From 748dce60cda8d77eece626541fa3b9733452f70c Mon Sep 17 00:00:00 2001 From: Nicolas Patry Date: Sun, 10 Nov 2024 14:14:10 +0100 Subject: [PATCH] h100 better name, and keep factor of 2 --- launcher/src/main.rs | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/launcher/src/main.rs b/launcher/src/main.rs index 109eb3eb..18658406 100644 --- a/launcher/src/main.rs +++ b/launcher/src/main.rs @@ -33,7 +33,7 @@ fn compute_optimal(config: Option<&Config>, compute: Option<&ComputeType>) -> Op if let (Some(config), Some(compute)) = (config, compute) { if let (Some(f16_max_compute), Some(model_compute)) = (compute.f16_flop(), config.flop()) { tracing::debug!("MAx compute {f16_max_compute} model compute {model_compute}"); - let optimal_size = (f16_max_compute / model_compute) as usize; + let optimal_size = (f16_max_compute / model_compute / 2) as usize; if optimal_size > 100 { // Ignore calculations that's too low // Most likely an error @@ -1508,7 +1508,7 @@ impl ComputeType { // https://www.nvidia.com/en-us/data-center/products/a10-gpu/ "nvidia-a10g" => Some(125 * 10u64.pow(12)), // https://www.nvidia.com/en-us/data-center/h100/ - "nvidia-h100" => Some(1900 * 10u64.pow(12)), + "nvidia-h100-80gb-hb3" => Some(1900 * 10u64.pow(12)), // https://www.nvidia.com/content/dam/en-zz/Solutions/Data-Center/a100/pdf/nvidia-a100-datasheet-us-nvidia-1758950-r4-web.pdf "nvidia-a100" => Some(312 * 10u64.pow(12)), card => {