byteLAKE team has been delivering solutions on Xilinx Alveo Data Center accelerator cards ever since they were officially announced (fall/winter 2018). During summer 2019, we made our first official announcement with Xilinx and stamped the partnership.
byteLAKE experts are available to advise about how to start your Alveo story as well as to help you adapt and optimize complex algorithms to these cards.
Our Alveo story started with a CFD (Computational Fluid Dynamics) project, an area we’d been working on for many years before. We are not your typical CFD user or adopter. Our take on the subject is more from the algorithmics perspective and its adaptation or optimization (if you like) to various, heterogeneous HPC configurations. We have been adapting and optimizing complex numerical algorithms to both: NVIDIA GPUs and Xilinx Alveo FPGAs. Projects like these, our partners, and clients eventually inspired us to invest in building a very innovative product, namely CFD Suite. CFD Suite leverages AI (Artificial Intelligence) to deliver new possibilities into the CFD world. A new chapter for our company and I warmly invite you to follow our efforts in the field by reading my other blog series started here: www.byteLAKE.com/en/AI4CFD-pt1. Also, if you find this topic interesting (or intriguing), join our AI4CFD community and participate in the exciting discussions we’re having there. I am personally looking forward to meeting you there as well! We’re building a community of partners and researchers striving to offer new value to fluid dynamics through AI.
With all the experiences we’ve gathered throughout the years, in both R&D as well as R&I projects, we decided to address a growing demand for Alveo expertise by creating our uniquely crafted: byteLAKE’s Alveo Solutions.
byteLAKE’s Alveo Solutions consists of 2 key elements:
- Alveo Products Marketplace: a collection of byteLAKE’s Alveo optimized solutions. Marketplace is available here: bytelake.accelize.com and through this portal you can easily deploy our products as well as gain access to Alveo-powered infrastructure. The first published accelerator is Alveo optimized MPDATA/Advection algorithm. It is a core algorithm for workloads like weather simulations, flood modeling, ocean modeling, etc. Technical details here. AI-enabled solutions will be gradually added.
- Expertise Software Services: our experts are available to help you adapt and optimize your workloads to Alveo-powered architectures. We know that projects like these are never about straight forward source code porting. Algorithms redesigning, mathematical formulas optimizations, data, and computations reorganization to make the most of the hardware are what we do almost daily. Therefore we have all it takes to deliver as well as advise about building efficient solutions with Alveo. Visit our website at www.byteLAKE.com/en/Alveo to learn more.
On top, and together with our partners, we also provide complete HPC infrastructures as well as instant VPN or Cloud access to Alveo accelerator cards. Simply reach out to us at: Alveo (at) byteLAKE.com to learn more. We want to be a partner to handle all your Alveo related topics.
Complex Algorithms Adaptation to Alveo
We have been adapting algorithms from Fortran-based desktop implementations to fully scalable, MPI-enabled C++ & accelerator optimized versions. From single nodes to clusters with Piz Daint being our largest deployment supercomputer. We showcased our solutions at Xilinx Developer Forum (XDF), Supercomputing Conference (SC) as well as other prestigious events. We are also very active in the relevant AI & HPC research projects, publishing for journals like Concurrency and Computation: Practice and Experience, Parallel Computing, Journal of Supercomputing. Beyond adaptation, we leverage AI to accelerate and enable new possibilities in HPC applications.
Solutions for the Financial sector (FinTech, Cryptocurrencies, etc.)
Working in the CFD space helped us jump on a fast lane into the financial sector and FinTech in specific. Programming techniques that are common in fluid dynamics are very similar to those used for instance in trading algorithms. Therefore our expertise in CFD algorithms design, optimization techniques, adaptation to FPGA/GPU skills turned out to be very helpful in the projects for the finance sector as well.
Alveo cards offer speedup but above all, they are incredibly efficient when it comes to energy consumption. And this became of huge importance for all crypto mining initiatives. We have started a couple of activities in the space. The first one is for a company from the cryptocurrency industry. The goal is to significantly lower the energy consumption for mining algorithms and eventually reduce the total cost of ownership. One of the algorithms we are working on is Ethash (Ethereum) but there are more which in many cases require a complete redesign to benefit from accelerator cards. Another area is obviously cryptocurrency trading. And as for a company specializing in AI, it could not be any other way than through offering an AI-enabled alternative for i.e. Monte Carlo based trading. In our case we are using Machine Learning models and Long Short-Term Memory, as well as Neural Network with Hidden Layers. We will present the results during the next edition of the ML Prague conference as communicated in one of our blog posts here: bytelake.com/en/mlprague2020
As reported by Xilinx, Alibaba uses Xilinx FPGAs and Xilinx Vitis AI development kit to deliver hundreds of pictures per-second, achieving a 3.5X performance improvement over initial GPU implementation and 75% savings in the total cost of ownership. While you can read more about this case study here: https://bit.ly/2V3Kd79, it is just one example that shows how outstanding results byteLAKE can help you deliver and in turn differentiate in the e-commerce market as well.
MedTech, AI for Medical Technologies
Here we offer a wide spectrum of solutions ranging from image analytics, through sound analytics, to algorithms optimization and adaptation to compact, portable Edge Devices. As a company, we have been working in the Computer Vision area for many years. One of our products, Ewa Guard is about collecting the data, working with domain experts to properly process it, extract features, and tag them to eventually prepare a dataset for our highly optimized Deep Neural Networks. Another project which follows a very similar path is about 2D and 3D graphics analytics and using AI for segmentation purposes. A solution that is being integrated into our client’s innovative dental diagnostic tools. Besides vision, we have also been working on a medical device prototype design for sound analytics. Again, the project is about new algorithm design and adaptation, edge device prototyping, data collection & processing, etc. Unfortunately, these 2 projects are under strict NDA but both are for healthcare industry companies.
There is a rapid increase in sensitive data, such as biomedical records or ﬁnancial. Protecting such data while in transit as well as while at rest is crucial. Consequently, suitable techniques and tools, including cryptographic ones, should be applied to provide such key security properties as conﬁdentiality, integrity, and authentication. As a company, we have spent a number of years mapping cryptographic algorithms i.e. symmetric-key cryptography (AES) and hash functions (i.e. Whirlpool hash, SHA), onto various architectures. Our researchers pioneered data processing parallelization (research paper published), an approach that allows us to exploit eﬃciently the computing resources of all processing units.
Our focus in the Automotive space is currently on leveraging Artificial Intelligence across designing, projecting, and manufacturing. We’re starting by proposing a machine and deep learning model to accelerate aerodynamics and fluid dynamics simulations. Our researchers are involved in AI-powered optimization of various aerodynamics characteristics, leading to building a model that will eventually land in our CFDSuite product.
Depending on the type of workload, Alveo certainly offers interesting acceleration numbers. As a company with many years of experience in the field, we can advise and help you roughly estimate the potential gains that Alveo can offer in your particular scenario. What is different when comparing Alveo with competitive accelerators though, is extremely low energy budget that these cards require to deliver results. Therefore if Green Computing is on your agenda, you should definitely consider Alveo cards. As mentioned, our experts are available to advise about how to start your Alveo story as well as to help you adapt and optimize complex algorithms to these cards. To learn more about Alveo applications across industries, visit this website: xilinx.com/applications.
We can advise and help you roughly estimate the potential gains that Alveo can offer in your particular scenario.
So, if you’re looking for Alveo expertise (FPGA), seek no more. Reach out to us at: Alveo (at) byteLAKE.com and let’s talk!