Intel Optane 900p vs 4800X

[oddball]

I've done some digging, and it looks like the 900p is a cheaper version of the 4800X.

The biggest difference in the two is endurance, the 900p can sustain 10 drive writes per day, whereas the 4800X can sustain 30 drive writes per day.

I'm looking at these as a fast storage for SQL Server logs. When writing we queue up and this becomes the bottle neck.

Does the endurance matter if I'm only writing .1 DWPD on this?

You can find these for about $1/GB on eBay brand new, which seems reasonable. It's slightly more expensive than a Samsung 960 Pro, but the performance is better too.

Thoughts?

 

[Monoman]

Confirmed PLP on the 4800x and DWPD seem to be it for the most part.

 

[MikeWebb]

I wish I had a 4800X to test with, I only have the 900P. So I'll put up some of my digging for that in the hope that someone with it's big brother could do the same. There appears to be more then just PLP.

root@pve:~# nvme list
Node             SN                   Model                                    Namespace Usage                      Format           FW Rev
---------------- -------------------- ---------------------------------------- --------- -------------------------- ---------------- --------
/dev/nvme0n1     PHMB742300FA280CGN   INTEL SSDPED1D280GA                      1         280.07  GB / 280.07  GB    512   B +  0 B   E2010325

root@pve:~# nvme id-ns /dev/nvme0n1 -H
NVME Identify Namespace 1:
nsze    : 0x209a97b0
ncap    : 0x209a97b0
nuse    : 0x209a97b0
nsfeat  : 0
  [2:2] : 0     Deallocated or Unwritten Logical Block error Not Supported
  [1:1] : 0     Namespace uses AWUN, AWUPF, and ACWU
  [0:0] : 0     Thin Provisioning Not Supported

nlbaf   : 0
flbas   : 0
  [4:4] : 0     Metadata Transferred in Separate Contiguous Buffer
  [3:0] : 0     Current LBA Format Selected

mc      : 0
  [1:1] : 0     Metadata Pointer Not Supported
  [0:0] : 0     Metadata as Part of Extended Data LBA Not Supported

dpc     : 0
  [4:4] : 0     Protection Information Transferred as Last 8 Bytes of Metadata Not Supported
  [3:3] : 0     Protection Information Transferred as First 8 Bytes of Metadata Not Supported
  [2:2] : 0     Protection Information Type 3 Not Supported
  [1:1] : 0     Protection Information Type 2 Not Supported
  [0:0] : 0     Protection Information Type 1 Not Supported

dps     : 0
  [3:3] : 0     Protection Information is Transferred as Last 8 Bytes of Metadata
  [2:0] : 0     Protection Information Disabled

nmic    : 0
  [0:0] : 0     Namespace Multipath Not Capable

rescap  : 0
  [6:6] : 0     Exclusive Access - All Registrants Not Supported
  [5:5] : 0     Write Exclusive - All Registrants Not Supported
  [4:4] : 0     Exclusive Access - Registrants Only Not Supported
  [3:3] : 0     Write Exclusive - Registrants Only Not Supported
  [2:2] : 0     Exclusive Access Not Supported
  [1:1] : 0     Write Exclusive Not Supported
  [0:0] : 0     Persist Through Power Loss Not Supported

fpi     : 0
  [7:7] : 0     Format Progress Indicator Not Supported

nawun   : 0
nawupf  : 0
nacwu   : 0
nabsn   : 0
nabo    : 0
nabspf  : 0
nvmcap  : 0
nguid   : 00000000000000000000000000000000
eui64   : 0000000000000000
LBA Format  0 : Metadata Size: 0   bytes - Data Size: 512 bytes - Relative Performance: 0x2 Good (in use)

That last bit is interesting. Basically a lbaf with a lbads:9. Not what I was expecting from a "NVMe SSD" and fdisk showed:

sectors of 1 * 512 = 512 bytes Sector size (logical/physical): 512 bytes / 512 bytes I/O size (minimum/optimal): 512 bytes / 512 bytes

(
Usually SSD reports Sector Sizes: 512 bytes logical, 4096 bytes physical i.e lbads:12. I tried to nvme format it 4096 and it would not have a bar of it. So a lbads of 512 would mean a bit more juggling if your use case was optimized for 4K and I would say, unlike a normal SSD, don't do ashift=12 when creating a zpool, this would create performance issues (I need to quantify this, but I am looking heavily into what is needed to optimize ZFS).

 

[Rand__]

On this thread : https://forums.servethehome.com/index.php?threads/intel-optane-p4800x-375gb-800.20266/#post-192418
@timd and @BackupProphet said they have (had) access to one ... maybe one of them still has ?

And I'd be interested in quantification of ashift - on my very first (physical) Freenas box i was looking heavily into optimization too but since I kind of skipped these - so I wonder whether it makes 10% difference or less

 

[BackupProphet]

I don't have the p4800x, but Patrick had one that he allowed me to do some testing on. Anyway, the difference between the p4800x and 900p is less than 1% when it comes to performance. However you get better DWPD and you can use the p4800x as a memory drive

Optane has no need for PLP.

For SQL Server logs, the Optane 900p is plenty.

 

[_alex]

For the ,memory drive‘, is this just a software/driver limitation?

 

[MikeWebb]

Found this on github:

namespace operation on optane 900P? · Issue #346 · linux-nvme/nvme-cli

Pretty much confirms what I was seeing and the link from there to the intel forum was helpful...more then the 900P product brief was. My assumption that the 900P would be 4K bs was based on my misinterpretation of the product briefs 4K random reads/writes performance (DUH?)

Still, the brute force performance makes up for any additional checksum overhead of 512 vs 4096 in ZFS for my needs (and budget)...I....Just..always..want..MORE.

I might use this as a reason to play with some variables etc now that I've just learnt about fio2gnuplot....after I find a way to move some data around.

 

[jerrytsao]

Found this on github:

namespace operation on optane 900P? · Issue #346 · linux-nvme/nvme-cli

Pretty much confirms what I was seeing and the link from there to the intel forum was helpful...more then the 900P product brief was. My assumption that the 900P would be 4K bs was based on my misinterpretation of the product briefs 4K random reads/writes performance (DUH?)

Still, the brute force performance makes up for any additional checksum overhead of 512 vs 4096 in ZFS for my needs (and budget)...I....Just..always..want..MORE.

I might use this as a reason to play with some variables etc now that I've just learnt about fio2gnuplot....after I find a way to move some data around.

Intel Consumer SSD (except 750) would never support variable sector sizes, it's always been DC only and can be easily changed in ISDCT.

 

[T_Minus]

Probably worthwhile to link to this to save some clicking too: set sector/block size in 900P in ubuntu. |Intel Communities