EU Microsoft Project Olympus Dual Epyc 7003 Boards - 87,12€ each
- Thread starter Prof Theorie
- Start date
Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.
This PSU should work from a single phase 200-240V AC source.
Three power supply modules seems to be fully independent on the input side (each module has its own neutral and live pin).
Also there is a note in documentation:
"Note that the power supply needs to be able to operate under three single phases to each module with the three sine waves all in phase (no 120
Tree sine waves in phase, no lag = the same phase
Last edited:
I've been looking for an affordable not too old epyc server for months now.
Besides the cpus and ram (32 core one around 420 eur new here, not worth to ebay for same price), what else do you need?
Heatsinks ? I don't see them in the picture/text of the whole server. So normal SP3 heatsinks ?
And the PSU, how is one supposed to start that, build your own custom connector ? Is the socket at least available to be 3d printed somwehere ? Or can you just buy these ? Or maybe it's enough to solder 240v, neutral and ground to some pins ?
Besides the cpus and ram (32 core one around 420 eur new here, not worth to ebay for same price), what else do you need?
Heatsinks ? I don't see them in the picture/text of the whole server. So normal SP3 heatsinks ?
And the PSU, how is one supposed to start that, build your own custom connector ? Is the socket at least available to be 3d printed somwehere ? Or can you just buy these ? Or maybe it's enough to solder 240v, neutral and ground to some pins ?
There is a part number in the OCP data sheets for the feed connector. It's a 13 power conductors (GND + 3 phase A + 3 phase B) + data connector (likely optional) and soldering it might still be a bit of work but should be doable.
The alternative would be to junk the PSU and just feed 12v directly into the custom ATX power connector on the motherboard (same ATX connector, different pin out). The easiest way is likely via mining PSU adapter boards.
if you take the entire server, the heatsinks should be included. i can confirm that later that day, mine should arrive today.
i'll also try, if the PSUs work with a single phase, dunno if ill get to that today though.
Edit: The Heatsinks are included in the chassis
I do not have CPUs yet, so I cannot really test the entire system yet.
For the mainboard connector, a 3x EPS12v or PCIE 8-pin to 24pin ATX sounds fun, maybe I try that as as well.
Would be cool to at least get a heartbeat from the ipmi.
Last edited:
its a long boy! comparison with a HP dl380 g10. its about 94.5cm long oO
psu says 200 - 240V or 346-415V, that sounds good to me
also says, it contains Li-ion batteries... lets see if thats correct
psu says 200 - 240V or 346-415V, that sounds good to me
also says, it contains Li-ion batteries... lets see if thats correct
Attachments
Also some size comparison of the mainboard with a regular beer bottle.
Main Power Connector is standard ATX24 pin as we know (e.g. connector Molex 39-01-2245)
SATA Power 8 pins are the same as EPS12V (e.g. connector Molex Minifit JR 39-01-3085)
SATA Power 4 pins are first half of EPS12V (e.g. connector Molex MiniFit JR 39-01-2045)
Haven't looked up the pinout for the SATA Power yet.
there is also an unpopulated FP_CONN, might be front panel?
we also got 8 SATA Lanes, four via SATA Plugs, one via Mini SAS HD (SFF8643) ?
Then we got a second SFF8643 for 4 PCIe lanes according to the silkscreen.
There is also an Oculink connector, maybe 8 PCIe lanes?
Another unpopulated Connector called BP_CONN next to an USB socket, maybe there is another version with a frontfacing backplane?
Main Power Connector is standard ATX24 pin as we know (e.g. connector Molex 39-01-2245)
SATA Power 8 pins are the same as EPS12V (e.g. connector Molex Minifit JR 39-01-3085)
SATA Power 4 pins are first half of EPS12V (e.g. connector Molex MiniFit JR 39-01-2045)
Haven't looked up the pinout for the SATA Power yet.
there is also an unpopulated FP_CONN, might be front panel?
we also got 8 SATA Lanes, four via SATA Plugs, one via Mini SAS HD (SFF8643) ?
Then we got a second SFF8643 for 4 PCIe lanes according to the silkscreen.
There is also an Oculink connector, maybe 8 PCIe lanes?
