TE
TechEcho
Home24h TopNewestBestAskShowJobs
GitHubTwitter
Home

TechEcho

A tech news platform built with Next.js, providing global tech news and discussions.

GitHubTwitter

Home

HomeNewestBestAskShowJobs

Resources

HackerNews APIOriginal HackerNewsNext.js

© 2025 TechEcho. All rights reserved.

Diffeq.jl v6.4: Full GPU ODEs, Neural ODEs with Batching on GPUs, and More

173 pointsby ViralBShahabout 6 years ago

8 comments

eigenspaceabout 6 years ago
Wow, that&#x27;s really cool. I feel like DiffEq is one of those libraries that seems to leverage every one of Julia&#x27;s cool features that I can think of.<p>Making differential equations &#x27;just work&#x27; on a GPU is super cool. I feel like if this were almost any other language, it&#x27;d be impossible to get this level of code-reuse where a user can just bring their own differential equation that may not even know about GPUs and then have the differential equation library solve it on the GPU by just giving a GPUarray as the initial condition.<p>________________________________<p>Also, since the blog post doesn&#x27;t seem to link to it, here&#x27;s a link to the actual DifferentialEquations.jl repo: <a href="https:&#x2F;&#x2F;github.com&#x2F;JuliaDiffEq&#x2F;DifferentialEquations.jl" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;JuliaDiffEq&#x2F;DifferentialEquations.jl</a>
评论 #19877127 未加载
评论 #19877142 未加载
pepijndevosabout 6 years ago
I&#x27;m contemplating writing some fun electromagnetic simulator, and so far had my eyes on Rust and&#x2F;or Futhark with some prototypes in Python even Matlab, but now I may need to consider Julia too...
adamnemecekabout 6 years ago
I&#x27;ve been writing julia a bunch lately. It&#x27;s a joy to use.
评论 #19875493 未加载
skdotdanabout 6 years ago
Noob question, having played around with CUDA and NVCC a bit: in languages such as Julia or Python, why does every algorithm have to be specifically adapted to be able to run on GPUs? Couldn&#x27;t algorithms be written from higher level building blocks? Eg. implement map, reduce and scanl in CUDA&#x2F;OpenCL, and have a default parameter like backend=&#x27;cpu&#x27; than can be set to &#x27;gpu&#x27;.
评论 #19877173 未加载
评论 #19877060 未加载
Certhasabout 6 years ago
Incredible work. Thank you everyone for the great release. This massively improves our usecase (sparse matrix equations) and I am looking forward to playing with it.
评论 #19877102 未加载
vsskanthabout 6 years ago
Anyone tried using Julia&#x27;s solvers to run FMI ModelExchange binaries ?
评论 #19877855 未加载
microcolonelabout 6 years ago
Which GPUs? I&#x27;m guessing by the &quot;CuArrays&quot; identifier that this is CUDA-only, it&#x27;d be nice if they said <i>NVIDIA</i> GPUs instead of getting my hopes up. :&#x27;- (
评论 #19875510 未加载
评论 #19875846 未加载
mleventalabout 6 years ago
this happens literally every single time i try to hop on the julia bang wagon: run example -&gt; example breaks.<p>1. fresh install of juno (1.1.0)<p>2. add latest packages<p>3. something goes wrong. this time it&#x27;s<p><pre><code> p = destructure(model) </code></pre> from example page gives<p>UndefVarError: destructure not defined top-level scope at none:0<p><a href="https:&#x2F;&#x2F;i.imgur.com&#x2F;8fCpVrR.png" rel="nofollow">https:&#x2F;&#x2F;i.imgur.com&#x2F;8fCpVrR.png</a><p>how the hell does anyone get anything when the codebase is shifting this fast!?
评论 #19880663 未加载