NetStor created the HL23T-Plus with one goal in mind: become the ultimate Radeon VII enclosure. Everything about this Thunderbolt 3 eGPU enclosure was tailored to accommodate the Radeon VII‘s shortcomings. AMD designed the Vega II PCB without considering external graphics use. When an active Thunderbolt connection from a host computer goes into suspended mode, the Radeon VII in an external GPU enclosure runs its fans full speed. The HL23T-Plus is smart enough to not only set itself to suspend mode but also to put the Radeon VII to sleep properly.
|PSU max power||400W|
|GPU max power
|Power delivery (PD)
|TB3 USB-C ports||1|
|Size (in/mm, LxWxH)
||14.6 x 4.3 x 8.66
371 x 110 x 220
|Max GPU len (in/cm)
|Updated firmware||33.1 ✔|
|TB3 cable length (cm)||50|
Sporting the identical footprint and chassis of the NetStor Hercules, the HL23T-Plus provides a bit more headroom through the use of a taller top lid. This revision was made to accommodate the taller height of the Radeon VII as well as the fitment of the PCIe power connectors. The top lid has a captive thumb screw for no-tool access. Build quality is excellent with all-aluminum panels reminiscent of the AKiTiO Node Pro. There’s a low-profile folding handle in the rear panel above the power supply plug.
You’ll notice the paint job is no longer a distraction. Unlike the original HL23T Hercules with its neon colors and a cartoon Hercules drawing, the HL23T-Plus has a flat grey finish. The only bright color is a blue status light indicator on the front fascia when there’s an active Thunderbolt connection with a host computer. This circular indicator may seem like a button because you can push on it. Bizarrely there’s no other functionality.
The front fascia also serves as an air vent with tiny square cutouts in the shape of a flame that’s never to be illuminated. To the left is the side panel with mesh vent for air intake. Directly behind the mesh screen is a 80mm fan, mounted on adjustable sliding rails. This is a rather unique feature to NetStor Thunderbolt 3 enclosures. The fan placement can be adjusted to match the blower fan on most reference graphics cards. A toggle switch is placed on top of the fan sliding rail bracket. Fan speed can be changed corresponding to the size of the three corresponding circles. When used with a multiple-fan cooler like the Radeon VII, this fan and its bracket can be removed to provide more airflow through the side vent.
Another noteworthy feature is a spare PCIe mounting bracket. While the PCIe slot board does not have a spare slot, this space can be used for a single slot add-in-card. One requirement is an adapter to power the AIC. In one test I used a PCIe riser adapter with a Y PCIe power splitter to power the GC-ALPINE-RIDGE. In this configuration I could feed two DisplayPort cables to the TB3 AIC for Thunderbolt 3 monitor output.
The Thunderbolt 3 mainboard is similar to AKiTiO enclosures in that it sits vertical to the base PCIe slot board. It occupies one of the two slots and is mounted to the rightmost edge. The other slot hosts the graphics card. Unlike the HL23T Hercules which has two Thunderbolt 3 ports, the HL23T-Plus only has a single upstream TB3 port to the host computer. A very crucial component found on this TB3 main board is the white 3-pin connector that gets power-state signals from the PSU. This is how the HL23T-Plus can properly put the Radeon VII to suspend mode.
Above this power signal connector is a strip of six vertically-placed LED diodes. They are labeled with different power voltage inputs and mainly used for diagnostics. I found them beneficial to observe the enclosure power state during testing. Also on this mainboard are components found on all Thunderbolt 3 enclosures. The single Alpine Ridge JHL6540 Thunderbolt 3 controller sits near the TB3 port. Next to it is the EEPROM firmware chip. The Texas Instrument TPS65983 USB-C controller sits near the PCIe pins.
In most Thunderbolt 3 eGPU enclosures the graphics card communicates with the enclosure’s USB-C controller to determine in which power state it should operate. The Radeon VII’s mainboard design somehow doesn’t have this functionality in place. It cannot interpret GPIO signal from the TI83 to go into suspend mode when the host computer goes to sleep or shuts off. Instead it goes hog wild, running three fans full speed when it thinks there’s no active Thunderbolt communication from a host.
Compared to Hercules’ 300W PSU, power output in the HL23T-Plus improves to 400W. They both share the same flex-ATX PSU form factor. This way the power supply can be placed at the bottom of the enclosure under the PCIe slot base board. The FSP400-60FGGBA is a 12V dual-rail unit. There are two 6 + 2-pin PCIe power cables for the graphics card. Other cables found are a 3-pin connector to the Thunderbolt 3 mainboard and a 4-pin Molex connector to power the base PCIe slot board. Equipped with one tiny 40mm fan, PSU noise was not as bad as I anticipated. During idle I could not distinguish GPU noise over the PSU noise and vice versa. During heavy load, the Radeon VII is definitely a screamer.
Thunderbolt 3 firmware is version 25.1 and provides full Thunderbolt 3 bandwidth at 22Gbps (capped by Intel). Power Delivery is 15W same as the HL23T Hercules. This limitation makes the HL23T-Plus a more suitable companion to a desktop such as the 2018 Mac mini and 2013 Mac Pro. I used these two host computers for a four-month long testing. Primarily I tested them in both macOS 10.14 Mojave and 10.15 Catalina with some Windows 10 Boot Camp action on the 2018 Mac mini.
