From 6566387ae37faee440a8be7169882e8b4a410f40 Mon Sep 17 00:00:00 2001
From: Jeffrey Morgan <jmorganca@gmail.com>
Date: Tue, 9 Jan 2024 00:28:03 -0500
Subject: [PATCH] add `TODO` for cuda overhead

---
 gpu/gpu.go | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/gpu/gpu.go b/gpu/gpu.go
index 2d32e9de..07a3bee3 100644
--- a/gpu/gpu.go
+++ b/gpu/gpu.go
@@ -131,8 +131,9 @@ func getCPUMem() (memInfo, error) {
 func CheckVRAM() (int64, error) {
 	gpuInfo := GetGPUInfo()
 	if gpuInfo.FreeMemory > 0 && (gpuInfo.Library == "cuda" || gpuInfo.Library == "rocm") {
-		// leave 20% of VRAM free for overhead
-		return int64(gpuInfo.FreeMemory * 4 / 5), nil
+		// leave 25% of VRAM free for overhead
+		// TODO: improve cuda memory allocation to avoid needing this large of an amount
+		return int64(gpuInfo.FreeMemory * 3 / 4), nil
 	}
 
 	return 0, fmt.Errorf("no GPU detected") // TODO - better handling of CPU based memory determiniation