Another unpopulated Connector called BP_CONN next to an USB socket, maybe there is another version with a frontfacing backplane?
Attachments
so, my PSU came with batteries included, dunno how I feel about that.
the 18650 still have charge, are at around 3.7V.
also the connector is pretty on there, dont know how one would modify that one :'D
the 18650 still have charge, are at around 3.7V.
also the connector is pretty on there, dont know how one would modify that one :'D
Attachments
Last edited:
Thanks for great info. Also please let us know when you manage to power it up.
Not sure if this is relevant for others, but sharing just in case -
piospartslap.de has the risers, heatsinks and PSUs matching the motherboards for sale (they also have the smartNICs and servers/MoBos, but sadly none of the m.2 PCIe adapter boards that I could find).
I suspect I could get more use out of the motherboards I got before the all-in-one server listing was discovered if I have these accessories. They accepted a pretty brave offer on my part (4 @ 33EU ea. for the PSUs, 3 @ 25EU ea. for the heatsinks, etc.).
I suspect they have a lot of these servers in stock and they're not really moving because they're so strange.
There's even one rack PDU on eBay but even if I could get a good offer accepted, shipping makes it too expensive for me (unless you could pick it up in person in VA).
I believe we can figure out the AC input connector and wire these PSUs.
I actually think the PSUs are interesting for those on 3-phase 230vac power in the EU. In theory, you get 2 AC feeds with automatic switchover, 1000W balanced 3-phase input and a tiny battery backup in one unit. Weird for sure, but also interesting...
piospartslap.de has the risers, heatsinks and PSUs matching the motherboards for sale (they also have the smartNICs and servers/MoBos, but sadly none of the m.2 PCIe adapter boards that I could find).
I suspect I could get more use out of the motherboards I got before the all-in-one server listing was discovered if I have these accessories. They accepted a pretty brave offer on my part (4 @ 33EU ea. for the PSUs, 3 @ 25EU ea. for the heatsinks, etc.).
I suspect they have a lot of these servers in stock and they're not really moving because they're so strange.
There's even one rack PDU on eBay but even if I could get a good offer accepted, shipping makes it too expensive for me (unless you could pick it up in person in VA).
I believe we can figure out the AC input connector and wire these PSUs.
I actually think the PSUs are interesting for those on 3-phase 230vac power in the EU. In theory, you get 2 AC feeds with automatic switchover, 1000W balanced 3-phase input and a tiny battery backup in one unit. Weird for sure, but also interesting...
The power connector appears to be an Amphenol 10106269-1C03001LF https://mou.sr/3LrNz0k
Might be interesting if we can source these easily and maybe design a small pcb for them.
Pinout and so on is here https://files.opencompute.org/oc/pu...s&t=ee6c3e775a7a8c6014f8eee71224ee02&download
Might be interesting if we can source these easily and maybe design a small pcb for them.
Pinout and so on is here https://files.opencompute.org/oc/pu...s&t=ee6c3e775a7a8c6014f8eee71224ee02&download
So, i got the thing working as far as I can test it!
I got some spare PCIe power cables from a Seasonic PSU, snipped of the ends and crimped the ATX24pin connector on.
Works nicely, however, the other PC has to stay on as well :'D
Hooked up to my network, BMC looked for IP via DHCP, no issues here.
nmap reported the following ports are open:
According to the following netapp kb netapp kb:
Port 623 is IPMI
Port 5120, 5123 are Remote Mount Image.
ssh reports an "Unable to negotiate with 10.101.10.64 port 22: no matching host key type found. Their offer: ssh-rsa,ssh-dss"
so I tried "ssh -o HostKeyAlgorithms=ssh-rsa admin@10.101.10.64" which lets me enter a password but stops responding.
Apparently, thats just serial redirection. After a third connection attempt, it kicks me out and says
"There is an active serial redirection session on the card.
Please terminate it first to be able to start a new session."
So, that i cannot try further as i do not have CPUs yet.
ipmitool reported "Error in open session response message : no matching cipher suite" at first.
apparently -C17 is default nowadays, tried with -C3, that got rid of the error.
I tried the credentials admin admin, and funnily enough, that worked
Going through the event list (ipmitool sel elist) shows *a lot* of Power Supply issues.
