A new (to me) Mac Pro

About my Mac Pro

My 3.33 GHz 6-core Mac Pro died after sitting for a month or so. I shut it down normally and when I went to turn it back on, nada. 

I found and watched a couple of potential replacement units on eBay and Facebook Marketplace, and then kind of back-burnered the project. Then a seller sent me an eBay offer, which I accepted because they were relatively local (vs. $95 in shipping otherwise, and these things have a habit of getting bent TF in shipping), and because it was a nicely equipped machine for $200: 12-core (2x Intel® Xeon® X5650 CPUs @ 2.67GHz), 28 GB RAM, 500GB SSD, and a USB 3.0/eSATA card. (It’s still running the ATI Radeon HD 5770 1024 MB GPU, but I can transplant the RX 580 Sapphire easily enough.)

Open Core Legacy Patcher

Getting the Installer with an Unsupported High Sierra / Safari Setup

First thing was to build an OpenCore Legacy Patcher-based installer on a USB flash drive. So I headed over to pick up the latest (1.0.1) OCLP app and ... 

Huh. The Mac Pro came loaded with High Sierra and a now ~5 year old version of Safari (11.1.2 (13605.3.8)). Looks like there’s some web dev trickery that that old browser can’t grok; the “Assets” section never expands:

Assets section, spinning

And of course it’s not ... l33t? enough ... to have the installer files simply linked in a plain old bullet list or something... Sigh.

So, over to download Chrome, which, to Google’s credit, automatically proffers the most recent version (116.0.5845.187) that will run on Mac OS X 10.13. And that browser gave me a list of files to download:

Assets section, listing files

You’d think a community built up around supporting ancient systems would degrade a little more gracefully when encountering older client software...

Okay. So I downloaded the latest Monterey installer (12.7) and now I’m waiting for it to finish writing to a flash drive.

Exploring the “New” Machine

I’m a little concerned that the machine is reporting an empty RAM slot, that seems ... unlikely ... And more like maybe one of the slots is bad:

One DIMM is missing...

If so, it’s not the end of the world. I can use the higher density RAM from the old Mac Pro, or just swap in its CPU board with the 3.33 GHz 6-core (which is likely to be better for my needs than the 2.66 GHz dual CPU 12-core - even though I did spend $19, free shipping, on eBay, for a couple of Intel Xeon X5675 3.06 GHz CPUs to max out‡ the dual processor card, and $5 on a long 3 mm hex tool on Amazon... Sigh.).

Okay. I took the CPU board out, popped out the DIMMs, gave them all a good blast with compressed air, hit them with DeoxIT just in case•, and reinstalled them, and the machine is seeing the full 32 GB RAM:

All eight RAM sticks recognized

Swapped in the RX 580 Sapphire GPU (with Apple firmware) and the Sonnet Allegro USB 3.0 4-port PCIe Port Expansion Card (USB3-4PM-E) (removing the CalDigit FASTA-6GU3 Pro that was installed originally), and installed the OWC Accelsior 1M2 NVMe card (and the Samsung SSD 970 EVO Plus 1TB installed on it). Still booting from the Samsung SSD 850 EVO 500GB drive on the SATA backplane for the moment, until I can do a fresh install on the PCIe-attached storage.

Ran Geekbench and was kind of shocked by the Metal compute score, which dropped from 49592 to 26495 - an almost 50% reduction! I’m going to re-run that test when I get things more settled, that seems really off to me. The CPU results were kind of what I was expecting, single core performance score dropped about 10.4%, from 627 to 562 (although on paper the 2.66 GHz core in the X5650 should be about 20% slower than the 3.33 GHz core in the X5680), while the multi-core score jumped about 77%, from 3271 to 5791. It will be interesting to see how it performs with the X5675s installed.

This is kinda cool to see, I’m not gonna lie:

CPU History of 12 multithreaded cores

Now, waiting for the new (to me) CPUs to arrive so I can bump this thing up, with a fresh application of Arctic MX-4 thermal paste. I’d forgotten how smooth these (ancient!) workstations feel in use. 

 Meanwhile, I’m “liking” the convenience (but not the typing experience!) of the cheap ($50) LTC NB681 Nimbleback Wired 65% Mechanical Keyboard (blue switches) I picked up for exactly this purpose. I normally run the USB connection to the Mac Pro through the USB3 card, but since that card isn’t “alive” at boot, you can’t, e.g., use a keyboard connected to one of those ports to access a boot selector. I can easily hook this up to one of the built-in USB 2.0 ports and have access, and since the keyboard has a built-in two port USB hub, I can also hook up a cheap wired mouse (etc.) to it for a clean, effective user input device.

KVM Switch

Speaking of I/O - years ago I had a couple of VGA/USB keyboard/video/mouse switches. My little workspace is tiny and I have a couple of laptops I’d like to swap in where I normally use the Mac Pro, so I just ordered this relatively inexpensive NAWEN USB 3.0 HDMI KVM Switch, 4 Ports KVM Switcher Selector Box with EDID Emulator Function, Support 4K@60Hz Resolution for 4 Computers Share Mouse Keyboard and Monitor. I’m a bit dubious but if it works it fits the bill perfectly, so ... 🤞(Update: So far so good, it’s working fine with the Thunderbolt 2 dock and this Mac Pro.)

