Intel Offers Up Royalty-Free Thunderbolt 3 To USB Promoter Group

[starshipeleven]

A GPU case would also not include its own GPU. That's bought separately.

I think I already mentioned that.

an eGPU only needs the box (obviously), the GPU and a PSU.

Although I think I saw both boxes with a GPU bundled and without, also with a PSU and without (requiring an ATX PSU), so there is variability.

[starshipeleven]

So far, we only had one company that was in charge of Thunderbolt: Intel. Even if we say that it was in collaboration with Apple, it was not the best situation.
Currently, we can expect that the following companies will work (or at least try to) on USB4 controllers:

Pfft, hahah sorry no.

Thunderbolt controllers are behemoths in size and complexity, so don't expect anyone on the list to make anything for a very long while, if at all (especially the cheap ones). Only one that might is AMD/ASMedia, and imho only if they leverage their agreements with Intel to have a look at their tech to make their own "copy".

Intel isn't removing royalties because they are nice, they are doing it because they know full well that once they integrate the controller in their CPU/Chipset the only way to make a thunderbolt-enabled device is to use Intel hardware for it, so any lost profit from licensing will be more than made up by increase of sales or marketshare (and by this I mean that while they are basically monopolizing the laptop and convertible segment, they desperately need a killer feature to entice people to buy new Intel ones when the CPU poerformance increases aren't enough to do it)

[tessio]

USB 4? Not USB 3.2.5.7.54.a.rc3?

It's USB4, withouth the space. See.. they can think of other ways to mess this up.

[starshipeleven]

It's USB4, withouth the space. See.. they can think of other ways to mess this up.

I'd beg to differ.

If I'm searching for a USB device around on ebay/amazon/aliexpress it's not as easy as I would like to just filter for 3.0 devices (so I can filter out the oceans of USB 2.0 hubs or flash drives or somesuch).

With "USB4" with no space it's very easy to find a device supporting USB4 even with rudimentary (read "crappy") search engines as those found in marketplace websites.

[kneekoo]

An eGPU will never be “cheap”. You basically have all the hardware a PC has combined with a extremely limited market. If anything the cost will likely go up in the future.

Due to the issue of cost and other factors I don’t expect to ever see eGPU’s as anything more than a niche product with niche pricing. In fact I’m kinda expecting the market to dry up over time.

An eGPU can great for people with multiple PCs (home desktop, travel laptop, home theater, spouse's/kids' PCs, etc). Buy one or two eGPUs and everyone is covered when no 2-3 PCs would require a powerful GPU at the same time. It's a convenient solution that also saves money and generates less heat and noise by only plugging-in the eGPU when required. Being able to move it between desktops and laptops would be its greatest feature.

[the_scx]

Pfft, hahah sorry no.

Thunderbolt controllers are behemoths in size and complexity, so don't expect anyone on the list to make anything for a very long while, if at all (especially the cheap ones). Only one that might is AMD/ASMedia, and imho only if they leverage their agreements with Intel to have a look at their tech to make their own "copy".

Intel isn't removing royalties because they are nice, they are doing it because they know full well that once they integrate the controller in their CPU/Chipset the only way to make a thunderbolt-enabled device is to use Intel hardware for it, so any lost profit from licensing will be more than made up by increase of sales or marketshare (and by this I mean that while they are basically monopolizing the laptop and convertible segment, they desperately need a killer feature to entice people to buy new Intel ones when the CPU poerformance increases aren't enough to do it)

I'm not saying that everything would happen at the same time. Please note that Cypress and Zhaoxin have obtained their own USB 3.1 Gen 2 controller in the middle of last year, 5 years after the specification was issued. For Zhaoxin it was like their first real USB 3.x controller, because the previous one was based on USB 3.0 controller (final specification released in 2008) from VIA.

When it comes to USB4 adaptation, I would split it into several phases.

Phase 0:

• Intel (Intel Corporation) - USA 🇺🇸

It should be obvious. Intel already has Thunderbolt 3, so it shouldn't be a problem at all.

