diff --git a/src/freedreno/ir3/ir3_validate.c b/src/freedreno/ir3/ir3_validate.c
index 810280aae2d..f6cc9ba159d 100644
--- a/src/freedreno/ir3/ir3_validate.c
+++ b/src/freedreno/ir3/ir3_validate.c
@@ -170,11 +170,21 @@ validate_instr(struct ir3_validate_ctx *ctx, struct ir3_instruction *instr)
 		case OPC_STL:
 		case OPC_STP:
 		case OPC_STLW:
-		case OPC_STIB:
 			validate_assert(ctx, !(instr->regs[1]->flags & IR3_REG_HALF));
 			validate_reg_size(ctx, instr->regs[2], instr->cat6.type);
 			validate_assert(ctx, !(instr->regs[3]->flags & IR3_REG_HALF));
 			break;
+		case OPC_STIB:
+			if (instr->flags & IR3_INSTR_B) {
+				validate_assert(ctx, !(instr->regs[1]->flags & IR3_REG_HALF));
+				validate_assert(ctx, !(instr->regs[2]->flags & IR3_REG_HALF));
+				validate_reg_size(ctx, instr->regs[3], instr->cat6.type);
+			} else {
+				validate_assert(ctx, !(instr->regs[1]->flags & IR3_REG_HALF));
+				validate_reg_size(ctx, instr->regs[2], instr->cat6.type);
+				validate_assert(ctx, !(instr->regs[3]->flags & IR3_REG_HALF));
+			}
+			break;
 		default:
 			validate_reg_size(ctx, instr->regs[0], instr->cat6.type);
 			validate_assert(ctx, !(instr->regs[1]->flags & IR3_REG_HALF));
diff --git a/src/freedreno/ir3/tests/disasm.c b/src/freedreno/ir3/tests/disasm.c
index e4dccdb9171..bd4bff47d46 100644
--- a/src/freedreno/ir3/tests/disasm.c
+++ b/src/freedreno/ir3/tests/disasm.c
@@ -143,6 +143,7 @@ static const struct test {
 
 	INSTR_5XX(c6e60000_00010600, "ldgb.untyped.4d.u32.1 r0.x, g[0], r1.x, r0.x"),
 	INSTR_5XX(d7660204_02000a01, "(sy)stib.typed.2d.u32.1 g[1], r0.x, r0.z, r1.x"),
+	INSTR_6XX(c0240402_00674100, "stib.b.untyped.1d.u16.1.imm.base0 r0.z, r0.x, 2"),
 
 	// TODO is this a real instruction?  Or float -6.0 ?
 	// INSTR_6XX(c0c00000_00000000, "stg.f16 g[hr0.x], hr0.x, hr0.x", .parse_fail=true),
diff --git a/src/freedreno/vulkan/tu_cmd_buffer.c b/src/freedreno/vulkan/tu_cmd_buffer.c
index 6e3a70d12d5..0de7c7dfd33 100644
--- a/src/freedreno/vulkan/tu_cmd_buffer.c
+++ b/src/freedreno/vulkan/tu_cmd_buffer.c
@@ -1812,7 +1812,7 @@ void tu_CmdPushDescriptorSetKHR(VkCommandBuffer commandBuffer,
    set->mapped_ptr = set_mem.map;
    set->va = set_mem.iova;
 
-   tu_update_descriptor_sets(tu_descriptor_set_to_handle(set),
+   tu_update_descriptor_sets(cmd->device, tu_descriptor_set_to_handle(set),
                              descriptorWriteCount, pDescriptorWrites, 0, NULL);
 
    tu_CmdBindDescriptorSets(commandBuffer, pipelineBindPoint, _layout, _set,
@@ -1851,7 +1851,7 @@ void tu_CmdPushDescriptorSetWithTemplateKHR(
    set->mapped_ptr = set_mem.map;
    set->va = set_mem.iova;
 
-   tu_update_descriptor_set_with_template(set, descriptorUpdateTemplate, pData);
+   tu_update_descriptor_set_with_template(cmd->device, set, descriptorUpdateTemplate, pData);
 
    tu_CmdBindDescriptorSets(commandBuffer, templ->bind_point, _layout, _set,
                             1, (VkDescriptorSet[]) { tu_descriptor_set_to_handle(set) },
diff --git a/src/freedreno/vulkan/tu_descriptor_set.c b/src/freedreno/vulkan/tu_descriptor_set.c
index 1bf28a9e445..e906cb2c95e 100644
--- a/src/freedreno/vulkan/tu_descriptor_set.c
+++ b/src/freedreno/vulkan/tu_descriptor_set.c
@@ -706,17 +706,23 @@ static uint32_t get_range(struct tu_buffer *buf, VkDeviceSize offset,
 }
 
