Brocade ICX Series (cheap & powerful 10gbE/40gbE switching)

[fohdeesha]

NOTE #1: do not PM me with switch questions, they will be ignored - post them in this public thread, where hundreds of other members can also answer, and the answer will be public for future users

NOTE #2 06-22-2023: Yes, this post is still up to date and nothing has changed: in fact judging by google analytics, they're more popular than ever. A v2 of the popular config guide has just been published, and I've added licensing information to it so you can unlock your switches on your own

Firmware + Docs + Feature Licenses:​

NOTE: If you buy any of these, use the setup guide at the following link. It will get you set up with a fresh slate, the latest firmware, & the latest documentation. It will also get you fully licensed if your switch requires it. Please do not post questions unless you've followed this first to get set up on the latest software:

---- Setup / Config / Licensing Guide v2 ----​

Starting a master thread for a few of my favorite affordable switches - I've been using and reverse engineering this switch line for years and figured it's time to share the goodies. The prices below fluctuate a lot with demand and with the best offer feature can usually be beaten. Switches are listed in order of raw capacity/capability, not price.

Choice Beef Cuts:

Brocade FCX - ~$30 on ebay

​

• original beef
• 24/48 1gbE copper (PoE available)
• I recommend these as a very cheap platform to learn networking with (as they support everything, including BGP), or a lab box. For switches to use in your permanent home network, you might want to look below at the newer models
• Full layer 3, IPV4 + IPV6 routing, L2/L3/L4 ACL's, VRRP, OSPF, SNMP, sflow, VRFs, tunnels, BGP
• No 10gbE by default. See Here for notes on adding 10gbE cards
• The non-PoE models are a little quieter than a dell R710 or ICX6610
• The PoE models are the loudest switch I've ever used - not recommended for home
• Fans cannot be modified
• ~40w power draw
• 2x redundant hot-swap PSU's
• Aggregate capacity: 152gbps / 114Mpps (wirespeed regardless of features enabled)
• Again, these are best for a lab environment on a budget where you don't care about a bunch of 10gbE or noise, but want advanced networking features such as BGP and VRFs
• 
  
• Datasheet

Brocade ICX7150-C12P - ~$210 on ebay

​

• microbeefer
• 12x 1gbE copper, all PoE
• 2x 10gbE SFP+
• Full layer 3, IPV4 + IPV6 routing, L2/L3/L4 ACL's, VRRP, OSPF, SNMP, sflow, all the usual
• 12w power draw
• totally fanless & silent
• single built in PSU
• Aggregate capacity: 68gbps / 51Mpps (wirespeed regardless of features enabled)
• 
  
• Datasheet (you will notice other cool looking 7150 models, but I have yet to see any list for less than $1k)
• STH post with detailed information

Brocade ICX6450 - ~$120 on ebay

​

• the beef snack
• 24/48 1gbE copper (PoE available)
• 4x 10gbE SFP+
• Full layer 3, IPV4 + IPV6 routing, L2/L3/L4 ACL's, VRRP, OSPF, SNMP, sflow, all the usual
• 25w power draw for the 24-port models with or without PoE
• 50w power draw for the 48-port models with or without PoE
• 1 small fan, nearly silent (48 port and PoE models have 2 or 3 fans, but they can be removed if you're not using near full PoE load)
• single built in PSU
• fans can be modified/replaced to run even quieter/near silent
• Aggregate capacity: 176gbps / 132Mpps (wirespeed regardless of features enabled)
• 
  
• Datasheet
• FAQ
• Architecture Brief

Brocade ICX7250 - ~$300 on ebay

​

• medium beef
• 24/48 1gbE copper (PoE available)
• 8x 10gbE SFP+
• Full layer 3, IPV4 + IPV6 routing, L2/L3/L4 ACL's, VRRP, OSPF, SNMP, sflow, all the usual, but also adding VRFs and tunnels (but no BGP)
• 50w power draw for the 24 port
• 65w power draw for the 48 port
• nearly silent (same sound level as the ICX6450)
• single built in PSU
• fans can be modified/replaced to run even quieter/near silent
• Aggregate capacity: 256gbps / 190Mpps (wirespeed regardless of features enabled)
• 
  
