wmf

52,078 karma · joined August 19, 2007

Wes Felter, expert in datacenter/cloud networking and containers. Previously worked at IBM Research and a large investment bank.

submissions comments

wmf··on Why Omarchy Exists

Ubuntu was curated by Mark Shuttleworth for years. GNOME OS and KDE Linux are also fairly curated. It's not clear that DHH's curation of Omarchy actually created a better distro.

wmf··on Linux support is coming to Snapdragon X2 Series

Is it more work to support a completely proprietary boot path or three different "standard" boot paths?

wmf··on Linux support is coming to Snapdragon X2 Series

Ironically that's being fixed by Microsoft forcing UEFI.

wmf··on Why Omarchy Exists

Putting politics aside, Omarchy contains a ton of bad technical decisions. I also question the non-discoverable UX but I realize that's subjective.

wmf··on SpaceX Starshield

Starshield includes cameras and other sensors that are not present on normal Starlink satellites.

wmf··on How did AMD Ryzen get 50% faster in two years?

Bulldozer was twice the die size of Intel 4C/8T but with lower performance.

wmf··on ReBarUEFI: Resizable BAR for almost any UEFI system

It allows the CPU to access all GPU VRAM. This improves gaming performance, especially for AMD GPUs.

wmf··on What Sun got wrong

Yet every UEFI server has VGA and remote KVM.

wmf··on What Sun got wrong

The alternative isn't driving to the DC; it's remote serial console through a console server or the BMC.

wmf··on What Sun got wrong

Path dependence again. x86 servers in the 1990s had to have VGA so it was never removed. (Except Oxide.)

wmf··on NASA’s Mars Sample Return mission is dead

Costs increased after it was already under construction so sunk costs kicked in. MSR was still in the planning stages when it was canceled.

wmf··on Fable 5 – Median thinking declined in August

...releasing a new model that’s marginally if at all better than the original...

This isn't what we see in benchmarks.

wmf··on HERMES radio enables voice and data communication over vast distances

AX.25 runs at 300 bps over HF while Mercury is 10x faster.

wmf··on What Sun got wrong

It's path dependence. If you built a Sun-syle x86 server with a serial console and inserted a random Linux installer CD, the installer would probably only output VGA and thus it wouldn't work.

wmf··on HERMES radio enables voice and data communication over vast distances

Coincidentally I looked into this the other day and the fastest mode is 3 kilobit/s. Great for email but it's not going to be an Internet connection.

wmf··on M5 Ultra Mac Studio Review

No, $40K is not comparable to $10K.

wmf··on Software sandboxing: The basics (2025)

VM tech has improved since then. Today you'd probably want to use Firecracker or Cloud Hypervisor with virtio-vsock instead of networking.

wmf··on Nobody pays for FOSS, we can force them to

My point is that the original maintainer can defuse the forks by including the same features. That leaves users no reason to bother using the forks.

wmf··on Nobody pays for FOSS, we can force them to

This new dependency adds very little over the original xz and xz could have just used a simple PR.

I kind of like the incentives created by this competition. This "what if someone takes your software, adds one feature, and sells it" argument has been made many times and I think the answer is that you, as the original maintainer, are in the best position to add features and if you squander that position you deserve to lose.

wmf··on Nobody pays for FOSS, we can force them to

"We encourage people who redistribute free software to charge as much as they wish or can. ... So if you are redistributing copies of free software, you might as well charge a substantial fee and make some money. Redistributing free software is a good and legitimate activity; if you do it, you might as well make a profit from it. ... Distributing free software is an opportunity to raise funds for development. Don't waste it!" https://www.gnu.org/philosophy/selling.html

wmf··on Samsung is expected to more than double output of its HBM4 and HBM4E DRAM

Supposedly OpenAI is paying 2025 prices.

wmf··on Samsung is expected to more than double output of its HBM4 and HBM4E DRAM

If they're charging one customer $X and another customer $5X for the same chips... there's got to be some law about this.

wmf··on How did Apple Silicon get 50% faster in three years? – Daniel Lemire's blog

AMD is also using N2 so it will be interesting to compare Zen 6 against M6.

wmf··on How OpenAI Used Its Own LLMs to Design Its Jalapeño Chip

There are some investigations starting about this but no rulings yet.

wmf··on How OpenAI Used Its Own LLMs to Design Its Jalapeño Chip

Back in the day you'd write the code before the chip came back but I guess today it's faster to wait.

wmf··on How OpenAI Used Its Own LLMs to Design Its Jalapeño Chip

The lawsuit appears to be about consumer devices, not NPUs or ASICs. If you think Jalapeno stole from anyone it would be Google.

wmf··on How OpenAI Used Its Own LLMs to Design Its Jalapeño Chip

Arm sells architecture licenses to anybody these days.

wmf··on Systemd is a suite of basic building blocks

Every now and then there's a blog post or presentation that combines several systemd features to show how the result is a better overall system. The All Systems Go conference https://all-systems-go.io/ is a good place to learn about systemd.

wmf··on Fujitsu launches made-in-Japan next-generation CPU FUJITSU-MONAKA

Honestly I think this is an HPC CPU that has been AI-washed. Then you might ask why not use GPUs for HPC and I think the simple answer is that Fujitsu just doesn't want to take on the effort of building a GPU.

wmf··on Goose Programming Language

Usually strict aliasing.

← PreviousPage 2 of 34Next →