 static void
-write_buffer_descriptor(uint32_t *dst, const VkDescriptorBufferInfo *buffer_info)
+write_buffer_descriptor(const struct tu_device *device,
+                        uint32_t *dst,
+                        const VkDescriptorBufferInfo *buffer_info)
 {
    TU_FROM_HANDLE(tu_buffer, buffer, buffer_info->buffer);
 
    assert((buffer_info->offset & 63) == 0); /* minStorageBufferOffsetAlignment */
    uint64_t va = tu_buffer_iova(buffer) + buffer_info->offset;
    uint32_t range = get_range(buffer, buffer_info->offset, buffer_info->range);
-   range = ALIGN_POT(range, 4) / 4;
-   dst[0] =
-      A6XX_IBO_0_TILE_MODE(TILE6_LINEAR) | A6XX_IBO_0_FMT(FMT6_32_UINT);
-   dst[1] = range;
+   /* newer a6xx allows using 16-bit descriptor for both 16-bit and 32-bit access */
+   if (device->physical_device->gpu_id >= 650) {
+      dst[0] = A6XX_IBO_0_TILE_MODE(TILE6_LINEAR) | A6XX_IBO_0_FMT(FMT6_16_UINT);
+      dst[1] = DIV_ROUND_UP(range, 2);
+   } else {
+      dst[0] = A6XX_IBO_0_TILE_MODE(TILE6_LINEAR) | A6XX_IBO_0_FMT(FMT6_32_UINT);
+      dst[1] = DIV_ROUND_UP(range, 4);
+   }
    dst[2] =
       A6XX_IBO_2_UNK4 | A6XX_IBO_2_TYPE(A6XX_TEX_1D) | A6XX_IBO_2_UNK31;
    dst[3] = 0;
@@ -784,7 +790,8 @@ write_sampler_push(uint32_t *dst, const struct tu_sampler *sampler)
 }
 
 void
-tu_update_descriptor_sets(VkDescriptorSet dstSetOverride,
+tu_update_descriptor_sets(const struct tu_device *device,
+                          VkDescriptorSet dstSetOverride,
                           uint32_t descriptorWriteCount,
                           const VkWriteDescriptorSet *pDescriptorWrites,
                           uint32_t descriptorCopyCount,
@@ -823,12 +830,12 @@ tu_update_descriptor_sets(VkDescriptorSet dstSetOverride,
             assert(!(set->layout->flags & VK_DESCRIPTOR_SET_LAYOUT_CREATE_PUSH_DESCRIPTOR_BIT_KHR));
             unsigned idx = writeset->dstArrayElement + j;
             idx += binding_layout->dynamic_offset_offset;
-            write_buffer_descriptor(set->dynamic_descriptors + A6XX_TEX_CONST_DWORDS * idx,
+            write_buffer_descriptor(device, set->dynamic_descriptors + A6XX_TEX_CONST_DWORDS * idx,
                                     writeset->pBufferInfo + j);
             break;
          }
          case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER:
-            write_buffer_descriptor(ptr, writeset->pBufferInfo + j);
+            write_buffer_descriptor(device, ptr, writeset->pBufferInfo + j);
             break;
          case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER:
          case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER:
@@ -915,7 +922,8 @@ tu_UpdateDescriptorSets(VkDevice _device,
                         uint32_t descriptorCopyCount,
                         const VkCopyDescriptorSet *pDescriptorCopies)
 {
-   tu_update_descriptor_sets(VK_NULL_HANDLE,
+   TU_FROM_HANDLE(tu_device, device, _device);
+   tu_update_descriptor_sets(device, VK_NULL_HANDLE,
                              descriptorWriteCount, pDescriptorWrites,
                              descriptorCopyCount, pDescriptorCopies);
 }
@@ -1023,6 +1031,7 @@ tu_DestroyDescriptorUpdateTemplate(
 
 void
 tu_update_descriptor_set_with_template(
+   const struct tu_device *device,
    struct tu_descriptor_set *set,
    VkDescriptorUpdateTemplate descriptorUpdateTemplate,
    const void *pData)
@@ -1049,11 +1058,11 @@ tu_update_descriptor_set_with_template(
             break;
          case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC: {
             assert(!(set->layout->flags & VK_DESCRIPTOR_SET_LAYOUT_CREATE_PUSH_DESCRIPTOR_BIT_KHR));
-            write_buffer_descriptor(set->dynamic_descriptors + dst_offset, src);
+            write_buffer_descriptor(device, set->dynamic_descriptors + dst_offset, src);
             break;
          }
          case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER:
-            write_buffer_descriptor(ptr, src);
+            write_buffer_descriptor(device, ptr, src);
             break;
          case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER:
          case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER:
@@ -1099,9 +1108,10 @@ tu_UpdateDescriptorSetWithTemplate(
    VkDescriptorUpdateTemplate descriptorUpdateTemplate,
    const void *pData)
 {
+   TU_FROM_HANDLE(tu_device, device, _device);
    TU_FROM_HANDLE(tu_descriptor_set, set, descriptorSet);
 
