brrrrrm

27 ms·

331.
by brrrrrm 4y ago
hmm, there's 1.6M lines of C++ in V8. And that's not a collection of simple mathematical functions - it's an entire compiler with many interdependent components. I think there's this weird but popular idea that people
332.
by brrrrrm 4y ago
More work than JIT compiling Python?
333.
by brrrrrm 4y ago
Non-trivial numerics are done through FFI in python. Python doesn’t support fp16, but a native library behind it does. The exact same premise applies to JavaScript
334.
by brrrrrm 4y ago
that's a good point, but it doesn't have full language support like the ones listed above
335.
by brrrrrm 4y ago
Do you think this will start to change? I often see posts on this site bemoaning the performance of Python (often referencing the order of magnitude performance differences of similarly dynamic languages like JavaScript). I'm hopeful
336.
by brrrrrm 4y ago
Do any of these JITs/compilers help with machine learning model building? (Mentioned explicitly in the s6 repo) e.g. PyPy doesn't work with PyTorch https://github.com/pytorch/pytorch/issues/17835
337.
by brrrrrm 4y ago
> S6 was aiming to do for Python what V8 did for Javascript > Python is slow, and a lot of researchers use Python as their primary interface to build models. Why not just build models in JS?
338.
Tesla assigned my Model 3 to some random person in Europe(old.reddit.com)
44 points by brrrrrm 4y ago|20 comments
339.
by brrrrrm 4y ago
The Python code is run every time.
340.
by brrrrrm 4y ago
It’s not clear a JIT compiled language would help much here unless the operations were baked into the JIT itself (which would have to identify the memory savings of an in-place call).
341.
by brrrrrm 4y ago
the CPU backend uses that. this is why the number is so high compared to more conventional CPU/GPU trade-offs
342.
M1 Pro CPU is faster than the GPU for anything less than 512x512(twitter.com)
18 points by brrrrrm 4y ago|2 comments
343.
Accelerate Python by Re-Writing It in JavaScript(jott.live)
2 points by brrrrrm 4y ago|0 comments
344.
by brrrrrm 4y ago
To answer the question, JavaScript is 31x faster out of the box on size 1000000 (compared to the 6x claimed in the post). 71x on 10M. bwasti@bwasti-mbp code % time python3 prime.py 78498 python3 prime.py 3.86s user 0.02s sys
345.
by brrrrrm 4y ago
how does this compare to a normal JIT compiled language? Hacking on performance to Python seems really brittle
346.
Profiling JavaScript/TypeScript in Bun 0.1.11(jott.live)
1 points by brrrrrm 4y ago|0 comments
347.
Show HN: Shumai – Fast ML in JavaScript(github.com)
3 points by brrrrrm 4y ago|0 comments
348.
by brrrrrm 4y ago
CPU-adjacent ML compute power. Today, models are trained on massive GPUs that live very far away from CPUs (microseconds). This means the models end up having extremely predictable workloads (dense linear algebra). If CPUs were closer t
349.
Libsysml: A collection of utilities for the intersection of ML and systems(github.com)
2 points by brrrrrm 4y ago|0 comments
350.
by brrrrrm 4y ago
I suspect it’s the latter. I would guess V8 with a thin wrapper and a ton of effort from someone who cares about perf to port all the native goodies of node would get similar wins.
351.
by brrrrrm 4y ago
I’m not familiar with deployment in the modern web world, but I’m curious how fast those specific features need to be in a production setting? Are they done dynamically for every request?
352.
by brrrrrm 4y ago
Congrats to Jarred! Bun has blown me away in terms of performance. Often (in other projects) performance claims out of nowhere are simply cherry-picked but I’ve found bun repeatedly impressing me with both speed and elegance. What doesn’t
353.
by brrrrrm 4y ago
number of exponent bits. bfloat16 has a larger dynamic range with lower precision. e.g. this would be infinity in fp16 https://float.exposed/b0x5f80
354.
by brrrrrm 4y ago
a popular trick is to encode information in NaNs
355.
by brrrrrm 4y ago
for lack of bfloat16 support
356.
by brrrrrm 4y ago
> the next release of Deno will include a new HTTP server. It is the fastest JavaScript web server ever built. I feel unease reading untestable claims (with no numbers) about future releases. I'm made more uneasy by the fact that t
357.
by brrrrrm 4y ago
JITs (as those found in V8, JSC, SpiderMonkey) can theoretically compile better code than static compilation approaches like C, C++, and Rust. It's because the runtime can collect information about what is being run and then generat
358.
by brrrrrm 4y ago
I suspect it's going to be the FFI integration. On my machine (M1 pro), Bun calling into a C library and running a no-op is 15x faster than Python calling a no-op function. V8 has never had stable bindings but Bun is changing the game
359.
by brrrrrm 4y ago
that uses Bing
360.
by brrrrrm 4y ago
RI/NY doesn't seem right

More ›