• Datasheet
• STH post with detailed information

Brocade ICX6610 - ~$200 on ebay

​

• the BEEF KING
• 24/48 1gbE copper (PoE available)
• 16x 10gbE (8x SFP+ in the front, 8x via 2 QSFP+ breakout ports on the rear)
• 2x 40gbE (separate from the previously mentioned breakout ports)
• Supports OpenFlow in hardware for SDN, including hybrid port mode
• SupportS MACSEC on the SFP+ ports for 80gbps of real time L2 AES-128 encryption
• Same OS features as ICX6450 but adds advanced protocols like BGP, VRFs, tunnels, everything
• 80w power draw for the 24 port models with or without PoE
• 110w power draw for the 48 port models with or without PoE
• audible - about the same as an R710, little quieter than LB6M
• 2x redundant hot-swap PSU's
• Fans cannot be modified
• Aggregate capacity: 528gbps / 396Mpps (wirespeed regardless of features enabled)
• 
  
• Datasheet
• FAQ
• Architecture Brief
• (note: when the above PDF's say the QSFP ports can only be used for stacking, they're lying)

Brocade ICX6650 - ~$600 on ebay

• beef overload
• 64x 10gbE (56x SFP+ on front, 8x via 2x QSFP breakout on the rear)
• 4x 40gbE via rear QSFP (separate from QSFP breakout ports)
• Supports OpenFlow in hardware for SDN, including hybrid port mode
• Same OS features as ICX6610 - everything including advanced protocols like BGP, VRFs, tunnels, etc
• 120w power draw
• audible - louder than the ICX6610, not screaming but would not want it next to you
• 2x redundant hot-swap PSU's
• Fans cannot be modified
• Aggregate capacity: 1600gbps / 1190Mpps (wirespeed regardless of features enabled)
• does NOT support stacking
• DOES support MCT / MC-LAG (cross-chassis LACP)
• These start to lose out cost wise compared to some Arista models with the same port count, such as the 7050T-64 which are a very good value - it's up to you if you prefer Arista's EOS or Brocade's FastIron OS
• 
  
• Datasheet
• STH post with detailed information

Software:​

These all run Brocade's FastIron OS. If you've done the LB6M flash, it's the same as that (but a much newer version). If you've ever used cisco gear, the CLI is about 90% identical. It's still under active development for all of these switches (most recent firmware release for them was 2 weeks ago).

The firmware/docs are freely available from Ruckus's site (who acquired these from brocade), you just need to make a free account. I also mirror an easy to grab ZIP of the latest FW and docu on my update guide linked above.

Since it's been asked before, yes - all of these are proper ASIC L3 routers that do everything in hardware at line rate, with no oversubscription. Nothing is punted to the management CPU.

These are enterprise/datacenter switches, so the main focus is the CLI interface. However they do have a web UI - granted it's very minimal. You can see screenshots here: Imgur

Optics:​

Brocade switches will take any make of optics, the brand does not matter. I've been using quite a few of the $7 dollar 10gbase-SR avago/jdsu/etc modules off of ebay with no issues for a few years. However since "official" Brocade SFP+ modules have come down in price to around $8, I would recommend those as they unlock optical monitoring, so you can see link strength, module temp, etc. They will take any manner of DAC's as well. Same goes for the 40gbE QSFP+ ports on the rear of the 6610 - optics or DACs, your choice.

NOTE: All Brocade switches are compatible with the Mikrotik S+RJ10 SFP+ 10gbase-T copper module, if you want to cheaply turn some of the sfp+ ports into rj45 10gbE copper ports: Mikrotik 6-Speed Rj-45 Module Up To 10Gbps Speeds

Other:​

I've tried to keep these a secret hoping to not wreck the used market for them, but there seems to be enough on ebay now it shouldn't matter. Many STH members have already been using them and discussing them in PM's, and I figured all the info and stuff we've shared privately would be of much more use out in the open. I have reverse engineered quite a bit of all these switches and found some things that are nowhere else on the internet - I will post the discoveries I think won't get me in trouble here as I find time

Useful Posts:

Tutorial videos for the FastIron OS

Basic guide on ACL's

Flash the cheap 649281-B21 NIC to 40gbE

Edit *nix SSH config to not warn for switch connections

Per-VLAN MTU info

Long distance ICX6610 stacking using MPO/MTP fiber

Mikrotik vs Brocade

Editing DAC/Optics vendor info

Configure ICX6610 QSFP ports for stacking AND breakout simultaneously

Extracting Linux Based Firmware Images​

 

[fohdeesha]

Power Supplies:​

The more powerful ICX6610 has the typical dual redundant PSU hotswap bricks you're all used to. However the ICX6450 only has one built in PSU, which initially worried me (I prefer redundancy), so I opened it up and did some poking around.

