add gpu power consumption chart

beszel/internal/agent/agent.go

@@ -78,7 +78,7 @@ func (a *Agent) Run(pubKey []byte, addr string) {
 	// initialize GPU manager
 	if os.Getenv("GPU") == "true" {
 		if gm, err := NewGPUManager(); err != nil {
-			slog.Error("GPU manager", "err", err)
+			slog.Warn("GPU", "err", err)
 		} else {
 			a.gpuManager = gm
 		}

beszel/internal/agent/gpu.go

@@ -93,9 +93,9 @@ func (gm *GPUManager) parseNvidiaData(output []byte) {
 				}
 				// update gpu data
 				gpu := gm.GpuDataMap[id]
-				gpu.Temperature += temp
-				gpu.MemoryUsed += memoryUsage / 1.024
-				gpu.MemoryTotal += totalMemory / 1.024
+				gpu.Temperature = temp
+				gpu.MemoryUsed = memoryUsage / 1.024
+				gpu.MemoryTotal = totalMemory / 1.024
 				gpu.Usage += usage
 				gpu.Power += power
 				gpu.Count++
@@ -120,7 +120,7 @@ func (gm *GPUManager) startAmdCollector() {
 
 // collectAmdStats runs rocm-smi in a loop and passes the output to parseAmdData
 func (gm *GPUManager) collectAmdStats() error {
-	cmd := exec.Command("/bin/sh", "-c", "while true; do rocm-smi --showid --showtemp --showuse --showpower --showproductname --showmeminfo vram --json; sleep 4.7; done")
+	cmd := exec.Command("/bin/sh", "-c", "while true; do rocm-smi --showid --showtemp --showuse --showpower --showproductname --showmeminfo vram --json; sleep 3.7; done")
 	// Set up a pipe to capture stdout
 	stdout, err := cmd.StdoutPipe()
 	if err != nil {
@@ -168,9 +168,9 @@ func (gm *GPUManager) parseAmdData(rocmSmiInfo *map[string]RocmSmiJson) {
 			gm.GpuDataMap[v.ID] = &system.GPUData{Name: v.Name}
 		}
 		gpu := gm.GpuDataMap[v.ID]
-		gpu.Temperature += temp
-		gpu.MemoryUsed += memoryUsage
-		gpu.MemoryTotal += totalMemory
+		gpu.Temperature = temp
+		gpu.MemoryUsed = memoryUsage
+		gpu.MemoryTotal = totalMemory
 		gpu.Usage += usage
 		gpu.Power += power
 		gpu.Count++
@@ -185,19 +185,14 @@ func (gm *GPUManager) GetCurrentData() map[string]system.GPUData {
 	gpuData := make(map[string]system.GPUData, len(gm.GpuDataMap))
 	for id, gpu := range gm.GpuDataMap {
 		// sum the data
-		gpu.Temperature = twoDecimals(gpu.Temperature / gpu.Count)
-		gpu.MemoryUsed = twoDecimals(gpu.MemoryUsed / gpu.Count)
-		gpu.MemoryTotal = twoDecimals(gpu.MemoryTotal / gpu.Count)
+		gpu.Temperature = twoDecimals(gpu.Temperature)
+		gpu.MemoryUsed = twoDecimals(gpu.MemoryUsed)
+		gpu.MemoryTotal = twoDecimals(gpu.MemoryTotal)
 		gpu.Usage = twoDecimals(gpu.Usage / gpu.Count)
 		gpu.Power = twoDecimals(gpu.Power / gpu.Count)
 		gpuData[id] = *gpu
-		// reset the data
-		gpu.Temperature = 0
-		gpu.MemoryUsed = 0
-		gpu.MemoryTotal = 0
-		gpu.Usage = 0
-		gpu.Power = 0
-		gpu.Count = 0
+		// reset the count
+		gpu.Count = 1
 	}
 	return gpuData
 }

