---
title: "Releasing a new GPU cloud offering based on NVIDIA Tesla V100"
description: "Start on-demand GPU virtual machines with up to 4 Tesla V100. Scale up your computational capabilities by reducing your costs."
date: 2019-05-21
author: "Ste"
tags: ["gpu"]
url: https://www.exoscale.com/blog/new-tesla-v100-gpu-offering/
---
# Releasing a new GPU cloud offering based on NVIDIA Tesla V100

Today we broaden our catalog with a new [GPU instance offering](https://www.exoscale.com/gpu/#gpu2), based on the top performing [NVIDIA Tesla V100](https://www.nvidia.com/en-us/data-center/v100/) graphic cards.

![Tesla V100 cloud virtual machines on-demand](https://www.exoscale.com/blog/new-tesla-v100-gpu-offering/tesla-v100-instances.jpg)

The new GPU2 offering lets Big Data scientists and Artificial Intelligence engineers scale up their computational capabilities dramatically, implementing even bigger and more complex deep learning algorithms.

Machine learning models can now be trained in days or even hours compared to weeks, and high performance computing applications can now tackle challenges that were once thought impossible.

With its 640 Tensor Cores and 5,120 CUDA Cores the V100 is the most advanced data center GPU ever built to accelerate AI, HPC, data science and graphics, and the world's first GPU to break the 100 teraflops barrier of deep learning performance.

With up to 4 Tesla V100 graphic cards per single instance, this means you can get up to 448 teraflops of mixed precision ML performance on a single virtual machine.

Combined with a performant setup of up to 48 CPU cores, 255 GB of RAM, fast SSD local storage, and [flexible pricing plans billed by the second](https://www.exoscale.com/pricing/#/gpu), the GPU2 offering is an extremely cost-effective solution for machine learning training,
offering you the opportunity to scale out your infrastructure and optimize your costs.

Supporting most of the leading deep learning frameworks out in the market, like
TensorFlow, PyTorch, Caffe2, CNTK, Theano, Keras, and more, the new GPU2 offering is available starting today in our [Vienna AT-VIE-1 zone](http://feature-gpu-sub-description.website.kubexo.space/datacenters/austria/).

## The Most Competitive, Production-grade Cloud GPU Offering on the Market

Good news don't stop there tough. In fact, we are dropping the price of our [GPU offering based on NVIDIA Tesla P100](https://www.exoscale.com/gpu/#gpu) by 15%, starting now at 1.0620 €/hr.

With its 18.7 teraflops per card and its new base price, the Tesla P100 GPU offering provides an accessible solution to get started and experiment in the field of high performance computing.

Cloud GPU pricing is largely inconsistent, and significant differences can exist between providers. Whereas Exoscale does not provide all possible combinations of GPU instance types, its pricing is almost always lower.

**Compared to a competitor like AWS, Exoscale is today as much as 78% less expensive**, with advantages in both On-demand and Long-term use cases. The tables below speak for themselves:

<table>
    <thead>
        <tr>
            <th colspan="5">
                <h3>On-demand, Tesla V100 instances</h3>
            </th>
        </tr>
        <tr>
            <th></th>
            <th>Model</th>
            <th>Hourly-price</th>
            <th>Monthly-price</th>
            <th>Difference</th>
        </tr>
    </thead>
    <tbody>
        <tr>
            <td>AWS</td>
            <td>p3.2xlarge (1 x Tesla V100)</td>
            <td>3.823 $/hr</td>
            <td>2'752.56 $/mo</td>
            <td></td>
        </tr>
        <tr>
            <td>Exoscale</td>
            <td>gpu2-small (1 x Tesla V100)</td>
            <td>1.375 $/hr</td>
            <td>990 $/mo</td>
            <td class="highlight">-64%</td>
        </tr>
    </tbody>
    <tbody>
        <tr>
            <td>AWS</td>
            <td>p3.8xlarge (4 x Tesla V100)</td>
            <td>15.292 $/hr</td>
            <td>11'010.24 $/mo</td>
            <td></td>
        </tr>
        <tr>
            <td>Exoscale</td>
            <td>gpu2-huge (4 x Tesla V100)</td>
            <td>3.31944 $/hr</td>
            <td>2'389.99 $/mo</td>
            <td class="highlight">-78%</td>
        </tr>
    </tbody>
    <tfoot>
        <tr>
            <td colspan="5">
                <ul>
                    <li>All prices in US Dollars, VAT excluded, as reported on providers websites on the 16th of May 2019.</li>
                    <li>AWS prices are evaluated from the European <code>EU Frankfurt</code> region. Exoscale has no regional price differences.</li>
                    <li>One month equals to 720 hours ($/hr x 720).</li>
                </ul>
            </td>
        </tr>
    </tfoot>
</table>

<table>
    <thead>
        <tr>
            <th colspan="5">
                <h3>On-demand, Tesla K80 (AWS) vs Tesla P100 (Exoscale) instances</h3>
            </th>
        </tr>
        <tr>
            <th></th>
            <th>Model</th>
            <th>Hourly-price</th>
            <th>Monthly-price</th>
            <th>Difference</th>
        </tr>
    </thead>
    <tbody>
        <tr>
            <td>AWS</td>
            <td>p2.2xlarge (1 x Tesla K80)</td>
            <td>1.326 $/hr</td>
            <td>954.72 $/mo</td>
            <td></td>
        </tr>
        <tr>
            <td>Exoscale</td>
            <td>gpu-small (1 x Tesla P100)</td>
            <td>1.16806 $/hr</td>
            <td>841.00 $/mo</td>
            <td class="highlight">-12%</td>
        </tr>
    </tbody>
    <tbody>
        <tr>
            <td>AWS</td>
            <td>p3.8xlarge (4 x Tesla K80)</td>
            <td>10.608 $/hr</td>
            <td>7'637.76 $/mo</td>
            <td></td>
        </tr>
        <tr>
            <td>Exoscale</td>
            <td>gpu-huge (4 x Tesla P100)</td>
            <td>2.82083 $/hr</td>
            <td>2'030.99 $/mo</td>
            <td class="highlight">-73%</td>
        </tr>
    </tbody>
    <tfoot>
        <tr>
            <td colspan="5">
                <ul>
                    <li>All prices in US Dollars, VAT excluded, as reported on providers websites on the 16th of May 2019.</li>
                    <li>AWS prices are evaluated from the European <code>EU Frankfurt</code> region. Exoscale has no regional price differences.</li>
                    <li>One month equals to 720 hours ($/hr x 720).</li>
                </ul>
            </td>
        </tr>
    </tfoot>
</table>