It's a very nice OEM Delta brick with high quality capacitors, well-heatsunk MOSFETs, etc. It also outputs a simple 12VDC with no extra sense pins - making this very easy to modify if you want to run it off of 12 volts. (note: this is for the non-PoE version. the PoE model will have a second 54VDC output).

If you want to run the ICX6450 off of 12 volts DC, just open it up and disconnect/ splice this wire bundle out the back, and that's it! You can see them in the picture below, there's four wires, but each pair is commoned to the same PSU pins (2x black grounds and 2x 12V positive). The power supply is very efficient - you can see it pulling 1.6a at 12v (19 watts) and it was only drawing 21 watts from the wall (plus or minus a couple watts):

 

[fohdeesha]

(ICX6450 & ICX7xxx) Accessing Linux​

The big guy, the ICX6610, runs on a PowerPC processor with it's own proprietary bootloader and FastIron OS. However Brocade eventually switched to ARM platforms (the little ICX6450 is one of them), and at the same time switched to using the popular u-boot bootloader, which then boots good ole Linux - Fastiron then runs as a monolithic application on top of Linux. This is all transparent of course, all you will ever see is FastIron like you're used to.

Note: because the ARM platform firmware is a heavily modified u-boot and linux, both of which are GPL, this means Arris (the parent company of Ruckus, who owns this switching line now) are legally obligated to provide the source code for said components. If you check Arris' sourceforge site, they are actually quite diligent about providing product source code and complying with open source licensing. I contacted their open source contact a month ago, and was told the FastIron-related sources are in the process of being added.


After some poking around in a dump of a running ICX6450's RAM, I stumbled upon the INIT script used to start the linux system. Of interest were these statements (trimmed for brevity):

#If noautostart is set in uboot, stop at linux prompt
cat /proc/cmdline |grep "noautostart" 1>/dev/null 2>&1 if [ $? -eq 0 ] then START_FI_AUTO=0

#If enabletelnet is set in uboot, then start telnet daemon
cat /proc/cmdline |grep "enabletelnet" 1>/dev/null 2>&1 if [ $? -eq 0 ] then
/usr/sbin/telnetd -l /bin/sh -f /etc/welcome

It's looking at the boot arguments passed by the bootloader for a noautostart or enabletelnet string. If it sees the noautostart string, it does not start the Fastiron app and drops you right into linux. If it sees the enabletelnet string, it starts a hidden telnet server with raw access to linux. These are our two ways in.

Method 1: Disable FastIron AutoStart​

It should be obvious, but you'll need a serial cable for both these methods. This method tells the switch not to boot the Fastiron app, and drops you right into linux. You will only have serial access to the switch, the usual networking software will NOT be loaded.

Get into the bootloader like usual (hit b during boot), then run the following:

setenv extra_bootargs noautostart
boot

note: this only sets this argument once and does not save it. So on subsequent reboots or power-ons, the string will be gone and it will boot normally.

It will begin booting like normal, but eventually you'll land in a shell:

BusyBox v1.18.3 (2013-11-22 14:33:50 IST) built-in shell (ash)
Enter 'help' for a list of built-in commands.

/ #

Press tab twice and it'll show you all possible commands (just typing help will not show you all commands), there's a couple hundred, most of the utilities you'd expect including wget, curl, vi, and ftpput.

Check out the filesystem:

/ # ls
FastIron.xz  etc          home         mnt          sbin         tmp
bin          fast_iron    init         proc         sys          usr
dev          ftp          lib          root         tftpboot     var
/ #

In the fast_iron directory, you'll find the startup configs you save from the normal switch CLI, as well as some logs. To reboot back to normal FastIron, just issue the reboot command.

