AI Deep Learning

AI Deep Learning

GPU accelerate up to 25x faster

Install a comprehensive Python and GPU powered deep learning suite. The Nvidia RTX tensor cores can accelerate deep learning tasks by up to 25x (2,500%) versus other popular developer laptops.

Please read the disclaimer before proceeding. This solution is updated regularly. Please write authorship with suggestions or requests.

Check Libraries

The Focus ships with libraries for deep-learning and gaming. However, adding repositories or packages can modify or break them. You may check and reset these libraries using a terminal as shown below.

/opt/kfocus-001/bin/fixup_run.sh 0 1 1 # If libraries do not match, run the following: /opt/kfocus-001/bin/fixup_run.sh

Run kfocus-001-conda

Use kfocus-001-conda to install an Anaconda suite. This is professionally vetted, comprehensive, and GPU accelerated. Open a terminal and enter the following.

kfocus-001-conda

Follow the prompts. CUDA, TensorRT, and cuDNN libraries are preinstalled.

Pycharm is an excellent IDE for Python. It can be installed through the Jetbrains toolbox which is preinstalled on all Focus laptops.

Tests are run at the end of kfocus-001-conda. It is non-destructive by default and can be rerun at any time. You may, for example, only install TensorFlow to start and then add PyTorch on a subsequent run. Run the AI suite at the end and compare to the results below.

Using Nvidia On-Demand

When running Deep Learning workloads you may notice that the software reports less available RAM than the GPU actually has. Part of this is due to driver overhead, but some memory is used for the actual display. One can use the Intel GPU for display purposes and reserve the Nvidia GPU for Deep Learning or other compute tasks. Start Nvidia Settings app and then select Prime Profiles > Nvidia On-Demand and then reboot. This mode does not significantly improve battery life so we recommend using Nvidia (Performance mode) or Intel (Power Saving Mode) as it suits your needs at other times.

Benchmarks

Comprehensive TensorFlow benchmarks can be run at the end of the kfocus-001-conda script. Please compare to the official results. RTX 2070 test results current 2020-09-28. Other results are estimates and will be replaced by actual values when testing is complete.

RTX 2080[1]RTX 2070[1]RTX 2060[1]Titan X[2]Titan XP W10[2]Titan XP Ubuntu[2]
Inferancce9,632(EST)8,8817,785(EST)7,4439,74111,948
Training9,766(EST)9,2958,671(EST)7,90810,37512,922
AI Score19,139(EST)18,17616,878(EST)15,35120,08924,870

[1] Kubuntu Focus 2.2Kg, 20mm thick

[2] Unconstrained Desktop GPU

Troubleshooting

Q: Starting Nvidia Visual Profiler results in an error. How can I fix this?

A: One must install Java 8 and then call nvvp with correct path.

sudo install openjdk-8-jdk nvvp -vm /usr/lib/jvm/java-8-openjdk-amd64/jre/bin/java

Revisions

This is a partial revision history. See the git repository for all entries.

Disclaimer

We try hard to provide a useful solution validated by professionals. However, we cannot anticipate every situation, and therefore cannot guarantee this procedure will work for your needs. Always backup your data and test the solution to determine the correct procedure for you.

THIS SOLUTION IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOLUTION, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

KFocus Logo