# CUDA.jl: Unexpected \`mapreduce\` error: threads per block exceed GPU limit (640 \> 512
**URL:** https://discourse.julialang.org/t/cuda-jl-unexpected-mapreduce-error-threads-per-block-exceed-gpu-limit-640-512/131778
**Category:** GPU
**Tags:** question
**Created:** [August 22, 2025, 3:11pm UTC](https://discourse.julialang.org/t/cuda-jl-unexpected-mapreduce-error-threads-per-block-exceed-gpu-limit-640-512/131778 "2025-08-22T15:11:00Z")
**Posts on this page:** 10
**Page:** 1
### Author:  [@ykkan](https://discourse.julialang.org/u/ykkan)
#### Post date: [August 22, 2025, 3:11pm UTC](https://discourse.julialang.org/t/cuda-jl-unexpected-mapreduce-error-threads-per-block-exceed-gpu-limit-640-512/131778/1 "2025-08-22T15:11:00Z")
Hello everyone,
I am using `mapreduce` from CUDA.jl to compute the sum of function evaluations over an array of `SVector{7,T}` data.
The code looks roughly like this:
```julia
function func(x1, x2, x3, x4, x5, x6, x7, x8, x9)
# very complicated calculation
out1 = ...
out2 = ...
out3 = ...
return SVector{3, Float64}(out1, out2, out3)
end
data = CuArray(zeros(SVector{7, Float64}, 2048))
mapreduce(vec -> func(2.0, 3.0, vec...), +, data)
```
However, I get the following error:
`ERROR: LoadError: Number of threads per block exceeds kernel limit (640 > 512).`
I am running this on our lab-maintained cluster. Interestingly, the same solver worked perfectly about a month ago, but now it suddenly fails with this error.
Does anyone know what might be causing this issue? Thanks.
---
### Author:  [@jling](https://discourse.julialang.org/u/jling)
#### Post date: [August 22, 2025, 6:29pm UTC](https://discourse.julialang.org/t/cuda-jl-unexpected-mapreduce-error-threads-per-block-exceed-gpu-limit-640-512/131778/2 "2025-08-22T18:29:12Z")
are you on CUDA 13.0? maybe downgrade it?
---
### Author:  [@jipolanco](https://discourse.julialang.org/u/jipolanco)
#### Post date: [August 25, 2025, 8:11am UTC](https://discourse.julialang.org/t/cuda-jl-unexpected-mapreduce-error-threads-per-block-exceed-gpu-limit-640-512/131778/3 "2025-08-25T08:11:36Z")
Without being an expert, I’d guess your kernel uses too many GPU registers, which limits the maximum number of threads per block.
A quick suggestion would be to replace your `mapreduce` call with AcceleratedKernel’s [`mapreduce`](https://juliagpu.github.io/AcceleratedKernels.jl/stable/api/mapreduce/), which allows you to tune the number of threads per block (via the `block_size` parameter).
---
### Author:  [@maleadt](https://discourse.julialang.org/u/maleadt)
#### Post date: [September 1, 2025, 11:14am UTC](https://discourse.julialang.org/t/cuda-jl-unexpected-mapreduce-error-threads-per-block-exceed-gpu-limit-640-512/131778/4 "2025-09-01T11:14:05Z")
This looks like [Invalid kernel config generated by `mapreducedim!` with `SubArray` input and output · Issue #2863 · JuliaGPU/CUDA.jl · GitHub](https://github.com/JuliaGPU/CUDA.jl/issues/2863). I’ll take a look.
---
### Author:  [@ykkan](https://discourse.julialang.org/u/ykkan)
#### Post date: [September 5, 2025, 2:27pm UTC](https://discourse.julialang.org/t/cuda-jl-unexpected-mapreduce-error-threads-per-block-exceed-gpu-limit-640-512/131778/5 "2025-09-05T14:27:07Z")
I am using CUDA 12.4.
---
### Author:  [@ykkan](https://discourse.julialang.org/u/ykkan)
#### Post date: [September 5, 2025, 2:30pm UTC](https://discourse.julialang.org/t/cuda-jl-unexpected-mapreduce-error-threads-per-block-exceed-gpu-limit-640-512/131778/6 "2025-09-05T14:30:44Z")
That’s what I suspected as well. In the end, I ended up writing my own kernel function. Thanks for pointing out _AcceleratedKernel_. I wasn’t aware of it, and I’ll definitely take a look.
---
### Author:  [@ykkan](https://discourse.julialang.org/u/ykkan)
#### Post date: [September 5, 2025, 5:28pm UTC](https://discourse.julialang.org/t/cuda-jl-unexpected-mapreduce-error-threads-per-block-exceed-gpu-limit-640-512/131778/7 "2025-09-05T17:28:11Z")
Thank you for taking a look. I saw the new commit. With the latest version (v5.9.0), `mapreduce` works for my case. However, I noticed it is significantly slower than before.
I also tested my own kernel, which uses `CUDA.reduce_block` at its core. The elapsed times were **780 μs** on v5.8.3 and **19.5 ms** on v5.9.0. Is this slowdown expected?
---
### Author:  [@maleadt](https://discourse.julialang.org/u/maleadt)
#### Post date: [September 9, 2025, 10:54am UTC](https://discourse.julialang.org/t/cuda-jl-unexpected-mapreduce-error-threads-per-block-exceed-gpu-limit-640-512/131778/8 "2025-09-09T10:54:10Z")
Some slowdown was expected, but that’s much too large. Can you open an issue?
---
### Author:  [@ykkan](https://discourse.julialang.org/u/ykkan)
#### Post date: [September 16, 2025, 11:57am UTC](https://discourse.julialang.org/t/cuda-jl-unexpected-mapreduce-error-threads-per-block-exceed-gpu-limit-640-512/131778/9 "2025-09-16T11:57:51Z")
Sorry for the late follow-up.
You suggested I open an issue last week, but it looks like this was already resolved at (if i understand it correct)
[mapreduce: reinstate and fix block optimization (#2880)](https://github.com/JuliaGPU/CUDA.jl/commit/c929405e57acc1a31ed005958403f342185d4273).
I re-ran my script on different versions, and the elapsed times are now much closer:
**780 μs** on v5.8.3 and **1.089 ms** on v5.9.0.
Do you still recommend that I open an issue, or can we consider this closed?
---
### Author:  [@maleadt](https://discourse.julialang.org/u/maleadt)
#### Post date: [September 18, 2025, 11:27am UTC](https://discourse.julialang.org/t/cuda-jl-unexpected-mapreduce-error-threads-per-block-exceed-gpu-limit-640-512/131778/10 "2025-09-18T11:27:07Z")
A 25% regression is still problematic, but this may be a measuring artifact, as our CI benchmarks indicate no issue. If you have the time, please take a look with NSight Compute and if that does indeed show an issue (e.g. a suboptimal launch configuration), please file an issue.