Method 2: Enable Hidden Telnet Server​

This method tells the switch to start a hidden telnet server attached to the underlying Linux OS. It's only accessible via the management port (this is the only networking port Linux can see). The IP for Linux (where the telnet server will be) is set by a bootloader variable.

The upside of this method is that FastIron will still start, so you'll still have a normal functioning switch. There will just be an (unsecure) telnet daemon running on the management port with raw access to the underlying Linux OS - great for live debugging.

Get into the bootloader like usual (hit b during boot), then run the following. For the IP, choose an IP that is not in use anywhere else, especially on the switch. If you have a management IP assigned to the switch, this is seperate from that and needs to be a different IP:

setenv extra_bootargs enabletelnet
setenv ipaddr 192.168.1.57
boot

note: this only sets these arguments once and does not save them. So on subsequent reboots or power-ons, the string will be gone and it will boot normally.

The switch will boot as normal, except now the underlying Linux OS has launched a telnet server at the specified IP, and is accessible via the management port only. When you telnet to it (default telnet port), you will have a full Linux CLI:

Welcome to FastIron Board's Telnet session
           _  _
          | ||_|
          | | _ ____  _   _  _  _
          | || |  _ | | | | / /
          | || | | | | |_| |/
          |_||_|_| |_|____|_/_/

          On Brocade's FastIron board


BusyBox v1.18.3 (2013-11-22 14:33:50 IST) built-in shell (ash)
Enter 'help' for a list of built-in commands.

/ # ls
FastIron   etc        home       mnt        sbin       tmp
bin        fast_iron  init       proc       sys        usr
dev        ftp        lib        root       tftpboot   var
/ #

What it took to figure this out:

​

 

[fohdeesha]

(ICX6610) Rear QSFP+ Ports​

Describing the rear ports can be a little confusing in text, so below is an extremely professional graphic showing how they break down. 2 of them are 40gbE only ports - for your freenas, hypervisors, whatever. The other 2 are 4x 10gbE breakout (each) only:

To break out those 2x breakout ports, you have many options:

You can use a DAC breakout cable: Dell Networking Cable QSFP+ 40GBE - 4X SFP+ 10GBE 3M 27GG5 | eBay

Or you can use a 40gbE optic to give you an MPO fiber connector: Brocade XBR-000232 QSFP 57-1000267-01 4*16Gb SWL 100m for DCX8510 EMC | eBay

Note: the above optic is a 4x 16gbps fibrechannel optic, which makes them available for super cheap. They work perfectly fine in brocade ethernet switches, as they are not picky and simply clock the optic down to 10gbps per channel. I have been using quite a few for a while now

Then using an optic like above, use MPO fiber like this: 5m 50/125 Standard MTP Multimode Fiber Optic Patch Cable Key-up to Key-down | eBay

To go to an MPO breakout panel like this: Systimax SCS InstaPATCH 24-Port Fiber Patch Panel Module 642337970740 | eBay

A single one of those panels will break out 4x MPO connectors - with 4x 10gbe LC connectors per MPO, that's 16x 10gbe ports - for instance if you had a stack of 2x ICX6610's - that's 4x total QSFP+ breakout ports, to one panel like this.

In each of the above, each broken out 10gbE is totally seperate, and can go to totally different devices. I know this is obvious to most of you, but some here have never had to deal with QSFP+ breakout before

you can see how the above ports (including breakout) show up in the OS as unique ports:

DEV-ICX5#show interface brief ethernet 1/2/1 to 1/2/10

Port       Link    State   Dupl Speed Trunk Tag Pvid Pri
1/2/1      Up      Forward Full 40G   None  No  N/A  0
1/2/2      Up      Forward Full 10G   None  No  N/A  0
1/2/3      Up      Forward Full 10G   None  No  N/A  0
1/2/4      Up      Forward Full 10G   None  No  N/A  0
1/2/5      Up      Forward Full 10G   None  No  N/A  0
1/2/6      Down    None    None None  None  No  1    0
1/2/7      Up      Forward Full 10G   1     No  1    0
1/2/8      Up      Forward Full 10G   1     No  1    0
1/2/9      Up      Forward Full 10G   1     No  1    0
1/2/10     Up      Forward Full 10G   1     No  1    0