With an estimated max GPU power of 350W, the HL23T-Plus can power the Radeon VII without issue. RX Vega 64 as usual is not suitable for eGFXs with a power supply producing less than 650W output. During one test that included an add-in card, I used one PCIe power cable to power a riser board and the other PCIe power cable to power the Radeon VII (through a Y-splitter). This was not sufficient because the graphics card didn’t have access to both 12V rails from the PSU.
My first extended test was with the 2013 Mac Pro’s Thunderbolt 2 connection through Apple Thunderbolt 3/USB-C to Thunderbolt adapter. Apple blocks Thunderbolt 1 & 2 Macs from using eGPU in macOS (since 10.13.4). Thanks to mac_editor’s Purge-Wrangler script, it’s possible to unblock this restriction. The NetStor HL23T-Plus and Radeon VII were working perfectly in 10.14. Sleep mode was normal as if the Mac Pro was working without any eGPU connected. The “Power Nap” feature in Energy Saver can trigger the eGPU to wake from sleep when the Mac system performs scheduled tasks. It’s best to disable this feature for predictable sleep behavior. Catalina 10.15 introduced a periodic enclosure wakeup which doesn’t wake the host computer.
An important reminder is that recent macOS versions tend to include a firmware update during the system update. It’s safest to disconnect all peripherals aside from the mouse and keyboard while the system undergoes updates. I learned this the hard way during my HL23T-Plus testing. I left the eGPU connected during a macOS Catalina Beta update, and the Mac Pro’s firmware was corrupted. I went through quite an ordeal to bring my Mac Pro back to life through the use of a solderless EFI chip. It essentially had a heart transplant.
While the trashcan was out of commission, my second test was with the 2018 Mac mini. This pairing was much less eventful thanks to native support in macOS for all components. As summarized in my macOS Catalina eGPU overview, the 2018 Mac mini gained boot screen support. I could use an external monitor through the HL23T-Plus + Radeon VII eGPU to switch between macOS and Windows. It’s worth noting that currently there’s a bug in Polaris drivers that causes white screen boot hang when the paired graphics card is RX 590/580/570/560/480/470/460 and Radeon Pro WX 7100/5100/4100. Until Apple resolves this issue through a macOS system and firmware update, it’s best to hot-plug the eGPU.
Sleep behavior testing was done throughout many builds of macOS 10.14 Mojave and 10.15 Catalina. I mixed the NetStor with different Radeon cards and, more importantly, paired the Radeon VII with different eGPU enclosures. The Radeon VII worked best with the NetStor HL23T-Plus in 10.14. When the Mac was set to sleep, the Radeon VII would go into sleep mode with all fans coming to a halt. On the Thunderbolt 3 mainboard, only LED2 (3V3) was on. LED3-LED6 turned off as expected. This indicated proper sleep. Once the NetStor eGPU enclosure is awake, LED2-LED6 are lit up green.
The ultimate sleep test was with two Radeon VII graphics cards. One was in the NetStor HL23T-Plus and the other in the AKiTiO Node Pro. They worked identically when paired to the same host system. The sole exception is during sleep and shut down. The Radeon VII cooling fans inside the HL23T-Plus would spin down to a full stop while the ones inside the Node Pro would spin up to full speed.
Some enclosures such as the Mantiz Venus mitigate fan issue during sleep by going into an idle sleep mode. In this mode, the Radeon VII thinks the Thunderbolt 3 connection is active but there’s no load so the cooling fans run at idle speed. In the latest builds of Catalina, I’ve observed this idle sleep mode on more enclosures compared to 10.14. I have reached out to my AMD XConnect contact as well as eGFX vendors for more information, and they confirmed the only permanent fix would need to come from AMD through a firmware update.
If you recall AMD was quick to issue a firmware update shortly after the Radeon VII launch in early February 2019. In my opinion this graphics card was rushed to production for the announcement of 7nm process. It’s a very strong compute card and yet AMD marketed it as the world’s first 7nm gaming card. It’s not by chance Apple decided to offer the 2019 Mac Pro BTO configuration with up to 4 Radeon Vega II GPUs.
Windows eGPU through Boot Camp remains a lot of work depending on different factors. First of all is the Mac firmware (Boot ROM version). The second crucial factor is Windows OS Build number. Recently we’ve learned Windows 10 1903 OS Build 18362.329 and later causes error 12 on many systems (some non-Mac as well) when paired with an AMD external graphics card. It’s best to run Windows 10 1809 or stay at 18362.295 and disable Windows Updates service to prevent automatic system updates. Another challenge is the newest Mac firmware following the Catalina update breaking Boot Camp eGPU workarounds such as rEFInd spoof settings and apple_set_os.efi in automate-eGPU EFI.
I commend NetStor for setting out to build the perfect eGPU enclosure for the Radeon VII. They have accomplished this goal. The problem is no one loves the Radeon VII as much as NetStor does. This singular focus narrows the HL23T-Plus’ target market to Mac professionals who need the computing power of the Vega II architecture.
Share this Post