Skip to content

vulkan: for small AMD GPUs, reduce submission threshold based on CU count #25240

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
0cc4m merged 1 commit into from
Jul 8, 2026
Merged

vulkan: for small AMD GPUs, reduce submission threshold based on CU count #25240

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
13 changes: 12 additions & 1 deletion ggml/src/ggml-vulkan/ggml-vulkan.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16308,7 +16308,18 @@ static ggml_status ggml_backend_vk_graph_compute(ggml_backend_t backend, ggml_cg
uint32_t submit_count = 0;
uint64_t batch_flops = 0;
uint64_t total_flops = 0;
uint64_t flops_per_submit = std::min(uint64_t(200'000'000'000), ctx->last_total_flops / 40u);
uint64_t flops_cap = 200'000'000'000ULL;

// On weaker AMD GPUs larger submissions can hit a driver timeout, submit more often to avoid this
if (ctx->device->vendor_id == VK_VENDOR_ID_AMD && ctx->device->shader_core_count > 0) {
if (ctx->device->architecture == AMD_GCN && ctx->device->shader_core_count < 32) {
flops_cap = 500'000'000ULL * ctx->device->shader_core_count;
} else if (ctx->device->architecture != AMD_GCN && ctx->device->shader_core_count < 24) {
flops_cap = 2'000'000'000ULL * ctx->device->shader_core_count;
}
}
uint64_t flops_per_submit = std::min(flops_cap, ctx->last_total_flops / 40u);

for (int i = 0; i < cgraph->n_nodes; i++) {
if (first_node_in_batch) {
submit_node_idx = i;
Expand Down
Loading