USB4 Version 2.0 Said to get 120 Gbps Asymmetric Mode

[TheLostSwede]

It was only yesterday that the USB Promoters Group announced the USB4 Version 2.0 spec with support for speeds of up to 80 Gbps, something TechPowerUp mentioned at the end of our USB4 article back in June. Now details of a 120 Gbps asymmetric mode has popped up, courtesy of Angstronomics and we've managed to confirm that it is indeed something that is coming from one of our own sources. We were in fact told back in June that the 80 Gbps mode was meant to be asymmetric, but this was not mentioned in the recent press release.

The 120 Gbps mode will use three of the four data pairs for upstream data and the fourth 40 Gbps data pair will be for downstream data from and to the host controller.Asymmetric data transfers are nothing new over USB Type-C cables, as the DP Alt Mode is already taking advantage of this. This is possible because USB4 Version 2.0 will move to PAM3 (Pulse-Amplitude Modulation) data encoding from today's 64/66- or 128/132-bit encoding. The next generation of Thunderbolt is also expected to use PAM3 encoding to reach the rumoured 80 Gbps speeds that were posted somewhat by mistake by an Intel executive last year.



View at TechPowerUp Main Site | Source

 

[ToTTenTranz]

120Gbps is 15GB/s which is just a tad lower than PCIe 3.0 x16.
At these speeds, we'd get eGPUs working with no performance deficit, if this USB4 v2 handles latencies well (which it probably doesn't).

 

[TheLostSwede]

120Gbps is 15GB/s which is just a tad lower than PCIe 3.0 x16.
At these speeds, we'd get eGPUs working with no performance deficit, if this USB4 v2 handles latencies well (which it probably doesn't).

Note that it's 120/40 Gbps, so it might not work for that, unless the GPU makers come up with some clever driver trickery.
USB4 supports PCIe tunelling, so it would be pure PCIe signals, so no latency penalty.

 

[Dristun]

Will USB naming ever get reasonable? "USB 4 2.0" is like a sick marketing joke

 

[TheLostSwede]

Will USB naming ever get reasonable? "USB 4 2.0" is like a sick marketing joke

It's USB4, not USB 4.

 

[ToTTenTranz]

Note that it's 120/40 Gbps, so it might not work for that, unless the GPU makers come up with some clever driver trickery.
USB4 supports PCIe tunelling, so it would be pure PCIe signals, so no latency penalty.

Perhaps the CPU-GPU interface would actually make sense if it was asymmetric?
Save for perhaps some select GPGPU tasks, I don't know why a GPU would need a whole lot of bandwidth to send data to the system memory. It's probably the flow of assets from system memory to VRAM that takes the most bandwidth anyway.

 

[TheLostSwede]

Perhaps the CPU-GPU interface would actually make sense if it was asymmetric?
Save for perhaps some select GPGPU tasks, I don't know why a GPU would need a whole lot of bandwidth to send data to the system memory. It's probably the flow of assets from system memory to VRAM that takes the most bandwidth anyway.

I'm not the right person to ask about that, as I'm not an expert on GPUs, but PCIe was designed to be symmetrical.
That said, it seems like someone has written a short paper on asymmetrical PCIe, that might help answer your question.

https://www.tdcommons.org/cgi/viewcontent.cgi?article=2864&context=dpubs_series

 

[Nanochip]

Is Intel continuing to donate its tech to the USB-IF for USB4? Donating thunderbolt (and its tunneling capabilities) was great for the ecosystem. If PAM3 is coming to USB4, and Intel mentioned PAM3 in that leaked slide, seems like Intel is continuing to contribute to the USB-IF to enable ultra high speed (my term) USB for the masses, which is a good thing.

