This role may have closed.
Listings older than 30 days are deprioritized in search. The link to the source is still below in case the posting is still open. For active openings, see similar jobs.
FriendliAI
Software Engineer – GPU Kernel
Seoul, Seoul, South KoreaPosted 3 months agofulltime
What you'd do
- Design,implement,and optimize GPU kernels for AI inference tasks like GEMM and attention
- Develop and maintain CUDA and C++ code including low-level assembly when required
- Implement reduced-precision and quantized kernels (FP8/FP4) for low-latency inference
What they want
- 3+ years experience in GPU programming,HPC,or performance-critical systems
- Strong proficiency in CUDA for NVIDIA GPUs or ROCm/HIP for AMD GPUs
- Deep understanding of GPU architecture: warps,threads,memory hierarchy,and synchronization
Nice to have
- Experience optimizing transformer or Mixture-of-Experts architectures at the kernel level
- Familiarity with GPU libraries like CUTLASS or Triton
- Open-source contributions to GPU performance or ML acceleration projects
Seoulstart's read
- Visa tier (likely)
- E-7 specialty / professional (likely)
- Korean language
- Korean not required
- English friendliness
- High: JD in English, foreign-friendly signals