Interesting points - do you have some runtime on a few 970 Pros to see if it’s wearing ok? (too @Quindor 's point above)
10 cores is a tough sell for me, when you can get 16 out of a 5950x… This is also 10th gen, not the fancier 11th gen. I think AMD basically owns this part of the market, since cores are everything in Chia plotting.
1 Like
I saw your post from a day or 2 ago to this effect and actually settled in my mind: “Jeff’s right, imma hookup the AMD build” and TOTALLY lost steam when I saw $1100 for the CPU vs $350 for the 10850k.
I even checked the older Zen 2 9-series for some big-core-count builds and same thing, $$$$.
Now on the bright side, if Chia farming isn’t a bust and this ends up working out - next build will be a threadripper because we’ll all be rich 
2 Likes
Oh well that’s pretty easy actually, TL;DR the 970 Pro is a MLC drive so in theory with the same capacity it should always last longer then a same capacity TLC equipped drive.
Now there are improvements in the process, flash chips, intelligence of the controller, etc.etc. but even with that, the per cell durability of NAND has been dropping over the years instead of increasing because we are going from SLC → MLC → TLC (Triplle Level Cell). It’s only that we now have 2TB drives instead of 256GB that the overall figures for the drives are still ok.
Durability /2 but x 2 NAND cells = 1 for drive durability, although actual durability per cell is lower. 
So big drives are still ok, but smaller drives have really suffered in this regard. Still, for desktop usage this is all perfectly fine, no sane user writes 600TB in 5 years or even 10 years of desktop usage.
Anyway, so yeah a 1TB MLC SSD should in theory always have a better endurance then a 1TB TLC SSD. And agreed, also less of a write hole because of SLC cache that gets filled up. The 980 Pro 1TB however still does quite good when it’s cache is full, about the same as the 970 Pro does normally too. The 980 Pro can just achieve much higher bursts (which is helpful with Chia).
In my country the 970 Pro 1TB is significantly more expensive then the 980 Pro 1TB hence I went with 2x 980 Pro 1TB in R0.
3 Likes
So far, I’ve opted to have my fstrim cronjob run hourly (instead of weekly, as you would for most workloads) instead of enabling continuous discards.
My gut tells me this ought to result in better performance overall since I’m not gonna touch every block in an hour and conventional wisdom says continuous TRIM has a significant performance impact. I haven’t actually compared these approaches head-to-head, though.
Would you expect continuous TRIM to be faster than hourly TRIM?
Yeah pricing and supply is a bit of a problem… also with GPUs, but we don’t care about GPUs for this purpose!
1 Like
Hmm, interesting. My idea was that continuous trim would be better because if you keep writing and writing to the drive at some point you are basically backing up the internal logic into a corner where it has to start clearing blocks so doing this real-time would keep everything fresh and well in free space for it to write into.
Doing it hourly, I guess that could also work if your drive is big enough to never exceed the threshold of no more free/fresh blocks available.
I personally have not measured real-time trim vs scheduled trim to see if that made a big difference, my volumes are performing well with discard (trim) enabled. Then again I am tuning towards a not too big IOwait so might incorporate that into the amount of streams I send to the drive (since trimming could cause a slightly higher IOwait).
Not sure if I’m going to test it because performance already seems optimal for me, but it’s not a bad point/idea!
–update More thoughts, still think real-time is the better choice, doing it hourly could maybe still cause more “stress” for the wear leveling algorithms then doing it real-time since it will basically have more choice of where to put your blocks.
2 Likes
I’m getting very confused here. It seems like there’s a lot of different point of views, and the understanding of SSDs regarding pooling is developing, and the recommendations changing day by day. At the beginning, it was all about TBW, and all respectable hardware reviewers compare these numbers directly between different manufacturers - yet here it is said that those numbers can be very misleading. The discussion has also been about speed - but now I’ve learned that the number 3000/5000/7000MBps is only half a story since we are dealing here with sustained writing performance etc. I’ve also seen a few Excel tables with user experiences but due to the different setups, the comparison was quite difficult.
A few questions:
- Hasn’t anyone tested in practice, what is the true endurance of different SSDs, i.e. whether Corsair MP600 and Firecuda (which were originally recommended thanks to high TBW) truly last longer than e.g. Samsung 980 PRO? Based on my understanding, the hardware costs are crucial when calculating ROI, so plotting fast might not be the best choice if it will end up costing twice as much (assuming that Samsung would plot double speed but also last half as long as Corsair and Firecuda, meaning that you get to your goal twice as fast but also need twice as many SSDs to get there)
- Based on the images in this discussion, Samsung SSD’s are from different planet compared to MP600 and Firecuda. Is this graph misleading or is this the truth?
