You are right. I only want my xserve to be seen as an XServe and not a mac pro. Can you post a picture of you "about this mac" ?
Is it easy to swap logic board ? thanks you !
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Xs 1,1-3,1 time for a sticky: Xserve upgrade guide
- Thread starter Flint Ironstag
- WikiPost WikiPost
- Start date
- Sort by reaction score
- Status
You are right. I only want my xserve to be seen as an XServe and not a mac pro. Can you post a picture of you "about this mac" ?
Is it easy to swap logic board ? thanks you !
yes, but a SOIC8 clip will do to unbrick.
About will show MacPro 5,1, but this depends on a few factors.
The Board ID is not changed - most Apple Software will use this to determine support.
I also like to get the original Xserve EFI upgraded. I did not try butI can imagine it's too difficult.
So far no one seems to have found the 6 core support. It's not just microcode.
The green Ethernet LED and blue Drive LED for activity work, - the green drive power indicator is not on.
White power LED is on too.
-> Goodie attached benchmark
Attachments
Hey, do you know where is located the eeprom on the logic board?
How did you get a serial number ?
Thanks for informations !
K that makes sense. Any progress from anyone in bringing back the front panel and accessing the other ram slots?
Like said before, without ROM expert such as tsialex or dosdude1 or rominator, this is a dead end.
plus the fact that microsoft droped support on those 56xx xeon combined to catalina being not vanila on cMP seal the end of support for those machine.
I will keep mine functional the way they are, and upgrade to rack MP7.1 when available.
I noticed the front USB port and the stock SSD board in the XServe 3,1 towards the front are of the same daughter board.
Does this USB port on the front panel work after replacing with cMP 5,1 firmware? What about the stock SSD? Does that boot? thanks
The front panel is seen by the OS as a USB device so in theory the port should work. The trick of getting the CPU lights to work on OS's newer than Lion takes advantage of that fact. With the SSD I'm not really sure. The only way to know is to flash and test it.
(even)cheap(er) on ebay... definitely need to snag a few more to max and run on my home network. Wake me up when all the front lights are working in MacOS or FreeBSD.
I run El Capitan since it is the end of support and one of the last OS X server versions that doesn't have all it's teeth pulled and all the front panel lights work just like lion.
I also runEl Capitan in my 2009 Xserve. it does everything i need it to do and stable as heck.
Sata is seen as it should, the lights are able to show CPU activity with IPMI kext from serveros on github.
Hello,
I have recently come to own an xserve 3,1, single CPU model
Have some background in tweaking, fiddling, etc.. with things like what has been discussed so far, that's some great progress with flashing the mac pro firmware.
My immediate goal is to get it working with a 6 core cpu with all 3 memory channels working if possible.
Later I intend to grab a dual cpu logic board and fit it, load up ESXi and use it with vmotion to offload my mac vms onto native hardware if indeed I can achieve 12 cores and 6 memory channels on the replacement logic board.
My unit was donated as I collect older hardware and work doesn't want to deal with the associated fan noise, so I'm quite happy to test and try things in what little spare time I have.
6 channels at 1333 could still hold its own versus single cpu lga2011/x79 stuff in the right circumstances, so this one's definitely worth investigating.
Edit: To me, it sounds like the Firmware for the Mac Pro 5,1, which as two memory channels, needs to be modified in the area that deals with memory channel configuration to enable the third. I wish I knew how to do this.
I have recently come to own an xserve 3,1, single CPU model
Have some background in tweaking, fiddling, etc.. with things like what has been discussed so far, that's some great progress with flashing the mac pro firmware.
My immediate goal is to get it working with a 6 core cpu with all 3 memory channels working if possible.
Later I intend to grab a dual cpu logic board and fit it, load up ESXi and use it with vmotion to offload my mac vms onto native hardware if indeed I can achieve 12 cores and 6 memory channels on the replacement logic board.
My unit was donated as I collect older hardware and work doesn't want to deal with the associated fan noise, so I'm quite happy to test and try things in what little spare time I have.
6 channels at 1333 could still hold its own versus single cpu lga2011/x79 stuff in the right circumstances, so this one's definitely worth investigating.
Edit: To me, it sounds like the Firmware for the Mac Pro 5,1, which as two memory channels, needs to be modified in the area that deals with memory channel configuration to enable the third. I wish I knew how to do this.
Last edited:
What??
You enable 3-channe mode simply by putting 3 sticks per socket instead of 4 or 2. There is no need for modification of the firmware for that. You do loose 2 slots per socket, but the remainig 4 are available same way as on a Mac Pro.
