[Unboxing & Teardown] Blackmagic eGPU Radeon Pro 580 Thunderbolt 3 Enclosure  

But I thought that the Mac Pro already had a good video chip inside it.

Awesome, I like it very much.
keep us updated.
i’ve noticed Packaging states Pro 580 but also RX 580 on the top side. Maybe a typo.

@ikir Yes, they misprinted on the box AMD-RX580 rather than Radeon Pro 580. Sometimes keeping things tightly under wraps can be bad. 😛

Considering the RX 580 and Pro 580 are the same GPU (just that "Pro" is the iMac variant), I have to wonder if the decision to call this one a Radeon Pro 580 was a late decision Apple pushed Blackmagic into so it sounds fancier, haha.

@detunedradios "More pro" is the way to go.  😉

Under Windows, Radeon Pro devices have unlocked drivers given better professional app performance compared to their Radeon counterparts.  Though this eGPU isn't supported under bootcamp. 

So it would seem the RP580 naming is merely a continuation of the RP560/RP580 used for the MBP and iMac for marketting purposes?

@nando4 I'm afraid so. I connected this Blackmagic eGPU to the Razer Blade Stealth and Radeon drivers could not identify the graphics card. Luckily there were drivers from previous setups so the eGPU was functional. When I tried to update the drivers to the latest version, it refused to proceed due to GPU identification error. I don't recommend this eGPU if you're planning on running Bootcamp.

As is typically with GPU inside a Mac, the Radeon 580 in this Blackmagic eGPU is clocked at 1200MHz. This is lower than the reference RX 480 that came out two years ago, clocked at 1266MHz. I ran GPU-Z, AIDA64, and HWiNFO64 to gather some specs. I also saved a dump of the ROM.

More bad new is how challenging this enclosure is to take apart. I managed to remove the internal block [PSU+GPU+cooler] from the shell. Both the top and bottom plastic vents have hidden Torx screws. Definitely not servicable at all. The power supply is a Mean Well EPP-400-12 with 400W output. What I found really interesting are these two mini-B USB ports.

That's it for tonight. I will separate the GPU and PSU from the block tomorrow.

Gutted Bootcamp is a bigger hurdle than usual....

Posted by: joevt

15ef JHL7540 Thunderbolt 3 Bridge [Titan Ridge DD 2018]

15f0 JHL7540 Thunderbolt 3 USB Controller [Titan Ridge DD 2018]

Maybe this USB thing is a feature of the DD type controller? 2C and 4C are for single and dual DisplayPort input controllers?

The product names in the pci ids database for Titan Ridge might be wrong. 2C might be JHL7340, 4C might be JHL7540, and DD might be JHL7440. The JHL7440 has been described as usable in docks or displays that can connect to a computer's Thunderbolt port or USB-C port. This new feature may explain how the USB in this eGPU works.

Hello,

@TheItSage, as always can you provide me a decent high resolution picture please.

For pcb and components analysis.

What i can tell just looking at the assembly:
-The psu + eGpu board seems well done, the gpu connection are designed to support high current as it should.
-If the air is pulled from the top to the bottom, i like to see the filtering cap ahead the VRM heatsink, to maximize the capacitor cooling, capability,  reliability.
-Either the gpu, the VRM, the ram seems own a heatsink. These are based on passive or low airflow heatsink design with quite a nice surface area.
-If i understand it well, the gpu is soldered on the eGPU board, it mean that the board have been implemented to hold the gpu, ram, power, TB3 I/O components packed into a same pcb. Pretty curious.
-The two covered USB are very interesting, are they working?

On overall at 1st glance it seems engineered well, looking at an electrical build side only, no clues about software.

Thanks you.

I accidentally overwrote my first reply. What I said was:

Looks like this eGPU uses Titan Ridge (looking at the ids at  https://pci-ids.ucw.cz/read/PC/8086  )

The eGPU's Titan Ridge controller has two Thunderbolt 3 ports AND a USB controller/port connected to a USB hub. This was not possible with Alpine Ridge which did not have a separate USB port (USB could only come out the two Thunderbolt ports or a separate USB controller connected via PCIe to the Thunderbolt controller).

If you click "USB 3.1 Bus" in System Information.app, does it show the Titan Ridge PCI ids?

