Hacker News Reader

Comment by ApolIllo | Hacker News Reader

> I appear to be paying a lot of money for products that I could buy for cheap on random websites is starting to piss me off a bit at this point.

The last 5+ years as an Amazon customer have me feeling this way.

> Realtek RTL8153

Could the Realtek issues be related to power?

I have a Tbolt dock which is powered using a DC barrel jack (6.5amps @ 20V). The comparison is not great as the dock (TBT3-UDZ) is not a USB C dock and uses the Intel i211 nic.

The author appears to have found stability with the Anker PowerExpand 8-in-1, which has the buck converter with power management components.

I've had terrible experiences with the Satechi Multiport Adapter V2 and ICY BOX USB-C type adapters.

I suspect that the issues can typically narrow down with any dock to bandwidth or power.

View on HN

At least with respect to macOS, the Realtek 8153 chipsets in these docks suffer from having to use the built-in ECM driver. If you're on Apple Silicon, that's your only option. If you're on Intel, there are some flaky but more performant drivers from Realtek available. The ECM driver will cause high CPU load and for many users, will result in performance that's worse than wifi because of it. You'll also get audio dropouts and system hitches from the CPU loading.

If you can find an adapter that uses the Realtek 8156 chipset, which I believe the CalDigit TS4 uses, macOS will utilize the NCM driver and your performance will be rock solid.

ECM is a very primitive USB protocol for ethernet, whereas NCM is a more modern and performant one. NCM is to ECM as UASP is to BOT mode, if you're familiar with USB external drives.

Power is a real concern - and if you have something pushing lots of pixels or data you should be suspicious of it if it doesn't have a power brick.

I am making these Type-C hubs for a living, and indeed RTL8153 is the most trouble free, and cheap chip on the market.

Though, I myself had misfortune buying broken RTL8153 based ethernet NICs. Mismatched coupling capacitors, PCB shorts, possibly broken firmwares.

40%-30% people who do EE in China are just above the Arduino level.

Chinese engineering companies can make good hardware if you pay them, and give time to fix self induced issues coming from rushed development schedules, but in that case, they are not really that ahead from any other EE companies around the world.

Though, I myself had misfortune buying broken RTL8153 based ethernet NICs.

One of the issues with the RTL8153 NIC is (as a sibling commenter points out) that on M1 Macs, you are confined to the CDC-ECM driver. RTL8153 with this driver usually cap at ~700MBit and cause a lot of CPU load (usually restricted to the efficiency cores, but it's still pretty bad anyway).

For USB docks, I can still understand the choice. But IMO this is inexcusable for Thunderbolt docks. You can tunnel PCI-E, so nothing holds dock makers from hanging a good NIC with well-supported drivers on the PCI-E bus. Some higher-end Thunderbolt Docks do this AFAIK. The Apple Thunderbolt 2 Ethernet adapter uses a Broadcom NIC on the PCI-E bus and can reach 1000MBit without causing high CPU loads. Unfortunately, it requires a Thunderbolt 3 -> 2 adapter, which is more expensive than the Ethernet adapter itself.

Reply on news.ycombinator.com