 

[nthu9280]

Thank you @fohdeesha for all the help with this switch to the STH community.
I just got a 48P POE unit few days ago and still have not connected the 10G and 40G interfaces. It came with 2PS and 1 Fan and fully licensed. I've disconnected the power to one of the PSU. There is a 15~20W additional draw when both are connected. Assume the PSU fan is major reason for that. Unlike the servers where 2nd PS is on hot standby (at least the ones I have), both the PSUs are providing 50% of the power on this switch. I didn't like the stress it would put on the rack ears due to it's weight. I has a spare UPS fixed rail set and used that. I'm sure it will be just fine for mounting on the rack ears. This belongs in some enclosed area / basement away from living areas due to heat and noise.

Here is a quick summary of different ICX 6610 models from the Brocade doc. While no guarantees on the used market on how accurate the sellers list their items, if you see ICX 6610-48P-PE or -PI that should have the ADV license with all the features.

Five models are available in the Brocade ICX 6610 family. Each model is orderable with either power supply side intake (-I) or power
supply side exhaust (-E) airflow:
••Brocade ICX 6610-24-I /ICX 6610-24-E with 24 10/100/1000 Mbps RJ-45 ports
••Brocade ICX 6610-48-I /ICX 6610-48-E with 48 10/100/1000 Mbps RJ-45 ports
••Brocade ICX 6610-24P-I/ICX 6610-24P-E with 24 10/100/1000 Mbps RJ-45 PoE+ ports
••Brocade ICX 6610-48P-I/ICX 6610-48P-E with 24 10/100/1000 Mbps RJ-45 PoE+ ports
••Brocade ICX 6610-24F-I/ICX 6610-24F-E with 24 100/1000 Mbps Small Form-Factor Pluggable (SFP) ports

 

[fohdeesha]

forgot to mention: licenses for these models are free for existing STH members - now that they are EOL & discontinued I don't really have an alternative to giving them away

 

[BackupProphet]

I've disconnected the power to one of the PSU. There is a 15~20W additional draw when both are connected. Assume the PSU fan is major reason for that. Unlike the servers where 2nd PS is on hot standby (at least the ones I have), both the PSUs are providing 50% of the power on this switch. I

Did you open the chassis and disconnect it internally or did you just not plug in the second power cable?

 

[nthu9280]

Just unplugged the cable. 24P would have been more than enough for me. When I got mine there were no 24P listed. I'll check if disabling is an option and if it makes any difference in heat & power draw.

Another thing to consider for folks installing this in the same rack with rest of the servers - they are exhausting to the back of the rack and if select E version at the top, the switch is pulling that warm air. I would have liked couple inches more space in the front of my rack so I could mount mine facing front so it also exhausts to the back of the rack.

 

[fohdeesha]

If you want to "disable" or disconnect the second power supply just pull it out of the switch no need to open it up, they're hot swap. Running on one power supply doesn't change anything about the functionality of the switch, it's still fully operational, you just wont have failover if the psu fails

 

[whitey]

Absolutely LUV my 6610-24P's...so much that I retired my Juniper EX4300, tip of the hat and BIG thanks to @fohdeesha for helping me out a month or so back to bring a couple of mine 'back to life' with a sideload of boot rom. These things are BEASTS, easy config cli, 264Gbps line rate switching non-oversubscribed...not much else to say other than these things are INSANE price/performance bang-for-buck!!!

 

[nthu9280]

I left the PSU in the unit but unplugged. My thought was not to alter the airflow pattern since I don't have a blank plate to cover the opening.

 

[anomaly]

This is great. You convinced me (as you already know!) to grab a ICX 6450-24. I was tempted to get the 48 port non PoE version but I need to know the power consumption of those (I might grab one down the line).

Got it for 130 USD SHIPPED to Europe, with GSP.... so that includes taxes. Awesome!

Also, easy peasy modification for the fans: you can use Sunon low noise 40mm fans to substitute the screamers. I can put some drop-in replacements listed here and where to grab them later.

The noise figures for all these:
https://www.flyteccomputers.com/ext/brocade/brocade-icx-6430-6450-ds.pdf

Power and acoustic specifications:

 

[fohdeesha]

Glad to hear you found a good deal!