The USB hub has USB vendor ID 1EDB = "BLACKMAGIC DESIGN PTY.".

It is unclear if a display connected to the eGPU is connected to the eGPU's Radeon or to the computer's GPU. What does IORegistryExplorer.app say the display is connected to? Maybe include a ioreg dump? lspci? Or try my pcitree.sh script at #223 (lspci and pcitree won't show displays and usb, but they do show pci devices).

Imagine a Thunderbolt 3 chain like this:
Computer -> dock 1 -> Blackmagic eGPU -> dock 2.
If a display is connected to dock 1, then I expect it to get video from the computer's GPU. If a display is connected to dock 2, then does the video come from the eGPU's GPU directly, or does it come from the computer's GPU?

Imagine a Thunderbolt 3 chain like this:
Computer -> NVMe device -> eGPU -> displays.
If two 4K displays (a configuration that uses the max display bandwidth of Thunderbolt 3) connected to the eGPU is getting video from the eGPU, then the NVMe device should be able to get near max write performance 2155 MB/s?

That is simultaneously the most beautiful and disgusting thing I have ever seen....

On the plus side, it is Titan Ridge and it appears to be an eGPU with daisy chaining.

On down side, only 85W PD and GPU does not appear to be upgradable. It is not modular at all. It also has an HDMI port.

I quoted the wrong device ID for the USB controller. It should be:
15ec JHL7540 Thunderbolt 3 USB Controller [Titan Ridge 4C 2018] I don't know why the bridges are 15ef (DD) and the USB controller is 15ec (4C). Are there two Thunderbolt controllers in the eGPU? It doesn't look that way from the HWiNFO screen shots. Maybe the 15ec is common to the 4C and DD and 15f0 is a special case for the DD only.

What does HWiNFO or IORegistryExplorer.app show when you connect a USB device to a Thunderbolt 3 port of the eGPU? I expect it to connect to Port1/Port3 of the USB Root Hub of the 15ec.

Posted by: karatekid430

On down side, only 85W PD and GPU does not appear to be upgradable. It is not modular at all. It also has an HDMI port.

It's an HDMI 2.0 port so not the biggest minus (except it costs $200 to convert that back to DisplayPort where as DisplayPort to HDMI 2.0 only costs $30 or $40 for HDR). Look for a DisplayPort to HDMI 2.0 chip (LSPCon?)on the PCB (Because I don't think DisplayPort to HDMI 2.0 is a function of the Thunderbolt controller but the Thunderbolt controller can do HDMI 1.4 by itself and the HDMI port just needs voltage level adjustment. But maybe Titan Ridge adds HDMI 2.0 because Titan Ridge supports DisplayPort 1.4 which is supposed to support HDMI 2.0 dual mode???).

Does this eGPU automatically go into a standby mode when not connected to a MacBook?

And does the same happen to any screens with an auto standby function that are connected?

Or do you have to be sure to switch this on and off - every time you wish to connect a MacBook and use it?

@joevt3 Nice observation of HWiNFO64 screen capture. You are right on. The Thunderbolt 3 main board packs a lot of goodies. It's the first one with Titan Ridge JHL7540 Thunderbolt 3 controller. It also sports two TI83 USB-C controllers. The tear down was not fun due to the amount of components and Torx screws. Before I could access the internal block, I had to remove 7 components and 27 Torx screws. There are plenty more screws holding the GPU & PSU together. Here are a couple of photos.

Posted by: itsage

@joevt3 Nice observation of HWiNFO64 screen capture. You are right on. The Thunderbolt 3 main board packs a lot of goodies. It's the first one with Titan Ridge JHL7540 Thunderbolt 3 controller. It also sports two TI83 USB-C controllers. The tear down was not fun due to the amount of components and Torx screws. Before I could access the internal block, I had to remove 7 components and 27 Torx screws. There are plenty more screws holding the GPU & PSU together. Here are a couple of photos.

I know you're still working on the tear down. The motherboard picture doesn't show the chips clearly. Can you confirm that it's a JHL7540 or is it the JHL7440 as I speculated? Then you have to put all back together again to answer all the other questions (DisplayPort 1.4? graphics from eGPU or computer?, etc.)

I wonder of macOS even supports DisplayPort 1.4 (from eGPU as anything with an Alpine Ridge can only do DisplayPort 1.2).

