SONY

Cloud Pricing

Deep learning is now available anywhere and anytime, with rich amount of resources on the cloud.
From the rich computing resources such as GPU, you can choose your proper resource according to your purpose.
Usage fee of CPU/GPU will be only charged for the number of seconds you will train or evaluate. While training and evaluation is not executed, CPU/GPU usage fee will not be charged and the cost can be kept to a minimum.
For payment, you can use a credit card and you will be charged on a monthly basis.

Pricing (excluding tax)

Compute Hours
CPU
  • CPU×185yen / hour
GPU Example of gpu usage fee
  • Standard
  • NVIDIA® TESLA® K80 GPUx1210yen / hour
  • NVIDIA® TESLA® V100 GPUx1560yen / hour
  • NVIDIA® TESLA® V100 GPUx42,900yen / hour
  • NVIDIA® TESLA® V100 GPUx85,800yen / hour
  • NVIDIA® TESLA® V100 GPU+x87,500yen / hour
  • ABCI
  • G.small300yen / hour
  • G.large1,650yen / hour
Additional Workspace
200 yen / 10GB per Month

* Usage of CPU and GPU is billed by the second.
* The amount of the largest workspace size selected within the month is charged.
* Regarding compute hours, usage fee of CPU/GPU will be only charged for the number of seconds when CPU/GPU is run, and there is no usage fee for CPU/GPU when training or evaluation is not executing, for instance, editing a network.
* There are no charges other than those listed in the above table, for instance, data transfer cost of uploading your dataset.
* See the FAQ for more detailed specification on Standard.
* See the ABCI Web page for more detailed specification on ABCI(AI Bridging Cloud Infrastructure).

Free Usage

You can get started with Neural Network Console Cloud for free within the following free usage limit.

  • Compute Hours CPU : 10 hours
  • Workspace : 10GB

Estimated usage fee for learning

In the Neural Network Console cloud version, the fee varies depending on the type of CPU / GPU used during learning.
An example of using two types of networks with different scales for CPUs and two types of GPUs will be introduced as an estimate of usage fees.

* Sample 1
Dataset: MNIST
Network: LeNet (from sample project)
epoch: 10
Environment
CPU
NVIDIA® TESLA® K80 GPU
NVIDIA® TESLA® V100 GPU
Learning execution time
903 seconds
105 seconds
107 seconds
Usage fee per hour
85 yen
210 yen
560 yen
Estimated usage fee
About 21 yen
About 6 yen
About 17 yen

In a small network like above, a higher-performance GPU (NVIDIA® TESLA® V100 GPU) cannot be worked efficiently, and a lower-performance NVIDIA® TESLA® K80 GPU is faster and costs less. In this way, GPU processing performance and usage fees may be reversed depending on the network or data set.

* Sample 2
Dataset: CIFAR 10
Network: ResNet-101
epoch: 300
Environment
CPU
NVIDIA® TESLA® K80 GPU
NVIDIA® TESLA® V100 GPU
Learning execution time
1,209,600 seconds
(336 hours)
14,976 seconds
(4.16 hours)
3,960 seconds
(1.1 hours)
Usage fee per hour
85 yen
210 yen
560 yen
Estimated usage fee
About 28,560 yen
About 874 yen
About 616 yen

In networks above, GPU performance differences manifest in processing time. NVIDIA® TESLA® V100 GPUs have faster processing times and lower prices.
The Neural Network Console cloud version can use multiple GPUs (multi-GPU) to further accelerate learning.

Payment Method

For payment, you can use your credit card.

  • VISA
  • mastercard
  • JCB
  • AMERICAN EXPRESS
  • Diners Club INTERNATIONAL

For corporations, we also have a menu that supports bill payment.
Click here for details.

The Cloud version does not support Internet Explorer.
Please open with Google Chrome or download the Windows version in here.

Sorry, the cloud is temporarily out of service.
Please try again later.

Which account do you sign in with?

Select an account to use with Neural Network Console.
The same function is available with both Google and Sony accounts.