I bought Corsair MP600 1TB (running together with Ryzen 7 5800x and 16GB RAM), and have been plotting 3 plots parallel. Plotting one set takes 5 hours, leading to 14.4 plots per day. I’m quite happy with the number. However, now I just ordered another similar SSD and started wondering, if I should buy instead Corsair MP600 PRO 1TB or Samsung 980 PRO 1TB, since all of these cost more or less the same.
I ended choosing originally MP600 just due to the TBW - I’d like them to last “forever” regardless if I end up plotting 30TB or 300TB (sure aware that 1600TBW means perhaps 800 plots).
I am planning on running the second SSD with enclosure on Macbook Pro that has no extra space but is lying around useless. Currently, I am also plotting on another Macbook Pro which seems to be able to produce about 9 plots per day. I’ve been hoping that the laptop wouldn’t mind if I plot a few TBs with it, being aware this isn’t recommended.
So, now I am wondering here, if I should return the other MP600 and buy Samsung instead. Also one option could be to return both SSDs and buy a 2TB SSD instead, (and double my RAM and run everything on PC) but I wasn’t sure which one is more efficient - taking 100 % out of my PC or sharing the load and running these 1TBs between the Macbook and the PC. Any thoughts? (To be honest, one of the main reasons for me not to return the MP600 is that the installation was one hell of a job for Asus rog strix x570-f, and I hope to never have to touch it again).
So it all depends on expectations. The MP600 inherently isn’t a bad drive, it’s just not very suited for this workload, running 2 to 3 parallel plots will probably have it perform ok as you have stated, but a lot of people are getting the MP600 2TB to run 8 plots at the same time and it just isn’t very good at that, it’s a very clearly “medium” quality desktop drive.
But the math is easy there really, any size SLC cache on these TLC NAND drives will be defeated/burned through by plotting Chia to it. That cache is to take in bursts and then write it away without you seeing/noticing it but Chia will continuously hammer the drive. With 2 or 3 plots parallel it might be able to recover a little bit here and there but if you want more (and most expect more) it’ll grind almost to a halt, doing max 500MB/sec sequentially and probably much much worse for random IO since it’s onboard processor is basically driven into a corner without a way out to arrange stuff to give you the stated speed values again.
In that regard and 970Pro or 980Pro can do 2000MB/sec in the same scenario as the MP600 does 500MB/sec thus meaning it can in theory handle 4x the load or parallel plots, hope that makes sense.
Regarding the TBW I believe it was just looked at too simplistic initially. SSDs have something associated called “Write Amplification Factor” for years now. This factor determines when you write 1TB to the drive if it’s going to have to write 1TB of NAND chips or like 5TB. (very short, if a NAND chip is 16KB and you manipulate 4KB, it will have to write the whole chip anyway thus having a WAF of 4, now do this thousands of times…).
It’s my belief that the WAF of Chia is pretty low because I believe it’s mostly doing 64KB writes and thus fitting better to the NAND layouts of NVMe SSDs (make sure your file system is aligned, etc.). This means it might only have a WAF of 2 instead of the 4 or 5 that a manufacturer has calculated their TBW value with and thus the NAND is going to “go dead” much less quick then the number would indicate.
Hope that maks a bit of sense. 
5 Likes
This applies similarly to MP600, 970Pro and 980Pro?
If yes, MP600 still outlast Samsung in term of warranty.
I also found that cost per TB of MP600 is much lower at least in my area.
Summary
MP600 (max 2 TB) - 3600 TBW, 5yrs, Sustained speed: 500MB/s
Samsung 970Pro (max 1TB) - 1200 TBW, 5yrs, Sustained speed: 2700MB/s
Samsung 980Pro (max 2TB) - 1200 TBW, 5yrs, Sustained speed: 1500MB/s
At 5 years with 80% online time, sustained speed x 5 yrs x online time
MP600 (2TB) = 7884 TB = 2.19X Warranty
Samsung 970Pro (1TB) = 42573 TB = 35.5X Warranty
Samsung 980Pro (2TB) = 23652 TB = 19.7X Warranty
Samsung970/980 Pro will surely “dead” or out of warranty before 5 years based on this estimation, while MP600 is 2.19X which in the WAF range you’ve said.