I wonder why manufacturers of Titan Ridge devices are not mentioning Titan Ridge in their product descriptions? Is that something Intel is telling them not to do? Maybe they want it to be about features and not chips (though mentioning the chip is the easiest way to describe features or the limits of the features). Well, a sample size of two is not large enough to make a conclusion.

@joevt3 The TB3 controller is JHL7540. I talked to a source at Intel and the primary benefit of the Titan Ridge controller is its ability to dynamically communicate with both Thunderbolt 3 hosts and USB-C hosts. Alpine Ridge devices require Thunderbolt hosts. In terms of eGFX, Thunderbolt has always been a prerequisite so another advantage of the new TB3 controller is the ability to route DisplayPort signal from the GPU through the controller so that there's Thunderbolt 3 monitor output for 5K displays. Here are a photo showing the traces from GPU die to Thunderbolt 3 controller then out to TB3 ports.

I put everything back together and all is working well. I don't have the LG Ultrafine 5K monitor so I can only test a 4K DisplayPort monitor via a USB-C to DisplayPort adapter (in transit). I'm also waiting on the 2018 13" MacBook Pro to arrive and confirm whether it has the newer Thunderbolt 3 controllers as well. From what I heard there's no performance difference between Alpine Ridge vs. Titan Ridge.

Thanks @itsage!
Other than the support of thunderbold monitors, it would be nice to know if there are any other differences. For example if the RX580 is used in another eGPU enclosure should it offer the same performance/ acceleration. I am pretty sure you were planning to test all these anyways...

Posted by: itsage

@joevt3 The TB3 controller is JHL7540. I talked to a source at Intel and the primary benefit of the Titan Ridge controller is its ability to dynamically communicate with both Thunderbolt 3 hosts and USB-C hosts.

According to all the Intel press releases, only the JHL7440 has the feature of being able to communicate to a USB-C host (I assume through one of it's Thunderbolt ports). Do all the Titan Ridge chips have the extra USB output port that this eGPU seems to be using? I think that didn't exist in Alpine Ridge. Can your Intel guy verify the pci ids in the pci ids database (which ids belong to the JHL7340, JHL7440, JHL7540)? I'd like to understand why your HWiNFO shows bridges of 15ef (DD) and a USB controller of 15ec (4C).

Posted by: itsage

In terms of eGFX, Thunderbolt has always been a prerequisite so another advantage of the new TB3 controller is the ability to route DisplayPort signal from the GPU through the controller so that there's Thunderbolt 3 monitor output for 5K displays.

All the Alpine Ridge chips had DisplayPort inputs. Maybe those inputs could only be used in a host configuration (motherboards and add-in cards)? Maybe use of those inputs in a peripheral device (e.g. this eGPU) is new to Titan Ridge?

I though of a third test. Imagine this Thunderbolt 3 chain:
computer -> dock 1 -> dock 2 -> eGPU -> dock 3
You could have a display connected to dock 1, another display connected to dock 2, and two displays after the eGPU because displays between the computer and eGPU would get video from the computer and displays after eGPU would get video from the GPU? In this configuration, all the displays can be 4K 60 Hz.

What if there were four displays after the eGPU? A Thunderbolt 3 controller can provide two DisplayPort streams (not counting MST). So maybe two could come from the computer and two could come from the eGPU but you would have no way of deciding which display gets which stream unless they were enumerated in a deterministic manner. In this configuration, the four displays would be limited to 2560 x 1440 @ 60 Hz.

Posted by: itsage

Here are a photo showing the traces from GPU die to Thunderbolt 3 controller then out to TB3 ports.

It's difficult to say that those traces are PCIe or DisplayPort or both (looks like 8 wide traces but could be 8 pairs of narrow traces - maybe there's multiple layers or something on the back side). If they are only PCIe, then graphics can only come from the computer. At least it appears the HDMI 2.0 is coming directly from the eGPU which means HDMI video can't come directly from the computer.

Posted by: itsage

I don't have the LG Ultrafine 5K monitor so I can only test a 4K DisplayPort monitor via a USB-C to DisplayPort adapter (in transit). I'm also waiting on the 2018 13" MacBook Pro to arrive and confirm whether it has the newer Thunderbolt 3 controllers as well. From what I heard there's no performance difference between Alpine Ridge vs. Titan Ridge.

