From dd8c62859c0f3c57f19493b4033ed8027df88a7b Mon Sep 17 00:00:00 2001 From: "Randall C. O'Reilly" Date: Fri, 22 Nov 2024 23:11:49 -0800 Subject: [PATCH] lvis results --- examples/bench_lvis/bench_results.md | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-) diff --git a/examples/bench_lvis/bench_results.md b/examples/bench_lvis/bench_results.md index 45fe28fd..2a388159 100644 --- a/examples/bench_lvis/bench_results.md +++ b/examples/bench_lvis/bench_results.md @@ -16,12 +16,16 @@ and performance is roughly similar. # V2.0.0-dev0.2.2 webgpu initial -Currently, can only handle -hiddenNeurs=4 due to memory limits! +Currently, can only handle -hiddenNeurs=4 due to memory limits! wgpu issues need fixed. ``` go test -gpu -hiddenNeurs=4 -verbose=false -ndata=1 -bench=. -run not ``` +* 2.7 with ndata = 1 on GPU, with ndata=1 di still inner-most. +* 9.7 on CPU = 3.6x speedup for GPU; CPU is sig slower due to atomic sendspikes. + + # v2.0.0-dev0.2.1 vgpu final 09/02/24 (linear SynCa approx) ## MacBook Pro M3