C++ Systems, Infrastructure, BizOps at Thunder Compute (YC S24)
Core
Building a userspace virtualization layer that decouples GPUs from physical machines to create a cluster-wide, schedulable GPU resource pool for AI workloads.
Role type
Senior IC C++ systems engineer (low-latency infrastructure)
Builds
A high-performance GPU virtualization shim that intercepts CUDA calls and manages GPU detachment/reattachment over the network.
Domain
Cloud infrastructure, GPU virtualization, AI/ML compute
Deliverable
production ML models | infrastructure
Required skills
C++, low-latency systems programming, CUDA, network programming, resource scheduling
Preferred skills
Quantitative development, HFT background, experience with high-frequency trading systems
Technologies
CUDA, LD_PRELOAD, C++
Responsibilities
Develop and optimize the low-level virtualization layer to ensure near-native performance for AI workloads; implement mechanisms for rapid GPU detachment and reassignment; manage network communication for GPU pooling and scheduling.
Seniority
Senior, hands-on IC
