GPU Engineer

We are looking

for an experienced GPU Programming Engineer to develop, optimize, and deploy

GPU-accelerated solutions for high-performance deep learning workloads. The

role focuses on GPU programming, parallel computing, performance tuning, and

deep learning system integration across platforms such as NVIDIA CUDA, AMD

ROCm/HIP, and OpenCL.

Responsibilities:

• Develop, optimize, and maintain GPU-accelerated

components for deep learning pipelines using CUDA, HIP, or OpenCL.

• Analyze and improve GPU kernel performance

through profiling, benchmarking, and resource optimization.

• Optimize memory access, compute, throughput, and

kernel execution to improve overall system performance on target GPUs.

• Port existing CPU-based implementations to GPU

platforms while ensuring correctness and performance scalability.

• Work closely with system architects, software

engineers, and domain experts to integrate GPU-accelerated solutions.

Requirements

Education:

Bachelor's or master's degree in Computer Science, Electrical Engineering, or a

related field.

Technical Skills (Must haves):

• 2+ years of hands-on experience in GPU

programming, preferably using CUDA, HIP, OpenCL, or similar GPU APIs.

• Strong understanding of GPU architecture, memory

hierarchy, shared memory, bank conflicts, and parallel programming models.

• Proficiency in C/C++ and hands-on experience

developing on Linux-based systems.

• Familiarity with profiling and tuning tools such

as Nsight, rocprof, or Perfetto.

Need to have (Can be bridged):

• Knowledge and experience in SIMD programming.
• Good understanding of neural network (NN)

operators and hands-on experience with PyTorch, TensorFlow, or TensorRT.

Good to have (Not essential):

• Exposure to deep learning concepts such as

quantization and pruning.

• Experience working with High Performance

Computing (HPC) systems.

Preferred Qualifications (Optional):

Additional

experience in GPU performance engineering, deep learning system optimization,

or heterogeneous computing would be advantageous.