Thunderbolt 5 Could Reach 80 Gbps According to Leaked Photo

[Uskompuf]

Intel Client Computing Group Executive Vice President & GM Gregory Bryant has recently published and deleted an image that appears to show the specifications of a next-generation Thunderbolt technology. The photo was taken at an Intel R&D facility in Israel and shows a poster highlighting "80G PHY Technology" which signifies that Intel is working on an 80 Gbps physical layer. This speed represents a doubling of the existing Thunderbolt 3/4 which runs at 40 Gbps while still working over a USB-C connection. These speeds can be achieved with novel PAM-3 Pulse Amplitude Modulation technology which allows for 50% more bits to be sent per cycle compared to the NRZ technology currently used while limiting the increased complexity that would be introduced with PAM-4. Intel appears to have a working test chip manufactured on the N6 node at TSMC which is showing promising results however we don't expect this technology to reach consumer devices for some time.



View at TechPowerUp Main Site

 

[Prima.Vera]

There are already mainstream network cards that can deliver 100Gbps and even 200Gbps. Yes, the protocol is different, but the technology beyond is similar.

 

[kayjay010101]

There are already mainstream network cards that can deliver 100Gbps and even 200Gbps. Yes, the protocol is different, but the technology beyond is similar.

Wouldn't exactly call it mainstream, but yeah, 100Gbps in the networking space is nothing new.
We're now at 400Gbps even.

 

[AnarchoPrimitiv]

80Gbps woulld bring eGPUs closer to having no bottleneck....I know a couple generations ago PCIe 3.0x8 was enough, but would that be true for the newest generation of GPUs? They should just throw caution to the wind and make a mainstream, consumer oriented ePCIe protocol/connector that has the option of doing 4.0x4/8/16 ( I know that have AICs and connectors in the enterprise market). Also, I wish Thunderbolt would be more...."upfront" about how their bandwidth really works, when they say 40Gbps, that's not actually what an eGPU or external drive can work with, that 40Gbps figure is an amalgamation of 32Gbps for data and 8Gbps which can be used ONLY for video. So, I'm going to assume that Thunderbolt 5 will actually have a 64Gbps data bandwidth with the remaining 16Gbps allotted for video....which it'll definitely need, and far more, if it intends to incorporate DisplayPort 2.0. It's reminiscent of how WiFI routers are currently marketed, claiming they have bandwidth of 7200Mbps or 11,000Mbps, which would imply to many that a single user, with capable hardware, could transfer data over a single connection at 7.2Gbps, but in reality this is a total figure of all the different bands and frequencies added together, and if it's a mesh system, a large portion of that figure could be attributed to a backhaul channel that doesn't get factored into actual speed at all.

 

[Chrispy_]

The problem with Thunderbolt is that the cost of the connection has alienated it from the mainstream market. There's still no guarantee that a USB-C port on an thunderbolt-enabled device is even fully capable of Thunderbolt 3.

As always, the lowest common denominator is cost and if 80Gbps needs even more expensive controllers and cabling then it'll be very very slow to arrive to the mass-market.

 

[Ja.KooLit]

wow.

AmD Should come up with something like this.

 

[cyrand]

There are already mainstream network cards that can deliver 100Gbps and even 200Gbps. Yes, the protocol is different, but the technology beyond is similar.

I think the difference is they don't want to require Optical cables. Doing 80Gbs/100/200/400 they could do over fiber but doing it over copper is way trickier. In network cards I pretty sure ethernet cards max out at 40gbs. 80 or higher are SFP+ or some other tech that support fiber transceivers/receivers.

I think there goal with thunderbolt 5 is to support passive copper like the majority of USB-C cables.

 

[Tomorrow]

80Gbps woulld bring eGPUs closer to having no bottleneck....

eGPU's are mostly bottlenecked by slow(er) CPU's in laptops compared to desktops. Having a faster interconnect does not solve this problem is paired with a slow CPU that results in high frametimes and stutter.

And yes there are laptops with some fast CPU's but ultimately even these are power constrained and a laptop with a high-end CPU usually already has a high end dGPU inside.

 

[henok.gk]

eGPU's are mostly bottlenecked by slow(er) CPU's in laptops compared to desktops. Having a faster interconnect does not solve this problem is paired with a slow CPU that results in high frametimes and stutter.

And yes there are laptops with some fast CPU's but ultimately even these are power constrained and a laptop with a high-end CPU usually already has a high end dGPU inside.

There's also the issue regarding latency depending on whether the PCIe lanes run from the chipset or directly from the CPU. So yeah increasing the bandwidth can only do so much before performance is held back by something else in the pipeline.

 

[Prima.Vera]

I think the difference is they don't want to require Optical cables. Doing 80Gbs/100/200/400 they could do over fiber but doing it over copper is way trickier. In network cards I pretty sure ethernet cards max out at 40gbs. 80 or higher are SFP+ or some other tech that support fiber transceivers/receivers.

I think there goal with thunderbolt 5 is to support passive copper like the majority of USB-C cables.

Now that if you're thinking, why not optical cabling?? Optical has come a long way since their debut. Now the cables are way more flexible and resistant to bending due to different materials used, also their price is on pair with the expensive coppers. I think technology should move forward.

 

[Minus Infinity]

I'm going to make a groundbreaking announcement. Thunderbolt 6 will hit 160Gb/s.

No please hold the applause, please make a donation to a charity of your choice instead.

 

[Chrispy_]

I'm going to make a groundbreaking announcement. Thunderbolt 6 will hit 160Gb/s.

No please hold the applause, please make a donation to a charity of your choice instead.

Thunderbolt will be superseded before it hits version 6 IMO.

It's an Intel-Apple collaboration that is struggling to gain unversal adoption because neither Intel nor Apple play nice with any other companies. Open standards will evolve in a divergent path to Thunderbolt, gain conflicting feature sets and capabilities to Thunderbolt, and force Intel/Apple to adopt a new proprietary standard as a way of oneupmanship over the wider tech industry instead.

Thunderbolt is making all the same mistakes that Firewire made because of simple corporate greed; They've updated their product but not the mindset that resulted in previous failures. Thunderbolt, for all intents and purposes, is just Firewire2. Now that Intel has finally made Thunderbolt royalty-free it'll be absorbed into USB4 and forgotten about as in time people will just use the term "USB" instead.