I'm using Tensorflow 2.5 to train a starGAN network for generating images (128x128 jpeg). I am using tf.keras.preprocessing.image_dataset_from_directory to load the images from the subfolders.
Additionally I am using arguments to maximize loading performance as suggested in various posts and threads such as loadedDataset.cache().repeat.prefetch
I'm also using the num_parallel_calls=tf.data.AUTOTUNE for the mapping functions for post-processing the images after loading.
While training the network on GPU the performance I am getting for GPU Utilization is in the picture attached below.
My question regarding this are:
Is the GPU utlization normal or is it not supposed to be so erratic for traning GANs?
Is there any way to make this performance more consistent?
Is there any way to improve the training performance to fully utlize the GPU?
Note that Ive logged my disk I/O also and there is no bottleneck reading/writing from the disk (nvme ssd).
The system has 32GB RAM and a RTX3070 with 8GB Vram. I have tried the running it on colab also; but the performance was similarly erratic.
It is fairly normal for utilization to be erratic like for any kind of parallelized software, including training GANs. Of course, it would be better if you could fully utilize your GPU, but writing software that does this is challenging and becomes virtually impossible when you are talking about complex applications like GANs.
Let me try to demonstrate with a trivial example. Say you have two threads, threadA and threadB. threadA is running the following python code:
x = some_time_comsuming_task()
y = get_y_from_threadB()
print(x+y)
Here threadA is performing lots of calculations to get the value for x, retrieving the value for y, and printing out the sum of x+y. Imagine threadB is also doing some kind of time consuming calculation to generate the value for y. Unless threadA is ready to retrieve y at the exact same time threadB finishes calculating it, you won't have 100% utilization of both threads for the entire duration of the program. And this is just two threads, when you have 100s of threads working together with multiple chained data dependencies, you can see how it becomes exponentially more difficult to eliminate any and all time threads spend waiting on other threads to deliver input to the next step of computation.
Trying to make your "performance more consistent" is pointless. Whether your GPU utilization went up and down (like in the graph you shared) or it stayed exactly at the average utilization for the entire execution would not change the overall execution time, which is probably the actually important metric here. Utilization is mostly useful to identify where you can optimize your code.
Fully utilize? Probably not. As explained in my answer to question one, it's going to be virtually impossible to orchestrate your GAN to completely remove bottlenecks. I would encourage you to try and improve execution time, rather than utilization, when optimizing your GAN. There's no magic setting that you're missing that will completely unlock all of your GPU's potential.
Related
Just a heads up: I'm fairly new to the worlds of both machine learning and parallel computing. I'll do my best to use the right terminology, but friendly corrections would be much appreciated.
I've been trying to tune hyperparameters for a Keras MLP (running on top of TensorFlow) by using sklearn's RandomizedSearchCV, wrapping my model in KerasClassifier as per numerous tutorials. I've been trying to parallelize this process via RandomizedSearchCV's built-in system for this. This was working more or less fine, but then when I started using around 6 threads the program would still run, but I would start getting NaN losses, which would lead to errors after the search had concluded.
Now, I know there are a bunch of usual suspects for gradient blow-ups, but there was something interesting here: this problem went away when I reduced pre_dispatch sufficiently (in this case down to n_jobs instead of 2*n_jobs).
Is there any reason why this would happen? My datasets are pretty large, but it seems that once the searching begins, each job is only using about 1-2% of available memory (also, this percent does not change when I cut the pre-dispatched jobs?), and I'm not getting any other memory use issues.
Secondly, on a more practical note, is there any easy way to get my code to break in RandomSearchCV as soon as an NaN loss comes up? It would save some time as opposed to letting it keep running until the end of the search--which, when fully implemented, will probably take a day or more--and then throw the error. I was also thinking of changing error_score to -1 or something. Would this actually be better? I think it is probably worth knowing that certain combinations of hyperparameters lead to gradient blow-ups, but not if its only because of this parallelization issue.
Is there a difference between the parallelization that takes place between these two options? I’m assuming num_workers is solely concerned with the parallelizing the data loading. But is setting torch.set_num_threads for training in general? Trying to understand the difference between these options. Thanks!
The num_workers for the DataLoader specifies how many parallel workers to use to load the data and run all the transformations. If you are loading large images or have expensive transformations then you can be in situation where GPU is fast to process your data and your DataLoader is too slow to continuously feed the GPU. In that case setting higher number of workers helps. I typically increase this number until my epoch step is fast enough. Also, a side tip: if you are using docker, usually you want to set shm to 1X to 2X number of workers in GB for large dataset like ImageNet.
The torch.set_num_threads specifies how many threads to use for parallelizing CPU-bound tensor operations. If you are using GPU for most of your tensor operations then this setting doesn't matter too much. However, if you have tensors that you keep on cpu and you are doing lot of operations on them then you might benefit from setting this. Pytorch docs, unfortunately, don't specify which operations will benefit from this so see your CPU utilization and adjust this number until you can max it out.
