Enterprise SSD "small deals"

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

kapone

Well-Known Member
May 23, 2015
1,890
1,267
113
luckylinux said:
Main Issue in increasing Power is the low-Load [lack of] Efficiency with the "big" Wattage PSU.
Click to expand...
Yup. Most Supermicro PSUs > ~700w are less than 70-75% efficient when you're idling in the two digits. Even power hungry CPUs like the Ivy Bridge era, idle at less than 40w with a single CPU. Even with dual CPUs they idle ~78-80w (I've done extensive measurements over the years).

The SM "platinum" PSUs are only efficient in the 30-90% range, and generally in a homely setting, we don't drive them that way.
 
  • Like
Reactions: luckylinux
L

luckylinux

Well-Known Member
Mar 18, 2012
1,567
501
113
kapone said:
Yup. Most Supermicro PSUs > ~700w are less than 70-75% efficient when you're idling in the two digits. Even power hungry CPUs like the Ivy Bridge era, idle at less than 40w with a single CPU. Even with dual CPUs they idle ~78-80w (I've done extensive measurements over the years).

The SM "platinum" PSUs are only efficient in the 30-90% range, and generally in a homely setting, we don't drive them that way.
Click to expand...
That's why I always remove the Redundant PSU and in most Cases I'm replacing the PWS-920P-SQ with the PWS-501P-1R (500W) (for those with the PDU) or similar Wattage with Cables. Still a Transition Phase though ...
 
K

kapone

Well-Known Member
May 23, 2015
1,890
1,267
113
luckylinux said:
That's why I always remove the Redundant PSU and in most Cases I'm replacing the PWS-920P-SQ with the PWS-501P-1R (500W) (for those with the PDU) or similar Wattage with Cables. Still a Transition Phase though ...
Click to expand...
Those are all of the reasons, the big guys (i.e. the hyperscalers and the likes of Oxide) don't even have individual PSUs in the systems. DC busbars are the name of the game. Way more efficient.
 
L

lopgok

Active Member
Aug 14, 2017
245
171
43
kapone said:
Yup. Most Supermicro PSUs > ~700w are less than 70-75% efficient when you're idling in the two digits. Even power hungry CPUs like the Ivy Bridge era, idle at less than 40w with a single CPU. Even with dual CPUs they idle ~78-80w (I've done extensive measurements over the years).

The SM "platinum" PSUs are only efficient in the 30-90% range, and generally in a homely setting, we don't drive them that way.
Click to expand...
All 80+ rated power supplies need to meet their efficiency goal at 20%, except for titanium which needs to meet the goal at 10%.

See the chart at 80 PLUS certification specifications and ratings | CLEAResult

Platinum needs to be 90% efficient at 20% load.
Titanium needs to be 90% efficient at 10% load.
That is for 115v rated power supplies.
I don't have 1U or 2U servers, so I buy reasonably efficient ATX power supplies. My biggest power supply in use an a 550w platinum unit I use for my 64 core EPYC milan system.
 
  • Like
Reactions: Talyrius
L

luckylinux

Well-Known Member
Mar 18, 2012
1,567
501
113
lopgok said:
All 80+ rated power supplies need to meet their efficiency goal at 20%, except for titanium which needs to meet the goal at 10%.

See the chart at 80 PLUS certification specifications and ratings | CLEAResult

Platinum needs to be 90% efficient at 20% load.
Titanium needs to be 90% efficient at 10% load.
That is for 115v rated power supplies.
I don't have 1U or 2U servers, so I buy reasonably efficient ATX power supplies. My biggest power supply in use an a 550w platinum unit I use for my 64 core EPYC milan system.
Click to expand...
Sure, but it doesn't say how fast the Efficiency drops below that Point (< 20%) :) .

Plus you'll have to add the Efficiency of the PDU to generate 5V and 3.3V.

Anyways, we are going [very] Off-Topic.
 
