wmf

5 ms·

91.
by wmf 1mo ago
That model would be larger and more expensive than the entire M7 chip.
92.
by wmf 1mo ago
SpaceX is going all in on Starship.
93.
by wmf 1mo ago
Existing CPUs have been extremely optimized by ~6 competing, well-funded teams. I expect AI to accelerate things somewhat but it's not clear that there is any low-hanging fruit available for AI to find.
94.
by wmf 1mo ago
Every company is designing their own chips so the dominant players will be one level down: Broadcom, TSMC, Hynix/Samsung/Micron, etc.
95.
by wmf 1mo ago
The bus is also 50% wider so the bandwidth is double.
96.
by wmf 1mo ago
I assume they'll have events for the MacBook Ultra and M7.
97.
by wmf 1mo ago
LPDDR6 is coming.
98.
by wmf 1mo ago
Speaking of "us" it seems that Samsung/Mediatek/Xiaomi/Huawei SoCs are not available in the US.
99.
by wmf 1mo ago
AFAIK it's mostly self-contained outside of basics like malloc. If you have any evidence to the contrary, post it.
100.
by wmf 1mo ago
No, M5 Ultra would have more DRAM chips visible.
101.
by wmf 1mo ago
Does Linux work on X2?
102.
by wmf 1mo ago
You can see the white/silver radiators; they're large but not outrageous. https://www.spacex.com/spacexai/starmind
103.
by wmf 1mo ago
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.
104.
by wmf 2mo ago
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.
105.
by wmf 2mo ago
Hugging Face was able to sign up for Glasswing but they just didn't.
106.
by wmf 2mo ago
Unfortunately AGPL has been a scam for years. It's now an anti-freedom license.
107.
by wmf 2mo ago
That's probably PCIe DMA initiated by the GPU so I don't think BAR is used.
108.
by wmf 2mo ago
People get killed crossing the street every day. File an insurance claim and move on.
109.
by wmf 2mo ago
1978?
110.
by wmf 2mo ago
You mean they were putting a thumb on the scale and now they're not.
111.
by wmf 2mo ago
Do you know about Jalapeno?
112.
by wmf 2mo ago
Noobs like to post summaries of the article as a comment but they don't know that we don't do that here.
113.
by wmf 2mo ago
Cerebras provides high-speed inference at high cost. It's never going to be the cheapest and thus it will probably remain niche.
114.
by wmf 2mo ago
Desktop SRAM isn't really viable because it could cost $100K just to load the model.
115.
by wmf 2mo ago
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.
116.
by wmf 2mo ago
I guess because I have actually set foot in a data center I don't imagine literally every product in my home.
117.
by wmf 2mo ago
No, but what's the alternative theory?
118.
by wmf 2mo ago
For every case like this, there's a patent on an optimal method which forces everyone to use suboptimal workarounds.
119.
by wmf 2mo ago
Cerebras is only claiming ~2x the performance of Groqvidia which usually isn't enough for people to switch.
120.
by wmf 2mo ago
162 kW

More ›