Dunno if i like the sound of that...
That went way better than I expected.
So yeah, I''m stuck right now. Unless somebody wants to send me some EPYC Milan CPUs :'D
I got some spare PCIe power cables from a Seasonic PSU, snipped of the ends and crimped the ATX24pin connector on.
Works nicely, however, the other PC has to stay on as well :'D
Hooked up to my network, BMC looked for IP via DHCP, no issues here.
nmap reported the following ports are open:
Code:
nmap 10.101.10.64 -p-
Starting Nmap 7.98 ( https://nmap.org ) at 2025-11-06 12:23 +0100
Nmap scan report for 10.101.10.64
Host is up (0.069s latency).
Not shown: 65530 closed tcp ports (conn-refused)
PORT STATE SERVICE
22/tcp open ssh
623/tcp open oob-ws-http
5120/tcp open barracuda-bbs
5123/tcp open unknown
50000/tcp open ibm-db2Port 623 is IPMI
Port 5120, 5123 are Remote Mount Image.
ssh reports an "Unable to negotiate with 10.101.10.64 port 22: no matching host key type found. Their offer: ssh-rsa,ssh-dss"
so I tried "ssh -o HostKeyAlgorithms=ssh-rsa admin@10.101.10.64" which lets me enter a password but stops responding.
Apparently, thats just serial redirection. After a third connection attempt, it kicks me out and says
"There is an active serial redirection session on the card.
Please terminate it first to be able to start a new session."
So, that i cannot try further as i do not have CPUs yet.
ipmitool reported "Error in open session response message : no matching cipher suite" at first.
apparently -C17 is default nowadays, tried with -C3, that got rid of the error.
I tried the credentials admin admin, and funnily enough, that worked
Code:
>> ipmitool -I lanplus -H 10.101.10.64 -C3 -U admin -P admin mc info
Device ID : 32
Device Revision : 1
Firmware Revision : 4.70
IPMI Version : 2.0
Manufacturer ID : 7244
Manufacturer Name : Quanta Computer Inc.
Product ID : 2020 (0x07e4)
Product Name : Unknown (0x7E4)
Device Available : yes
Provides Device SDRs : no
Additional Device Support :
Sensor Device
SDR Repository Device
SEL Device
FRU Inventory Device
IPMB Event Receiver
IPMB Event Generator
Chassis Device
Aux Firmware Rev Info :
0x00
0x00
0x00
0x00
Code:
>> ipmitool -I lanplus -H 10.101.10.64 -C3 -U admin -P admin fru print
FRU Device Description : Builtin FRU Device (ID 0)
Chassis Type : Blade Enclosure
Chassis Part Number : M1069035-001
Chassis Serial : QTFCUE82400CF
Chassis Extra : 3.0
Chassis Extra : A
Board Mfg Date : Wed Jun 20 13:29:00 2018 CEST
Board Mfg : Microsoft
Board Product : C202A
Board Serial : JUE82400004
Board Part Number : M1039059-001
Board Extra : B
Product Manufacturer : Microsoft
Product Name : C2025
Product Part Number : M1088444-001
Product Version : 4.0
Product Serial : 206793870557
Product Asset Tag : 10925200
Product Extra : 5
Product Extra : MSF-033672Going through the event list (ipmitool sel elist) shows *a lot* of Power Supply issues.
Dunno if i like the sound of that...
That went way better than I expected.
So yeah, I''m stuck right now. Unless somebody wants to send me some EPYC Milan CPUs :'D
Attachments
Ah, i forgot port 50000, that simply shows me some XML data via HTTP:
Don't know if thats an API or so?
