The quantum industry has assumed enterprise-grade GPUs are mandatory for quantum error correction. We measured it. 👇 With an FPGA controller submitting a 128-bit payload over RDMA across roughly a million samples: 2.3 µs mean end-to-end to a stock AMD Ryzen Threadripper PRO CPU, 4.4 µs to an AMD Instinct MI210 GPU. The commodity CPU cleared the co-processing budget with room to spare. For workloads that don't require heavy parallel GPU processing, that takes the GPU off the critical path — no accelerator allocation to wait on, and no specialized system mandating shared CPU/GPU memory. We built Backline for photonic timing budgets, one of the tightest of any modality. Infrastructure built for photonics means over-specified for many others. Learn more: 📷 ↧