jcjohnson / neural-style

Torch implementation of neural style algorithm
MIT License
18.31k stars 2.7k forks source link

What is the max resolution possible and best settings for g2.2xlarge and g2.8xlarge spot requests? #220

Open ProGamerGov opened 8 years ago

ProGamerGov commented 8 years ago

https://aws.amazon.com/ec2/instance-types/

g2.2xlarge 1 8 15 1 x 60

g2.8xlarge 4 32 60 2 x 120

ProGamerGov commented 8 years ago

The answer is 700px for VGG and 1200px for NIN.

monik3r commented 8 years ago

We would need to know what kind of vram there is in each box for this, along with the details of drivers/cublas version

ProGamerGov commented 8 years ago

For NIN with Cudnn, autotune cudnn, and Optimizer ADAM, you can get up to 2500px.

th neural_style.lua -style_image style.jpg -content_image content.jpg -output_image out.png -tv_weight 0.0001 -image_size 2500 -save_iter 50 -content_weight 10 -style_weight 1000 -num_iterations 1000 -model_file models/nin_imagenet_conv.caffemodel -proto_file models/train_val.prototxt -content_layers relu0,relu1,relu3,relu5,relu7,relu9,relu12 -style_layers relu0,relu1,relu3,relu5,relu7,relu9,relu12 -backend cudnn -cudnn_autotune -optimizer adam

ProGamerGov commented 8 years ago

g2.2xlarge:


+------------------------------------------------------+
| NVIDIA-SMI 346.46     Driver Version: 346.46         |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  GRID K520           Off  | 0000:00:03.0     Off |                  N/A |
| N/A   48C    P0    35W / 125W |     10MiB /  4095MiB |      0%      Default |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID  Type  Process name                               Usage      |
|=============================================================================|
|  No running processes found                                                 |
+-----------------------------------------------------------------------------+

ubuntu@ip-172-31-6-102:~$ free -m
             total       used       free     shared    buffers     cached
Mem:         15039       1517      13522          0         23       1094
-/+ buffers/cache:        399      14640
Swap:            0          0          0
ekg commented 8 years ago

@ProGamerGov Do you have an ec2 machine image of your setup available to share? I would be super grateful if you do!

ProGamerGov commented 8 years ago

@ekg Search for: ami-1dbd2c6e on the EU (Ireland) region. I made it public. Once you have it, save it yourself so you have a copy in case I mess the ami up again in the future. You can also move it to another region if you save it as your own ami.

It has 3 models, Places205, NIN, and VGG-19.

If you have any questions just ask.

dxmachina commented 8 years ago

@ProGamerGov is your image still available?