Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Low Performance on i9-9900K processor #419

Open
lsyang1111 opened this issue Sep 9, 2019 · 2 comments
Open

Low Performance on i9-9900K processor #419

lsyang1111 opened this issue Sep 9, 2019 · 2 comments

Comments

@lsyang1111
Copy link

Hi
I'm running TensorFlow benchmark on a single Motherboard Pro WS C246-ACE(Processor: i9-i9900K, DRAM: 64GB), just trying out the performance without using GPUs. But the performance is quick low. 2.5~3.5 image/sec. Even after 20000 epoch of training, the max top-1/top-5 accuracy is about 0.031/0.094. Learning_rate is increasing at a same speed, from 0 to 1.7e-3(still increasing while training).
I wonder is this the expected performance or I put the wrong parameters.

tf.version=1.15.0-dev20190821
No MKL is used.

Below is my script:
python tf_cnn_benchmarks.py --num_gpus=1 --batch_size=32 --model=resnet50_v1.5 --variable_update=parameter_server --device=cpu --data_format=NHWC --data_name=imagenet --data_dir=F:\ILSVRC2017_CLS-LOC --save_summaries_steps=10 --train_dir=C:\Users\C246_0710\Anaconda3\envs\tensorflow1\benchmarks-master\scripts\tf_cnn_benchmarks\training --summary_verbosity=1 --print_training_accuracy --benchmark_log_dir=C:\Users\C246_0710\Anaconda3\envs\tensorflow1\benchmarks-master\scripts\tf_cnn_benchmarks\training --save_model_steps=1000

@JMadgwick
Copy link

Running the same command and version of TF (1.15.0) I get this with a Ryzen 3700X:

Running warm up
Done warm up
Step	Img/sec	total_loss
1	images/sec: 9.2 +/- 0.0 (jitter = 0.0)	7.791
10	images/sec: 9.3 +/- 0.0 (jitter = 0.1)	7.952

Unless there is something wrong with the data your CPU seems to be underperforming.

@MatPoliquin
Copy link

Did you try on Linux distro such as Ubuntu?
From experience I found performance is much better on Ubuntu (18.04) than Windows 10 for these type of tasks

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

3 participants