ncnn · Issues· 1210 open

YOLOv8 detection results on Mali-G720 Vulkan GPU (CPU results are correct)

feature plan

AMD CPU SDPA result error

python 版后处理例子基本上都跑不通了。。

ARM fp16 arithmetic produces all-NaN output for a ViT-tiny graph / model (ncnn 20260526)

Inference results are not deterministic due to uninitialized padding memory in Mat::create()

Vulkan compute output corrupted on Imagination/PowerVR with opt.lightmode=1 blob recycling — allocator bookkeeping proven correct by internal tracing during a corrupted inference; barriers and submit boundaries ruled out

ncnn KV cache mem opt ncnn 的 KVcache 模式优化

Android更新到20250503版本后偶现__kmp_abort_process崩溃

关于使用vulkan时的初始化时间

any case about tinypose and movenet running on the mali gpu with opencl

ncnn/examples/yolov4.cpp, yolov4->opt.use_vulkan_compute = true , detection is Error.

To use Vulkan on RK3399

在windows安卓子系统中，不能使用GPU

[RFC] vulkan on webgpu native