zdw

A YouTuber automated ROCm version testing for Strix Halo/R9700 + LLM workloads:

https://kyuz0.github.io/amd-strix-halo-toolboxes/

https://kyuz0.github.io/amd-r9700-ai-toolboxes/

What hardware (CPU/GPU/memory) and network was used for this? What quantization for GLM 5.2? How much tuning of the split was needed?

While Thiel's recent Antichrist talks could come to mind, this could just as easily be business speak, like in the Collins/Porras "Built to Last" where one of the points of great companies is "Cult-like culture".

Blocking specific service providers in this way should be made illegal.

It's an abuse of market power primarily used to eliminate competition.

I post a lot of links to HN. It's a hobby - I have lots of RSS feeds in NetNewsWire.

Double posts do happen, HN's algorithm allows it. Occasionally a post that got little attention will get boosted up to the front page after a few days.

But I didn't post another copy of this link a few hours ago, and I've never had HN create a duplicate post way later - no clue what's going on here...

Someone should make a sport out of this.

We already have sailing sports where people race all kinds of wind-powered vessels, and they push the envelope of tech development, just like F1 and the car industry.

Also rich people love this sort of thing. Give them something to do with all that money that has some sort of chance of improving things.

I feel like the first part of this explanation reminds me of stories about mainframe computers in the 1950-70s - highly expensive special gear that needs special operators. Is there something about the physics of the process standing in the way?

If you could dramatically reduce cost and improve availability, would this still be a problem?

What's the limiting factor that prevents medical imaging from getting cheaper and more available?

Dunno, this seems fully functional?

Given the age of TrueType, wouldn't nearly all patents be expired already?

Apache2's license I've heard described as mutually-assured-patent-destruction - if you use the code and make a patent claim, your rights to use the code go away.

So Apache2 offers little benefit here, and MIT may get it into more hands?

FWIW, the O2's UMA let it handle far more textures than almost any other contemporary system with reasonable performance.

Most other SGIs had single or low double-digit megabytes of texture memory, whereas the O2 could host one gigabyte of unified memory and use a huge chunk of that for textures.

Video: https://www.youtube.com/watch?v=zyoOwf0pp80

A good link on this are what technology different sects allow:

https://en.wikipedia.org/wiki/Template:Amish_technology_comp...

Nearly everyone allows the washing machine, which having wrung out laundry by hand when ours broke gave me a whole new respect for folks who do without the tool.

Having been through an hiring cycle recently and prior to AI, the entire process has been pretty broken for a long time, but AI is definitely breaking it (and a whole lot of other things) in new and novel ways.

The only reliable and high quality signal is a positive referral, but those are gated by your personal network, which may not be well developed.

Would be great if it could do other things, like be a volume knob, or do media play/pause, given that scrolling and mouse button press seem like things you already have a mouse to do?

Microsoft's volume licensing, from the perspective of sysadmins and other folks trying to actually obtain software for use, is known to be some cross product of "byzantine" and "kafkaesque".

I fail to see how this is a win for the vast majority of folks impacted by the licensing process...

This seems to have happened about a year before "The IT Crowd" episode "Smoke and Mirrors" aired.

In that episode Moss, one of the IT denizens, goes to a TV studio where he is mistakenly put on a news program and interviewed about a war.

I wonder if they're related...

There's a youtube channel that is documenting the entire japanese PC Engine game library: https://www.youtube.com/playlist?list=PL1sb8k4ZPagYwRxx-s3mr...

They previously did the entire famicon library.

I highly doubt that anyone would expect that a chat transcript from a fairly insular group to be made public in this way decades later...

MTP support is being addedto llama.cpp, at least for the Qwen models ( https://github.com/ggml-org/llama.cpp/pull/20533) and I'd imagine Gemma 4 will come soon.

The performance uplift on local/self-hosted models in both quality and speed has been amazing in the last few months.

The HN story autoformatter changes titlecase on well known terms, which I'm betting is what happened here.

Ideally the fees would be similar to the Norway model, where some tickets are tied to the income of the driver, in this case the pre-tax earnings of the company that created the driverless car.

It's like someone should make a file... maybe in /etc ... and put short names for services in it... maybe it could be called /etc/services...

No unencumbered epub, no sale.

macOS has a Network location in the sidebar that will show other SMB devices discovered on the network.

That's a 16" (from the size of the speaker grille on each side of the keyboard), so even more claustrophobic.

IMO, Kubernetes isn't inevitable, and this seems to paint it as such.

K8s is well suited to dynamically scaling a SaaS product delivered over the web. When you get outside this scenario - for example, on-prem or single node "clusters" that are running K8s just for API compatibility, it seems like either overkill or a bad choice. Even when cloud deployed, K8s mostly functions as a batteries-not-included wrapper around the underlying cloud provider services and APIs.

There are also folks who understand the innards of K8s very well that have legitimate criticisms of it - for example, this one from the MetalLB developer: https://blog.dave.tf/post/new-kubernetes/

Before you deploy something, actually understand what the pros/cons are, and what problem it was made to solve, and if your problem isn't at least mostly a match, keep looking.

This is the most esoteric post I've seen on HN in a while.

How many museum curators who need non-yellowing flexible thermoplastic are there on here?

Eh, Gundam is Japan's Star Wars. Released about the same timeframe (late 70's), tons of sequels and spinoffs, etc.

Evangelion is what happens when someone does a very successful riffs on the genre that Gundam is the most prolific example of.

AMD's Hawaii architecture had 320GB/s on a 512b GDDR5 bus in 2013.

The Fiji XT architecture after it had 512GB/S on a 4096b HBM bus in 2015.

The Vega architecture did have 400GB/s or so in 2017, which was a bit of a downgrade.