USB

Fixed: Built In USB Ports Not Working Under OCLP / Monterey

Oddball things learned: The Mac Pro doesn’t recognize the Tokyo60 keyboard, or the receiver for the Logitech M585 mouse (which requires a Bluetooth version higher than this 2010 hardware supports), on the built-in USB2 ports. Have to use the USB3 ports on the Sonnet card. Which also means the keyboard can’t wake the Pro from sleep. Edit: Apparently this is a known issue with newer operating systems on this old hardware. Stick a USB 2.0 (or better) hub between the on board USB ports and a USB 1.1 device and it works. I’m using this cheap thing since I had it handy.

Logitech M585 with USB Receiver

Getting the Logitech mouse to sync to the bundled receiver requires using a web application (that can disappear at any time), and, I couldn’t get the Logitech Options+ software to install (it would get stuck at the granting input monitoring permissions stage, even though I followed the instructions). But just now, when I went to pull up the installer again to get the text I was looking for, now it’s working (... or not, now it’s stuck in the same loop at the step: “Grant Bluetooth Permission”). 🤬🤷🏻‍♀️ Still, as much as I like the feel of these mice, I shouldn’t need a problematic, 250MB+ app just to enable a couple of buttons.

New CPUs, New Firmware, PCIe Devices Running Full Speed

Update: The CPUs (etc) arrived and I installed them. The results were great, single core performance is on par with the 3.33 GHz machine and multi-core performance blows the other machine away.

Also, despite the eBay seller proclaiming the dual-proc box had “latest possible firmware update,” it was in fact running MP51.0089.B00 (which I didn’t immediately catch, since OCLP spoofs firmware 9999.999.999.999.999). This resulted in the NVMe drive (a Samsung 970 EVO Plus 1TB M.2 unit OWC Accelsior 1M2 M.2 SSD to PCIe Adapter Card) running at like ~700 MB/s and being seen by the system as a 2.5 GT/s device. I was getting less than half the performance I should have been seeing (the 2010 Mac Pro is a PCIe 2.0 device, so I was never going to get close to the Samsung’s maximum performance).

Updating the firmware to 144.0.0.0.0 was a PITA, I had to install High Sierra on a spare drive, re-install the old ATI Radeon HD 5770 GPU (the Sapphire won’t work in High Sierra, and the ATI won’t work in Mojave since it’s not Metal capable), use High Sierra to build a Mojave installer (using Greg Neagle’s installinstallmacos.py script (discussion)), reinstall the AMD GPU, boot the Mojave installer, update the firmware, zap the PRAM ... And then booted back into Monterey (OCLP), everything was copacetic (and fast!) again; 5.0 GT/s x4 in System Information (and the GPU is now x16 at 5.0 GT/s, up from 2.5), and the benchmarks were what they should be on this old host (AmorphousDiskMark showing up to 1600+ MB/s sequential read/write speeds).

pci144d,a808:

  Type:                NVM Express Controller
  Driver Installed:    Yes
  MSI:                 Yes
  Bus:                 PCI
  Slot:                Slot-2
  Vendor ID:           0x144d
  Device ID:           0xa808
  Subsystem Vendor ID: 0x144d
  Subsystem ID:        0xa801
  Revision ID:         0x0000
  Serial Number:       [omitted]
  Link Width:          x4
  Link Speed:          5.0 GT/s

BlackMagic Disk Speed Test results (>1400 MB/s)

NVMe is always 4 PCIe lanes (x4), which has a theoretical limit of 2 GB/s throughput on a PCIe 2.0 system like the Mac Pro 2010 (5GT/s (~500MB/s) for PCIe 2.0). That’s 2,000 MB/s (again, theoretically). PCIe 2.0 has a roughly 20% overhead (newer versions are much lower; PCIe 3.0 is <2.0%), so 1,400–1,500 MB/s read/write speeds are pretty much spot on for what I’d expect.

And now I’m comfortable everything on this machine is as tuned and tweaked as it’s going to be (at least for now), and nothing is running in a sub-optimum configuration. Huzzah. (It’s like when I first installed a bus mastering PCI Ethernet card - a 3Com 3C905 - or my first properly setup UDMA hard drives, decades ago...)

Time to get some actual work done on it!

‡ Folks on the Interwebs have reported running dual X5680s successfully, but Apple never offered that configuration and I’d worry about heat, etc. So I’m sticking to the max spec configuration Apple shipped to consumers, which is dual 3.06 GHz chips.

• Love that stuff, it saved my Porsche a few years ago when one of the oxygen sensors threw a code and lit the Check Engine light - which would have caused it to fail smog check. Getting to the sensor to replace it unnecessarily would have been expensive, but pulling the connector and dousing it with DeoxIT was all it needed - the error code was thrown due to some light corrosion causing a bad connection.