Improve multi-gpu handling at the limit

Still not complete, needs some refinement to our prediction to understand the
discrete GPUs available space so we can see how many layers fit in each one
since we can't split one layer across multiple GPUs we can't treat free space
as one logical block

gpu/amd_linux.go

@@ -27,7 +27,7 @@ const (
 	GPUTotalMemoryFileGlob = "mem_banks/*/properties" // size_in_bytes line
 
 	// Direct Rendering Manager sysfs location
-	DRMDeviceDirGlob   = "/sys/class/drm/card[0-9]/device"
+	DRMDeviceDirGlob   = "/sys/class/drm/card*/device"
 	DRMTotalMemoryFile = "mem_info_vram_total"
 	DRMUsedMemoryFile  = "mem_info_vram_used"
 

gpu/gpu.go

@@ -246,10 +246,6 @@ func GetGPUInfo() GpuInfoList {
 			return GpuInfoList{cpus[0].GpuInfo}
 		}
 
-		// TODO - implement
-
-		// TODO refine the discovery to only gather total memory
-
 		// On windows we bundle the nvidia library one level above the runner dir
 		depPath := ""
 		if runtime.GOOS == "windows" && envconfig.RunnersDir != "" {

gpu/types.go

@@ -44,14 +44,14 @@ type CPUInfo struct {
 
 type CudaGPUInfo struct {
 	GpuInfo
-	index int // device index
+	index int // nolint: unused
 }
 type CudaGPUInfoList []CudaGPUInfo
 
 type RocmGPUInfo struct {
 	GpuInfo
-	usedFilepath string // linux
-	index        int    // device index on windows
+	usedFilepath string // nolint: unused
+	index        int    // nolint: unused
 }
 type RocmGPUInfoList []RocmGPUInfo