From: Chris Duncan <chris@zoso.dev>
Date: Tue, 22 Apr 2025 21:17:11 +0000 (-0700)
Subject: Reduce workgroup size to 64 and update comments with reasoning.
X-Git-Url: https://zoso.dev/?a=commitdiff_plain;h=4646ec891cf97c367e094477227188428e62c255;p=nano-pow.git

Reduce workgroup size to 64 and update comments with reasoning.
---

diff --git a/src/lib/gpu/compute.wgsl b/src/lib/gpu/compute.wgsl
index ba14501..e06df02 100644
--- a/src/lib/gpu/compute.wgsl
+++ b/src/lib/gpu/compute.wgsl
@@ -139,13 +139,12 @@ var<workgroup> m4: vec2<u32>;
 
 /**
 * Search compute function
-* Calls main with a workgroup size of 96 which in testing was the lowest value
-* that would saturate the GPU active thread count and warp occupancy which
-* provides a decent balance with the power-sensitive requirements of mobile
+* Calls main with a workgroup size of 64 which balances warps between NVIDIA and
+* AMD cards while still considering the power-sensitive requirements of mobile
 * devices. The entire workgroup exits immediately if a nonce was already found
 * by a previous workgroup.
 */
-@compute @workgroup_size(96)
+@compute @workgroup_size(64)
 fn search(@builtin(global_invocation_id) global_id: vec3<u32>, @builtin(local_invocation_id) local_id: vec3<u32>) {
 	if (local_id.x == 0u) {
 		found = atomicLoad(&work.found) != 0u;