Search results

Add a note suggesting users prefer PTX MMA over WMMA

cuda kernelsStuff about writing CUDA kernels.Stuff about writing CUDA kernels.

documentationImprovements or additions to documentationImprovements or additions to documentation

performanceHow fast can we go?How fast can we go?

Status: Open (in progress).

JuliaGPU/CUDA.jlnumber 2816

#2816 In JuliaGPU/CUDA.jl;

· kshyatt opened on Jul 23, 2025

Summing a vector is faster than summing a multidimensional array

cuda arrayStuff about CuArray.Stuff about CuArray.

good first issueGood for newcomersGood for newcomers

performanceHow fast can we go?How fast can we go?

Status: Open.

Enhancement

#2764 In JuliaGPU/CUDA.jl;

· lanceXwq opened on Apr 28, 2025

Huge slowdown when performing <code>fft</code> over the second dimension of a 3D array

cuda librariesStuff about CUDA library wrappers.Stuff about CUDA library wrappers.

performanceHow fast can we go?How fast can we go?

Status: Open.

Enhancement

#2641 In JuliaGPU/CUDA.jl;

· marcsgil opened on Feb 5, 2025

Try fast linear indexes for KA

enhancementNew feature or requestNew feature or request

needs changesChanges are needed.Changes are needed.

performanceHow fast can we go?How fast can we go?

Status: Draft (not ready).

JuliaGPU/CUDA.jlnumber 2612

#2612 In JuliaGPU/CUDA.jl;

· vchuravy opened on Jan 9, 2025

CUDA.@profile: DataFrames post-processing needs to be optimized

enhancementNew feature or requestNew feature or request

help wantedExtra attention is neededExtra attention is needed

performanceHow fast can we go?How fast can we go?

Status: Open.

#2567 In JuliaGPU/CUDA.jl;

· Technici4n opened on Nov 28, 2024

Improve launch latency

good first issueGood for newcomersGood for newcomers

performanceHow fast can we go?How fast can we go?

Status: Open.

#2456 In JuliaGPU/CUDA.jl;

· cibinjoseph opened on Jul 31, 2024

Replace unsafe_free! with finalize?

performanceHow fast can we go?How fast can we go?

speculativeNot sure about this one yet.Not sure about this one yet.

Status: Open.

#2444 In JuliaGPU/CUDA.jl;

· maleadt opened on Jul 12, 2024

Explore early finalization

enhancementNew feature or requestNew feature or request

performanceHow fast can we go?How fast can we go?

Status: Open.

#2443 In JuliaGPU/CUDA.jl;

· maleadt opened on Jul 12, 2024

FFT on real array with <code>plan_fft(x)</code> is 50% slower than PyTorch

help wantedExtra attention is neededExtra attention is needed

needs informationFurther information is requestedFurther information is requested

performanceHow fast can we go?How fast can we go?

Status: Open.

#2419 In JuliaGPU/CUDA.jl;

· roflmaostc opened on Jun 17, 2024

Base.stack is underperforming.

good first issueGood for newcomersGood for newcomers

performanceHow fast can we go?How fast can we go?

Status: Open.

#2248 In JuliaGPU/CUDA.jl;

· rcalxrc08 opened on Jan 21, 2024

Add stream pool to avoid cuStreamCreate overhead

enhancementNew feature or requestNew feature or request

good first issueGood for newcomersGood for newcomers

performanceHow fast can we go?How fast can we go?

Status: Open.

#2012 In JuliaGPU/CUDA.jl;

· simonbyrne opened on Jul 29, 2023

Use GemmKernels.jl for fallback mul!

enhancementNew feature or requestNew feature or request

performanceHow fast can we go?How fast can we go?

Status: Open.

#1990 In JuliaGPU/CUDA.jl;

· maleadt opened on Jul 5, 2023