wmf

10 ms·

361.
by wmf 4mo ago
FWIW the original IBM BladeCenter platform was developed before BMCs so it had a much smaller service processor per blade connected over an out of band RS-485 management bus to a chassis management module. It was closer to Oxide than what&#
362.
by wmf 4mo ago
That's not what I meant. IBM published some scientific papers which explain the details. If you don't have the background to understand the papers and you don't trust the marketing that still doesn't justify rhetorically
363.
by wmf 4mo ago
https://rfd.shared.oxide.computer/rfd/0003
364.
by wmf 4mo ago
99% of data centers do not have liquid available to the rack. Liquid within the rack doesn't really help.
365.
by wmf 4mo ago
I know what it means. Something isn't automatically bullshit because it's outside your field of expertise.
366.
by wmf 4mo ago
The vendors did make blades in the 2000s.
367.
by wmf 4mo ago
https://en.wikipedia.org/wiki/There%27s_Plenty_of_Room_at_th... https://en.wikipedia.org/wiki/Landauer%27s_principle
368.
by wmf 4mo ago
It's the equivalent performance of a 0.7 nm planar transistor. It's not about the feature size.
369.
by wmf 4mo ago
The marketing nm better represent the density and performance of the transistors than the actual feature size, especially in this case.
370.
by wmf 4mo ago
They licensed 2 nm to Rapidus so yes.
371.
by wmf 4mo ago
Using multiple chips seems to work fine for Cerebras and Groq so it should also work for Taalas. It does sounds challenging to reach >10K tok/s but latency could be below 1 us which is a small part of the token budget.
372.
by wmf 4mo ago
The next step would be forcing them to pay out FCF as dividends. When that doesn't work Bernie will come up with an even worse idea.
373.
by wmf 4mo ago
28nm chips is just about suitable to run frontier models I doubt it. 28 nm is 4-5 generations back so inferencing would need a large number of chips with very high power consumption. Maybe you're thinking more of 7 nm which is what C
374.
by wmf 4mo ago
I love those 128 GB dGPUs.
375.
by wmf 4mo ago
If GPUs have 75% margin then 50% cheaper is no surprise.
376.
by wmf 4mo ago
Don't forget Stargate. Update: Somebody on Twitter said it's going to be hosted 50/50 at Microsoft and Oracle.
377.
by wmf 4mo ago
https://dl.acm.org/doi/10.1145/3785362 https://developer.nvidia.com/culitho https://www.synopsys.com/blogs/chip-design/analog-layout-syn... https://arxiv.org&
378.
by wmf 4mo ago
I don't see any evidence of this. Open source software thrived on proprietary hardware for decades.
379.
by wmf 4mo ago
“People who are really serious about software should make their own hardware.” ― Alan Kay
380.
by wmf 4mo ago
Good models will require multiple Taalas chips but Groq and Cerebras also require a lot of chips and that hasn't stopped them.
381.
by wmf 4mo ago
I assume OpenAI has been buying memory and "giving" it to Nvidia in exchange for a discount.
382.
by wmf 4mo ago
It's probably the same speed but cheaper.
383.
by wmf 4mo ago
Memory bandwidth is the bottleneck in the Spark. If you replace the SoC with an optimized ASIC but keep the same 256-bit LPDDR5 the performance will be the same. You can increase performance by using wider memory but that's also more e
384.
by wmf 4mo ago
Was his goal to provide for his people? I thought his goal was to bring down the Great Satan.
385.
by wmf 4mo ago
The Synapse fintech collapse.
386.
by wmf 4mo ago
At the time 90% of people had never even touched a Mac.
387.
by wmf 4mo ago
It depends what you mean by solid. System 7.5 was slower and less reliable than Windows 95 but its usability was slightly better.
388.
by wmf 4mo ago
Don't give him ideas.
389.
by wmf 4mo ago
"...the signal is what it tells us about..." AAAAAAAHHHHHHHH!
390.
by wmf 4mo ago
Also these proxies are against the terms of service of every ISP.

More ›