Phase 1 (up to 3 years from release):

• AMD (Advanced Micro Devices, Inc.) - USA 🇺🇸 and/or ASMedia (ASMedia Technology Inc.) - Taiwan 🇹🇼
• Apple (Apple Inc.) - USA 🇺🇸

AMD will definitely be interested in USB4. Currently, Thunderbolt 3 is one of Intel's killer features. And ASMedia designed the Promontory AM4 chipsets for AMD's Zen micro-architecture, as well as USB 3.0 host controller for Intel's Sandy Bridge motherboards.
Apple can use a USB4 controller from Intel in their Macs, but if they opt for ARM-based Macs, they will surely want to obtain their own controller. They have funds for this and they are surprisingly effective. They managed to create their own GPU just like that, so this shouldn't be a problem neither. Please note that Thunderbolt was developed by Intel, but in collaboration with Apple.

Phase 2 (up to 3-5 years from release):

• TI (Texas Instruments Inc.) - USA 🇺🇸
• NVIDIA (Nvidia Corporation) - USA 🇺🇸

Texas Instruments is one of the major manufacturers of USB 3.x controllers. Moreover, it is a member of the USB 3.0 Promoter Group.
NVIDIA may be interested in USB4 mainly because of its Tegra platform. Please keep in mind that their Project Denver was originally intended to support both ARM and x86 code using code morphing technology from Transmeta, but was changed to the ARMv8-A ISA (AArch64), only because NVIDIA could not obtain a license to Intel's patents. This, however, may change in the future and then USB4 will be extremely desirable.
Moreover, both companies have funds for that.
I believe that there is still room for IBM and HP, if only they would express such interest.

Phase 3 (up to 5-10 years from release):

• Qualcomm (Qualcomm Incorporated) - USA 🇺🇸
• ST (STMicroelectronics) - EU 🇪🇺
• Renesas/NEC (Renesas Electronics Corporation/NEC Corporation) - Japan 🇯🇵
• Samsung (Samsung Electronics Co., Ltd.) - South Korea 🇰🇷
• VIA (VIA Technologies, Inc.) - Taiwan 🇹🇼 and/or Zhaoxin (Shanghai Zhaoxin Semiconductor Co., Ltd.) - China 🇨🇳

Samsung have already expressed interest in the idea of device convergence. In 2017, they released DeX (Desktop eXperience) - an accessory that extends the functionality of their flagship smartphones to allow desktop-like functionality. Later that year, they also announced "Linux on Galaxy". Today, DeX functionality is naively supported by the Note 9 and the Tab S4, but with support only for external display. USB4 would be the Holy Grail for them.
Qualcomm would certainly not be left behind Samsung in that race. Moreover, together with Microsoft, they are involved in ACPC (Always Connected PC) devices, and USB4 would be definitely a nice feature.
STMicroelectronics and Renesas are both members of the USB 3.0 Promoter Group. NEC introduced its first USB 3.0 host controller in 2009. In 2010 HP released a laptop featuring a Renesas USB 3.0 host controller. It was several months before some of their competitors. What is more, AMD worked with Renesas to add its USB 3.0 implementation into its chipsets for its 2011 platforms. It obvious to me that Renesas will be interested in USB4. STMicroelectronics should be as well, if they want to remain competitive.
VIA announces its SuperSpeed USB 3.0 Hub Controller in 2010, and Host Controller in 2011, so it was a quite fast. On the other hand, they were in better shape then. Zhaoxin has now taken over the baton. As I said before, they have obtained their own USB 3.1 Gen 2 controller (1d17:9203) in the middle of last year. I believe that it was developed from their USB 3.0 controller (1d17:9202), which was based on the USB 3.0 controller from VIA (1106:9201), and it has nothing to do with the VIA Labs VL808 USB 3.1 Gen 2 Host Controller (it was a test chip, after all), so we can say that it's their own design. Anyway, in 2012 this company doesn't even exist, and now they has a modern SoC with x86_64 CPU (SSE4.2, AVX, AVX2, AES), integrated Northbridge (memory controller, DDR4 support), DX11-capable GPU with hardware encoding/decoding for H.264/H.265, PCIe 3.0, built-in USB 3.0 Host Controller as well as USB 3.1 Gen 2 Host Controller in ZX-200 IOE chip, etc. Of course, they relied heavily on VIA's IP, but they have also achieved a lot on their own for the last few years. Their roadmap has always been very ambitious, but they have always succeeded in achieving their goals. KX-7000, the next processor in the series, will adopt a new architecture, hoping to achieve performance level of AMD Zen. It should support PCIe 4.0 and DDR 5. If it really works out, USB4 shouldn't be a problem neither.

