I’m using an AMD system as an OBS Studio streaming system, and Linux was not great.
I first set it up with a Ryzen 5 3600 and Radeon HD 6750, running Ubuntu 20.04 LTS, because I thought I didn’t really need that much processing power. After un-blacklisting the driver for such an old GPU, I discovered I was using upwards of 80% CPU and dropping frames while streaming at 1440p, so I decided to upgrade.
Then, I tried a Ryzen 7 5700g with integrated Vega 8. First, I needed to upgrade to Ubuntu 21.10 for such a new GPU, and then OBS Studio was randomly crashing while switching between scenes. Also, hardware video encoding wasn’t working well, so it was still taking upwards of 80% CPU while streaming at 1440p. And the video outputs were finicky, sending windows to the wrong screen on power up. Random crashing is unusable, so I switched to Windows.
With Windows 11 on the Ryzen 7 5700g, the hardware video encoding works well, so the same scenes are taking less than 50% CPU while streaming at 4K (2160p) and not dropping any frames. Now I can do other things on the stream.
In short, Apple does things in non-standard ways without explaining how to get another OS to work.
Apple doesn’t prioritize lack of binary blobs. The EFI firmware is all proprietary. All their Wi-Fi have been switched to Broadcom.
They do weird non-standard things to the Thunderbolt controller, e.g., you have to lie to the firmware and claim to be macOS in order for it not to disable the Thunderbolt controller.
Newer MacBooks hide a bunch of hardware behind the proprietary T2, and whatever embedded OS runs the Touch Bar.
MacBooks are not ideologically pure, and sunk efforts to get an OS working on other machines are often wasted on MacBooks because Apple does things in bizarrely different ways.
The design challenge of high-performance CPUs today is heat. Too much heat induces throttling. Putting low-priority stuff onto efficiency cores reserves the power budget of the SoC for high-priority tasks.
My impression was more than a little regionalism. KDE came from Europe. GNOME came from America. Most distros that I know come from America.
In addition, Richard Stallman made lawyers nervous of KDE by claiming that using GPL libraries with Qt (which at the time was under a proprietary license with source code available via FreeQt, definitely not a FSF-approved license) meant the KDE developers had no legal right to use GPL programs at all, even after KDE and Qt were relicensed under GPL.
Stallman eventually relented and granted permission to use the FSF’s own software, but the KDE project didn’t ask for it and didn’t bother to get forgiveness from other GPL license owners. Whether they need to go through that ceremony was never tested in court. GPLv3 switched to a much more reasonable standard, just provide the source under a compatible license.
By the time the license tempest-in-a-teapot reached a conclusion, GNOME and KDE already had substantial-enough installed bases and corporate sponsors to keep going indefinitely.
Those negative effects were not necessarily linked. Prop 13 was written by a lobbyist for commercial landlords, and corporations have always benefited the most from it.
There are other ways to provide relief for rising property taxes. For example, Washington has a property tax deferral program, so seniors don’t have to pay the increased property tax until they or their heirs sell the house; the capital gains that increased the property tax will pay for the deferred property tax.
When we have developers building luxury condos amidst squalor, that’s a second-order effect of policies intended to create such contrasts. The market is constrained by the laws.
Most of the stress on land comes from exclusionary laws. People with power don’t want poorer people (and especially people of another race) to live near them—attributing their success to their efforts and labeling their own servants as parasites[1] and deadbeats—and pass laws that make housing affordability impossible. Thus, people need to drive long distances to jobs and live in disinvested communities in order to participate in regional economies.
The worst offenders are actually not cities, but the inner suburbs surrounding cities. Cities are filled with guilty settlers who want to help less fortunate people but are unable to let go of the systems of power, while their suburbs are usually created for the purpose of segregation and repeatedly reaffirm their desire for segregation.[2]
> Also, doing burnouts and donuts on public streets sounds fun but it leaves the road covered in tire rubber and I imagine it smells awful while they’re doing it.
It indeed does smell awful, and the tire dust is filled with toxins that you probably don’t want in your system; that have already been implicated in the deaths of salmon.
It’s also very loud, well over 100 dB. That’s enough to cause permanent hearing loss.
I don’t believe the moral high ground is with the car club. Their behavior harms people outside of their in-group, with the most severe harms falling on people less privileged than themselves.
Correct, you don’t need to care. But the commons only needs to be a tragedy if you let it be a tragedy.
Just because others behave badly or worse doesn’t give you the moral right to do badly.
We don’t need to be perfect. If everybody did as politely as we could, and did what we could to help others behave as politely as they could, then I think everybody would be better off. Well, “polite” is not quite a direct translation of the concept I have in mind… I’m not sure how to communicate it in English.
…So it can route 2.5 Gbps, by cutting out Linux’s entire networking stack and rebuilding the minimum necessary in eBPF. Not slowed down by NAT or TC yet.
How do you do the I/O? As I understand the MOX, it has one SGMII interface for the built-in 1G Ethernet port, SDIO and PCIe for the WiFi interfaces, and a single 2.5 Gbps SGMII interface to the rest of the Ethernet ports. To get 24 ports, you connect 8-port modules together via their 2.5 Gbps SGMII interfaces.
Seems like the I/O should be enough for 1 Gbps full-duplex, which is enough for a home router with a gigabit Internet connection, but it can’t do 2 Gbps full-duplex.
In terms of raw data rate, each byte of data on 802.11g takes the airtime of about 3 bytes on 802.11n. Since 2.4 GHz WiFi penetrates obstacles so well, that’s airtime that you’re excluding from up to several nearby homes if you generate traffic on 802.11g.
For me, it’s primarily because this device has stood still while faster, cheaper, more powerful devices with lower power consumption and more fully open-source firmwares have entered the marketplace.
