From: Chris Duncan <chris@zoso.dev>
Date: Tue, 7 Jan 2025 21:38:18 +0000 (-0800)
Subject: Throttle back workgroup size since apparently this value works with GPUs better and... 
X-Git-Url: https://zoso.dev/?a=commitdiff_plain;h=b7c4ce7bce5d4a11e148e83e9095633711d8b82c;p=libnemo.git

Throttle back workgroup size since apparently this value works with GPUs better and we're still checking over a billion nonces per dispatch; the probability of finding a nonce at this level is over 99.9%
---

diff --git a/src/lib/workers/powgpu.ts b/src/lib/workers/powgpu.ts
index b60ce03..65620c6 100644
--- a/src/lib/workers/powgpu.ts
+++ b/src/lib/workers/powgpu.ts
@@ -172,7 +172,7 @@ export class PowGpu extends WorkerInterface {
 		* 8-byte work is split into two 4-byte u32. Low 4 bytes are random u32 from
 		* UBO. High 4 bytes are the random value XOR'd with index of each thread.
 		*/
-		@compute @workgroup_size(256)
+		@compute @workgroup_size(64)
 		fn main(
 			@builtin(workgroup_id) workgroup_id: vec3<u32>,
 			@builtin(local_invocation_id) local_id: vec3<u32>