wmf

The bus is also 50% wider so the bandwidth is double.

I assume they'll have events for the MacBook Ultra and M7.

LPDDR6 is coming.

Speaking of "us" it seems that Samsung/Mediatek/Xiaomi/Huawei SoCs are not available in the US.

AFAIK it's mostly self-contained outside of basics like malloc. If you have any evidence to the contrary, post it.

No, M5 Ultra would have more DRAM chips visible.

Does Linux work on X2?

You can see the white/silver radiators; they're large but not outrageous. https://www.spacex.com/spacexai/starmind

But running Linux on top of seL4 does not take advantage of the capability model at all. You need a new non-unix-like userspace for that.

There isn't really an ELI5 version of computer architecture but you could start with the book Inside the Machine by Jon Stokes. Then you can get into SIMT.

Hugging Face was able to sign up for Glasswing but they just didn't.

Unfortunately AGPL has been a scam for years. It's now an anti-freedom license.

That's probably PCIe DMA initiated by the GPU so I don't think BAR is used.

People get killed crossing the street every day. File an insurance claim and move on.

1978?

You mean they were putting a thumb on the scale and now they're not.

Do you know about Jalapeno?

Noobs like to post summaries of the article as a comment but they don't know that we don't do that here.

Cerebras provides high-speed inference at high cost. It's never going to be the cheapest and thus it will probably remain niche.

Desktop SRAM isn't really viable because it could cost $100K just to load the model.

Linux loves to leave papercuts unfixed or undocumented for decades. The solution is to build against an older version of glibc but no one tells you that or how to do it.

I guess because I have actually set foot in a data center I don't imagine literally every product in my home.

No, but what's the alternative theory?

For every case like this, there's a patent on an optimal method which forces everyone to use suboptimal workarounds.

Cerebras is only claiming ~2x the performance of Groqvidia which usually isn't enough for people to switch.

162 kW

I've only seen the Pollock movie but he definitely dripped the paint himself. Are you confusing Pollock with someone else like Sol LeWitt?

If you mix proprietary human-written code with uncopyrightable public domain code, the result is still proprietary. There's no reverse virality. There's no problem to fix here.

For example, every proprietary OS includes public domain SQLite and it's fine.

Mix in some human-written code and dare the haters to prove that it's 100% AI. They won't bother.

Memory fabs make a mix of DDR/LPDDR/GDDR/HBM and they continuously adjust the mix to market demand. If demand for HBM drops they will start making more consumer RAM.