The WRT54GL doesn’t have a fully open firmware—the WiFi remains closed—and has so little RAM and flash that OpenWRT, that started with the WRT54G, no longer supports it. DD-WRT is creeping in the same direction.
IPv6 support is more off than on. It’s never going to support DoT, WPA3, or other modern security measures. Most of the world is urbanized, and in an urban setting it’s a bit rude to use 54 Mbps 802.11g on the 2.4 GHz channels.
If you’re using it for an internal network on a farm, it’s fine, but if you’re in today’s world then you need to support today’s protocols.
I have one of their dual-channel mini routers, the GL-AR750. The preinstalled OS is a fork of OpenWRT that comes bundled with Wireguard, Samba, and several other useful programs, but the package repository doesn’t have the variety of the OpenWRT repository.
The hardware is fully supported by the OpenWRT project, so installing plain OpenWRT is easy.
But it’s still a relatively slow device (650 MHz MIPS 24Kc) with throughput well below 100 Mbps when encrypting. The faster GL-AR750S is still well below 100 Mbps when encrypting. That’s the tradeoff of a small and cheap device that can be powered by a USB 5V power adapter.
The more annoying part is that the 5 GHz WiFi has not been certified for DFS, and therefore DFS channels are disabled.
Benchmark sites like userbenchmark.com and cpu-monkey.com rate the 7-year-old 3.2 GHz Haswell as having roughly the same single-core performance as the 2-year-old 2.1 GHz (3.7 GHz single-core turbo) Zen+ CPU in the DMAF5. But that doesn’t cover the I/O performance, especially with Realtek Ethernet interfaces.
Most mini-PCs are using 2-GHz-or-lower Gemini Lake CPUs with much worse single-core performance, but much better video I/O performance. But that still doesn’t cover network I/O, and Realtek vs Intel. (And Intel i40 vs i211.)
Relatively few people have gigabit Internet, but it can happen suddenly. 3 years ago, I upgraded from 20 Mbps DSL to 1 Gbps symmetric fiber. Bye-bye to my old MIPS 24K router. But all these years later, my brother 1 mile away is still waiting.
The limitation is only on the DFS channels. Channels 36–48 and 149–165 are free of radar. And also crowded in a modern urban environment.
I’m not sure what the device manufacturers are doing to prevent unauthorized use of DFS channels, but I have a cheap router with a QCA9887 (GL.iNet GL-AR750) and it doesn’t use DFS channels no matter what firmware you put on it.
I tried prioritizing “open.” I had a customized firmware on a fully open-source (even open-source WiFi firmware) Atheros-based router. And “open” turned out to mean, “enough rope to hang yourself.” I didn’t dare update that thing for years because I installed it in an inconvenient location and I couldn’t trust that it would continue to work if I installed upstream updates.
Now I emphasize update automation. The closed firmwares of ISP routers are not great, but the ISPs take charge of maintaining them. I don’t recommend plain OpenWRT to non-technical users because it doesn’t auto-update.
Maybe a Turris router, because they have the CZ.NIC people in charge of updates. Even there, the transition from Turris OS 3 to Turris OS 5 has been disruptive because of the upgrade from OpenWRT 15 to OpenWRT 19 and its migrations to Device Tree and Distributed Switch Architecture. At least CZ.NIC is still updating the Turris OS 3 packages.
Thin clients were fun, but they aren’t nearly fast enough for modern multi-gigabit Internet connections. They can route maybe up to a couple hundred Mbps.
I’m wondering whether a mini-PC can route even 1 Gbps at line rate. Lots of people are using mini-PCs as routers, but most of them have only 1×1Gbps Ethernet interface and no PCI slots. The Minisforum DMAF5 has 2×1Gbps Ethernet interfaces, but that’s an off-label use and I haven’t seen any benchmarks.
Of course, if you go all the way up to mini-ITX, then there are plenty of options for various performances of CPUs and network interfaces.
Isn’t that basically Keyboardio? Except it’s a San Francisco company selling them for $150; expensive, but still within reason for boutique mechanical keyboards.
I think the big motivation for the Omnia is the Turris project, not open source per se. Security threat analysis and automatic updates from the nonprofit organization that runs the Czech DNS registrar. LXC, Wireguard, and the customization options from the mini-PCIe slots are a bit of a bonus.
The Omnia doesn’t have great OpenWRT upstream support, and the wireless performance sucks. 2GB of RAM seems enormous for a router, but when I put a medium-size number of clients on it (100-ish), its security monitoring features overran the memory and oom-killed essential services. Fortunately, that can be turned off.
And the Turris project seems to be retreating from modern Internet speeds. The Omnia can’t keep up with 1Gb full-duplex fiber, but they’ve moved onto their next product: The MOX/Shield is even slower. (1.6 GHz CPU vs 1.0 GHz CPU)
I first set it up with a Ryzen 5 3600 and Radeon HD 6750, running Ubuntu 20.04 LTS, because I thought I didn’t really need that much processing power. After un-blacklisting the driver for such an old GPU, I discovered I was using upwards of 80% CPU and dropping frames while streaming at 1440p, so I decided to upgrade.
Then, I tried a Ryzen 7 5700g with integrated Vega 8. First, I needed to upgrade to Ubuntu 21.10 for such a new GPU, and then OBS Studio was randomly crashing while switching between scenes. Also, hardware video encoding wasn’t working well, so it was still taking upwards of 80% CPU while streaming at 1440p. And the video outputs were finicky, sending windows to the wrong screen on power up. Random crashing is unusable, so I switched to Windows.
With Windows 11 on the Ryzen 7 5700g, the hardware video encoding works well, so the same scenes are taking less than 50% CPU while streaming at 4K (2160p) and not dropping any frames. Now I can do other things on the stream.