A

asgardthor

New Member
Jan 24, 2020
10
5
3
luckylinux said:
That's why I always remove the Redundant PSU and in most Cases I'm replacing the PWS-920P-SQ with the PWS-501P-1R (500W) (for those with the PDU) or similar Wattage with Cables. Still a Transition Phase though ...
Click to expand...
This is interesting to me, I have 2x PWS-1K28P-SQ in my SuperMicro CSE 847 36 bay that idles roughly at 350 watts.
 
T

tubeit

New Member
Aug 1, 2021
14
22
3
asgardthor said:
This is interesting to me, I have 2x PWS-1K28P-SQ in my SuperMicro CSE 847 36 bay that idles roughly at 350 watts.
Click to expand...
I have a g292-z20 (which is a power hog), removing one of the two power supplies lets me idle at around 270W instead of 290W.
I'm quite sure that most of it is the CPU frequency/voltage being lowered by the system (I have the P-state managed by Linux, but removing the PSU changes the frequency a bit), but probably also some efficiency gain by using a single PSU.
 
N

nexox

Well-Known Member
May 3, 2023
1,986
990
113
asgardthor said:
This is interesting to me, I have 2x PWS-1K28P-SQ in my SuperMicro CSE 847 36 bay that idles roughly at 350 watts.
Click to expand...
If you have accurate power metering at the wall (IPMI might be close enough, mine matches my meter, others don't trust theirs much) you can quickly test pulling one PSU to see how much the idle power drops from pushing the load over 20% on the remaining PSU. I personally run my 826 with one PSU pulled out a half inch so I can manually swap to it in the rare case that the primary PSU dies.
 
  • Like
Reactions: luckylinux
L

luckylinux

Well-Known Member
Mar 18, 2012
1,567
501
113
tubeit said:
I have a g292-z20 (which is a power hog), removing one of the two power supplies lets me idle at around 270W instead of 290W.
I'm quite sure that most of it is the CPU frequency/voltage being lowered by the system (I have the P-state managed by Linux, but removing the PSU changes the frequency a bit), but probably also some efficiency gain by using a single PSU.
Click to expand...
That's NOT why the Power Drops by removing one PSU.

If you remove one PSU, you take away a "Offset" Power Consumption that every switching Power Supply (or any Power Supply for that Matter) has: Auxiliary Power Consumption. Controller, Fan, the switching Effect and the losses it Generates in MOSFETs, Inductors, Capacitors, etc.

Furthermore, at low Load (in %), the PSU is NOT very efficient. In particular below 10% but also between 10% and 20%. For the same Reason (you still have Auxiliary and Filter Currents).

So if you have 2 x PSU each operating at 10% the efficiency will be FAR below that of a single PSU (of the same Model) being operated at 20% Load. Similar story but even bigger change if you do 2 x 5% Load << efficiency of 1 x 10% Load. The story reverses to some extent at high load, where 2 x PSUs can be more efficient, but then we are talking about each PSU being operated at > 40 - 50% compared to a single PSU being operated at 60-100%, the latter of which has its own Set of Issues ;).

The efficiency curve follows (approximately) an inverted Parabola. Reality is more complex than that of course ;) .

Just one Example of many:
1755550094130.png
 
A

asgardthor

New Member
Jan 24, 2020
10
5
3
This is getting above my head haha
I would say this my measurement is decently accurate, it's from the unifi PDU, and adding up both wattages
 
  • Haha
Reactions: luckylinux
C

Cruzader

Well-Known Member
Jan 1, 2021
953
944
93
luckylinux said:
How good are these Prices ? It seems VERY good to me. Where's the Catch ? Old Generation, Compatibility Issues, etc ?
Click to expand...
The catch is that they are lower performance and endurance than the typical 1.6tb/3.2tb cards.

There are also still a bunch of surplus/new stock optane hitting the market, so you can get a 905p 280gb at 90-100$ while you tend to not know the wear of used cards.
For endurance there are also surplus/new P4800x 375gb u.2 drives at around 120-130$ with a nice 20.5PBW (and x4 cards at 180-190$).
 
  • Like
Reactions: nexox and luckylinux
You must log in or register to reply here.
Top Bottom