81–90 of 358 posts
Earlier quoted context omitted.
I did bitcoin but used it for beer money. RIP the 150 or so coins I mined.
This is where most of my “what if” thoughts about Bitcoin go. I heard about it when it was pretty cheap — I want to say single digit dollars, but it was a while ago. But if I’d have gotten into it at the time, I surely would have decided “it’ll never go higher” around $200 at most.
I have a tangential question. I remember distributed processing being a "next big thing" back in the day. The idea was that a company could run complex operations by borrowing some power from all their employee's computers. Seti@home proved the concept was possible. So what happened? Why can't I train a random forest across all the computers currently logged into my company's VPN?
This makes me sad because it reminds me of something similar I just went through. For the last few years, I've been running >120k gpus to mine ETH. Before we turned them off after ETH switched to PoS, I tried really hard to find an alternative workload for them. A large number of the GPUs were based on the PS5 chip, which is an APU (CPU+GPU) and ran as blade computers. It is actually a pretty power efficient chip, bu…
You ran 120,000 graphics card to mine ETH and didn't end up with lambos or like... ether.. to donate to the professor along with the hardware?
Is it possible to distribute AI training like this so that people could donate their idle processing power to creating a truly open source unencumbered LLM?
Yes. You'd have to package up slices of the training set and the prior epoch as work units and host them. Our hypothetical GPT@home clients download the training set, train just those examples, and then send back their gradients. The server then averages all the gradients together. As far as I can tell this isn't any different conceptually from the distributed inference setup that PyTorch/Huggingface Accelerate/etc s…
The main problem I believe is that you would need extremely high bandwidth. For a 100B model you're looking at 400GB of gradients you need to send each batch. There's a reason why super fast networking is used.
Earlier quoted context omitted.
Bitcoin has nothing to do with "cracking SHA256 hashes", for what it's worth. The only thing remotely related is how it brute-forces inputs until it generates hashes that begin with a certain number of bits set to 0.
> hashes that begin with a certain number of bits set to 0. This myth was always intriguing to me. Where does it come from? It's not a number of 0s (which would imply difficulty could only be doubled or halved), it's just finding a hash that is less than the target.
This makes me sad because it reminds me of something similar I just went through. For the last few years, I've been running >120k gpus to mine ETH. Before we turned them off after ETH switched to PoS, I tried really hard to find an alternative workload for them. A large number of the GPUs were based on the PS5 chip, which is an APU (CPU+GPU) and ran as blade computers. It is actually a pretty power efficient chip, bu…
> The problem was finding funding to pay for the operating expenses, megawatts of power, so it never got off the ground and the hardware has been turned off and decommissioned. Such a bummer. You ran 120,000 graphics card to mine ETH and didn't end up with lambos or like... ether.. to donate to the professor along with the hardware?
This makes me sad because it reminds me of something similar I just went through. For the last few years, I've been running >120k gpus to mine ETH. Before we turned them off after ETH switched to PoS, I tried really hard to find an alternative workload for them. A large number of the GPUs were based on the PS5 chip, which is an APU (CPU+GPU) and ran as blade computers. It is actually a pretty power efficient chip, bu…
> I've been running >120k gpus to mine ETH So when the sea levels rise and destroy my home, I'll know who to blame. Thank you.
misdirected angst
It seems the aliens are already here, so not much need for SETI anymore. More seriously though, it appears there was some unresolved difficulty in searching and analyzing the data returned from the network, to the point the scientists working on it gradually threw in the towel and moved on to other projects. https://setiathome.berkeley.edu/forum_thread.php?id=85996 https://einsteinathome.org/content/can-we-run-setiho…
https://www.militarytimes.com/off-duty/military-culture/2023...
The only time a senior DOD / military official uses the word "possible" is when it's either happened or is going to happen.
"It's possible you guys will be deployed within the next year." means "You guys are going to be deployed sometime within the next year."