The Thunderbolt in this case is coming from the chipset on the motherboard, not the CPU. He lists the motherboard as an X570 Phantom Gaming-ITX/TB3 which supports 2nd gen Ryzen, and which has that support built in.
This is incorrect, no AMD chip (CPU or chipset) comes with thunderbolt built in. The x570 Phantom Gaming board uses an Intel Titan Ridge (see the obligatory AnandTech block diagram) controller hanging off of the chipset's PCIe lanes to provide thunderbolt 3. This is functionally equivalent to just buying a thunderbolt add-in card, other than the space requirements.
Ah my bad, you are right, I didn't meant to imply that it was actually provided by the x570 chipset, just through it as a part of the motherboard. I just worded it poorly.
Does this explain the author’s surprise that apple’s fanless MacBooks can handle thunderbolt 3 without getting hot? If they’re doing it on cpu then I suppose that could bring about a lot of cpu i/o efficiency.
The X570 chipset is strange. It's just a Zen 2 IO die backported to GlobalFoundries 14nm. Presumably they were able to remove the unnecessary parts (memory controller) while backporting, but maybe they're reusing excess chips if they chose late in the process to move the IO die to 12nm. I'd guess there's some inefficiency from being a clever hack and not exactly designed for purpose.
Additionally, the X570 chipset supports a lot of peripherals:
- 16x PCIe 4.0 lanes + 4x lanes to the CPU
- 12 SATA 3 ports (some PHYs are shared with PCIe lanes so you can't have all of both)
- 8 USB 3.0 10Gbps (I'm not going to google the "official" nomenclature, ugh)
- 4 USB 2.0
- Other things like SuperIO and SMBus which don't use much power
Far more peripherals than M1. (Probably, it seems there's still no detailed specs about the thing anywhere).
Then there's the process advantage, 14nm GF to 5nm TSMC is a huge difference for all the logic involved. The advantage is dampened somewhat for IO because a significant portion of the power is used to drive the signals out of the chip, which is not affected by process changes. Apple's PCIe 4.0 PHYs may be more efficient considering X570 is a first generation PCIe 4.0 design.
Also, the power consumption of X570 is not that bad for what it is, 7-10W [0]. AMD must be a stickler about the 15W TDP because the fan seems unnecessary. This article is basically the worst case scenario: small heatsink on a cramped ITX motherboard in a passive case with no airflow. Maybe BCLK overclocking would run the fan? Or a dust clogged no-airflow case in a hot climate running all the peripherals at full bore?
The lack of airflow is probably to limit coil whine.
I think the main fault with X570 is that it doesn't save power at idle (PCIe supports downclocking links to save power, actually hot-swappable), but it appears to run all the lanes at full speed. Going from a Intel 6700K to AMD 3950X double the power draw at idle (50 watts to 90+) and that's with the same components.
On my system the chipset will quickly heat to 57c (when the fan starts) within two minutes regardless of usage. I swear Asus have the cooler upside down (semi-passive heatsink blows air downwards), worse still it ingests warm are from the heatsink. I have been meaning to modify it, right now I have a 140mm fan pointing at it to stop it hitting 1500rpm (where it irritates me).
Chasing silence does have diminishing returns. Manually setting the 3950X to 3.6ghz 0.95 volts does make a massive difference. I could run the PC semi passive (no case fans, just CPU <500rpm) if it wasn't for the damn PCH fan!
AMD also sets the voltage too high for their GPUs, making them actually run slower (the one-click downvolt works well), I think AMD dislikes the planet...
With the exception of Intel's ice lake CPUs (10th gen Core), everyone does thunderbolt on a discrete controller (in the case of this ASRock board it's a Titan Ridge controller). The x570 chipset runs very hot and is actively cooled regardless of the presence of an off-die thunderbolt controller hooked up to some of it's PCI-express lanes.