Phase 4 (up to 10-15 years from release):

• the rest

15 years ago, Huawei doesn't even exist on the mobile phone market. Their first smartphone was unveiled in 2009. Today they are one of the biggest players on this market. They even own HiSilicon, the largest domestic designer of integrated circuits in China. This is a quite impressive.
We all know that the Chinese government supports Chinese companies. We may like it or not, but it is a fact. Five years ago they didn't have any significant Flash memory semiconductor companies. Today, after several years of so-called Big Fund, we have Innotron Memory and Fujian Jin Hua Integrated Circuit. Just like that.

https://www.anandtech.com/show/12681/chinese-dram-industry-spreading-its-wings-two-more-dram-fabs-ready

Anyway, my point is that in 10 years, even Rockchip could be like Motorola/Freescale from the 80s.

Of course, everything here is just my speculation.

[starshipeleven]

Of course, everything here is just my speculation.

I'm just saying that any extrapolation based on USB 3.something adoption is unrealistic, I'll go into more depth.

USB controllers are orders of magnitude simpler than anything Thunderbolt. Even those with alternate Displayport or Thunderbolt 3 mode are rare as it's significant complexity addition.

External controllers are prohibitively large as you need to pipe them a displayport, PCIe lanes, USB and whatnot to this controller. Current Intel ones are as large as a smartphone SOC, that's a hard sell even for larger devices like a laptop.
This is not something that can change as this is interface size. If your BGA chip needs XXX pads under it to carry electrical signals to the board, you can't make it smaller even if the silicon inside is tiny. Same issue for board design. A ton of traces need to be routed to this controller.

Only way to place them into any product at decent scales is to integrate them in the CPU/Chipset/SoC so it is sitting on the same fast interconnect bus the other components in the SoC are, this already excludes everyone that does not make CPU/Chipsets/SoC.

Also, it's a large and power-hugry thing for anything in the mobile segment, this seriously limits its availability for mobile. Again this is not a thing that can change much as controllers get more efficient as it is a bus power factor, to sustain a high speed 4x USB 3.2 connection or route PCIe lanes you need a seriously big amount of power, for a mobile device's power budget.

Which leaves us with a list with Intel, AMD and possibly NVIDIA for their ARM platforms, but I doubt they care as it's not oriented to consumers anyway.

[microcode]

Of course you can secure any arbitrary big attack surface. But that doesn't mean that it is a good idea to expose this attack surface in the first place.

I mean, you can literally just whitelist memory regions for DMA; that reduces your attack surface to much smaller than USB devices (which can mess with all sorts of out-of-band mechanisms). If you're concerned that you may expose more data in the DMA region than you should, then your software is the problem, not Thunderbolt. In the case of Linux, apparently (I may have misread this) it directly exposes packets not destined for a physical device (e.g. packets that have not yet been encrypted for IPSec, and packets from UNIX domain sockets); the simple (conceptually) solution is to separate these memory regions.

[chithanh]

microcode I'm not that optimistic. With Thunderbolt you depend on the correct implementation of the IOMMU in hardware, firmware and operating system drivers. In the past that wasn't very secure. I don't know how you came to believe that hardware designers, firmware vendors and operating system / driver programmers suddenly turned around and made their stuff robust.

[microcode]

microcode I'm not that optimistic. With Thunderbolt you depend on the correct implementation of the IOMMU in hardware, firmware and operating system drivers. In the past that wasn't very secure. I don't know how you came to believe that hardware designers, firmware vendors and operating system / driver programmers suddenly turned around and made their stuff robust.

With USB the story is the same: if the controller is not implemented correctly, you are screwed. With Thunderbolt, in theory only the operating system and the CPU vendors need to get it right; everyone else can screw up.