I am running a simple deep learning model on Google's colab, but it's running slower than my MacBook Air with no GPU.
I read this question and found out it's a problem because of dataset importing over the internet, but I am unable to figure out how to speed up this process.
My model can be found here. Any idea of how I can make the epoch faster?
My local machine takes 0.5-0.6 seconds per epoch and google-colabs takes 3-4 seconds
Is GPU always faster than CPU? No, why? because the speed optimization by a GPU depends on a few factors,
How much part of your code runs/executes in parallel, i.e how much part of your code creates threads that run parallel, this is automatically taken care by Keras and should not be a problem in your scenario.
Time Spent sending the data between CPU and GPU, this is where many times people falter, it is assumed that GPU will always outperform CPU, but if data being passed is too small, the time it takes to perform the computation (No of computation steps required) are lesser than breaking the data/processes into thread, executing them in GPU and then recombining them back again on the CPU.
The second scenario looks probable in your case since you have used a batch_size of 5.
classifier=KerasClassifier(build_fn=build_classifier,epochs=100,batch_size=5), If your dataset is big enough, Increasing the batch_size will increase the performance of GPU over CPU.
Other than that you have used a fairly simple model and as #igrinis pointed out that data is loaded only once from drive to memory so the problem in all theory should not be loading time because the data is on drive.
i have the problem to understand where tensorflow executes the variables. In my case, i have a large word embedding matrix which is used by a RNN to generate text. During the text generation process, there is a lookup of the embeddings and i know that this lookup needs to be executed on CPU because GPU doesn't support it.
I want to expand my system so that there are calculations with the large embedding matrix, but this operation is very slow. I think this will also be executed on cpu, although a calculation on GPU is possible. When i loop into the tool GPU-Z during the calculation, i can see that the most time the bus interface load is very high (>60%) and the GPU load is very low (<10%).
It is very difficult to post a minimal example, i hope the problem is clear. I don't know how to debug this. Are the embedding automatically placed on the cpu because of the lookup operation? Do you have any idea how to overcome this problem?
I am reading this performance guide on the best practices for optimizing TensorFlow code for GPU. One suggestion they have is to place the preprocessing operations on the CPU so that the GPU is dedicated for training. To try to understand how one would actually implement this within an experiment (ie. learn_runner.run()). To further the discussion, I'd like to consider the best way to apply this strategy to the Custom Estimator Census Sample provided here.
The article suggests placing with tf.device('/cpu:0') around the preprocessing operations. However, when I look at the custom estimator the 'preprocessing' appears to be done in multiple steps:
Line 152/153 inputs = tf.feature_column.input_layer(features, transformed_columns) & label_values = tf.constant(LABELS) -- if I wrapped with tf.device('/cpu:0') around these two lines would that be sufficient to cover the 'preprocessing' in this example?
Line 282/294 - There is also a generate_input_fn and parse_csv function that are used to set up input data queues. Would it be necessary to place with tf.device('/cpu:0') within these functions as well or would that basically be forced by having the inputs & label_values already wrapped?
Main Question: Which of the above implementation suggestions is sufficient to properly place all preprocessing on the CPU?
Some additional questions that aren't addressed in the post:
What if the machine has multiple cores? Would 'cpu:0' be limiting?
The post implies to me that by wrapping the preprocessing on the cpu, the GPU would be automatically used for the rest. Is that actually the case?
Distributed ML Engine Experiment
As a follow up, I would like to understand how this can be further adapted in a distributed ML engine experiment - would any of the recommendations above need to change if there were say 2 worker GPUs, 1 master CPU and a parameter server? My understanding is that the distributed training would be data-parallel asynchronous training so that each worker will be independently iterating through the data (and passing gradients asynchronously back to the PS) which suggests to me that no further modifications from the single GPU above would be needed if you train in this way. However, this seems a bit to easy to be true.
MAIN QUESTION:
The 2 codes your placed actually are 2 different parts of the training, Line 282/294 in my options is so called "pre-processing" part, for it's parse raw input data into Tensors, this operations not suitable for GPU accelerating, so it will be sufficient if allocated on CPU.
Line 152/152 is part of the training model for it's processing the raw feature into different type of features.
'cpu:0' means the operations of this section will be allocated on CPU, but not bind to specified core. The operations allocated on CPU will run in multi-threads and use multi-cores.
If your running machine has GPUs, the TensorFlow will prefer allocating the operations on GPUs if the device is not specified.
The previous answer accurately describes device placement. Allow me to provide an answer to the questions about distributed TF.
The first thing to note is that, whenever possible, prefer a single machine with lots of GPUs to multiple machines with single GPUs. The bandwidth to parameters in RAM on the same machine (or even better, on the GPUs themselves) is orders of magnitude faster than going over the network.
That said, there are times where you'll want distributed training, including remote parameter servers. In that case, you would not necessarily need to change anything in your code from the single machine setup.