Code:
<root>
<specVersion>
<major>1</major>
<minor>0</minor>
</specVersion>
<device>
<deviceType>urn:schemas-upnp-org:device:management:1</deviceType>
<friendlyName>AMID8C49769F062</friendlyName>
<manufacturer>American Megatrends Inc</manufacturer>
<manufacturerURL>http://www.ami.com</manufacturerURL>
<modelName>AMI MegaRAC SPX</modelName>
<modelNumber>SPX 1.0</modelNumber>
<modelDescription>
AMI's MegaRAC SPX, the award winning full-featured Service Processor Software Stack
</modelDescription>
<modelURL>http://www.ami.com</modelURL>
<UDN>uuid:d24ad5d1-696a-f644-5682-127e741056c2</UDN>
<serviceList> </serviceList>
<presentationURL>http://10.101.10.64</presentationURL>
</device>
</root>Hey! I work with these servers (blades as we call them) at a Microsoft datacenter, so I might be able to help you with some stuff
I mostly only do breakfix tasks on these blades, and this exact C2020/C2025 isn't too common in my DC, but all blades work fundamentally the same. I have been eyeing the Open Compute Project docs and they are pretty similar to what's used internally.
It goes without saying that I am under strict NDAs and can only provide some guidance based on my experience and the open source docs (no I won't be leaking any Microsoft Intellectual Property, so please don't ask for BIOS binaries or firmware)!
The BMC management interface is used by engineering, and datacenter techs don't connect through it, so I don't have much information about how to access it, but I might be able to explore it a bit.
There is a way to access a serial console and IPMI through the PMDU (PDU with builtin management, it's explained in the OCP docs as well). These PMDUs house a Rack Manager (RM), which is the device we connect to at the rack level, and then the RM routes the IPMI requests or serial interface to the specified blade in the rack by index (U number). It would be interesting to see if someone can figure out a way to connect to the PSU data interface (which has its pinout documented in the OCP docs) to get a serial session going without needing a Rack Manager.
As many of you have noticed too, they do not have power buttons. To power on or off these blades we either make a request through a web portal that I believe goes through the BMC port, or manually set the power state through the Rack Manager CLI (so using the PSU data interface).
I might be able to help with some hardware troubleshooting as well since I've worked on thousands of blades like these, they can be a bit finicky, but I always thought they were pretty cool pieces of hardware!
Let me know if you have any questions, I will do my best to help!
I mostly only do breakfix tasks on these blades, and this exact C2020/C2025 isn't too common in my DC, but all blades work fundamentally the same. I have been eyeing the Open Compute Project docs and they are pretty similar to what's used internally.
It goes without saying that I am under strict NDAs and can only provide some guidance based on my experience and the open source docs (no I won't be leaking any Microsoft Intellectual Property, so please don't ask for BIOS binaries or firmware)!
The BMC management interface is used by engineering, and datacenter techs don't connect through it, so I don't have much information about how to access it, but I might be able to explore it a bit.
There is a way to access a serial console and IPMI through the PMDU (PDU with builtin management, it's explained in the OCP docs as well). These PMDUs house a Rack Manager (RM), which is the device we connect to at the rack level, and then the RM routes the IPMI requests or serial interface to the specified blade in the rack by index (U number). It would be interesting to see if someone can figure out a way to connect to the PSU data interface (which has its pinout documented in the OCP docs) to get a serial session going without needing a Rack Manager.
As many of you have noticed too, they do not have power buttons. To power on or off these blades we either make a request through a web portal that I believe goes through the BMC port, or manually set the power state through the Rack Manager CLI (so using the PSU data interface).
I might be able to help with some hardware troubleshooting as well since I've worked on thousands of blades like these, they can be a bit finicky, but I always thought they were pretty cool pieces of hardware!
Let me know if you have any questions, I will do my best to help!
I was initially hoping this could be related to redfish, but I tried looking elsewhere and I tend to think it's actually related to uPnP discovery:
(screenshot is from here, which is unrelated other than the use of MegaRac)
the intel blades have pwr-btn soldered and they work but PSU is not controlled(PS_EN pulled low at pwr management connector),
these AMD blades have also space for a SMD pwr-btn.
I have seen some blades with the power button populated indeed (not even all intel ones have it populated, I'm not sure why some have it and some don't), and I have tried to turn on a few with it but I never had success in the datacenter, so I am not sure exactly how the power button works, we just never use it even if the blade has it.
ah nice, that shouldn't be too hard to solder something on there
They also have a reset switch next to the BMC Ethernet Jack
Last edited:
Hi, can anyone confirm what types of CPUs can be run on this board, the only thing I found is that this server specifically came with an EPYC 7551, can anyone try an EPYC 7003 or EPYC 7002 series and one that has a TDP of more than 180W?