OpenCL/CUDA Developer

RequirementsThis is a full time, in-house contractor job. We need a fully dedicated resouce, you may not have side jobs.Key QualificationsCandidates must have a strong technical background and be capable of coming up to speed on new technologies quickly University degree (BSc, MSc, PhD) in a technical area (Computer Science, Electrical Engineering or related) 5+ years of experience in software design and development Proficiency in CUDA/OpenCL (3+ years) Strong GPU kernel design experience using OpenCL, CUDA Experience performing timing analysis, power analysis and benchmarking GPU kernels and applications Strong debugging skills Good command of English Collaborative, open-minded, team player attitude Hands-on experience in Linux Good communication and problem-solving skills and the ability to work both individually and collaboratively in a team environment are required Job DescriptionWhat you’ll be working onThe OpenCL/CUDA Developer is responsible for: Designing and testing complex, high speed, GPU designs implementing DSP algorithms for high speed network routing and encryption applications. Translating requirements into GPU architectures. Implementing and documenting GPU designs. Supporting systems integration and testing. Performing timing analysis, power analysis and benchmarking GPU kernels and applications. Experience with C/C++, bash, Rust, etc. Bachelor's degree in electrical engineering, computer engineering, or closely related field.ResponsibilitiesWork directly with the CTO to develop strategies for restructuring high divergence algorithms into vectorizable code. Creating hypothetical performance reports, developing PoC code to test real world performance, and then using experimentation to understand the cause of any significant discrepancy Porting existing kernels to target different platforms (NVidia, AMD, Intel, ARM Mali) Implementing RDMA logic for direct GPU/GPU, GPU/NIC, and GPU/SSD communication Creating development environments and testing infrastructure to allow for scalable GPU codebases