GPU Software Engineer (Graphics / ML)

Навыки

• C++
• CUDA / ONNX / TensorRT
• Отладка и поиск ошибок
• Jaeger / OpenTelemetry
• Machine Learning
• Python
• PyTorch

О компании и продукте

• We are looking for hybrid engineers - real-time graphics AND machine learning in one person - to join a GPU software team working at the intersection of rendering and image ML (upscaling, denoising, artifact suppression for interactive visual applications).

Задачи

• Develop and optimize rendering and ML inference components for real-time visual pipelines (DX12, Vulkan, ONNX-based stacks)
• Profile GPU workloads and tune for latency, memory and throughput
• Integrate ML models (super-resolution, denoising) into graphics pipelines
• Evaluate output quality using objective and perceptual metrics (PSNR/SSIM, LPIPS) and visual regression tooling
• Author clean, testable, reproducible code
• collaborate with graphics, ML and platform teams
• 4+ years of professional software engineering experience (C++ primary
• strong Python on the ML side)
• Solid GPU fundamentals: pipeline, synchronization and memory models, performance trade-offs
• BOTH areas below are mandatory
• A profile covering only one of them is not a match
• super-resolution, denoising, artifact suppression, segmentation or comparable CNN/transformer work
• including inference deployment and optimization on GPU (ONNX Runtime or TensorRT, quantization)

Требования

• (a) GPU / graphics programming: hands-on production experience with DX12 and/or Vulkan (or CUDA/HIP GPU compute), shader/kernel authoring (HLSL/GLSL/compute), GPU debugging and profiling (RenderDoc, PIX, Radeon GPU Profiler, Nsight)
• (b) ML on images: hands-on project experience with PyTorch (or equivalent) on vision models

Будет плюсом

• Ray tracing (DXR/VKRT), game engines (Unreal, Unity) or rendering middleware
• Color and image processing fundamentals (sRGB vs linear, HDR, resampling/filtering)
• CUDA/HIP compute experience
• Render fidelity testing, SSIM/PSNR-based visual regression tooling
• CI-driven development, automated test harnesses

Паспорт вакансии

История публикации

Откуда что взялось

Отмечено то, что вывели мы. Без пометки — значение назвал работодатель.

Почему на этом месте в выдаче

Порядок выдачи объявлен контрактом: свежесть решает между днями, полнота и зарплата — внутри дня.