-   tu_update_descriptor_set_with_template(set, descriptorUpdateTemplate, pData);
+   tu_update_descriptor_set_with_template(device, set, descriptorUpdateTemplate, pData);
 }
 
 VkResult
diff --git a/src/freedreno/vulkan/tu_device.c b/src/freedreno/vulkan/tu_device.c
index 2ce0e268874..5fe163be27d 100644
--- a/src/freedreno/vulkan/tu_device.c
+++ b/src/freedreno/vulkan/tu_device.c
@@ -340,6 +340,8 @@ void
 tu_GetPhysicalDeviceFeatures2(VkPhysicalDevice physicalDevice,
                               VkPhysicalDeviceFeatures2 *pFeatures)
 {
+   TU_FROM_HANDLE(tu_physical_device, pdevice, physicalDevice);
+
    pFeatures->features = (VkPhysicalDeviceFeatures) {
       .robustBufferAccess = true,
       .fullDrawIndexUint32 = true,
@@ -393,7 +395,7 @@ tu_GetPhysicalDeviceFeatures2(VkPhysicalDevice physicalDevice,
       switch (ext->sType) {
       case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_1_FEATURES: {
          VkPhysicalDeviceVulkan11Features *features = (void *) ext;
-         features->storageBuffer16BitAccess            = false;
+         features->storageBuffer16BitAccess            = pdevice->gpu_id >= 650;
          features->uniformAndStorageBuffer16BitAccess  = false;
          features->storagePushConstant16               = false;
          features->storageInputOutput16                = false;
@@ -489,7 +491,7 @@ tu_GetPhysicalDeviceFeatures2(VkPhysicalDevice physicalDevice,
       case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_16BIT_STORAGE_FEATURES: {
          VkPhysicalDevice16BitStorageFeatures *features =
             (VkPhysicalDevice16BitStorageFeatures *) ext;
-         features->storageBuffer16BitAccess = false;
+         features->storageBuffer16BitAccess = pdevice->gpu_id >= 650;
          features->uniformAndStorageBuffer16BitAccess = false;
          features->storagePushConstant16 = false;
          features->storageInputOutput16 = false;
diff --git a/src/freedreno/vulkan/tu_extensions.py b/src/freedreno/vulkan/tu_extensions.py
index aa69ae0fd1a..ad83b8d2704 100644
--- a/src/freedreno/vulkan/tu_extensions.py
+++ b/src/freedreno/vulkan/tu_extensions.py
@@ -112,6 +112,7 @@ EXTENSIONS = [
     Extension('VK_KHR_pipeline_executable_properties',    1, True),
     Extension('VK_KHR_shader_float_controls',             1, True),
     Extension('VK_KHR_shader_float16_int8',               1, True),
+    Extension('VK_KHR_16bit_storage',                     1, 'device->gpu_id >= 650'),
 ]
 
 MAX_API_VERSION = VkVersion(MAX_API_VERSION)
diff --git a/src/freedreno/vulkan/tu_private.h b/src/freedreno/vulkan/tu_private.h
index edc34f19473..2db0958503e 100644
--- a/src/freedreno/vulkan/tu_private.h
+++ b/src/freedreno/vulkan/tu_private.h
@@ -1552,7 +1552,8 @@ uint32_t
 tu_subpass_get_attachment_to_resolve(const struct tu_subpass *subpass, uint32_t index);
 
 void
-tu_update_descriptor_sets(VkDescriptorSet overrideSet,
+tu_update_descriptor_sets(const struct tu_device *device,
+                          VkDescriptorSet overrideSet,
                           uint32_t descriptorWriteCount,
                           const VkWriteDescriptorSet *pDescriptorWrites,
                           uint32_t descriptorCopyCount,
@@ -1560,6 +1561,7 @@ tu_update_descriptor_sets(VkDescriptorSet overrideSet,
 
 void
 tu_update_descriptor_set_with_template(
+   const struct tu_device *device,
    struct tu_descriptor_set *set,
    VkDescriptorUpdateTemplate descriptorUpdateTemplate,
    const void *pData);
diff --git a/src/freedreno/vulkan/tu_shader.c b/src/freedreno/vulkan/tu_shader.c
index 9c3bef21501..4c144c0df2e 100644
--- a/src/freedreno/vulkan/tu_shader.c
+++ b/src/freedreno/vulkan/tu_shader.c
@@ -75,6 +75,7 @@ tu_spirv_to_nir(struct tu_device *dev,
          .runtime_descriptor_array = true,
          .float_controls = true,
          .float16 = true,
+         .storage_16bit = dev->physical_device->gpu_id >= 650,
       },
    };