My conclusion is that:
If speed is important, go for Samsung 970/980 Pro but it’s relatively more expensive and dead quickly
If longevity and cost are important, go for MP600.
Cannot get my hand on the intel one. That would solve all this problem.
Why not external SSD. No TBW. Cheaper cost per TB.
Sustained speed at 1000 MB/s but have not tested it yet.
3 Likes
From what I’ve read/watched, it seems as if the onboard controller could use some cooling, but the NAND chips actually perform better hot. But I couldn’t find much on what if they’re under this kind of full load all the time, does that change the recommendation?
I don’t think is quite that good.
An external SSD should suffer the same inherent problems as internal ssd’s.
The only thing really different is the way it is connected to your system. But the speed problem is not due to the bandwith of the connection but rather to the internal layout and quality of the NAND chips.
Or am I wrong here?
2 Likes
In the user perspective, what matter are plotting speed, price and warranty.
I also think that form factor or connection type are not that important, given the speed/price and warranty/price ratio are good.
If true(can sustain decent speed, warranty with NO TBW, better price ratio)
use it
From my understanding, type/quality of that chip in detail is already reflected in price and sustain speed.
1 Like
Well sure, although we still don’t know with which WAF Corsair calculated it’s TBW with and which Samsung did. If corsair calculated with 3 and Samsung with 5, they are in reality much closer together in actual TBW for Chia then the numbers would suggest but as said, we don’t really know, only time and actual usage data will tell.
Also, if you are buying an NVMe for plotting, you shouldn’t really care about warranty because if you burn out your NAND cells, you aren’t going to get any anyway. 
Also, we’re not going to be plotting 24Hrs a day for 5 years. You are going to be plotting until your storage is filled up and currently filling it up faster in theory nets you quicker/more wins since the netspace is growing with giant amounts each day. So if you are 10 days in and are at 150TB plots available with the Samsung drives or 50TB plots available with the MP600, that could make a difference there (although I have not done the calculations).
But it’s all about how you want to do this. I know people who are plotting on single HDDs, but then 60 at the same time. Sure a plot takes 24Hrs, but they still get 60 plots a day. And if you aren’t in it to get as much TB as soon as possible and are just going to see what it does in the long run, that’s a fine strategy too. But most questions here are about “Which SSD should I get to plot 20+ in parallel” and well then the MP600 is a poor choice in regards to speed (and we’ll have to see in regards to endurance, I’m willing to bet the Samsung 970Pro with it’s MLC chips will outlast it many fold. ). So there is nothing inherently wrong with it, it all depends what you are expecting. And people buying a 5950x generally don’t then want to be limited to 2 or 3 plots in parallel because they bought the “wrong” NVMe to match the rest of their system.
p.s. I am not sure what you mean with External SSD not having a TBW, it’s just a housing with a drive in there talking over a different protocol? Inherently no difference in TBW and generally it will be slower because of extra translation layers (USB vs NVMe (which is direct PCIe).
3 Likes
I agree. If the system is super high-end, it is better to go with super fast NVME to match the rest of the system.
I’ve read through their term and confirmed with their support a couple times on this point.
Unlike internal SSD, manufacturer simply does not state TBW for external SSD in their warranty term. Why? I’ve no idea. Do they think we cannot burn out their SSD? ofc, Chia plotting will do just that
If I was able to rechoose my ssd pick, I would go for the Samsung 970 pro. Unfortunately at the time I wasn’t aware of the sustained write speed issues so I just figured, pcie gen 4 is fastest, and then got the drive with the highest endurance.
Right now it’s still a bit of a scramble and everting is new and uncertain.
Besides the biggest bottleneck of all is your storage space and more specifically your ability(cashflow) to add more storage space.
So with that in mind I would opt right now for faster plotting and when it burns out and plotting is still interesting, and I have cash to buy more HDD then I’d go get myself some real enterprise ssd for the long haul
Edit: I just looked at the price…in that case you might as well get an entry level enterprise drive straight away…
Something like this perhaps?
I did read that it can be difficult to get Samsung drivers though, because unlike intel they don’t support end-users for these products
1 Like
Well they’ll burn out like any other SSD, and you could of course try to claim warranty but that’s kind of a dick move…
2 Likes
Just made an account here to tell you guys : This thread is gold. Thanks!
9 Likes
You nailed it, this is why I went with SSD’s for plotting instead of just a bunch of cheaper disks.
1 Like