Tests using a Thunderbolt 3 to Dual DisplayPort adapter would be useful. It allows testing both streams with one inexpensive device but it's only useable at the end of the Thunderbolt 3 chain.

Posted by: kotlos

Other than the support of thunderbold monitors, it would be nice to know if there are any other differences. For example if the RX580 is used in another eGPU enclosure should it offer the same performance/ acceleration. I am pretty sure you were planning to test all these anyways...

You mean a different RX580 of the same model, since this eGPU has a RX580 soldered on to the Thunderbolt PCB.

@joevt3, using this and this image as reference, in the second image we see:

1. top:( 16 wires) = 4 PCIe lanes leading to the GPU
2. left-top (4 wires) = 2 TB3 ports
3. right: (20 wires) = is it even wired? If it is then could be a displayport set of lanes:  DP_1, DP_2, DP_3, DP_4, DP_AUX

Without seeing a direct RP580 GPU DP <> TB3 controller link, it looks like it's simply the same as before:

TB3 display <> TB3 controller <> notebook TB3 controller <> notebook GPU

Only difference here is this new TB3 controller supports DP 1.4 and this RP580 has no Windows AMD drivers.

@TheItSage
Thx for the pictures, can you take one that cover the whole pcb with the heatsink nearby.
To have a general view of the pcb.
I can't say because not leveraged each chip and the circuit in details, but on overall it seems a solid build!

I want take the time to explain again how i think TB works and how it is wired. Maybe worth build a dedicated thread for the purpose.
First take some time looking at how the TB IC is connected to other components. Maybe it changed but i think it's a good stating point.

-See here for the TB IC configuration. 
-See here for the TSP65983 configuration.
-See here for the ASMEDIA USB 3.1 configuration.

I draw this little scheme, hope it is right and help explain a bit more what going on.

Unfortunately i couldn't find any implementation using the updated JHL7540.
I'm pretty curious how the USB hub works, i mean there is a daughter board with a UBS controller.
There is also an Atmel ARM based microcontroller is suppose, near the daughter board socket, wonder what it is used for.

I’m curios about performance too

@Ikir, this uses the same GPU as found in a RX580 card, eg: Gigabyte Gaming Box, but it is downclocked in this way:

GPU clock: 1200 vs 1340 (-10.4%)
memory clock: 1695 vs 2000 (-15.3%)

Presumably to keep temperature and power consumption in check.

This means you can review any of these TB3 RX580 build's benchmarks and subtract ~10% to get it's performance level.

https://egpu.io/builds/?date?table_filter ="-TB3 "RX 580""

Posted by: nando4

@joevt3, using this and this image as reference, in the second image we see:

1. top:( 16 wires) = 4 PCIe lanes leading to the GPU
2. left-top (4 wires) = 2 TB3 ports
3. right: (20 wires) = is it even wired? If it is then could be a displayport set of lanes:  DP_1, DP_2, DP_3, DP_4, DP_AUX

Without seeing a direct RP580 GPU DP <> TB3 controller link, it looks like it's simply the same as before:

TB3 display <> TB3 controller <> notebook TB3 controller <> notebook GPU

Only difference here is this new TB3 controller supports DP 1.4 and this RP580 has no Windows AMD drivers.

Oh, right. I should have seen that the zoomed in picture clearly shows the 8 wide traces as 8 pairs of narrow traces.

It would be sad if there's no DisplayPort connection to the GPU. The "Graphics/Displays" screenshot shows a wide screen display (32:9 3840 x 1080) connected to the eGPU's GPU as a Television. I had originally thought it was connected as DisplayPort and that it might be a clue that video was coming directly from the eGPU's GPU. But since it says "Television" and itsage said he's waiting for a USB-C to DisplayPort adapter, does that mean the display is connected to the HDMI 2.0 port which does seem to be connected directly to the eGPU's GPU? The refresh rate is only 59 Hz (HDMI 1.4?) instead of the max 120 Hz (HDMI 2.0). Do you need a macOS AMD pixel clock patch (I don't know if one exists)? I am assuming C49HG9x refers to the C49HG90DMM?