<table>
    <thead>
        <tr>
            <th colspan="5">
                <h3>Long-term, Tesla V100 instances</h3>
            </th>
        </tr>
        <tr>
            <th></th>
            <th>Model</th>
            <th>Upfront billing</th>
            <th>Monthly price</th>
            <th>Difference</th>
        </tr>
    </thead>
    <tbody>
        <tr>
            <td>AWS</td>
            <td>p3.2xlarge (1 x Tesla V100)</td>
            <td>32'352 $</td>
            <td>886.32 $/mo</td>
            <td></td>
        </tr>
        <tr>
            <td>Exoscale</td>
            <td>gpu2-small (1 x Tesla V100)</td>
            <td class="highlight">None</td>
            <td>841.50 $/mo</td>
            <td class="highlight">-5%</td>
        </tr>
    </tbody>
    <tbody>
        <tr>
            <td>AWS</td>
            <td>p3.8xlarge (4 x Tesla V100)</td>
            <td>12'9410 $</td>
            <td>3'545.28 $/mo</td>
            <td></td>
        </tr>
        <tr>
            <td>Exoscale</td>
            <td>gpu2-huge (4 x Tesla V100)</td>
            <td class="highlight">None</td>
            <td>2'031.49 $/mo</td>
            <td class="highlight">-43%</td>`
        </tr>
    </tbody>
    <tfoot>
        <tr>
            <td colspan="5">
                <ul>
                    <li>All prices in US Dollars, VAT excluded, as reported on providers websites on the 16th of May 2019.</li>
                    <li>AWS prices are evaluated from the European <code>EU Frankfurt</code> region. Exoscale has no regional price differences.</li>
                    <li>
                        The AWS offering has been evaluated considering the lowest possible price:
                        <ol>
                            <li>3 years Standard Reserved Instance Pricing</li>
                            <li>Full upfront payment</li>
                            <li>AWS monthly price equals to the upfront total / 36 months</li>
                        </ol>
                    </li>
                    <li>The Exoscale offering has been evaulated considering the highest tier of our 12 months committment discount (15%).</li>
                    <li>One month equals to 720 hours ($/hr x 720).</li>
                </ul>
            </td>
        </tr>
    </tfoot>
</table>



<table>
    <thead>
        <tr>
            <th colspan="5">
                <h3>Long-term, Tesla K80 (AWS) vs Tesla P100 (Exoscale) instances</h3>
            </th>
        </tr>
        <tr>
            <th></th>
            <th>Model</th>
            <th>Upfront billing</th>
            <th>Monthly price</th>
            <th>Difference</th>
        </tr>
    </thead>
    <tbody>
        <tr>
            <td>AWS</td>
            <td>p2.2xlarge (1 x Tesla K80)</td>
            <td>15'146 $</td>
            <td>414.72 $/mo</td>
            <td></td>
        </tr>
        <tr>
            <td>Exoscale</td>
            <td>gpu-small (1 x Tesla P100)</td>
            <td class="highlight">None</td>
            <td>841.00 $/mo</td>
            <td>+3%</td>
        </tr>
    </tbody>
    <tbody>
        <tr>
            <td>AWS</td>
            <td>p3.8xlarge (4 x Tesla K80)</td>
            <td>12'1164 $</td>
            <td>3'319.92 $/mo</td>
            <td></td>
        </tr>
        <tr>
            <td>Exoscale</td>
            <td>gpu-huge (4 x Tesla P100)</td>
            <td class="highlight">None</td>
            <td>2'030.99 $/mo</td>
            <td class="highlight">-39%</td>`
        </tr>
    </tbody>
    <tfoot>
        <tr>
            <td colspan="5">
                <ul>
                    <li>All prices in US Dollars, VAT excluded, as reported on providers websites on the 16th of May 2019.</li>
                    <li>AWS prices are evaluated from the European <code>EU Frankfurt</code> region. Exoscale has no regional price differences.</li>
                    <li>
                        The AWS offering has been evaluated considering the lowest possible price:
                        <ol>
                            <li>3 years Standard Reserved Instance Pricing</li>
                            <li>Full upfront payment</li>
                            <li>AWS monthly price equals to the upfront total / 36 months</li>
                        </ol>
                    </li>
                    <li>The Exoscale offering has been evaulated considering the highest tier of our 12 months committment discount (15%).</li>
                    <li>One month equals to 720 hours ($/hr x 720).</li>
                </ul>
            </td>
        </tr>
    </tfoot>
</table>

Exoscale loses against AWS only on Long-term use-case using a `p2.2xlarge` instance, but with no upfront compared to a whopping $15'146, and a commitment to use an older K80 GPU for 3 years to come, which will be largely obsolete by the end of the term.

As it stands today, we are confident to assess that Exoscale's GPU offering is the most competitive production-grade Cloud GPU offering on the market, providing [certified and compliant resources for HPC industries](https://www.exoscale.com/compliance).