I'm looking forward to more powerful docks, peripherals, eGPUs, DP2.0 monitors, faster external storage coming to market. And with an open USB4 implementation, prices of such technology should come down, as many manufacturers won't need to get thunderbolt certification. Case in point: I have an Anker USB4 cable, certified by the USB-IF for 40 Gbps data, and 240W charging. The cable was only $26.99 on Amazon, which less than half of the price of TB4 cables... and this cheaper cable can charge a compatible device at 2.4x the charging speed capability as the more expensive TB4 cables. 240W vs 100W. I have a 140W battery, and TB4 cables can only charge it up at 100W... the cheaper USB4 cable can charge it up at full tilt.

My main question here is: what generation of PCIe will USB4 v2 enable? Is it PCIe4.0? 5.0? And how many lanes of PCIe bandwidth? Finally, we've heard that with PAM3 signaling, existing TB4 (and USB4 40 Gbps?) cables will be able to take advantage of the new data speeds, no problem. But what about existing TB4 controllers...will a firmware update enable them to pump data at faster than 40 Gbps? I doubt this: my thought is TB4 controllers are connected to the PCIe root complex at PCIe 3.0... foreclosing any opportunity for a speed bump via a firmware update.

 

[ymdhis]

There won't be a single combination of motherboard, chipset, driver, and cable, where this mode will be usable. The USB-C cables sold here aren't even wired for USB3, but only for USB2. It's considered the new phone charger cable here. And this is not even getting into the clusterfuck that is the usb3.1.2 usb-power gen2.3 mode 2x2 naming scheme.

 

[aQi]

Dont we need usb4 devices first to test all the might first ?

 

[mohammed2006]

Can’t wait for USB4 (v2X2)2

really how hard is it to call it USB5 it has x2 the bandwidth

 

[thewan]

There won't be a single combination of motherboard, chipset, driver, and cable, where this mode will be usable. The USB-C cables sold here aren't even wired for USB3, but only for USB2. It's considered the new phone charger cable here. And this is not even getting into the clusterfuck that is the usb3.1.2 usb-power gen2.3 mode 2x2 naming scheme.

I have both 10Gbps and 20Gbps usb-c cables. I have verified their speeds with a Kingston XS2000 that can use the maximum 20Gbps it supports when connected to a few different PCS/Laptops in my family with 10Gbps (Ryzen X470 Desktop and 10th Gen Dell Laptop) and 20Gbps (Intel B660 Desktop). Both 10Gbps and 20Gbps cable work as advertised, although depending on the host system, my XS2000 can't seem to reach full sequential write speeds for the Dell Laptop and the Intel B660 Desktop. I do not blame the cables, the original cable that came with it also exhibit the same behavior, so I blame the system manufacturer.

Those cables are not that hard to find. But yes you will have more luck finding generic or "dodgy" brand cables than well known brand cables. But they do work. How long they last I will never know though. I can also find USB4 cables but I don't need those and I can't test them anyway because they are too expensive, comparable to TB4 cables in pricing. Not to mention I don't have a Ryzen 6000 laptop anyway.

Dont we need usb4 devices first to test all the might first ?

Just buy a Ryzen 6000 series laptop for a host. Then buy a good NVME SSD and a USB 4 (USB4? whatever) external enclosure. They exist. Then come back with a review. Thank you.

 

[Assimilator]

I think Thunderbolt is officially dead now. I don't see any advantage it has, or can have, over USB4 now. And that's a good thing for the industry because Thunderbolt is a royal PITA.

 

[Haile Selassie]

I wonder what power does it require to run 2-3 ft over copper at this speed.

 

[The Von Matrices]

Perhaps the CPU-GPU interface would actually make sense if it was asymmetric?
Save for perhaps some select GPGPU tasks, I don't know why a GPU would need a whole lot of bandwidth to send data to the system memory. It's probably the flow of assets from system memory to VRAM that takes the most bandwidth anyway.

Think laptops with high resolution displays and an external GPU. A lot of data has to flow asymmetrically from the external GPU to the iGPU to drive the internal display.

