Memory Speed matters for ZFS? (and samsung -yh9Q2?? mem)

[james23]

I'm aware of the general (limited) research and various opinions out there that: small bumps in memory speed do not equate too much real word performance gain (I'm referring to same generation, IE DDR3 1066 vs 1333 or vs 1600) I'm aware D3 vs. DDR4 *is* quite a real-world performance benefit (for other reasons too).

I also know this from my overclocking days and heavy memory benchmarking related to overclocking, that d3 1066 vs 1333mhz , even in synthetic benchmarks might give you 5 to 10% boost, if that. (but that may not translate at all into real-world apps/benefit)

For normal desktop or even server OSs memory speed may not be super critial, however I do know how important (and utilized) memory (ecc) is to ZFS. ie the 128gb D3 ECC 1066mhz I currently have in my very large Freenas server is always 100% in use by ARC. Does anyone know if upgrading from 1066 to 1333 makes much of an improvement for ZFS disk speeds specifically? (FreeBSD / freenas zfs)

Currently, I can add another 128g ( so would be a total of 8 sticks of 32GB ECC d3 1066), for $240.

I have another offer where I can buy 8x sticks of the same memory but at 1333 megahertz (not 1066) for $360, (thus replace my 128g 1033 sticks, with 256g of 1333 sticks.)

Thus my question of- if a small bump in memory speed actually translates to real performance improvements for something as memory-intensive as ZFS. Does anyone know?

I realize most people/posts say small mem speed bumps are not much of a realworld performance improvement (ie i read 1% to 4% if that), but there has to be a reason that faster memory is so expensive on servers (ie demand), and that people Spec servers for faster mem, and that all the major motherboard manufacturers put such detailed info / docs / tables on how to arrange your sticks to properly achieve a certain (higher) memory speed. My point is, all of that tells me there is a reason that ram speed does matter in servers. (or is it just the Database admins for whom ram speeds really effect?)

The other question I had is-
I know through experience (and learning the hard way) that on many supermicro boards, specifically the x9 series, it's super important to stick to exact memory models that are on the boards QVL. My question is on samsung sticks, i see this model (which is on the QVL):

M393B4G70DM0-YH9 (Samsung 32GB PC3L-10600R DDR3-1333 Registered ECC Memory)

but i also keep seeing this physical memory label out there:

M393B4G70DM0-YH9Q2 (Samsung 32GB PC3L-10600R DDR3-1333 Registered ECC Memory)

What is that Q2 at the end? I cannot find any info, anywhere on what the Q2 at the end means. Does anyone know? Even Samsung's very thorough PDF specs don't mention 02 at all (see links below).

Thanks!

Samsung M393B4G70DM0-YH9 links:

M393B4G70DM0-YH9 | RDIMM | Samsung Module | Semiconductor

datasheet1:
https://www.samsung.com/semiconduct...7/11/DS_DDR3_4Gb_D_die_RDIMM_Rev06_135V-0.pdf

ds2:
https://www.samsung.com/semiconductor/global.semi/file/spd/M393B4G70DM0-YH903_YK003-0.pdf

EDIT; i found this thread after i posted, although its not directly related/answered-
https://forums.servethehome.com/index.php?threads/how-much-does-memory-speed-matter.8347/

 

[T_Minus]

I'm aware of the general (limited) research and various opinions out there that: small bumps in memory speed do not equate too much real word performance gain (I'm referring to same generation, IE DDR3 1066 vs 1333 or vs 1600) I'm aware D3 vs. DDR4 *is* quite a real-world performance benefit (for other reasons too).

Did you miss the other thread where NEtFlix was able to pick up (iirc) 10-15Gbit by upgrading from 1866 to 2400?
I believe they also said going from DDR3 to DDR4 was no difference when they went 1600 to 1866 too.

There's no "blanket statement" of this vs. that due to the specifics of RAM (latency and frequency).

I also know this from my overclocking days and heavy memory benchmarking related to overclocking, that d3 1066 vs 1333mhz , even in synthetic benchmarks might give you 5 to 10% boost, if that. (but that may not translate at all into real-world apps/benefit)

Depending what you're doing 5 to 10% is a huge performance boost. If you have more than 1 device you're experienceing this on too, then obviously that's even greater increase in performance.

For normal desktop or even server OSs memory speed may not be super critial, however I do know how important (and utilized) memory (ecc) is to ZFS. ie the 128gb D3 ECC 1066mhz I currently have in my very large Freenas server is always 100% in use by ARC. Does anyone know if upgrading from 1066 to 1333 makes much of an improvement for ZFS disk speeds specifically? (FreeBSD / freenas zfs)

Going from 1066 to 1333 is not worth it.

Currently, I can add another 128g ( so would be a total of 8 sticks of 32GB ECC d3 1066), for $240.

I have another offer where I can buy 8x sticks of the same memory but at 1333 megahertz (not 1066) for $360, (thus replace my 128g 1033 sticks, with 256g of 1333 sticks.)

Thus my question of- if a small bump in memory speed actually translates to real performance improvements for something as memory-intensive as ZFS. Does anyone know?

I realize most people/posts say small mem speed bumps are not much of a realworld performance improvement (ie i read 1% to 4% if that), but there has to be a reason that faster memory is so expensive on servers (ie demand), and that people Spec servers for faster mem, and that all the major motherboard manufacturers put such detailed info / docs / tables on how to arrange your sticks to properly achieve a certain (higher) memory speed. My point is, all of that tells me there is a reason that ram speed does matter in servers. (or is it just the Database admins for whom ram speeds really effect?)

You should get 1866mhz DDR3 that's a significant boost over 1333.

The other question I had is-
I know through experience (and learning the hard way) that on many supermicro boards, specifically the x9 series, it's super important to stick to exact memory models that are on the boards QVL.

This is not true at all. Not sure where you heard that or what your experience is but I've literally used 100 of DIMMS in X9 and X10 and not once have I ever made sure it was on an approved list. I have generic with samsung\hynix\micron chips. I have branded samsung, hynix, micron, and a couple lower-tier brands I can't even recall. I've even gotten DIMMS via USPS with absolutely 0 packaging in a box, and all of those worked too (x9 \ DDR3).

I think you may be thinking of E3 x9 \ Kingston problems people have, but have never heard that with X9 E5 systems. I have kingston UDIMM and RDIMM that work too FWIW.

My question is on samsung sticks, i see this model (which is on the QVL):

M393B4G70DM0-YH9 (Samsung 32GB PC3L-10600R DDR3-1333 Registered ECC Memory)

but i also keep seeing this physical memory label out there:

M393B4G70DM0-YH9Q2 (Samsung 32GB PC3L-10600R DDR3-1333 Registered ECC Memory)

What is that Q2 at the end? I cannot find any info, anywhere on what the Q2 at the end means. Does anyone know? Even Samsung's very thorough PDF specs don't mention 02 at all (see links below).

Thanks!

Samsung M393B4G70DM0-YH9 links:

M393B4G70DM0-YH9 | RDIMM | Samsung Module | Semiconductor

datasheet1:
https://www.samsung.com/semiconduct...7/11/DS_DDR3_4Gb_D_die_RDIMM_Rev06_135V-0.pdf

ds2:
https://www.samsung.com/semiconductor/global.semi/file/spd/M393B4G70DM0-YH903_YK003-0.pdf

EDIT; i found this thread after i posted, although its not directly related/answered-
https://forums.servethehome.com/index.php?threads/how-much-does-memory-speed-matter.8347/