Notifications
Clear all

USB4 Specification Released  

  RSS

SuperCoolBeaver
(@supercoolbeaver)
New Member
Joined: 1 year ago
 

The USB4 Specification was released today. It optionally supports interoperability with Thunderbolt 3 products. You can get from the USB Document Library.

To do: Create my signature with system and expected eGPU configuration information to give context to my posts. I have no builds.

.

ReplyQuote
Ray
 Ray
(@ray)
Eminent Member
Joined: 3 years ago
 

When can we start seeing this make its way into products? And can machines without USB4 but with Thunderbolt 3 (like my 2018 MBP) somehow be updated to support it? I doubt do because I’d imagine it would need a new controller, BUT if it’s interoperable with Thunderbolt 3, maybe the Thunderbolt controller can be double as a USB 4 controller. Only time will tell I suppose.

To do: Create my signature with system and expected eGPU configuration information to give context to my posts. I have no builds.

.

ReplyQuote
C Films
(@c-films)
Active Member
Joined: 11 months ago
 

Hopefully it means egpus will work with it as well.

To do: Create my signature with system and expected eGPU configuration information to give context to my posts. I have no builds.

.

ReplyQuote
nu_ninja
(@nu_ninja)
Reputable Member
Joined: 2 years ago
 

I think USB4 with thunderbolt 3 capability will be eGPU compatible, whereas USB4 without thunderbolt/pcie will not. I would hope nearly all USB4 products add the thunderbolt/pcie functionality.

From what I've read, USB4 will be backwards compatible with previous USB's. Combined with TB3 interoperabiltiy, I think this means USB4 devices will work with TB3 ports OOB, at least in some capacity.

Mid-2012 13" Macbook Pro (MacBookPro9,2) TB1 -> RX 460/560 (AKiTiO Node/Thunder2)
+ macOS 10.15+Win10 + Linux Mint 19.1

 
2012 13" MacBook Pro [3rd,2C,M] + RX 460 @ 10Gbps-TB1 (AKiTiO Thunder2) + macOS 10.14.4 [build link]  


ReplyQuote
odin
 odin
(@odin)
Trusted Member
Joined: 2 years ago
 
Posted by: @ray

When can we start seeing this make its way into products? And can machines without USB4 but with Thunderbolt 3 (like my 2018 MBP) somehow be updated to support it? I doubt do because I’d imagine it would need a new controller, BUT if it’s interoperable with Thunderbolt 3, maybe the Thunderbolt controller can be double as a USB 4 controller. Only time will tell I suppose.

Probably late 2020 early 2021, and then consider a long period of trickling out of products and refining software.

LG Gram 17 | Sonnet Breakaway Box 550 | Asus Strix RTX 2070 OC Edition | Win 10 Pro 2004 + Fedora 32 Dual Boot
Build Link

 
2018 17" LG Gram 17 [8th,4C,U] + RTX 2070 @ 32Gbps-TB3 (Sonnet Breakaway 550) + Win10 [build link]  


ReplyQuote
kumowoon1025
(@kumowoon1025)
Active Member
Joined: 2 years ago
 

Woah, it looks like USB's actually going to be an actual bus...
I guess that's somewhat expected from it tunneling TB3, but now I wonder if hubs are going to be really expensive with tcams and everything basically being multilayer switches??

2018 Mac mini I7 - Sonnet BB550W + Gigabyte RX570 4GB - Alpine Ridge AIC – (TB3 to TB adapter) – Apple Thunderbolt Display
2019 15" MBP i9 also works as expected in above setup
2012 15" rMBP works with purge-wrangler
All on 10.14.6 (18G84/b) for debug kernel...
iMac14,2, no Built-in LCD, GT 755M dGPU - Bricked after Catalina DP update tried to flash nonexistent T2 chip. Mucking around with Pomona clip, looking for BootROM dump if anyone has one 🙁


ReplyQuote
joevt
(@joevt)
Noble Member
Joined: 3 years ago
 

Some differences between Thunderbolt 3 and USB4:
- USB4 adds USB3.x tunneling (can replace USB PCIe controller with USB hub/controller).
- USB4 has a slightly slower clock (10 or 20 Gbps instead of 10.3125  or 20.625).
- USB4 can have multiple downstream ports (maybe Thunderbolt can too, but Intel only ever produced Thunderbolt chips with two ports and I think a Thunderbolt host is limited to a single PCIe downstream port per USB4 device).
- USB4 max depth is 5 instead of 6.
- USB4 host can have more devices (wider instead of deeper tree).

I tried to find in the spec information on PCIe bandwidth limit (without reading the entire thing). There was mention of balancing latency with performance by changing the amount of buffering (applies to USB and PCIe tunneling).

Mac mini (2018), Mac Pro (Early 2008), MacBook Pro (Retina, 15-inch, Mid 2015), GA-Z170X-Gaming 7, Sapphire Pulse Radeon RX 580 8GB GDDR5, Sonnet Echo Express III-D, Trebleet Thunderbolt 3 to NVMe M.2 case


ikir and itsage liked
ReplyQuote
chx
 chx
(@chx)
Estimable Member
Joined: 3 years ago
 

 Even as the USB4 specification introduces a new underlying protocol, compatibility with existing USB 3.2, USB 2.0 and Thunderbolt 3 hosts and devices is supported; the resulting connection scales to the best mutual capability of the devices being connected

That's the official word. I think USB 4 devices using the Thunderbolt mode (which seems to be optional) will work TB3 hosts but if they sport USB hubs / devices attached to those then those simply won't work. But, a plain eGPU would work. In the other direction, a TB3 eGPU should work with a full capable USB 4 port.

Lenovo ThinkPad 25 -- GALAX SNPR TB3 1060 -- Lenovo Graphics Dock -- Benq BL2411PT - - two PackedPixels - Dasung not-eReader backer


ReplyQuote
karatekid430
(@karatekid430)
Estimable Member
Joined: 3 years ago
 

To add to discussion: if PCIe tunnelling is defined in the USB4 spec, then it is possible that manufacturers will be able to make USB4 PCIe devices, without using Thunderbolt branding. I am hopeful. It does say that all hubs must have a PCIe switch, IIRC.

Other notes: I believe it is possible to create a Thunderbolt 3 hub which splits one port into up to 8, but Intel clearly stuck with daisy chaining because of the restrictions it placed. It prevented you daisy chaining eGPUs (more than one GPU per Thunderbolt 3 port) which would probably use more PCI resources than are reserved per Thunderbolt 3 port. Using an add-in card on Linux with my kernel patches, I reserve 64G prefetchable / 256M MMIO under each port, which is vastly more than Intel reserves by default in firmware. Once I put a GC-TITAN RIDGE into an eGPU enclosure and had it splitting one port into two, in a cumbersome way. Linux does seem to retain a bus assignment bug which is only revealed by such strange configurations. I tried to fix it but there is a double recursive function call for bus assignment in drivers/pci/probe.c which I could never quite understand.

To do: Create my signature with system and expected eGPU configuration information to give context to my posts. I have no builds.

.

itsage liked
ReplyQuote