Accelerated Parallel Computing
with NVIDIA Tesla and GPU Compute
Peak delivers the highest possible compute performance into the hands of developers, scientists, and engineers to advance computing enabled discovery and solution of the world's most challenging computational problems.
Puget Systems has over 20 years experience designing and building high quality and high performance PCs. Our emphasis has always been on reliability, high performance, and quiet operation. We take this experience to the HPC sector with our Peak family of workstations and servers. Through in-house testing we do not blindly follow the industry -- we help lead it. We provide the products below as starting points that we feel cover some of the most compelling areas that we can contribute to the HPC community. Do you have a project that needs some serious compute power, and you don't know where to turn? Let us help, it's what we do!
Minimum noise and maximum performance, reliability and usability. Puget Peak is an evolutionary step from our custom systems experience. Genesis performance post-production, Summit server stability, Serenity silent design, Obsidian reliability and even the diminutive Echo have influenced Peak.
TeraFLOPS. Using Intel Xeon CPU's and the Intel MKL library, or the well established CUDA platform and libraries, there is tremendous potential for applications leveraging the computing power of both the CPU and the GPU.
Ready for use. Peak systems are installed, configured and tested under load before they ship and will (optionally) arrive with the setup and tools you need to get started. Our CentOS setup will provide a configuration that can be the basis of your working environment.
Part of what makes our cooling both effective and quiet is that we specifically target the hot spots of each system. We place fans only where they are needed and only when they are needed. We then verify the final configuration with extensive testing, full load stress testing, and thermal imaging to ensure excellent cooling.
We know that these PCs are intended for heavy, long duration workloads. We have designed them for long life with 24/7 load, and that is our primary design goal. Through targeted cooling and high quality thermal solutions, we are able to achieve an excellent low noise level while maintaining the cooling necessary for long term high load. Even better, since we are implementing a custom cooling plan for each order, if you have a preference of whether you'd like us to tune more aggressively in either direction (towards even quieter operation, or more extreme cooling), all you have to do is let us know!
The second new NVIDIA RTX30 series card, the GeForce RTX3090 has been released. The RTX3090 is loaded with 24GB of memory making it a good replacement for the RTX Titan... at significantly less cost! The performance for Machine Learning and Molecular Dynamics on the RTX3090 is quite good, as expected.
The much anticipated NVIDIA GeForce RTX3080 has been released. How good is it with TensorFlow for machine learning? How about molecular dynamics with NAMD? I've got some preliminary numbers for you!
WSL2 offers improved performance over version 1 by providing more direct access to the host hardware drivers. Recent "Insider Dev Channel" builds of Win10 even allows access to the Windows NVIDIA display driver for GPU computing applications for WSL2 Linux applications! The performance improvements with WSL2 are largely because this version is running as a privileged virtual machine on to of MS Hyper-V. This means that at least low level support for the Hyper-V virtualization layer needs to be enabled to use it. In particular, the Windows feature "VirtualMachinePlatform" must be enabled for WSL2. We tested to see if there was any negative application performance impact.
The current JupyterHub version 2.5.1 does not allow user installed extension for JupyterLab when it is being served from JupyterHub. This should be remedied in version 3. However, even when this is "fixed" it is still useful to be able to install extensions globally for all users on a multi-user system. This note will show you how.
WSL on Windows 10 does not (currently) provide a direct way to copy a Linux distribution that was installed from the "Microsoft Store". The following guide will show you a way to make a working copy of an installed distribution with a new name.
In this note I'll go through creating self-signed SSL certificates and adding them to a JupyterHub configuration running on a LAN or VPN. This will allow encrypted access to the server using https in a browser.
This is a quick note about setting up a JupyterHub server and JupyterLab using conda with Anaconda Python.
On March 19, 2020 I did a webinar titled, "AMD Threadripper 3rd Gen HPC Parallel Performance and Scaling ++(Xeon 3265W and EPYC 7742)" The "++(Xeon 3265W and EPYC 7742)" part of that title was added after we had scheduled the webinar. It made the presentation a lot more interesting than the original Threadripper only title! This is a follow up post with the charts and plots of testing results presented in that webinar.
I am the Scientific Computing adviser here at Puget Systems. Many of us are sharing our work-from-home stories for coping with the changes to our normal life and work routines during the COVID-19 quarantine. I have already been working from home for a few years and wanted to share my setup and offer some tips that I hope will encourage you if you are suddenly finding yourself bewildered by disruption to your work-life.
Is 32-cores enough? I had some testing time again on an AMD Threadripper 32-core 3970x and thought it would be interesting to compare that to the 64-core 3990x. In this post I take a comparative look at parallel performance and scaling for HPL Linpack, Python numpy and the NAMD molecular dynamics program.