Intel's Maple Ridge (JHL8540) Thunderbolt 4 Controller Now Shipping
by Ganesh T S on December 23, 2020 8:15 AM EST- Posted in
- Thunderbolt
- Intel
- DisplayPort
- Thunderbolt 4
- Maple Ridge
Following the launch of Thunderbolt 4 earlier this year as part of Intel's Tiger Lake CPUs, the next piece of the TB4 hardware stack has dropped this week with the release of Intel's first stand-alone Thunderbolt 4 controller, Maple Ridge (JHL8540). Previously announced back in July as part of Thunderbolt 4's reveal, Intel this week updated their Ark database to add a product page for the Maple Ridge controller family and flag that the first part is now shipping. With the release of the discrete Thunderbolt 4 controller, it will now be possible for hardware vendors to build TB4 hosts with additional ports, or in devices not using Intel's Tiger Lake Silicon.
This late-December launch follow's Intel's previous roadmap, which had the launch of standalone controllers set to take place before the end of 2020. These included the Goshen Ridge (JHL8440) device controller – for use in docks and peripherals, and the Maple Ridge (JHL8540 and JHL8340) host controllers – for use in computers, tablets, and other client devices. Goshen Ridge went into production soon after the announcement. And with the release of Maple Ridge Intel has also kept its promise here, getting it out just prior to the end of the year.
For quite some time, Thunderbolt ports were found only on systems with Intel processors. However, last year we saw vendors such as ASRock innovate with the introduction of a Thunderbolt 3 port on the X570 Phantom Gaming-ITX/TB3, an AMD Ryzen platform motherboard. This was followed a few months back by the introduction of M1-based Macs featuring Thunderbolt 3 (backed by Apple's in-house controllers). The use of Maple Ridge will now enable motherboard vendors to create systems with Thunderbolt 4 ports that do not necessarily need to be based on Intel processors.
The JHL8540 Maple Ridge controller interfaces with the host processor using a PCIe 3.0 x4 link and also takes in two Display Port 1.4a inputs. On the downstream side, the controller enables two Thunderbolt 4 ports, which along with their native Thunderbolt (packet encapsulation) abilities can also be used as straight-up USB4 ports, or as DisplayPorts via USB-C's DP alt mode.
The PCIe switch and, in general, the PCIe support in Maple Ridge has been updated to work with many optional features, keeping security in mind and the rich variety of PCIe devices coming into the market. For example, Maple Ridge includes PCIe peer-to-peer support which allows two PCIe devices connected to the two Thunderbolt 4 ports to exchange data with each other without having to make it travel upstream to the host RAM. From a security viewpoint, Access Control Services (ACS) is also supported to provide isolation between different sets of PCIe devices and make them always go through the IOMMU. Precision Time Measurement (PTM) is also a supported feature, allowing different downstream PCIe devices to accurately synchronize with each other and the host system.
It must be noted that Thunderbolt 4 brings more guaranteed bandwidth to end-users. With Thunderbolt 3, device vendors could skimp on the connection of the controller to the host processor – using only a PCIe 3.0 x2 upstream link instead of PCIe 3.0 x4, but still obtain Thunderbolt 3 certification. This reduced the minimum available PCIe data bandwidth to just 16 Gbps. With Thunderbolt 4, that is no longer possible. Vendors are mandated to use a full PCIe 3.0 x4 link if they desire Thunderbolt 4 certification. Thunderbolt 3's bandwidth sharing mechanism between video and data also put in some dampeners – even in the absence of tunneling DisplayPort streams, 18 Gbps of bandwidth was always reserved for video traffic, and only 22 Gbps available for actual data transfer. Thunderbolt 4 apparently fixes that with up to 32 Gbps of data traffic (full PCIe 3.0 x4 bandwidth) available, allowing devices such as Thunderbolt 4 SSDs to provide 3GBps+ speeds.
Intel has not published official pricing of the new Maple Ridge controller, however Mouser Electornics is listing the controllers for as cheap as $11.34 in bulk quantities. As for the availability of devices featuring the JHL8540, I suspect we're going to see them sooner than later. Intel's next-generation desktop platform, Rocket Lake-S, is not expected to have built-in support for Thunderbolt 4, as this feature was noticeably absent from Intel's Rocket Lake reveal back in October. So adding Thunderbolt 4 to Rocket Lake-S will likely require using Maple Ridge.
This would be consistent with other documentation from Intel, such as the Intel 500 series chipset guidelines, which apparently point to instructions to use a discrete USB4-compliant Intel Thunderbolt 4 controller connecting to four PCIe 3.0 lanes from the chipset for USB4/Thunderbolt 4 support. To that end, we expect that the development of actual hardware by Intel’s partners using the Maple Ridge controller should be well under way by now.
