NVIDIA V100
Studies measured on NVIDIA V100-SXM2 32GB GPUs (Volta, SM70), the hardware expXa is tuned for.
- No change
expXa vs PXA 4 on Qwen3.6-35B-A3B Q6_K
Matched dual-V100 comparison on Qwen3.6-35B-A3B Q6_K: both engines with and without MTP, 512–65,536 input tokens, every run and correctness checks.
- Adopted
expXa vs PXA 4 on Qwen3-30B-A3B Q4_K_M
Matched V100 comparison of expXa and PXA 4.0.1 on Qwen3-30B-A3B Q4_K_M with 40 CPU expert layers: per-run results, method, and how expXa went from 49 s to 24 s.
- Not adopted
Warp sharing in expert and attention kernels
Testing warp-level sharing of expert IDs, weight fragments and attention scalars in expXa's Volta kernels on two V100s. No change earned adoption.
- Not adopted
Reordering expert work for weight reuse
Grouping MoE expert jobs by expert and traversing row bands on V100: about 4% less operator time at 255 active jobs, slower for small batches and the paired kernel.
- No change
Dense models across three engines
Qwen3.5-9B Q4_K_M and Qwen3.5-0.8B Q8_0 on custom PXA 3.1, PXA 4.0.1 and expXa: 54 matched requests on V100, with correctness and cancellation checks.
Other topics: Attention AVX-512 VNNI CPU kernels CUDA kernels MoE CPU offload PXA comparison Q4_K_M Q6_K