Neanderthal vs. ND4J – Native Performance, Java and CPU
dragan.rocks
dragan.rocks
Deeplearning4j is written around F ordering and ND4J supports this. Admittedly, our ordering API is not obvious to the average user.
Here's an example test you can run yourself that demonstrates ordering: https://gist.github.com/raver119/92b615704ca1bf169aa23a6a6e7...
o.n.i.TensorFlowImportTest - Orders: CCC; Time: 11532 ns;
o.n.i.TensorFlowImportTest - Orders: CCF; Time: 2101 ns;
o.n.i.TensorFlowImportTest - Orders: CFC; Time: 10202 ns;
o.n.i.TensorFlowImportTest - Orders: CFF; Time: 1960 ns;
o.n.i.TensorFlowImportTest - Orders: FCC; Time: 10744 ns;
o.n.i.TensorFlowImportTest - Orders: FCF; Time: 1717 ns;
o.n.i.TensorFlowImportTest - Orders: FFC; Time: 10097 ns;
o.n.i.TensorFlowImportTest - Orders: FFF; Time: 1716 ns;
1. This is the benchmark that you provided, so while this might be not obvious to the average user, it also seems to not be obvious to the above average user that wrote the benchmark. I was just using what you proposed, assuming that you used the right thing in your library.
2. It still does not explain how you got better performance with ND4J with the same non-optimal call, which was what started the discussion, and inspired this post.
3. Neanderthal supports both Row and Column oriented order with cuBLAS with the same performance, and won't have those problems that you mention for ND4J.
I'm, of course, interested in following up on this. Please decide on what cases you'd like to compare, post the (optimal) code, and the ND4J and Neanderthal numbers that you get, and I'll respond with my comments.
https://github.com/treo/benchmarking_nd4j/blob/master/src/ma...
The difference is rather huge with the newer versions of nd4j.
While the numbers in the following gists do not contain the measurements I took for neanderthal, they do contain the numbers that I got for ND4J.
Without f ordering: https://gist.github.com/treo/1fab39f213da26255cf4f75e383ff90...
With f ordering: https://gist.github.com/treo/94fe92c9417b5c8b24baa12924a35b0...
As you can see something happened in the time between the 0.4 release (I took that as the comparison point since that was when I ran my own benchmarks the last time) and the 0.9.1 release that introduced additional overhead.
Originally I planned to create my own write-up on this, but I wanted to first to find out what happened there.
Given that ND4J is mainly used inside of DL4J and the matrix sizes it is used with usually are rather large, the performance overhead difference that I've observed there for tiny multiplications isn't necessarily that bad, as the newer version performs much better on larger matrices.
Although, to my defense, the option in question here is very poorly documented. I've found the ND4J tutorial page where it's mentioned, and even after re-reading the sentence multiple times, I still do not connect its description to what it (seems to) actually do. It also does not mention that it affects computation speed.
Anyway, I'm looking forward to reading your detailed analysis, and especially seeing your Neanderthal numbers.
When I originally set out to compare ND4J and Neanderthal, I've ran into the issue that I bottomed out at: they basically both call MKL (or Openblas) for BLAS operations.
We will be sending out a doc for this by next week with these updates. Thanks a lot for playing ball here.
Beyond that, can you clarify what you mean? Do you mean just the gemm op?
For that, that's the only case that mattered for us. We will be documenting the what/how/why of this in our docs.
Beyond that, I'm not convinced the libraries are directly comparable when it comes to the sheer scope of the libraries to each other.
You're treating nd4j as a gemm library rather than a fully fledged numpy/tensorflow with hundreds of ops and support for things you would likely have no interest in building.
A big reason I built nd4j was to solve the general use case of building a tensor library for deep learning, not just a gemm library.
Beyond that - I'll give you props for what you built. There's always lessons to learn when comparing libraries and making sure the numbers match.
Our target isn't you though, it's the likes of google,facebook, and co and tackling the scope of tasks they are.
That being said - could we spend some time on docs? Heck yeah we should. At most we have java doc and examples. We tend to help people as much as we can when profiling.
Could we manage it better? Yes for sure. That's partially why we moved dl4j to the eclipse foundation to get more 3rd party contributions and build a better governance setup. Will it take time for all of this to evolve? Oh yeah most definitely.
No project is perfect and always has things it could improve on.
Anyways - let's be clear here. You're a one man shop who built an amazingly fast library that scratches your own itch for a very specific set of use cases. We're a company and community tackling a wider breadth of tasks and trying to focus more on serving customers and adding odd things like different kinds of serialization, spark interop,.. etc.
We benefit from doing these comparisons and it forces us to document things better that we normally don't pay attention to. This little exercise is good for us. As mentioned, we will document the limitations a bit better but we will make sure to cover other topics like allocation and the like as well as the blas interface.
