Tiered storage sizing recommendations

[CyklonDX]

I just checked. You don't need to over provision that much.
1.92T 1DWPD drives have about 15% over provision
3DWPD drives have 37% over provision. You don't need to cut the drive by half to get 3 DWPD!

Go ahead and test that.
I know Samsung claimed some stuff - but if you are reading pdf meant for 845DC - i think its only meant for that specific model. (as i recall its for 10 dwpd disk); they already overprovisoned that disk many times over (it has 8x 256G memory chips) (2TB overall for 512/400G disk) so if you cut out additional 10% it has enormous benefits in terms of endurance. (giving you 2x original dwpd - if you over-provision additional 6.7%, and 5x original dwpd if you over-provision 28% additionally. (i personally been running op pm983a's nvme's, and their endurance has only risen 2x at most for half the space)

 

[DaveLTX]

Go ahead and test that.
I know Samsung claimed some stuff - but if you are reading pdf meant for 845DC - i think its only meant for that specific model. (as i recall its for 10 dwpd disk); they already overprovisoned that disk many times over (it has 8x 256G memory chips) (2TB overall for 512/400G disk) so if you cut out additional 10% it has enormous benefits in terms of endurance. (giving you 2x original dwpd - if you over-provision additional 6.7%, and 5x original dwpd if you over-provision 28% additionally. (i personally been running op pm983a's nvme's, and their endurance has only risen 2x at most for half the space)

Go check TPU's SSD database.
Even D7-P5810 is a pSLC with 37% overprovisioning

 

[CyklonDX]

got pcb diagram/shots? (btw this u.2 uses SLC chips so their endurance on their own is likely a killer too)

 

[DaveLTX]

got pcb diagram/shots? (btw this u.2 uses SLC chips so their endurance on their own is likely a killer too)

They have info on the number of nand used and etc
5810 doesn't use SLC. It uses QLC as qSLC which isn't the same thing as Kioxia XL flash or Micron XTR. Slower! Lower endurance as well

Which is a increasing problem as I realized their 61.44TB SSD is actually a PLC running as pQLC
Ditto for the other QLC drive

If you think they use that many dies, you're talking out of your rear unfortunately
Best example is PM1733 and PM1735, they're EXACTLY THE SAME. Same nand and everything. You can even go into CLI and reconfigure the 1735 1.6 back to 1.92 so on and so forth.

 

[CyklonDX]

They have info on the number of nand used and etc
5810 doesn't use SLC. It uses QLC as qSLC which isn't the same thing as Kioxia XL flash or Micron XTR. Slower! Lower endurance as well

Which is a increasing problem as I realized their 61.44TB SSD is actually a PLC running as pQLC
Ditto for the other QLC drive

If you think they use that many dies, you're talking out of your rear unfortunately
Best example is PM1733 and PM1735, they're EXACTLY THE SAME. Same nand and everything. You can even go into CLI and reconfigure the 1735 1.6 back to 1.92 so on and so forth.

(in vid they also say slc)

( P5336 uses QLC )



// i typically check pcb, memory chip part number to see the actual specification (how many writes), and capacity of the chips, add them together - not what onboard management chipset tells me.

 

[DaveLTX]

View attachment 35213

(in vid they also say slc)

( P5336 uses QLC )



// i typically check pcb, memory chip part number to see the actual specification (how many writes), and capacity of the chips, add them together - not what onboard management chipset tells me.

Go check TPU SSD database. It's fully pseudo SLC made from QLC flash.
It uses N38A flash which is very specifically originally, all I said is that it's not true SLC but marketing can indeed say it's SLC
Is it actually SLC? No it isn't.
N38A is also used as pseudo TLC in P5510

 

[CyklonDX]

TPU db is not trustworthy if device doesn't include all details. (if they didn't have actual hardware on hand)
Its basically editorial of what they think it is, based on article/detail they could found online.


(it doesn't even fully line up with details in product brief)

https://www.solidigm.com/content/dam/solidigm/en/site/products/data-center/d7/p5810/documents/solidigm-d7-p5810-product-brief.pdf

 

[DaveLTX]

TPU db is not trustworthy if device doesn't include all details. (if they didn't have actual hardware on hand)
Its basically editorial of what they think it is, based on article/detail they could found online.

View attachment 35216

All you need to do is. Understand that you were wrong in thinking you needed that much over provisioning. That hard?
Now you're most insistent site is wrong blablabla

 

[CyklonDX]

All you need to do is. Understand that you were wrong in thinking you needed that much over provisioning. That hard?
Now you're most insistent site is wrong blablabla

I never looked at it from right / wrong -
I still stand on part that your typical 1DWPD SSD won't become 3DWPD if you overprovison additional 20-30% of that disk. The PBW endurance comes from multiple factors from chips cell endurance, to how much the disk originally had been overprovisioned on the disk. (Not based on some calculation or data available online for another disk ~ this is especially a thing in enterprise ~ but by physical check of the disk.)
(and it all differs by the disk you are looking at)

Next part, TPU database is not always accurate if there are blanks / if they haven't gotten good source - and I've shown you that even product brief from manufacturer doesn't match the specs they had outlined in TPU specs it supposedly has. TPU is not always accurate if there's no data.

(at this moment i have 2 unused on hand sas ssd 1.6tb px04svb160 from toshiba - and i'm more happy to open it up and show you what i see.)
(they are 8.7PBW endurance disks | 3DWPD for 5 years - an older cheap part)