Posted by: wimpzilla

I want take the time to explain again how i think TB works and how it is wired. Maybe worth build a dedicated thread for the purpose.
First take some time looking at how the TB IC is connected to other components. Maybe it changed but i think it's a good stating point.

-See here for the TB IC configuration. 
-See here for the TSP65983 configuration.
-See here for the ASMEDIA USB 3.1 configuration.

I draw this little scheme, hope it is right and help explain a bit more what going on.

 

Unfortunately i couldn't find any implementation using the updated JHL7540.
I'm pretty curious how the USB hub works, i mean there is a daughter board with a UBS controller.
There is also an Atmel ARM based microcontroller is suppose, near the daughter board socket, wonder what it is used for.

The PD IC shows USB 2.0. Isn't that just for the Billboard Device and that when the Billboard Device is not needed then USB 2.0 comes from the TB controller along with the USB 3.1? I didn't look too closely at the documentation or I don't understand it all so I'm not sure. There does appear to be USB2 signals from the TB controller, and the PD IC has a lot of muxing going on. 

The Alpine Ridge in your motherboard schematic has these main parts:
1) TBT Port A
2) TBT Port B
3) DP Sink Port 0
4) DP Sink Port 1
5) DP Source Port
6) PCIe 3.0 x4

From the HWiNFO screenshot, we are assuming that Titan Ridge has another part:
7) USB 3.1 port

We don't know for sure if the Titan Ridge is using the DP Sink Ports. It appears the HDMI is coming from the GPU directly, and not the DP Source Port. As an example, the GC-ALPINE RIDGE uses all 6 parts (DP Source is used in DP++ dual mode for HDMI 1.4 output).

The Titan Ridge USB port must then be connected to a USB 3.0 hub somewhere. Is the USB hub on the daughterboard containing the four USB ports (there is a picture but it's not zoomed in)? The daughterboard connector seems to have more wires than a single USB connection requires. Maybe the connector has signals for all four USB ports, meaning that the USB hub is on the main board with the TB controller. Maybe the hub is the ATMEL chip (I can't read the markings - I think it says USB near the end)?

To be honest i do not know how an USB Hub controller could be linked to the TB controller, if i take the example of all the enclosures using docking feature without doubling the TB controller.
Also bear in mind that the TB controller is also an USB 3.1 controller on it'sown.

I would suppose that an USB IC Hub need a PCI-E lane connection to be linked to the TB controller, however all the lanes are already occupied by the gpu.
I did not understood how the TB controller accept accept additional USB Husb without doubling the TB scheme.

I would suppose, as you said, some data muxing must occur either between the IC PD chip or the TB controller and the USB IC Hub.
In any case there are a lot of possible implementation, mixing a lot of protocols, hard to tell whats going on, especially because i did not spent time on this.

The only thing that i would check, is the usual issue when using USB devices meanwhile playing or heavy gpu loads.
To verify that no issues occur and works as a featured double TB controller. 

I did more testing this morning. To get Thunderbolt 3 monitor output, I connected the Mantiz Titan TB3 dock to the second TB3 port then run a DisplayPort cable to a LG 4K monitor. The monitor is accelerated by the Radeon Pro 580 eGPU. Thunderbolt tree in System Information also showed the nice arrangement 20 Gb/s upstream to the nMP and 40 Gb/s downstream to the Mantiz dock.

Here are two photos I took of the Blackmagic eGPU PCBs.

It’s beginning to evolve into a really bright future. 
Now we need a aggregated TB3 solution and more CPU lanes.

@itsage, if you boot Windows and run an app like Unigine Heaven off the Mantiz Titan attached LG 4K LCD, does performance reflect it being eGPU accelerated?

That's a test outside of macOS as Apple may be applying some tweaks to make the TB display appear attached to the eGPU.

Posted by: wimpzilla

To be honest i do not know how an USB Hub controller could be linked to the TB controller, if i take the example of all the enclosures using docking feature without doubling the TB controller.
Also bear in mind that the TB controller is also an USB 3.1 controller on it'sown.

I would suppose that an USB IC Hub need a PCI-E lane connection to be linked to the TB controller, however all the lanes are already occupied by the gpu.
I did not understood how the TB controller accept accept additional USB Husb without doubling the TB scheme. 

