diff --git a/ggml/src/ggml-vulkan.cpp b/ggml/src/ggml-vulkan.cpp
index c677a2728..00ad13bb9 100644
--- a/ggml/src/ggml-vulkan.cpp
+++ b/ggml/src/ggml-vulkan.cpp
@@ -1164,11 +1164,11 @@ static void ggml_vk_load_shaders(vk_device& device) {
     // mulmat
     std::initializer_list<uint32_t> warptile_l = { 128, 128, 128, 16, device->subgroup_size * 2, 64, 2, 4, 4, device->subgroup_size };
     std::initializer_list<uint32_t> warptile_m = { 128,  64,  64, 16, device->subgroup_size, 32, 2, 4, 2, device->subgroup_size };
-    std::initializer_list<uint32_t> warptile_s = { device->subgroup_size,  32,  32, 16, 32, 32, 2, 2, 2, device->subgroup_size };
+    std::initializer_list<uint32_t> warptile_s = { std::max(device->subgroup_size, 16u),  32,  32, 16, 32, 32, 2, 2, 2, device->subgroup_size };
 
     std::initializer_list<uint32_t> warptile_mmq_l = { 128, 128, 128, 32, device->subgroup_size * 2, 64, 2, 4, 4, device->subgroup_size };
     std::initializer_list<uint32_t> warptile_mmq_m = { 128,  64,  64, 32, device->subgroup_size, 32, 2, 4, 2, device->subgroup_size };
-    std::initializer_list<uint32_t> warptile_mmq_s = { device->subgroup_size,  32,  32, 32, 32, 32, 2, 2, 2, device->subgroup_size };
+    std::initializer_list<uint32_t> warptile_mmq_s = { std::max(device->subgroup_size, 16u),  32,  32, 32, 32, 32, 2, 2, 2, device->subgroup_size };
 
     std::array<uint32_t, 3> l_wg_denoms = {128, 128, 1 };
     std::array<uint32_t, 3> m_wg_denoms = { 64,  64, 1 };