# Flux model on CPU runs slowly
**URL:** https://discourse.julialang.org/t/flux-model-on-cpu-runs-slowly/47723
**Category:** Performance
**Tags:** question, flux
**Created:** [October 4, 2020, 2:22am UTC](https://discourse.julialang.org/t/flux-model-on-cpu-runs-slowly/47723 "2020-10-04T02:22:41Z")
**Posts on this page:** 4
**Page:** 1
### Author:  [@AquaIndigo](https://discourse.julialang.org/u/AquaIndigo)
#### Post date: [October 4, 2020, 2:22am UTC](https://discourse.julialang.org/t/flux-model-on-cpu-runs-slowly/47723/1 "2020-10-04T02:22:41Z")
I find the `Flux` model on CPU runs more slowly than that on GPU:
```julia
julia> m = Chain(
Dense(2250, 500, σ),
Dense(500, 50, tanh),
Dense(50, 7, σ));
julia> X = Float32.(X);
julia> size(X)
(2250, 484)
julia> @btime m(X);
10.718 ms (10 allocations: 2.06 MiB)
julia> X_gpu = X |> gpu;
julia> m_gpu = m |> gpu;
julia> @btime m_gpu(X_gpu);
21.864 μs (134 allocations: 3.80 KiB)
```
And training the model on CPU was untolarably slow, and I am really confused.
---
### Author:  [@ChrisRackauckas](https://discourse.julialang.org/u/ChrisRackauckas)
#### Post date: [October 4, 2020, 2:36am UTC](https://discourse.julialang.org/t/flux-model-on-cpu-runs-slowly/47723/2 "2020-10-04T02:36:45Z")
That’s a really big matmul for a GPU 🤷♂️. It should be about the same cost on any CPU implementation since its all going to be in a BLAS kernel.
---
### Author:  [@AquaIndigo](https://discourse.julialang.org/u/AquaIndigo)
#### Post date: [October 4, 2020, 2:52am UTC](https://discourse.julialang.org/t/flux-model-on-cpu-runs-slowly/47723/3 "2020-10-04T02:52:02Z")
And when I trained the same Pytorch model on CPU, it seemed much faster than the Flux model on CPU.
---
### Author:  [@Oscar\_Smith](https://discourse.julialang.org/u/Oscar_Smith)
#### Post date: [October 4, 2020, 2:54am UTC](https://discourse.julialang.org/t/flux-model-on-cpu-runs-slowly/47723/4 "2020-10-04T02:54:10Z")
Julia’s tanh is fairly slow, but the bottleneck really should be the matmul