I was very intrigued to see the Blackmagic eGPU announcement today (and, of course, disappointed that it only has an RX 580 inside).
However, what surprised me the most was the fact that the enclosure includes USB ports and a downstream Thunderbolt 3 port due to the potential for that decision to impose a substantial performance hit (which is why many TB3 eGPU enclosures like the Sonnet one I own do not include any ports beyond the 1 required upstream TB3 port):
My (albeit amateur) understanding is that PCI Express lanes are discrete (cannot be sub-divided). Therefore, when you add any accessory (even a simple USB hub) to a Thunderbolt 3 enclosure, you lose one entire PCIe lane (out of four) that could be used to connect the GPU, even if the USB Hub only uses a small percentage of the bandwidth afforded by that lane. If this is the case, that would seem to represent a substantial loss (25% of total bandwidth) especially considering that Thunderbolt 3's 4 PCIe lanes are already only 25% of the lanes that would be used to connect a desktop GPU (16) (or 50% of the 8 typically used for laptop GPUs). And when you consider that you would be giving up 25% of your already limited bandwidth to a GPU just for the sake of a few USB ports, this would seem to be a fairly bad trade.
I was wondering if any of the Blackmagic staff (or anyone else for that matter) could clarify:
1. Do the USB Ports on the Blackmagic eGPU utilize 1 of the 4 PCIe lanes of the Thunderbolt 3 connection thereby limiting the GPU to 3 PCIe lanes? (Or does the USB hub use a different part of the Thunderbolt protocol ride outside of the PCIe lanes?)
2. If the USB hub does use one of the PCIe lanes, does it relinquish that lane to the GPU when no USB devices are connected?
Bonus questions:
3. When daisy-chaining a Thunderbolt 3 or DisplayPort display off of the Blackmagic eGPU, does that display have its signal routed directly from the GPU downstream to the display (as would be the case when connecting via the HDMI port) or is it routed back to the host computer before being redirected back down the Thunderbolt connection to the display?
4. Any shot at a Vega 16GB model? (It is 2018, after all!)
I am very glad to see so many companies, especially Blackmagic and Apple embracing and investing in eGPU technologies. I absolutely love being able to use a powerful GPU at home, unplug a cable, and bring a compact laptop with me on the road (as opposed to copying files between my desktop and laptop every time I traveled – which is what i used to do). Now if only we could get some modern GPUs on the Mac side!
However, what surprised me the most was the fact that the enclosure includes USB ports and a downstream Thunderbolt 3 port due to the potential for that decision to impose a substantial performance hit (which is why many TB3 eGPU enclosures like the Sonnet one I own do not include any ports beyond the 1 required upstream TB3 port):
My (albeit amateur) understanding is that PCI Express lanes are discrete (cannot be sub-divided). Therefore, when you add any accessory (even a simple USB hub) to a Thunderbolt 3 enclosure, you lose one entire PCIe lane (out of four) that could be used to connect the GPU, even if the USB Hub only uses a small percentage of the bandwidth afforded by that lane. If this is the case, that would seem to represent a substantial loss (25% of total bandwidth) especially considering that Thunderbolt 3's 4 PCIe lanes are already only 25% of the lanes that would be used to connect a desktop GPU (16) (or 50% of the 8 typically used for laptop GPUs). And when you consider that you would be giving up 25% of your already limited bandwidth to a GPU just for the sake of a few USB ports, this would seem to be a fairly bad trade.
I was wondering if any of the Blackmagic staff (or anyone else for that matter) could clarify:
1. Do the USB Ports on the Blackmagic eGPU utilize 1 of the 4 PCIe lanes of the Thunderbolt 3 connection thereby limiting the GPU to 3 PCIe lanes? (Or does the USB hub use a different part of the Thunderbolt protocol ride outside of the PCIe lanes?)
2. If the USB hub does use one of the PCIe lanes, does it relinquish that lane to the GPU when no USB devices are connected?
Bonus questions:
3. When daisy-chaining a Thunderbolt 3 or DisplayPort display off of the Blackmagic eGPU, does that display have its signal routed directly from the GPU downstream to the display (as would be the case when connecting via the HDMI port) or is it routed back to the host computer before being redirected back down the Thunderbolt connection to the display?
4. Any shot at a Vega 16GB model? (It is 2018, after all!)
I am very glad to see so many companies, especially Blackmagic and Apple embracing and investing in eGPU technologies. I absolutely love being able to use a powerful GPU at home, unplug a cable, and bring a compact laptop with me on the road (as opposed to copying files between my desktop and laptop every time I traveled – which is what i used to do). Now if only we could get some modern GPUs on the Mac side!
Resolve Studio × Micro Panel, MacBook Pro 16" × eGPU (Vega Frontier Edition), Canon C200, Pocket Cinema Camera 4K × Metabones, Pocket Cinema Camera × Metabones, Eizo CG248-4K