Platforms

Every platform in the series.

The current processors and accelerators on the same finance workloads, all in throughput: paths or options priced per second, at every problem size.

All 6 platforms were measured in November 2026, and each has its full measurement record. Earlier results stay in the insight posts, as published.

Intel CPUs

PlatformLaunchedMachineNotes
Intel Xeon 6 (Granite Rapids)20242 sockets, 192 threads, 755 GiBBuilt with Intel oneAPI icpx 2026.2 -O3 -qopenmp -march=graniterapids -mprefer-vector-width=512, oneMKL.

AMD CPUs

PlatformLaunchedMachineNotes
AMD EPYC Turin20242 sockets, 192 threads, 375 GiBBuilt with GCC 16.2 -O3 -ffast-math -fopenmp -march=znver5 -mprefer-vector-width=512, oneMKL.

NVIDIA GPUs

PlatformLaunchedMachineNotes
NVIDIA H2002024One NVIDIA H200 GPUBuilt with CUDA 12.9.1, sm_90, cuRAND. One whole GPU, on the 96 host threads of its memory node. Default power limit, ECC on, clocks not locked.
NVIDIA B2002025One NVIDIA B200 GPUBuilt with CUDA 12.9.1, sm_100, cuRAND. One whole GPU, on the 96 host threads of its memory node. Default power limit, ECC on, clocks not locked.

AMD GPUs

PlatformLaunchedMachineNotes
AMD Instinct MI350X2025One AMD Instinct MI350X GPUBuilt with ROCm 7.2.4, gfx950, hipRAND/rocRAND. One whole GPU, unpartitioned, with 192 host threads. Power cap 1,000 W, the maker's rated board power, ECC on, clocks not locked.
AMD Instinct MI355X2025One AMD Instinct MI355X GPUBuilt with ROCm 7.2.4, gfx950, hipRAND/rocRAND. One whole GPU, unpartitioned, with 192 host threads. Power cap 1,400 W, the maker's rated board power, ECC on, clocks not locked.