NVIDIA RTX 5880 Ada 48GB

The RTX 5880 Ada Generation is the cut-down AD102 workstation card NVIDIA introduced for markets where the RTX 6000 Ada cannot be sold. It keeps the things that make the 6000 useful, 48 GB of ECC GDDR6 on a 384-bit bus at 960 GB/s in a 285 W dual-slot card, and trims the compute: 14,080 CUDA cores and 440 fourth-generation Tensor Cores against the 6000's 18,176 and 568, for 69.3 TFLOPS FP32 and 554.2 TFLOPS of dense FP8 tensor throughput. That works out to roughly three-quarters of the RTX 6000 Ada's arithmetic at the same memory capacity and bandwidth, which is the trade that matters for local inference, where a model either fits in the frame buffer or it does not. It has no NVLink, like every Ada card, so it scales out over PCIe Gen 4 rather than pairing. NVIDIA publishes only two throughput figures for it, single-precision and a sparse FP8 tensor number, so the rows below are deliberately short.