Positive change has come out of this and I'd like to thank you for the work you put in. We will make sure to re run some of the comaprisons on our side.
Please also note that Neanderthal also has hundreds of operations. The set of use cases where it scratches itches might be wider and more general than you think.
The reasons I'm showcasing matrix multiplications are:
1. That's what you used in the comparison. 2. It is a good proxy for the overall performance. If matrix multiplication is poor, other operations tend to be even poorer :)
Anyway, as I said, I'll be glad to compare other operations that ND4J excells at, or that anyone think are important.
I would also like to see ND4J's comparisons with Tensorflow or Numpy, or PyTorch, or, JVM based MXNet.
That being said, while gemm is one op, it's a lot more than just jni back and forth that use other libraries. What matters here are also things like convolutions, pair wise distance calculations, element wise ops, etc.
There's nuance there.
There are multiple layers here to consider:
1. The JNI interop managed via javacpp (relevant to this discussion)
2. Every op has allocation vs in place trade offs to consider
3. For our python interface, we have yet another layer to benchmark there (we use pyjnius for jumpy the python interface for nd4j)
4. Op implementations for the cuda kernels and the custom cpu ops we wrote. (That's where our avx512 and avx2 jars matter for example)
For the subset we are comparing against, it's basically making sure we wrap the blas calls properly. That's definitely something we should be doing.
We've profiled that and chose the pattern you're seeing above with f ordering.
That is where we are fast and chose to optimize for. You are faster in those other cases and have laid that out very well.
Again, there's still a lot that was learned here and I will post the doc when we get it out there to make that less painful next time.
You made a great post here and really laid out the trade offs.
I wish we had more time to run benchmarks beyond timing for our own use cases, if we had smaller scope we would definitely focus on every case you're mentioning here. We likely will revisit this at some point if we find it worth it.
In general, our communications and docs can always be improved (especially our internals like our memory allocation)
Re: your last point we do do this kind of benchmarking with tensorflow. For example: https://www.slideshare.net/agibsonccc/deploying-signature-ve... (see slide 3 and also the broader slides for an idea of how we profile deep learning for apps using the jvm)
We need to do a better job of maintaining these things though. We don't keep it up to date and don't profile as much as we should. It has diminishing returns after a certain point vs building other features.
I'm hoping a CI build to generate these things is something we get done this year so we can both prevent performance regressions and have consistent numbers we can publish for the docs.
Once the python interface is done that will be easier to do and justify since most of our "competition" is in python.
Benchmarking ND4J and Neanderthal (Scientific Computing in Java and Clojure)
https://www.dubs.tech/blog/benchmarking-nd4j-and-neanderthal...
https://github.com/treo/benchmarking_nd4j/tree/master/src/ma...
deeplearning4j.org/native http://deeplearning4j.org/workspaces
We have our own garbage collection as well as native config and off heap memory management.
We use this
https://github.com/deeplearning4j/deeplearning4j/blob/master...
to specify use of MKL.
The source is in the text, but conveniently accessible here: https://github.com/uncomplicate/neanderthal/blob/master/exam...
ND4J returns the `mmuli` result in F order, and DL$J is aware of that. We design our algorithms around this. It is possible to get the `mmuli` result in C order, but it'll cause an additional conversion operation call, which is expensive. That's exactly what Dragan did. So on one hand - he's right: in the CCC case, Neanderthal is faster, because he doesn't have to do the F->C conversion of the result.
We make mmuli always return F because of Cuda. The cuBLAS implementation has no option for C ordered output.
`mmuli` CCC always introduces a dup for us. We expect the gemm result to always be F, so if result is not F, each operation allocates tempResult array, F ordered, and does result.assign(tempResult).
Thanks for the explanation. I used the code that ND4J guys provided, assuming that they'd do the right thing. Also read my response to crockpotveggies: provide the ND4J code that you think is optimal, and ND4J and Neanderthal results on your machine, and I'll be happy to write a follow up.
transpose is the only option.
cublasStatus_t cublasSgemm(cublasHandle_t handle, cublasOperation_t transa, cublasOperation_t transb, int m, int n, int k, const float alpha, const float A, int lda, const float B, int ldb, const float beta, float *C, int ldc)
For completeness sake, could you also provide some information about your machine specs and operating system (and if on linux your glibc and kernel version)?
https://neanderthal.uncomplicate.org/articles/getting_starte...
And just to be sure, choose option 2 under that heading (add the folder with appropriate dlls to PATH)
This seems a little unfair :). I’m pretty sure that if you’re using NumPy linked against MKL, it would be exactly as fast as Neanderthal running on MKL.
Matrix multiplication benchmarks themselves just aren’t that interesting when all they are doing is testing an underlying library.