If you are losing slots per cpu socket, you are losing memory channels, by definition.
You appear to be under the misunderstanding that the CPU can just assign memory channels to whatever ram you throw in whatever socket, unfortunately that is not how the x86 platform works, at all.
Memory channels are assigned to pairs of sockets, this is actually a hard wiring thing, it cannot be changed as there is no fancy memory channel switching hardware in the x86 platform for doing so, and doing so would probably introduce latency and needless complexity into the memory controller design.
The consequence is, if you lose a memory channel due to either a hardware fault or firmware issue, you lose two memory slots.
As I understand it, we lose a memory channel when we apply the mac pro 5,1 firmware due to the differences in the mac pro 5,1 firmware and the xserve 3,1's hardware, specifically the memory controller configuration part of things.
In the end it's commodity hardware, it's not as smart or capable as it could be or that one would assume it is, due to cost reasons, memory channels are tied to specific memory slots for the same reason you can't hot swap a CPU on any kind of computer you've probably ever used. (You can on old mainframes, but they're super expensive, to give an example.)
The 5,1 has 3 channels, just like Xserve.
Slots 1 and 2 on each CPU are their own channels then slots 3 and 4 are shared on each CPU.
It may make more sense if numbered like a typical server board. So think of each cpu having this layout in a 5,1
CPU-A
DIMM-1A — slot 1
DIMM-2A — slot 2
DIMM-3A — slot 3
DIMM-3B — slot 4
CPU-B
DIMM-1A
DIMM-2A
DIMM-3A
DIMM-3B
in the Xserve we have 6 slots per CPU
CPU-A
DIMM-1A — slot 1
DIMM-1B — slot 2
DIMM-2A — slot 3
DIMM-2B — slot 4
DIMM-3A — slot 5
DIMM-3B — slot 6
CPU-B
DIMM-1A
DIMM-1B
DIMM-2A
DIMM-2B
DIMM-3A
DIMM-3B
What I would suspect is that what is lost is the second slot (B) for each DIMM except DIMM-3B, as that would like up to the 5,1 config. So we’d be able to use slots 1,3,5,6 on each CPU in a triple channel configuration like Is done on the cMP.
There are other boards that run Nehalem / westmere chipsets that use the same layout I reference (SuperMicro X8DT6, and I think X8DTH is the variant with 18 slots)
in the 18 slot variant you get 3 DIMMS per channel so DIMM-1A,1B,1C for example are all on memory channel 1.
Now one one thing we could test would be to see if we can use those other slots using tools from the hack into an world ... like clover (not recommended) or OpenCore (which is fast becoming something anyone with a cMP should use to replace UGA EFI flashed GPUs and replace the limited Apple bootloader)
both of these options allow for custom SMBIOS configurations to be applied which may allow for one to manually configure a 5,1 cMP smbios to have 6 slots per cpu.
Slots 1 and 2 on each CPU are their own channels then slots 3 and 4 are shared on each CPU.
It may make more sense if numbered like a typical server board. So think of each cpu having this layout in a 5,1
CPU-A
DIMM-1A — slot 1
DIMM-2A — slot 2
DIMM-3A — slot 3
DIMM-3B — slot 4
CPU-B
DIMM-1A
DIMM-2A
DIMM-3A
DIMM-3B
in the Xserve we have 6 slots per CPU
CPU-A
DIMM-1A — slot 1
DIMM-1B — slot 2
DIMM-2A — slot 3
DIMM-2B — slot 4
DIMM-3A — slot 5
DIMM-3B — slot 6
CPU-B
DIMM-1A
DIMM-1B
DIMM-2A
DIMM-2B
DIMM-3A
DIMM-3B
What I would suspect is that what is lost is the second slot (B) for each DIMM except DIMM-3B, as that would like up to the 5,1 config. So we’d be able to use slots 1,3,5,6 on each CPU in a triple channel configuration like Is done on the cMP.
There are other boards that run Nehalem / westmere chipsets that use the same layout I reference (SuperMicro X8DT6, and I think X8DTH is the variant with 18 slots)
in the 18 slot variant you get 3 DIMMS per channel so DIMM-1A,1B,1C for example are all on memory channel 1.
Now one one thing we could test would be to see if we can use those other slots using tools from the hack into an world ... like clover (not recommended) or OpenCore (which is fast becoming something anyone with a cMP should use to replace UGA EFI flashed GPUs and replace the limited Apple bootloader)
both of these options allow for custom SMBIOS configurations to be applied which may allow for one to manually configure a 5,1 cMP smbios to have 6 slots per cpu.