Also, easy peasy modification for the fans: you can use Sunon low noise 40mm fans to substitute the screamers.

have you actually heard the stock fan yet? Screamer isn't exactly the term I would use, there's a stack of 3 of these switches 4 feet behind me and I literally can't hear them. The noise measurements in the spec sheet are for the fan at max speed, which is what it would ramp up to if you started doing line rate switching on every port simultaneously (160gbps or so). Needless to say I doubt that will happen, I've never seen the fan go over 30% or so

 

[anomaly]

Another, different topic:

POWER SUPPLY OPTIONS:

1. MeanWell: they make 1U ready PFC-capable, very efficient 12V (and multi output) power supplies. They have superb filtering at both AC and DC stages so the EMI is non existent (and I have equipment to test, I do RF stuff as a hobby).
   1. Over 90% efficiency at 12v: https://www.tme.eu/en/Document/920b959fa94fff6729ddb7d92aa45670/HEP-100-spec.pdf (there are others!)

I will probably power mine through a custom built PDU for 12v I am designing, using a MeanWell RSP-320-13.5 (active PFC, can be regulated to 12V), a built-in Raspberry Pi with a MCU controlling a relay board for switchable outputs and current sensing. Much more efficient than any of the PSUs built into these, and also removes the need for other AC-DC converters, and I can power assorted 12V devices directly.

The LRS series are also 1U and very efficient, though 12v efficiency will always be worse than higher voltages. The 48V ones can be regulated/configured to output above 50V, FYI. All these have a small potentiometer that affects the output voltage. The closer to the AC-DC conversion voltage, the more efficient it will be.

I can help anyone here pick the right supply to switch, just PM me, or ask here.

 

[fohdeesha]

Much more efficient than any of the PSUs built into these

Well....not exactly The efficiency of the MeanWell RSP-320 is listed at 88%, while measuring the Delta in the ICX6450 with an Agilent power meter, it's a little over 91% (and it's very hard to do much better than this). Power supplies have come a long way.....MeanWell is essentially a "cheaper" OEM compared to Delta with cheaper capacitors ETC, I have quite a few in the field on different projects. They're not bad if they're in a temperature controlled environment

 

[anomaly]

Glad to hear you found a good deal!



have you actually heard the stock fan yet? Screamer isn't exactly the term I would use, there's a stack of 3 of these switches 4 feet behind me and I literally can't hear them. The noise measurements in the spec sheet are for the fan at max speed, which is what it would ramp up to if you started doing line rate switching on every port simultaneously (160gbps or so). Needless to say I doubt that will happen, I've never seen the fan go over 30% or so

I will test, but I was referring to the PoE versions. See the acoustics in the datasheets I referred to. Everyone has different noise acceptable levels... 37 dB is not silent by any stretch of the word at the frequency and characteristics of fan noise (whirring, etc). I'm not saying these are terrible, but they will not be quiet in a small or relatively small lab room that is *quiet*. My lab is built to be as silent and self-contained as possible, so I had to do lots of aftermarket changes, as well as building my own ducts, tunnels and fan holders (mostly out of aluminum). I definitely can tell 37dB as it sits a good >17 dB louder than anything else. Also, dBs are not linear in sense that, a small increase DOUBLES the perceived loudness. ~2dB higher is twice as loud.

 

[fohdeesha]

Again, the acoustic specs are for max load. For the PoE models, that's not only max ASIC load, but max PoE power consumption (20 watt devices on every single port). At idle or with a couple PoE ports active, it should run the fan at around 30% just like the non-PoE model. In fact, I need to double check, but I'm pretty sure the PoE model has the exact same fan part # as the non-PoE switch

 

[anomaly]

Well....not exactly The efficiency of the MeanWell RSP-320 is listed at 88%, while measuring the Delta in the ICX6450 with an Agilent power meter, it's a little over 91% (and it's very hard to do much better than this). Power supplies have come a long way.....MeanWell is essentially a "cheaper" OEM compared to Delta with cheaper capacitors ETC, I have quite a few in the field on different projects. They're not bad if they're in a temperature controlled environment

Hmm. Did you measure it correctly? Was this efficiency at the wall (after AC-DC conversion etc) or at DC load? I'm curious about your measuring setup. You usually need to properly analyze them, meaning same load configuration. How you connect your test equipment, what the load is configured to be, etc, affects that. If you have the part number for the Delta I would like to check, but measured at the wall I doubt it will be better than the RSP, too. You did not say if this was at 12V output. Most open frame or closed frame PSUs will not go higher than 88% at 12V. Higher voltage with a highly efficient DC-DC conversion stage (ex. multi output) tends to be better.

