Thunderbolt on the M1 Mac mini – when 2 equals 4 | Hacker News Reader

Thunderbolt on the M1 Mac mini – when 2 equals 4

eshop.macsales.com

23 pointsmy12323 comments

> (I usually suggest that displays being connected over Thunderbolt not be combined on the same chain as storage devices to improve the reliability of your mounted volumes.)

This is how you get cargo cults... Thunderbolt isn't a "lossy" channel; the real concern is ultimately what bandwidths you need. The most difficult factor in this is determining whether your storage is write-heavy, read-heavy or both. (If it's a SSD, that is - HDDs don't even get remotely near TB bandwidths.)

The general rules are:

  - Displays: output bandwidth, and quite a lot of that
  - eGFX: mostly output (textures) + a reasonable amount of input
  - 10G/40G NICs: generally input bandwidth, somewhat use case dependent
  - Storage: no general rule, strongly use case dependent

Also, the PCIe TB<>CPU connection needs to be considered, particularly in some older laptops it's only a PCIe 3.0×2 link, which caps out at 16GT/s.

If anything, it'd be nice to get good per-device bandwidth counters somewhere; not sure if the PCIe bridges can provide that data somehow (through QoS/AER features maybe?)

I purchased an M1 Mac Mini (really tired of the instability of my thunderbolt hub, although the TB3 hub has been great). But I can't find info anywhere about whether I can just hook up two 4k monitors at 60Hz using the two thunderbolt ports (in DisplayPort mode).

There's literature about hooking up one of them to TB, and then the other to HDMI.

But this article suggests there should be more than enough bandwidth to use two TB ports in DisplayPort mode.

The Mac Mini page on the Apple site specifically calls out "Connect one external display up to 6K and a second with HDMI 2.0 up to 4K." Under the thunderbolt 4 section, if it had more output capability they would have been proud to say it.

I don't think it's so much a bandwidth question, display outputs aren't encapsulated in Thunderbolt they are an alternate mode for some lines in the cable and the device still has to have the display output uniquely routed internally.

Bandwidth wise it’s possible but the M1 doesn’t support it. It only supports a single display “scan out” on thunderbolt. Plus a second that must be via HDMI.

The issue as I understand it is the lack of ability for 2 “scan out engines” to run and output at the same time.

Given the MacBook Air/Pro support only 1 external display I imagine the HDMI port connects to the hardware that would otherwise connect to the internal LCD. Hence why the mini can do 2 but the MacBooks only 1.

I hope this limitation to be lifted in the next generation. It’s the only reason I haven’t purchased one. I want 3 external monitors on a MacBook Pro.

For an actual Thunderbolt display (contrasted to the DisplayPort alternate mode), the sole DisplayPort controller shouldn't be a limitation, as render offscreen + DMA to the device would work just fine in that case.

Think that Apple did this to not have to say "you can use two of those displays that are plugged in over Type-C, but not those two other displays... that are also plugged in over Type-C".

> For an actual Thunderbolt display (contrasted to the DisplayPort alternate mode), the sole DisplayPort controller shouldn't be a limitation, as render offscreen + DMA to the device would work just fine in that case.

No, that's not how this is implemented. The difference between a Thunderbolt display and a DP-alt-mode display is that the TB one decodes the additional DP-in-TB layer to get at the DisplayPort data. And since it does TB to do that, it can (but doesn't have to) implement PCIe & USB functions too.

TB displays do not implement a PCIe endpoint for the actual display contents and don't have any DMA capability. The DisplayPort stream is wrapped inside TB packets (with some empty bits being left out, e.g. H/Vblank); the Display (or TB dock/hub) just decodes the packets back into DisplayPort.

... which means the host needs to be able to encode a particular number of DP streams into the TB connection, and depending on the silicon and its connections this may only be 1 stream. The only way around that would be with DisplayPort MST support, since that's only one GFX->TB stream for DP :)

Thunderbolt essentially transports 3 kinds of packets over its connection:

  - TB native packets (bit of a FireWire followup, kinda)
  - PCIe packets
  - DisplayPort packets

Very interesting... I wonder why they went that way instead of PCIe... (I assumed that TB always meant that PCIe endpoints were used for Thunderbolt)

The DisplayPort over TB link is very much synchronous/isochronous to the source; it's transported over TB at explicitly scheduled intervals and there are almost no buffers at the source and destination. PCIe would've needed larger buffers at the very least, and would need quite a bit of QoS mangling (there isn't really a bandwidth reservation system on PCIe, so best you could do would be to just give it highest priority)... which also means more places to fail & get wonky.

Also, there's no preexisting standard for framebuffers over PCIe.

What's the thunderbolt hub and what's the TB3 hub?

I had the OWC Thunderbolt 2 hub for nearly five years. It’s been mostly fine until the past year when I hooked up two 4K displays to it.

I just bought the CalDigit TB3 plus.

Part of this article seems incorrect. My Intel Mac Mini (2018) is reporting two Thunderbolt hubs in System Report, just like the M1 they’re demonstrating. That part at least is not unique to the new chip.

4 Thunderbolt ports, with each pair having a shared bus is what the Mac mini 2018 seems to have.

TLDR Apple provide the capability to service more than two ports performatively, but choose to only include two physical ports so you'll need to buy accessories.

Thanks Apple.

Proof of your statement? Have a memo or email that says “Tim, I suggest we only include 2 ports so we can drive up the sales of thunderbolt docks (oh, we should maybe make one too, so we can take more money, because we don’t right now).”

You comment attributes the design to malice with zero proof and Apple does not even make an external thunderbolt dock to make money on.

1. Check a review, no .. why not check a few.

2. Look at the main disadvantages listed.

3. Notice lack of ports is mentioned as a major one.

Then .. as a thought experiment, consider the article we're commenting on.

It's describing the fact that there are only 2 ports, and commenting that it isn't actually a bad thing .. it's spun positively, because there's bandwidth to support four devices.

Can you imagine any other manufacturer getting away with this?

A professional user _will_ need more than two ports. Apple know this, and you're still writing an apologia for them.

Crazy.

I'm not sure we have enough information to draw that conclusion. The thunderbolt ports on the M1 are driven directly from the SOC, so adding more would have meant adding a hub chip, which may have been deemed not worth the cost either financailly or in terms of power/thermal budget.

I don't see how this differs from what I'm saying?

I think the parent is saying that throughput isn’t the only “performance” metric.

Cost/thermals/power are also performance metrics.

It should be noted, however, that the article does not claim that Apple left throughout performance on the table, only that Apple chose -not- to use a HUB this time which could potentially oversaturate their thunderbolt controllers. This has the benefit of being significantly more stable.

I wouldn't agree that cost is a performance metric - unless we're talking about value for Apple shareholders.

I can't imagine including two more ports would significantly increase the heat profile of a device.

Surely the power requirements are a moot point, if devices can be chained?

I think that even if you debate cost as a performance metric the others hold true.

And including a thunderbolt hub in the machine would be an additional (quite warm) micro controller.

Even external USB hubs can get quite toasty.

We're talking about a device which is lauded for being very very cool.

You're stretching credulity.

I think we’re agreeing, actually.

A device optimised for operating “cool”-ly has obvious been optimised for temperature control. That doesn’t happen by accident.

Perhaps, but I'm definitely left wondering how the previous generation Mac mini .. with an integrated hub and an old school 'hot' processor ever managed to function.

Reply on news.ycombinator.com