Thunderbolt in a Classic Mac Pro
Deprecation of the Enumeration caused Consternation in the Migration
There's always that scene in the action hero movie where the protagonist is told that something "simply cannot be done". That generally sets into motion the thumping tense strings with the subtle backbeat to tell the audience that in the mind of the hero, a vision has formed that will not be denied. Yeah, that's how I see this mission. I was told it cannot be done.
In a larger context, we can't do anything in the computer business without the cooperation of a multiple skill sets. Mechanical design, logical and physical layout, chip manufacturing, firmware, drivers, software, and on it goes to make stuff that we blithely use and too flippantly criticize. I say that because I hope the outcome of this article is that Apple issues a kext to make this work natively without the grooming by Windows 10.
At the left is a cell phone photo of my test rig. It's a 2010 Classic Mac Pro with an AKiTio (not my capitalization, talk to them!) Node Duo case. It's attached via an Apple Thunderbolt-3 0.8m-long cable because I didn't know a cable was included in the ship group.
For the impatient, here are the essentials:
The AKiTio must be seen by Windows 10 and the macOS must be soft-booted
There is a UAD-2 DSP Duo card and an RME MADIface ExpressCard/34 in the AKiTio
The macOS High Sierra doesn't report any Thunderbolt devices in the System Report, but they're visible in the PCIe section of the report
So far testing has shown that Windows 10 doesn't always immediately discover the AKiTio. Performance of the Thunderbolt devices seems to be the same as it was on internal PCIe slots.
The Back Story
I've decided to work hard at concert videos using one Panasonic DMC-FZ2500, three Panasonic DMC-FZ1000's and one Mevo. Post production had been done initially with PowerDirector 16, then 17, but the multi-cam and smooth integration of DaVinci Resolve was enough to prompt me to buy the Studio version.
I'm running an Nvidia GTX 1070 bought during the Bitcoin mining bubble. It has the EFI upgrade/hack expensively done by MacVidCards.com and therefore there's no macOS Mojave in my future. Sticking with the DaVinci Resolve Studio post-production processing, a second GTX 1070 would be glorious, and I've got one waiting in the wings.
Since I concurrently decided that I wanted to join the Cult of the RME Interfaces, I obtained a relatively affordable ExpressCard34 RME MADIface, and connected via fiber optic multimode to the surprisingly low-priced DN32-MADI in my M32 mixer. The ExpressCard34 connects via an SIIG ExpressCard to PCIe adapter that costs about $30. Although there is a nice little bus-powered Thunderbolt-2 box for ExpressCard made by Sonnet, as you'll see below, that is a solution I cannot use. So, two additional PCIe slots had to come from somewhere. An entirely new motherboard and computer is the simple answer that requires no heroics. Alas, the Apple corporation doesn't offer anything new that has the profile that I want.
I had briefly flirted with a Hackintosh when my Mac G5 was pushed into obsolescence by the powers that be. It was a valuable learning experience that may come in handy now that the i9-9900K with RAM and motherboard makes more economic sense than anything else. Rather than trekking into the sparsely populated lands of the Hackintosh community, I decided to try for combining the best features of what I have with what I need. So here we are.
Thunderbolt-2 Doesn't Work
The best box for what I needed is the Rocstor XR4. It has four PCIe slots, a 430+ watt power supply, and two 6-pin PCIe cables. The picture of its internals are shown at the bottom left. Each of its two Thunderbolt-2 ports serves two PCIe slots. Each slot is mechanically x16 and electrically x4. A new one cost over $1600, but some savvy eBay work brought it to me because the seller had no idea whether it would even power up. I tested the Rocstor on my 2011 MacBook Pro (Thunderbolt-1) and on a 2017 iMac (Thunberbolt-3) through an Apple Tbolt3-to-Tbolt2 adapter. It worked perfectly.
However, when attached to the GC-Titan Ridge card, nothing within the Rocstor was visible to Win10 or macOS High Sierra. It turns out that there is much more firmware-guided negotiation happening on a Thunderbolt link than we're willing to read about in one sitting. I'll spare you the drama and simply state that the GC-Titan Ridge card is a less-than-full implementation of the Thunderbolt protocol, and is likely to remain that way. Therefore, no downshift from 3-to-2 is going to happen.
With that defeat, I ordered the AKiTio purely on reputation garnered from a shaky YouTube video. The AKiTio arrives beautifully packaged, and the mechanical design is "top shelf" as they say in the ethanol dealer business. The Thunderbolt-3 interface in this box was the key to my limited success to this point.
Testing PCIe Connections
Part of the MADIface connection is enabled by the fact that the ExpressCard standards is a repackaged PCIe x1. I've included a handy image at the left that shows the commonly shared data and power lines of the two implementations.
The UAD-2 PCIe setup was made odd by the fact that Win10 could "see" the hardware in the Device Manager, even in the Rocstor, but the brains loaded by the UAD driver would not work. That is still true in Win10 as the Restart to soft-boot into High Sierra is done. There's no hint that this will work, but it's perfectly happy in High Sierra so far. I'll have to do a few projects to shore up that statement, so everything stated here is an early-days assessment.
For the record, I've included the Oblique RTL results of a full ADA loopback path through my MADIface and M32. It's done at 48k because I do practically everything at 48k. That's Video Life.
Next Steps and Market Observations
Unfortunately, I have to disassemble the testing topology because I have a couple of projects in Cakewalk by BandLab to finish. That's Win10-only, and since the RME can't work in Win10 within the AKiTio box, I'll have to revert to my conventional configuration for a while.
Hopefully Apple will be interested enough to answer some pointed questions about the opportunity this GC-Titan Ridge brings to them as a corporation. To that end, I submit the following observations:
Video editing products will continue to be centered around the ever-expanding capabilities of video cards. High end production of 4k and 8k materials are is compatible with mere 1.4x or 1.9x increases in compute capability. Highly parallel processing on specialized video devices demands a throughput and power supply profile that is only marginally supported by Thunderbolt-3. It's a somewhat hidden fact that the PCIe-only throughput of Thunderbolt-3 is 22Gbit/sec. Think of it as about an x3 PCIe 3.0 slot. We an survive for a while with BRAW 4.6k and lightly compressed 8k, but nobody should pretend that there's a lot of headroom in post-production computing.
Also, nobody should pretend that the 5k or 6k 27-inch screens offered in the iMac Pro is the preferred way of working in this arena. A modular implementation that offers a growth path, convenient reconfiguration (no soldered RAM or SSD's/NVMe's) and an infrastructure for video card "farms" is a good way forward from here. The option to do this in a Linux or Windows-based solution already exists, so Apple will need to offer a better value proposition that appeals to both the content creators and the accountants.
A good way to begin the narrative of that value proposition is to rescue the diminished value of its previous Greatest Hit, the Classic Mac Pro. Supporting a Thunderbolt solution in this box and offering a series of better-priced PCIe boxes ($1000 for 3 external slots doesn't feel like money well spent) will cause some friendly breezes to blow in the right direction.
- Ted Gary of TedLand
May 7, 2019