The previous enclosure examples all used Alpine Ridge (as far as I know). What we're talking about here is the new Titan Ridge which might have a new method of getting USB (an extra USB port in addition to the ports included in Alpine Ridge or maybe instead of some ports of Alpine Ridge - we don't know.)

I think the External GPU Buyer's Guide could add a couple extra rows of information describing the I/O ports. For example, for this eGPU:
I/O ports : 4 port USB 3.0 hub
I/O method : Titan Ridge USB

For the AKiTiO Node Duo:
I/O ports : 2nd PCIe slot
I/O method: TB PCIe split x2,x2

For the Sonnet Echo Express III-D:
I/O ports: 2nd & 3rd PCIe slots
I/O method: PCIe switch x8,x8,x4

For the Razer Core V2:
I/O ports: 4 USB 3.1, Ethernet
I/O method: 2nd TB PCIe split x1,x1,x1

Posted by: itsage

I did more testing this morning. To get Thunderbolt 3 monitor output, I connected the Mantiz Titan TB3 dock to the second TB3 port then run a DisplayPort cable to a LG 4K monitor. The monitor is accelerated by the Radeon Pro 580 eGPU. Thunderbolt tree in System Information also showed the nice arrangement 20 Gb/s upstream to the nMP and 40 Gb/s downstream to the Mantiz dock.

Depending on how many docks you have, you could try some of the Thunderbolt chains I suggested. With a single dock, you could try the following thunderbolt chain:
computer -> dock -> eGPU
Connect the display to the dock. Does macOS show the display as connected to your computer's GPU?

The chain with a NVME device should definitely prove there is no graphics communication between the computer's GPU and displays connected after the eGPU (only if you have two 4K displays). Well, if the read/write speeds are < 2000 MB, then it means there is graphics communication too/from the computer's GPU or there is a design flaw somewhere (Titan Ridge or macOS).

Thanks you for the pictures, give a nice look at the whole board.

Tbh, i seems nicely done, what need to be cooled have an heatsink, the VRM on the back of the pcb are placed on the opposite side of the gpu vrm.
They have a plenty of pcb surface to dissipate heat, should be ok since they are not designated to high power draw components.
The pcb on overall is well deigned, i just do not understand why not extend the cooling also the TB IC, maybe i'm wrong thinking it heat quite a lot and could throttle!
It should perform well, further in depth testing will tell, tho not sure i would overclock too much, since the power plane is shared by the other components.
Even if the gpu pack a full 6+1 or 3(doubled)+1 phases, it should be possible to hit AIB clocks if cooled enough.

On overall i like the minimalist design, even if components are packed close together, it seems have been put together with common sense.
Will not comment on price, marketing,  software have no knowledge about.

To inquire deeply how the USB hub is connected one would need high res pics of the pcb to follow the traces.
And the IC's part number to check which component it belong, the daughter board pack 3 major IC's, the Atmel chip is also interesting.
Could be interesting to do the same with other enclosures, to compare further the update from the DSL6540 to the JHL7540.

Posted by: nando4

@itsage, if you boot Windows and run an app like Unigine Heaven off the Mantiz Titan attached LG 4K LCD, does performance reflect it being eGPU accelerated?

That's a test outside of macOS as Apple may be applying some tweaks to make the TB display appear attached to the eGPU.

I tried the same arrangement with the Razer Blade Stealth as you suggested. There was absolutely no monitor output even through the HDMI port directly from Blackmagic eGPU to LG monitor. The monitor detected there's a video input when I plugged it in and the internal display of RBS flickered indicating there's an external monitor. There was nothing showing on the LG monitor however.

I've thought of another test to prove that graphics are not going through the computer's GPU:
PC with Thunderbolt 3 add-in card -> eGPU -> display connected to eGPU's Thunderbolt port.

With the add-in card, you can disconnect the DisplayPort inputs. This will prove the display is driven by the eGPU.

A couple issues:
1) Windows doesn't like this eGPU so you need to use a Hackintosh.
2) Hackintosh has it's own issues with Thunderbolt (hot plug, etc.) but when it works, it works.
3) Hopefully the add-in card doesn't actually require use of the DisplayPort inputs to enable PCIe tunneling over Thunderbolt.
4) If you have a PC with PCIe slots then you don't need a eGPU, but maybe you have a desktop computer to go with your laptop.

are you getting the boot issue with most egpus on macs with windows?