Facing the highest value business problems, no one wants to wait for weeks to finish the work, as the speed of computations may directly impact time to market. When a few weeks later may mean a few weeks too late, you want to speed things up no matter what, which means scale, or so you were told… In practice, one may realise that not all computational problems are natural candidates for parallelism across multiple compute nodes. Even in that case, significant speed-up may be possible using powerful server GPUs. In either case, TechnoLynx is here to help! Skillful R&D Software Engineers at TechnoLynx can speed up your High-Performance Computing problems by order of magnitudes by researching novel algorithmic solutions and developing highly optimised, parallelised GPU code. Our team can target heterogeneous architectures from our partner Intel’s latest Scalable Xeon series (including CPU, GPU and even FPGA “cores”) using OpenCL as well as the most potent Tesla A100 and H100 GPUs from NVIDIA using CUDA. Our AWS-certified team can deliver said novelties to our clients in the shape of Docker containerised solutions, using skeletons of high-performance C++ code with cross-platform CMake build systems, minimal dependencies, and the use of OpenCL or CUDA for further GPU acceleration, targeting our partner AWS’s EC2 compute instances. For expanded scalability over several compute nodes, we can also utilise OpenMPI.
To validate or review our previous works in this field, feel free to check our recent projects:
Read more about Accelerated Physics Simulation using GPUs.
Read more about Accelerating Crypto-Currency Mining.
Our hard work has been reflected in the excellent reviews we have received from our clients on multiple platforms, as well as in being chosen the “Software Consultancy of the Year” by the Corporate Live Wire Awards/Magazine.
They have a seniority beyond their age. The fact they run low-powered PCs to maximise code efficiency speaks plenty about their dedication to their trade and originality in an age where owning a Macbook is confused with IT literacy.” Alex Farrant - Founder @ Farrant Consulting
Read our Clutch reviews
Our Founder, Balázs Keszthelyi, started his career as a real-time graphics programmer and has retained his enthusiasm for GPUs ever since. Having worked on multiple complex GPU projects like SPH fluid simulation, AES acceleration, and path tracing, he ended up helping Broadcom in Cambridge develop their own GPU IP, the VideoCore V, which is now in use in Raspberry PI. During that tenure, he represented Broadcom at the Khronos standard body on matters concerning OpenCL and became named contributor to the SYCL standard. He later leads his V-Nova team to the delivery of the first GPU-compute-friendly video codec, now known as the SMPTE ST 2117-1 VC-6 standard. Read more about GPU programming on our Medium.com blog!
It is often possible for practical simulational problems like Simulation of Wave Propagation (e.g. light) and other physical issues like Computational Fluid Dynamics (CFD) to use approximate solutions instead of exact, and TechnoLynx can help you research such models either via analytical means or via the use of surrogate models. Once derived, we can help optimise said model’s implementation close to metal in a scalable manner.
Both for Drug Discovery and Genomics, the speed and scale of processing are of crucial importance as they may ultimately determine the time to market and the efficiency to convert a unit of computational resources to business value, which could quickly become what sets apart competitors from each other.
Problems like Risk Analysis, Stock Price Prediction, and Sentiment Analysis are all in need of significant computational resources, whether it be because of the need for real-time processing of massive streams for High-Frequency Trading (HFT) or just the sheer amount of computing power needed for the training of large language models and other NLP applications.