Also, the RSP-320 is not the most efficient MeanWell PSU around.

Edit: I picked the RSP for assorted reasons, but one was cost. I'm in the market for a 1U power supply that is very efficient with low loads. Checking if MeanWell has anything new.

Edit 2: Just to be clear, load affects the efficiency so much that you need to simulate exactly the same conditions. I have to pull up the certificate for the RSP where they show the load/efficiency charts,. If you test a 320W rated at 89% under a 10W load vs a 20W PSU at the same wattage, odds are, unless the 20W PSU is absolute garbage, the lower wattage PSU will win. The 80 standard specifies how they test things, but this is not 80 PLUS gear, so you have to double check exactly how efficiency was calculated. And then apply the same proportional loads.

 

[fohdeesha]

To measure efficiency, you measure energy out of the power supply (DC side) versus the energy into the power supply (AC side), to then get the total loss, which is then integrated to find the efficiency percentage. Better explained here - Power Topics for Power Supply Users: Efficiency Calculations for Power Converters

AC input was measured with a Continental Controls Wattnode Modbus, an AC power meter with +/-0.5% accuracy. the DC side (voltage and current) was measured with a pair of agilent U1242B's, which have something like +/-0.1% DC base accuracy. Given how low the power draw of the switch is (something like 22 watts), a single watt lost would move the efficiency up or down 3 or 4 points, so in my mind it's honestly not even worth worrying about hehe. Even giving it a magical 100% efficiency power supply would save you about a single watt. However you are definitely right, there's a big advantage to having everything in a lab on a single power supply rail, doing away with a bunch of switchmode power supplies certainly helps out

 

[anomaly]

To measure efficiency, you measure energy out of the power supply (DC side) versus the energy into the power supply (AC side), to then get the total loss, which is then integrated to find the efficiency percentage. Better explained here - Power Topics for Power Supply Users: Efficiency Calculations for Power Converters

AC input was measured with a Continental Controls Wattnode ModBus, an AC power meter with +/-0.5% accuracy. the DC side (voltage and current) was measured with a pair of agilent U1242B's, which have something like +/-0.1% DC base accuracy. Given how low the power draw of the switch is (something like 22 watts), a single watt lost would move the efficiency up or down 3 or 4 points, so in my mind it's honestly not even worth worrying about hehe. Even giving it a magical 100% efficiency power supply would save you about a single watt. However you are definitely right, there's a big advantage to having everything in a lab on a single power supply rail, doing away with a bunch of switchmode power supplies certainly helps out

Sweet setup with nice "toys".

I think unless we pick a LOW wattage model, the switch is not going to help at all. You need to load up a PSU at least at 50% to reach peak efficiency usually. The best thing about a single rail is that in my case I only have one source of EMI to worry about. I have a radio station next to the lab and I do my RF work in the same room. Some of it in the microwave frequency range so, it is crucial to have low RFI.

I'm passing the DC through a differential pi filter for the 12V line. Ripple so low I don't need to care. No RFI out too. AC is filtered with an additional double mode filter, though this is totally overkill. MeanWell does not cheap out on the AC stage filters, so they don't leak crap into the AC line.

I have a bit of an obsession with PSU efficiency at this point, as we build more low power systems, it becomes quite relevant. And every watt wasted is more heat generated. My rack is a 60cm short depth cabinet meant for communication setups. Basically I had to customize and cut rails to fit, weld them, etc. I should write about it one day, but it was painful. Case selection is also super limited and more complicated with that short depth. Basically no case comes with how swap fan boxes or backplanes at that size.

Edit: this is one of the >90% efficiency supplies from these guys: https://www.tme.eu/en/Document/920b959fa94fff6729ddb7d92aa45670/HEP-100-spec.pdf