.
Definitely keeping all options open.
I am also considering a RAID 0 array of 3x1TB drives, saves a few bucks that way as well.
I’ve heard about this and I haven’t done the HW. I will also keep this option open.
What would you gain with K35 plots compared to 32 or 33?
Fair question. It boils down to choices made at the start to leave all options on the table in the future. I’m planning a fairly large operation so thinking a few years out makes sense.
Right now the only viable option to jump this hurdle is a single 4TB drive, 2x2TB or even 3x1TB drives drives in raid 0.
There are several other considerations I had in mind if you are interested. Looking at how that Adaptec 71605 card holds up against 4 port pcie expanders. Your experience? Went with fractal design define XL as the case to start with.
You can do k.33 or k.34 of course, all within your current 2TB temp space.
That way you’d still be way ahead of the music.
And by the time those get out of fashion we’ll be several years further and your HDD’s will be replaced at least once.
Thanks, I am definitely still in the planning stages for all this to hopefully start operations either late next month or early january.
Quick question while I have people here.
In terms of temp storage required. I am under the assumption that the temp storage is what the TBW output is per each plot?
So in this instance a single K35 plot = 2.34 TBW on a particular SSD?
No, 2.34 TB is the minimum temporary disk space to plot a k.35. The total number of bytes written (TBW) during the plotting process is (much) larger.
For instance k.32 needs a minimum temporary space of ~256GB during plotting of a ~108GB k.32 plot, but writes over a TB (e.g. 1 TBW).
I have no idea if the ~4 factor also applies to higher k-numbers but that seems logical.
You can save ~75% of the bytes written by using RAM as temporary storage with the madmax plotters option of the chia 1.2.11 client. 25% of the writes will still be on SSD/NVME. Or save all of the writes with the bladebit option.
But you’ll need large sizes of RAM, 128GB for k.32, don’t even want to think about the RAM requirements for k.35… (don’t know either, maybe somebody else can fill in).
Just a tad is quite accurate.
I have found that the total space used on the NVME for plotting is just a little under what the documentation and client claim.
The space required according to the client to plot a K35 is a just a tiny bit larger than the available space on a 2TB NVMe.
I would suggest you try to plot one.
It will either work, or it will choke, but it will not do any damage trying.
If it does choke you will have to delete all the files on the NVME manually. Sometimes after a Chia plot choke, ghosts are left that cannot be seen and/or deleted. You will see space taken on the drive with no files to account for the usage. If this happens you must quick format the drive. I have never heard this causing any further problems.
Please let us know if you try and the result! 
@xkredr59 WOW! That helps a ton, so we are potentially looking at ~8-9 TBW per K35 plot. That’s lunacy, but I appreciate the information. Maybe the PNY LX3030 drives do indeed make sense even with the very high price tag.
@Aspy68 haha I am not going to bother if you have done this testing for us already 
Thanks, this so far has been extremely helpful. I thought I made some good choices with the Firecuda 530 drives, but these might not last as long as I anticipated.
Your TBW will be almost exactly the same for one K35 vs two K34s.
Don’t worry, we cannot kill NVMes despite the hype. Check this thread: Has anyone here personally killed an NVME ? (NVME Endurance)
All of our NVMes are lasting far past warranted TBW. I believe that TBW causes problems for NVMEs that result in non-volatile memory loss if powered down. You may make the NVME useless for any other purpose but as you never power down while plotting this is a non-issue for Chia plotting drives. If you have a quality NVMe it will probably outlast your Chia plotting. In any case, NVMe cost is pennies per plot if the NVMe is dedicated to plotting.
There are a number of good manufacturers, but my favorite is the Samsung Plus and Pro lines for their low temperature reliability without an added heat sink and for the Samsung Magician software.
I appreciate all the knowledge @Aspy68 & @xkredr59
Regarding the SSDs, that’s encouraging to see the responses in that thread. I went with the Firecuda 530s because it apparently has double the endurance at 1200+ compared to the 600 rated for the 980 Pro at the same capacities. It’s a bit odd that Samsung dropped that number on their pro line compared to the 970 series, but I am suspecting that many were returning SSDs with an unusually high usage rate for whatever reason
As much of a history seagate has, I actually had about the same failure rate in the past with burstcoin than with the others such as WD.
Good choice I think, have a (one) 2TB firecuda 510 myself and no problems so far. ~1800TBW, well below the guaranteed 2600TBW but NVME’s can go beyond those guaranteed limits.
The 530 series are Gen4 so faster again but only on a motherboard that supports Gen4.
I take it you’ll go for the 3x1TB in raid-0 then to plot k.35’s.
Could you please come back and inform us about plotting speed.
Did a k.34 on my i7-9700 CPU and the single 2TB NVME… 5 hours and 30 minutes.
k.32 with a 110GB ramdisk on my system does 37 minutes so times 4 is about 2,5 hours for roughly the same plotted space. I’ll stick to k.32’s for now
Awesome, yeah I am going to give it a shot at K35s on these systems I am building up. I am under the assumption that as long as your system is powerful enough that the total amount of time via the client itself will average out to about the same per TBW regardless of plot size selected, but maybe I am missing something.
Question regarding RAM/ramdisk. What kind of wear are you saving on your drive by utilizing this for these larger K plots? I CAN install up to 128GB on these machines if this means it can help save some of the wear and tear on the drives. I am assuming I have this option to set ramdisk on the chia client itself? That could be fun to play with as well if I have the option.
A full 18TB drive plotted in 4.25 days is pretty impressive @xkredr59
I can’t wait to get started soon.
edit: It appears I have to do the entire K35 plot set via ramdisk if I wanted to go down this rabbit hole? Yeah, at 128GB that’s simply not going to be enough lol.
You can combine an NVME with a ram tempdrive to reach slightly more than 2TB. But I think k34 is enough, you’re really pushing the limits on what can be done economically past that point. As bigger ssd become cheaper, you’ll be able to plot k35 easily in the future. Also to consider : bigger ssds usually have more tbw endurance, so bigger is definitely better.
Regarding the total amount written per plot, it’s ~13x the size of the plot : about 1.3tib for a k32 file, and about 11tib for a k35 file.
Yes, and no. The endurance per TB is exactly the same (for the same model). However, a bigger SSD has more TBs …
Yeah that’s been the case for a while now. Double the SSD storage, double the TBW coverage under warranty.
Bigger drive are overprovisionned and/or tailored for entreprise use, hence the improved endurance. You can’t find small drives with high endurance (except the new plotting SSDs but they are extremely expensive).