hedgehog

23 ms·

511.
by hedgehog 2y ago
In my case spawn overhead is negligible relative to the jobs themselves, my existing experience with AWS Batch has been pretty good. I took another look at the Dokku documentation and extending `run:detached` would work well enough, though
512.
by hedgehog 2y ago
Are there plugins or anything that will allow Dokku to run queues of batch jobs? Skimming the docs there is clear support for Heroku-style apps and one-offs that can be launched via cron, not so much for batch although it would be quite use
513.
by hedgehog 2y ago
Often the product is "hard IP" which is a design which has been tuned and validated for a specific manufacturing process. A rough software analogy would be buying a license for a proprietary library that's been validated on a
514.
by hedgehog 2y ago
It depends on the API used, whether the data is in the region considered "GPU memory" or whether it's shared with the compute API from the app's memory space. Support is somewhat in flux and I haven't been following
515.
by hedgehog 2y ago
U2 was Skunkworks, they started with an existing airframe (F-104?) and worked from there. Significant compromises, for example the wings droop so much they skid on the ground on landing and it's use was limited by the improved anti-air
516.
by hedgehog 2y ago
Monocrystalline turbine blades are one of those things that make me appreciate how deep the state of the art is for a lot of "simple" things when you look closely.
517.
by hedgehog 2y ago
Is Mojo actually getting significant uptake in research? I haven't been following closely but new tooling at that layer seems much more useful when cost-optimizing deployment.
518.
by hedgehog 2y ago
The 10nm debacle was something to behold but they're still at it and shipping some products on Intel 4 now. I think the question is "at what cost?"
519.
by hedgehog 2y ago
I suspect AMD's assessment at the time was similar so they offloaded fab and hopped on the TSMC bandwagon. It's worked out well for them. Interestingly Global Foundries decided not to keep chasing leading edge and instead invest d
520.
by hedgehog 2y ago
The problem with keeping the foundry business is they're probably never going to get back to competitive at leading edge and that dooms end products that rely on having the best process to be competitive. In times past Intel dominated
521.
by hedgehog 2y ago
It's not about the syntax, it's all the knowledge, tools, existing code, etc that make Python so attractive.
522.
by hedgehog 2y ago
Ah true. I live in a hilly region so the water issues tend to be rain water moving down hill (which is really mostly laterally). In that situation the exterior drain solution works well, totally dry basement. I don't know how common it
523.
by hedgehog 2y ago
He's arguing for more competition on the customer side (with Apple) but starting more domestic fabs wouldn't directly help unless they have backing with patience and absurdly deep pockets. A big reason we're in this situation
524.
by hedgehog 2y ago
Yeah it's not a cheap approach but unless the water is coming in really fast it'll get the basement really dry. Key to have the drain and pump well below the footings otherwise yes the water will just be under the slab.
525.
by hedgehog 2y ago
Intel was the last company to have domestic leading edge fabs and they used to be one of (the?) highest volume producers by wafer count so they're a pretty big part of any conversation about US semiconductor manufacturing. Another is A
526.
by hedgehog 2y ago
I thought it was weak, his core argument seems to boil down to the idea that if there were more domestic competition then device manufacturers would somehow be willing to buy worse chips for more money. It just doesn't make any sense.
527.
by hedgehog 2y ago
As far as I can tell inside waterproofing is a bad idea, it leaves your foundation soaking wet, excess humidity indoors, and as you point out it doesn't alleviate the pressure from outside. Waterproofing the exterior, french drains dee
528.
by hedgehog 2y ago
I think this attitude is what is driving his complaints. Most engineering work exists in the context of towering teetering piles of legacy decisions, organizational cultures, partially specified problems, and uncertainty about the future. P
529.
by hedgehog 2y ago
That argument would make more sense if such a language was widely available but today in practice it isn't so we live in the universe of less ideal solutions. Actually it doesn't really respond to DJB's point anyway, his case
530.
by hedgehog 2y ago
There is a fundamental difference of priorities between the two worlds. For most general application code any optimization is fine as long as the output is correct. In security critical code information leakage from execution time and resou
531.
by hedgehog 2y ago
For those interested in going deeper I think the classic reference in this area is the GotoBLAS paper: https://www.cs.utexas.edu/~pingali/CS378/2008sp/papers/gotoP...
532.
by hedgehog 2y ago
Yes and no. Read his notes about his goals and what he didn't implement, and note that his job when he wrote that was as part of a team that ports models to different hardware. That's a smaller job than writing a shippable framewo
533.
by hedgehog 2y ago
It's worse than that, thermal management is part of the puzzle. Think of that as heat generation happening across three dimensions (X + Y + time) along with diffusion in 3D through the package.
534.
by hedgehog 2y ago
Hah, ok, yeah except for small org + tiny models no ML team is training on laptops. On the other hand I think we're already seeing more app devs using language models while not doing any training at all, and those people will probably
535.
by hedgehog 2y ago
The above poster is mostly happy with a 7840HS which tells us they're a laptop user and likely looking at workloads that would work on a laptop with spare time hobby levels of effort. Modern laptop chip memory bandwidths are in the 100
536.
by hedgehog 2y ago
No need for that, you can get pretty far with a $500 desktop GPU added to your existing machine or a $2000 gaming laptop (for inference workloads).
537.
by hedgehog 2y ago
One thing to keep in mind is that most of the "NPU" type on-chip accelerators are designed to minimize power consumption for small networks (think speech recognition and video call background removal) rather than supporting big la
538.
by hedgehog 2y ago
I don't know about the NPU but 6.10 fixes one of the big memory issues that affected deep learning workloads on APUs.
539.
by hedgehog 2y ago
I'm not sure what the overall calculation looks like but if you remove 10-15 liters per minute per occupant to match normal respiration that's a pretty small loss of heat. Maybe a net win if it reduces need for other ventilation a
540.
by hedgehog 2y ago
This is a pretty cool project, it's basically a retrofit approach to what in industry is known as "displacement ventilation" or "stratification ventilation". In the US our typical approach is "mixing ventilatio

More ›