Drop Vulkan throughput claim: smoke-tested only

README.md

@@ -75,11 +75,11 @@ Mainline llama.cpp does not load `clover_moe`. This build adds the converter, th
 | Linux x86_64 | NVIDIA sm_80 / 86 / 89 / 90 (A100, A10G, L40S, RTX 4090, H100/H200) | CUDA 12.0 | shipped |
 | Linux x86_64 | NVIDIA Blackwell sm_100/120, AMD, Intel | none | not in this build |
 | Windows x86_64 | NVIDIA sm_50 through sm_120 (Maxwell to Blackwell) | CUDA 12.8 | shipped |
-| Windows x86_64 | AMD and Intel GPUs | Vulkan 1.4.363 | shipped, low throughput |
+| Windows x86_64 | AMD and Intel GPUs | Vulkan 1.4.363 | shipped, smoke-tested on one Intel iGPU |
 | Windows x86_64 | Any x86-64 CPU | per-CPU DLLs, AVX-512 off | shipped |
 | macOS, ARM64, ROCm on any OS | - | none | not published |
 
-Measured with the NanoPlus GGUF in this build: 6.75 t/s on CPU, 8.51 t/s on CUDA at `-ngl 8`, 1.20 t/s on Vulkan over an Intel iGPU. The builds are meant to run the model everywhere x86-64 exists, not to be fast on every backend.
+Throughput measured with the NanoPlus GGUF in this build: 6.75 t/s on CPU, 8.51 t/s on CUDA at `-ngl 8`. Vulkan was only smoke-tested on a basic Intel iGPU to confirm it loads and generates; no throughput claim is made for AMD or Intel GPUs. Coverage here means the model runs, not that every backend is benchmarked.
 
 Windows is native. The ZIP carries its CUDA 12.8 libraries and uses the Vulkan loader from the GPU driver; extract it and run `set PATH=%~dp0bin;%PATH%` from the archive root. It ships the same tool set as the Linux archives, including `llama-speculative-simple`, `llama-quantize` and `llama-imatrix`.