My most recent $75 overnight via FedEx Express was not delivered on time (day late which mattered and why I spent so much) and they straight up lied about it. Only anecdata but sad to hear so many concurring stories.
FedEx lost the paperwork I overnighted to try and get my dog unstuck from customs during an international (California to EU) move and gave me the runaround for 3 days telling me "it'll be delivered tomorrow!"
UPS got it there in under 24h.
Weeks later I got a "your package was delivered!" email from FedEx. Fantastic stuff.
Needless to say I'm never proactively choosing them again.
Summary Statement: He summarizes the results of the "sabotaging" attempts by saying, "we just couldn't for the life of ourselves get a nonprotected modern lithium battery to do anything" [02:40:27], in terms of causing a fire. (referring to 18650s)
There are certainly problems out there with cells but it was a surprising statement from someone with a ton of real-world experience. Also they are a very conscientious company so they don't deal with dodgy stuff if they can avoid it.
Unfortunately, for products with batteries in non-standard form factors, we rarely have a choice of a manufacturer. For example, with home robot vacuums, we can only hope that brands will use top quality cells, but the information what cells are used is not even available to the customer. My Neato vacuum still runs great, but now that the company is out of business, my only battery replacement options are from no-name brands, with zero visibility of what cells are used internally.
> For example, with home robot vacuums, we can only hope that brands will use top quality cells, but the information what cells are used is not even available to the customer.
Hardware used to come with diagrams & schematics, for self-service and repair. This was a courtesy, but I guess manufacturers figured out it's better to make the consumer buy a new model. (AKA part of the reason why Commodore went bankrupt.)
Meanwhile we've been talking and implementing measures such as lists of ingredients and nutrients in food, SBOMs in software, privacy/tracking transparency, etc. Let's push it a little further.
You also have the option of building your own battery pack for these, or to disassemble an existing pack and replacing the cells. How difficult that is depends on the manufacturer, but from a quick look at the Neato packs I see it looks relatively trivial.
I'm in the process of replacing the battery in my old robot vacuum right now, and for the amount of time spent doing that, I could buy a whole new vacuum. It's a pity, because the battery itself can be replaced just by plugging it in, but all the batteries I could find were AliExpress fakes, so my two options are either "more e-waste" or "spend $500 in time doing it myself".
Some of us enjoy fixing things like these over and above the hypothetically fungible billable hour, and whether you can substitute some time in an evening with a billable hour or three is highly dependent on your employment situation.
But in this case the relevant cost under discussion isn't that of a replacement vacuum cleaner, but what value you assign to your house not burning down due to a crappy 18650 cell, or the anxiety of worrying that that'll happen.
Unfortunately the cheap "fatbikes" every kid has here these days have no premium cells and even worse chargers and usually poor or no balancing in the BMS. I don't expect them to not catch fire and my HOA is considering to ban charging ebikes in the basement (where you put your bike) due to that. End result is likely that people with more expensive bikes and cells adhere to this and charge in the house while the cheap ass bikes will still be charged in the basement and the fire department can flood it once again when it all goes up in flames.
I have one in 21700 and it seems a great battery so far. Otherwise I'm a Panasonic guy.
I ask because it sits on my forehead, inside a headlamp. Having my hands tied, addressing a problem quickly might be difficult. And them rascals get toasty fast when they ignite.
Molicel is top tier - their P50B is the best 21700 cell available on the open market. Eve, BAK and Ampace also make some really good stuff, although they can't match the performance of Molicel.
I came to the same conclusion when upgrading from 18650s to 21700s, I spent a long time trying to weed out the poorly binned rewards, and the minefield of lithium batteries. They weren't cheap but have been happy with their performance.
There are a lot of people out there that see 18650 or 21700 and think a lithium ion battery is a lithium ion battery and they're all the same (i.e. trying to pull 20 amps from a 2 amp peak battery). I miss one of the father's of liion battery education, Mooch (from ecigarettes forum) who had a whole methodology of testing, and educating people.
I've seen other videos of people driving nails into 18650s and "spicy pillow" square cells, and most of them didn't even let off a puff of smoke. What the vaping community was doing 10-15 years ago to launch batteries into the ceiling is beyond me.
Huh, I did not get that from the article. The main takeaway for me was doing ALU operations in memory resulting in massive energy savings. There is still a von Neumann architecture running the show.
"In recent inference tests run on a 3-billion-parameter LLM developed from IBM’s Granite-8B-Code-Base model, NorthPole was 47 times faster than the next most energy-efficient GPU and was 73 times more energy efficient than the next lowest latency GPU."
It's also fascinating that they are experimenting with analog memory because it pairs so well with model weights
Yeah, analog memory fits so incredibly well. Who cares if it's not "exact" and fuzzes around a bit if it's only used for weights and has massive efficiency advantages. Weights are never "exact" themselves, and it doesn't matter if they don't always read exactly the same. You basically just get some extra "temperature" for free!
A bit beautiful that we might end up partially going back to analog computers, which were quickly replaced by digital ones.
> A bit beautiful that we might end up partially going back to analog computers, which were quickly replaced by digital ones.
How long till we get a Ben Eater-style video about someone making a basic analog neural network using some DACs, analog multipliers[1] and bucket-brigade chips[2] for intermediate values?
Their NorthPole chip doesn't look much different than the Groq LPU or Tenstorrent's hardware or even just AMD's NPU design. The tenstorrent cards have a pretty big amount of SRAM considering their price.