Thanks for clarifying. so it is specifically the slot configuration and not channel configuration causing the issue, in short. That's the only thing that didn't seem to be clearly explained prior.
So a triple channel config would be 1A 2A (3A or 3B)?
*sigh* at least it's documented clearly now: triple channel works, you just lose the first two B memory slots.
So a triple channel config would be 1A 2A (3A or 3B)?
*sigh* at least it's documented clearly now: triple channel works, you just lose the first two B memory slots.
Last edited:
In order for 3B to work 3A must be populated. At least this is how the typical server board works.
i plan to flash my single cpu Xserve soon to verify this behavior. I wanted to get the LOM tested out better first Before flashing the cMP 5,1 firmware. I also have a misbehaving fan according to Macs fan control. The in/out4 pair will only intermittently go to Max rpm when manually setting them to max rpm.
Last edited:
Just for future reference, the how-to can be found on ifix.com. https://www.ifixit.com/Guide/How+to+Upgrade+the+Firmware+of+a+2009+Mac+Pro+41/98985
I’ve got two Dual CPU Xserve3,1 flashed with v144 bootrom. ESXi v6.5 recognizes all 12 RAM slots populated with 8GB RAM for a total of 96GB. The other one has six 16GB RAM modules installed and also recognizes 96GB.
So this means that we should be able to correct macos using OpenCore smbios changes should fix this behavior in macOS.
Anyone know the best way to put a os on a 3,1 that has no drives. I’d like to use an external drive (usb) and I don’t have another Mac.
I'm not sure if this is the right spot...
Documenting my xserve upgrade process
Goal is 6-8 relatively quick mac VMs for iOS builds and automated UI testing at minimal cost
Dual CPU Xserve 3,1 -> found on shelf
96GB (16GB*6) RAM -> $100
3 x 960GB SATA SSD -> $200 ea
DELL LSI MegaRAID 9280-4i4e raid card -> $13
G5 drive backplane -> $12
SAS trays (to eliminate the SATA2 limitation) -> $40
Temporarily using SATA trays for power only and just plugged the SATA data cables directly into the drives from the MegaRAID breakout cable.
Custom power wiring from 14pin power distro board to 12pin G5 backplane
G5 drive backplane pinout:
1 - GND, 2 - GND, 3 - GND, 4 - GND, 5 - GND, 6 - UNKNOWN
7 - +12, 8 - +12, 9 - +5, 10 - +3.3, 11 +3.3, 12 - UNKNOWN
intel power board pinout:
1 - GND, 2 - +5 unswitched, 3 - GND, 4 - +12, 5 - +12, 6 - +12, 7 - GND
8 - GND, 9 - +5, 10 - +5, 11 - GND, 12 - +3.3, 13 - +3.3, 14 - GND
I left pins 6 and 12 on the G5 board disconnected, and pins 2, 3, 4, and 9 disconnected on the intel end. cable needs to be about 20 inches long, give or take.
I still haven't figured out how best to route the cable or how to modify the hold-down bolts.
If this holds up, I'll upgrade the CPUs and flash the bios in the fall...
Currently running ESXi 6.5
Documenting my xserve upgrade process
Goal is 6-8 relatively quick mac VMs for iOS builds and automated UI testing at minimal cost
Dual CPU Xserve 3,1 -> found on shelf
96GB (16GB*6) RAM -> $100
3 x 960GB SATA SSD -> $200 ea
DELL LSI MegaRAID 9280-4i4e raid card -> $13
G5 drive backplane -> $12
SAS trays (to eliminate the SATA2 limitation) -> $40
Temporarily using SATA trays for power only and just plugged the SATA data cables directly into the drives from the MegaRAID breakout cable.
Custom power wiring from 14pin power distro board to 12pin G5 backplane
G5 drive backplane pinout:
1 - GND, 2 - GND, 3 - GND, 4 - GND, 5 - GND, 6 - UNKNOWN
7 - +12, 8 - +12, 9 - +5, 10 - +3.3, 11 +3.3, 12 - UNKNOWN
intel power board pinout:
1 - GND, 2 - +5 unswitched, 3 - GND, 4 - +12, 5 - +12, 6 - +12, 7 - GND
8 - GND, 9 - +5, 10 - +5, 11 - GND, 12 - +3.3, 13 - +3.3, 14 - GND
I left pins 6 and 12 on the G5 board disconnected, and pins 2, 3, 4, and 9 disconnected on the intel end. cable needs to be about 20 inches long, give or take.
I still haven't figured out how best to route the cable or how to modify the hold-down bolts.
If this holds up, I'll upgrade the CPUs and flash the bios in the fall...
Currently running ESXi 6.5