My concern with USB4 is the ever shortening length of cables. All the high speed USB-C cables are 1 meter long and you really have to search to find a longer cable. Connecting my webcam from the top of my monitor to the desktop PC on the floor was a pain because I needed a 3m cable and had to buy and return 5 different ones before I found one that reliably works at 5gbps.

 

[Lycanwolfen]

USB 4 great more USB that cannot do USB speeds. USB 2.0 was 60 mb/s and today we do have some USB drives that can finally get that speed. USB3 speed is 600 mb/s a sec well some drives now can get that but most drives maybe 30 mb/s Now USB 4 comming out. Give me a break. How long will it take before devices can actually do 2400 mb/s or 4800 mb/s or even 120 GB a sec. Nothing maybe in 40 years something will get that. I have some samsung bar USB 3.1 drives and the fastest transfer on those is maybe 250 mb/s not even USB 3.0 speeds.

 

[Sabotaged_Enigma]

USB4 1.0 2.0... Gen 1, Gen 2, Gen 2x2, Gen 2x2x2...
WTF USBIF??? Make it easy, will you?

 

[Mussels]

Note that it's 120/40 Gbps, so it might not work for that, unless the GPU makers come up with some clever driver trickery.
USB4 supports PCIe tunelling, so it would be pure PCIe signals, so no latency penalty.

the wiring distance adds a latency penalty, it's small but its there - PCI-E is sensitive to distance


I do like the concept of the assymetrical connections, if only they could negotiate in realtime - allowing external hard drives for example to swap around in real time depending if you're reading or writing

 

[Readlight]

120Gbps is 15GB/s which is just a tad lower than PCIe 3.0 x16.
At these speeds, we'd get eGPUs working with no performance deficit, if this USB4 v2 handles latencies well (which it probably doesn't).

That's 10 seconds to transfer 100 GB game file.

 

[TheLostSwede]

Think laptops with high resolution displays and an external GPU. A lot of data has to flow asymmetrically from the external GPU to the iGPU to drive the internal display.

My concern with USB4 is the ever shortening length of cables. All the high speed USB-C cables are 1 meter long and you really have to search to find a longer cable. Connecting my webcam from the top of my monitor to the desktop PC on the floor was a pain because I needed a 3m cable and had to buy and return 5 different ones before I found one that reliably works at 5gbps.

This is why there's work on active copper and fibre cables, which will obviously be pricey to start with, but cheaper than Thunderbolt cables.

 

[zlobby]

I wonder what specs we'd get for the USB4.1 Gen 6 Rev.2 PD Alt-mode Cat9 Class 8 High Speed???
This has gone beyond anything sane...

USB 4 great more USB that cannot do USB speeds. USB 2.0 was 60 mb/s and today we do have some USB drives that can finally get that speed. USB3 speed is 600 mb/s a sec well some drives now can get that but most drives maybe 30 mb/s Now USB 4 comming out. Give me a break. How long will it take before devices can actually do 2400 mb/s or 4800 mb/s or even 120 GB a sec. Nothing maybe in 40 years something will get that. I have some samsung bar USB 3.1 drives and the fastest transfer on those is maybe 250 mb/s not even USB 3.0 speeds.

Exactly! Plus, these speeds can't be achieved over even moderate distances with USB4.
Pretty much useless spec

 

[Wirko]

The proper name for this asymmetrical thing would be...
USB4 [major version dot minor version] [generation] [revision] x(3+1)
optionally followed by PD [major version dot minor version].

 

[thegnome]

Just call it USB5, the USB standard really has gone downhill since 3.1...

 

[DeathtoGnomes]

Just call it USB5, the USB standard really has gone downhill since 3.1...

It's USB4, not USB 4.

Soon it will be USB4 v2 rev3 wer3.5 at6.125

 

[zlobby]

The proper name for this asymmetrical thing would be...
USB4 [major version dot minor version] [generation] [revision] x(3+1)
optionally followed by PD [major version dot minor version].

The proper name would be 'KYS'! I mean the USB guys.