Source: Intel
79 Comments
View All Comments
lmcd - Wednesday, December 23, 2020 - link
I don't really understand what "full bandwidth" means to you. 22 Gbps will already get you a PCIe 3 SSD's worth of bandwidth in anything but pure sequential scenarios. Nothing says your external SSD is obsolete.repoman27 - Wednesday, December 23, 2020 - link
Then it should blow your mind to learn that there *never was* an arbitrary 22 Gbit/s cap on PCIe throughput.PCIe tunneling to any single Thunderbolt endpoint is limited to the equivalent of a PCIe Gen3 x4 link. That’s 4 x 8 Gbit/s, or 32 Gbit/s, which is how Intel comes up with their advertised number.
Given an 128 B maximum TLP payload size and 16 B TLP headers, that should yield up to 25.92 Gbit/s of throughput after accounting for encoding and protocol overhead. The observed PCIe throughput over Thunderbolt seems to hit a wall at around 21.32 Gbit/s. However, that would mean that at least 25.92 Gbit/s of PCIe packets are being tunneled... which seems an odd coincidence.
So at best the only improvement in PCIe throughput you could possibly see would be from 21.32 to 25.92 Gbit/s, but I’m pretty sure you won’t even see that because there’s something else at play here.
hubick - Wednesday, December 23, 2020 - link
You are incorrect. Thunderbolt 3 reserves bandwidth for DisplayPort, even when it's not in use.See Figure 7. https://thunderbolttechnology.net/sites/default/fi...
The maximum bandwidth for data, when only data is being transmit, is 22Gbps.
repoman27 - Wednesday, December 23, 2020 - link
No, I’m not. That bandwidth isn’t being reserved for anything. The maximum PCIe throughput is limited by encoding, protocol overhead, and implementation details.Benchmarks clearly show that up to 21.32 Gbit/s PCIe throughput is possible over Thunderbolt. The PCIe and USB4 (which includes everything necessary for Thunderbolt 3 interoperability) specifications are freely available. You can do the math yourself if you like and see that 21.32 Gbit/s of payload data when packetized and framed out for transport over either a PCIe or Thunderbolt link would require at least 25.92 Gbit/s of bandwidth. That infographic from Intel marketing does not account for protocol overhead, and is merely designed to set realistic expectations for maximum PCIe throughput.
There is simply no mechanism anywhere in the specs for limiting PCIe bandwidth to accommodate DisplayPort traffic in the way that everyone seems to insist is happening.
vlad42 - Friday, December 25, 2020 - link
No reoman27, you are wrong. According to the author of this very article, "Thunderbolt 3's bandwidth sharing mechanism between video and data also put in some dampeners – even in the absence of tunneling DisplayPort streams, 18 Gbps of bandwidth was always reserved for video traffic, and only 22 Gbps available for actual data transfer."ganeshts - Saturday, December 26, 2020 - link
Actually, I should clarify that 22 Gbps is the maximum practical data throughput for *TB3*. With TB4, Intel seems to have pushed this up to around 25 Gbps (given their 3000 MBps claim for SSDs over TB4). There is some bandwidth always reserved for video, but Intel never indicated it is *18 Gbps*. I have reached out to Intel for further clarification and may update the article when I get a response. It could just be that up to 10 Gbps is lost in overheads.hubick - Tuesday, December 29, 2020 - link
Clarification would be great. I don't even know anymore. I read their tech doc (linked above), lots of articles, and I don't know enough to know if the math is right, and it is just overhead, or if there actually is a hardcoded limit, or if it changed, or WTF is going on.What I do know is, for the amount of generations this has been around, and now with it becoming a standard, and the amount of money I've spent on the tech, that Intel has done a really **itty job at being even the slightest bit forthcoming on what's actually going on here, and what's changing from 3 to 4, etc. Having some BS marketing to sell stuff is one thing, but can we get some ACTUAL ANSWERS somewhere Intel? Anyone?
hfm - Friday, December 25, 2020 - link
I'm more interested in TB4 eGPU devices as giving 50% more bandwidth to an eGPU would be a huge win. I currently use an LG Gram 17 + Sonnet Breakaway Box 550 with a 2070 OC installed in it. If I can get a TB4 laptop and TB4 eGPU enclosure, that would be a huge win for setups like mine.lmcd - Wednesday, December 23, 2020 - link
I personally want to see this integrated onto low-end graphics cards for docking purposes (as well as, of course, use with VR headsets). While docking a PC sounds useless, mini-PC form factors are actually pretty portable these days and avoid a lot of the tradeoffs found in laptop form factors.It'd be ironic if that arrangement also supported an eGPU, but possibly occasionally useful. The benefits are mostly elsewhere, and imo the cost of implementation shouldn't be much higher than the chip itself and I guess some sort of PCIe bridge chip. With PCIe 3, mid-range graphics cards already do not saturate the x16 link, and barely saturate an x8. Power delivery can be increased over the minimums just off of excess from a single 8-pin. And the DisplayPort out is already onboard.
This would primarily benefit the ITX crowd, but I think it would be an extremely sought-after part. There's very few ways to combine an ITX board with Thunderbolt and an AMD CPU.
headeffects - Thursday, December 24, 2020 - link
What’s the actual USB spec for their 4.0 implementation. Is it 20gbps?