Micron 6600 ION 245TB SSD Announced

Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.

ca3y6

Well-Known Member
Apr 3, 2021
826
813
93
Sequential operations only. 40k IOPS, that's slower than many old SATA SSDs.
 

nabsltd

Well-Known Member
Jan 26, 2022
810
609
93
Sequential operations only. 40k IOPS, that's slower than many old SATA SSDs.
As the article said, it's only about storage density, because you can easily outperform this SSD with 10-20TB of fast NVMe as cache for 250TB of spinning rust.

The 6600 ION at max capacity will give you at least 4PB per 1U, while spinning rust maxes out at around 400TB per 1U (assuming a ~50-drive 4U chassis, with 30TB drives).
 
  • Like
Reactions: Patrick

i386

Well-Known Member
Mar 18, 2016
4,901
1,931
113
36
Germany
This is insane :D
My main fileserver/nas at home has 291 TByte usable space and the generation after this ssd could replace it with a single device...
 

ca3y6

Well-Known Member
Apr 3, 2021
826
813
93
maybe you have overseen the @ QD128 ?
not sure what you mean. If the drive caps at 40k iops with a QD128 that a sata ssd can achieve with QD32 or less, if anything it makes the matter slight worse (ie you would expect potentially even lower IOPS at a lower queue depth)?
 

RolloZ170

Well-Known Member
Apr 24, 2016
9,983
3,204
113
germany
not sure what you mean.
expected, but IOPS is generaly lower at enterprise/server SSD, but they buy it for some reason.
i doubt an old sata SSD controller can handel 245TB and if, will be less IOPS than this one.
 

ca3y6

Well-Known Member
Apr 3, 2021
826
813
93
The bottleneck here is unlikely to be the controller, it is the QLC and the likely large write amplification which kills endurance and IOPS.
 
  • Like
Reactions: nabsltd

nabsltd

Well-Known Member
Jan 26, 2022
810
609
93
It does appear the this drive is truly aimed at the massive source data required by AI, in that you could put all the data needed on the drive and treat it as WORM.

You could then use the data for training different models, giving each model very fast ingest.
 
  • Like
Reactions: nexox