I am not an expert on this but reading Groq's description of their hardware it still has a compute/memory split. They make the memory super fast so it can fully feed the CPU without latency (80 terabytes second!). In the end is it much different than moving the ALU into memory like IBM is doing? The goal for both is to eliminate the memory bottleneck so there can be a variety of valid approaches.
How does Cerebras WSE-3 with 44GB of 'L2' on-chip SRAM compare to Google's TPUs, Tesla's TPUs, NorthPole, Groq LPU, Tenstorrent's, and AMD's NPU designs?
Where should the delivery trucks park if there is no infrastructure for them and the public has an ever increasing appetite for delivered products? Try to think about it from the delivery drivers point of view and their safety. The roads are not any one users exclusive resource. We all pay for them and they must be shared.
One 5 minute delivery spot is as good as many regular parking spots since it won't be taken up by long term parkers. You could probably eliminate 90% of parking spots and turn the other 10% to 5 minute spots and it would be easier for delivery drivers than the status qou
I agree. And on top of that it would be great to have spots with ALPRs that delivery companies can pay for their use and discourage or tow non-compliant vehicles.
Covid time encouraged new food pickup priority parking spots but I don't see a lot of new thinking around emergent street use needs. We have massively increased delivery culture and micro mobility shares and city planning is lagging. (I think delivery is great - fewer car trips and just overall more efficient - my opinion).
In commercial loading zones! We've allocated the color yellow for this in SF.
If commercial drivers petitioned SFMTA to convert more private parking spaces into commercial zones I'd be signing petitions and backing them in their goal 100% of the way.
But generally I've found that commercial drivers would rather just violate the law and endanger others rather than engaging in activism for better infrastructure on our streets, so it's hard to feel sorry for them if they're cited and fined as a result.
I can't speak for SF I'm in Seattle. I don't think it is incumbent on delivery drivers to do activism for their employers. That's my opinion. And I still don't understand why people don't see that the delivery driver on foot is a vulnerable user of the roads and sidewalks. We aren't perfect but we are there for the public not because we like it.
Someone else mentioned "externalizing" the cost of parking via citations. Those are expensive and a trove for the city. That sounds more like subsidising than externalizing.
As far as feeling sorry for "them" - that is a disconcerting view of a servant class.
UPS drivers don't need to write their congressperson. UPS the company can just get their lobbyists to pressure city officials to convert more street parking spaces to commercial only spaces.
Bike lanes reduce the number of cars on the road and therefore make it easier for vehicles that are actually necessary (delivery, work, emergency, etc) to travel and park, not harder. So do all viable alternatives to driving. That in 2025 people still unironically say "just one more lane bro, and we'll solve traffic" is almost unbelievable.
Ok. I understand bike lanes. I didn't say anything about more lanes, bro. I am talking about delivery vehicles and the challenges they face in urban environments. Keep in mind that a delivery driver can spend as much time on foot as they do in their vehicle. This means interacting with vehicles and bikes as a laden pedestrian. They are compromised and it can be dangerous.
Delivery vehicles reduce the number of cars on the road and therefore make it easier for bicycles to negotiate the roads. One delivery vehicle can easily replace dozens of car trips.
It’s the same in the US in college towns. Many people live in shared housing as you describe. My kids and my nieces and nephews have all done it. But I do recognize the point of the article.
I would be truly interested if you could expand on this. I know I can do my own research but I'm starting down the path of what could be called performance python or something similar and real world stories help.
My use case is realtime audio processing (VST plugins).
Metal.jl can be used to write GPU kernels in Julia to target an Apple Silicon GPU. Or you can use KernelAbstractions.jl to write once in a high-level CUDA-like language to target NVIDIA/AMD/Apple/Intel GPUs. For best performance, you'll want to take advantage of vendor-specific hardware, like Tensor Cores in CUDA or Unified Memory on Mac.
You also get an ever-expanding set of Julia GPU libraries. In my experience, these are more focused on the numerical side rather than ML.
If you want to compile an executable for an end user, that functionality was added in Julia 1.12, which hasn't been released yet. Early tests with the release candidate suggest that it works, but I would advise waiting to get a better developer experience.
I'm very interesting in this field (realtime audio + GPU programming). How do you deal with the latency? Do you send or multiple single vectors/buffers to GPU?
Also I think because samples in one channel need to be processed sequentially, does that mean mono audio processing won't benefit a lot from GPU programming. Or maybe you are dealing with spectral signal processing?
You need to find parallelism somewhere to make it worth it. This can be multiple independent channels/voices, one large simulation, one high quality simulation, a large neural network, solving PDEs, voxel simulation (https://www.youtube.com/watch?v=1bS7sHyfi58), additive synthesis, a multitude of FFTs...
Are they producing less food? Migration from the country to the city has been going on for a long time. You just don't need as many people to produce food as you used to.
I'm probably just nostalgic, but to me this hardware is a piece of history that's mostly forgotten or overlooked by anyone who wasn't working in IT at the time. That's why I've been trying to get museums to take it, because my hope would be they'd do something educational with the hardware. Alas, selling is indeed probably my only recourse if I don't want these things to end up on the heap (right away.)
I was there wishing I could afford those amazing machines. My point with the pricing is to hopefully find someone who values the machine as more than a sum of its parts. Speaking of which it reminds me of the youtube channel save it for parts. Saving old tech is not very popular. We lost Living Computers: Museum + Labs after Paul Allen died.
reply