beszel/site/src/components/charts/gpu-power-chart.tsx

@@ -0,0 +1,112 @@
+import { CartesianGrid, Line, LineChart, YAxis } from "recharts"
+
+import {
+	ChartContainer,
+	ChartLegend,
+	ChartLegendContent,
+	ChartTooltip,
+	ChartTooltipContent,
+	xAxis,
+} from "@/components/ui/chart"
+import {
+	useYAxisWidth,
+	cn,
+	formatShortDate,
+	toFixedWithoutTrailingZeros,
+	decimalString,
+	chartMargin,
+} from "@/lib/utils"
+import { ChartData } from "@/types"
+import { memo, useMemo } from "react"
+
+export default memo(function GpuPowerChart({ chartData }: { chartData: ChartData }) {
+	const { yAxisWidth, updateYAxisWidth } = useYAxisWidth()
+
+	if (chartData.systemStats.length === 0) {
+		return null
+	}
+
+	/** Format temperature data for chart and assign colors */
+	const newChartData = useMemo(() => {
+		const newChartData = { data: [], colors: {} } as {
+			data: Record<string, number | string>[]
+			colors: Record<string, string>
+		}
+		const powerSums = {} as Record<string, number>
+		for (let data of chartData.systemStats) {
+			let newData = { created: data.created } as Record<string, number | string>
+
+			for (let gpu of Object.values(data.stats?.g ?? {})) {
+				if (gpu.p) {
+					const name = gpu.n
+					newData[name] = gpu.p
+					powerSums[name] = (powerSums[name] ?? 0) + newData[name]
+				}
+			}
+			newChartData.data.push(newData)
+		}
+		const keys = Object.keys(powerSums).sort((a, b) => powerSums[b] - powerSums[a])
+		for (let key of keys) {
+			newChartData.colors[key] = `hsl(${((keys.indexOf(key) * 360) / keys.length) % 360}, 60%, 55%)`
+		}
+		return newChartData
+	}, [chartData])
+
+	const colors = Object.keys(newChartData.colors)
+
+	// console.log('rendered at', new Date())
+
+	return (
+		<div>
+			<ChartContainer
+				className={cn("h-full w-full absolute aspect-auto bg-card opacity-0 transition-opacity", {
+					"opacity-100": yAxisWidth,
+				})}
+			>
+				<LineChart accessibilityLayer data={newChartData.data} margin={chartMargin}>
+					<CartesianGrid vertical={false} />
+					<YAxis
+						direction="ltr"
+						orientation={chartData.orientation}
+						className="tracking-tighter"
+						domain={[0, "auto"]}
+						width={yAxisWidth}
+						tickFormatter={(value) => {
+							const val = toFixedWithoutTrailingZeros(value, 2)
+							return updateYAxisWidth(val + "W")
+						}}
+						tickLine={false}
+						axisLine={false}
+					/>
+					{xAxis(chartData)}
+					<ChartTooltip
+						animationEasing="ease-out"
+						animationDuration={150}
+						// @ts-ignore
+						itemSorter={(a, b) => b.value - a.value}
+						content={
+							<ChartTooltipContent
+								labelFormatter={(_, data) => formatShortDate(data[0].payload.created)}
+								contentFormatter={(item) => decimalString(item.value) + "W"}
+								// indicator="line"
+							/>
+						}
+					/>
+					{colors.map((key) => (
+						<Line
+							key={key}
+							dataKey={key}
+							name={key}
+							type="monotoneX"
+							dot={false}
+							strokeWidth={1.5}
+							stroke={newChartData.colors[key]}
+							isAnimationActive={false}
+						/>
+					))}
+					{colors.length < 12 && <ChartLegend content={<ChartLegendContent />} />}
+				</LineChart>
+			</ChartContainer>
+		</div>
+	)
+})

beszel/site/src/components/routes/system.tsx

@@ -24,6 +24,7 @@ const MemChart = lazy(() => import("../charts/mem-chart"))
 const DiskChart = lazy(() => import("../charts/disk-chart"))
 const SwapChart = lazy(() => import("../charts/swap-chart"))
 const TemperatureChart = lazy(() => import("../charts/temperature-chart"))
+const GpuPowerChart = lazy(() => import("../charts/gpu-power-chart"))
 
 const cache = new Map<string, any>()
 
@@ -285,6 +286,7 @@ export default function SystemDetail({ name }: { name: string }) {
 
 	// if no data, show empty message
 	const dataEmpty = !chartLoading && chartData.systemStats.length === 0
+	const hasGpuData = Object.keys(systemStats.at(-1)?.stats.g ?? {}).length > 0
 
 	return (
 		<>
@@ -455,6 +457,7 @@ export default function SystemDetail({ name }: { name: string }) {
 						</div>
 					)}
 
+					{/* Swap chart */}
 					{(systemStats.at(-1)?.stats.su ?? 0) > 0 && (
 						<ChartCard
 							empty={dataEmpty}
@@ -466,6 +469,7 @@ export default function SystemDetail({ name }: { name: string }) {
 						</ChartCard>
 					)}
 
+					{/* Temperature chart */}
 					{systemStats.at(-1)?.stats.t && (
 						<ChartCard
 							empty={dataEmpty}
@@ -476,10 +480,22 @@ export default function SystemDetail({ name }: { name: string }) {
 							<TemperatureChart chartData={chartData} />
 						</ChartCard>
 					)}
+
+					{/* GPU power draw chart */}
+					{hasGpuData && (
+						<ChartCard
+							empty={dataEmpty}
+							grid={grid}
+							title="GPU Power Draw"
+							description="Average power consumption of GPUs"
+						>
+							<GpuPowerChart chartData={chartData} />
+						</ChartCard>
+					)}
 				</div>
 
 				{/* GPU charts */}
-				{Object.keys(systemStats.at(-1)?.stats.g ?? {}).length > 0 && (
+				{hasGpuData && (
 					<div className="grid xl:grid-cols-2 gap-4">
 						{Object.keys(systemStats.at(-1)?.stats.g ?? {}).map((id) => {
 							const gpu = systemStats.at(-1)?.stats.g?.[id] as GPUData
@@ -489,7 +505,7 @@ export default function SystemDetail({ name }: { name: string }) {
 										empty={dataEmpty}
 										grid={grid}
 										title={`${gpu.n} ${t`Usage`}`}
-										description={t`Total utilization of ${gpu.n}`}
+										description={`Average utilization of ${gpu.n}`}
 									>
 										<AreaChartDefault chartData={chartData} chartName={`g.${id}.u`} unit="%" />
 									</ChartCard>
@@ -497,7 +513,7 @@ export default function SystemDetail({ name }: { name: string }) {
 										empty={dataEmpty}
 										grid={grid}
 										title={`${gpu.n} VRAM`}
-										description={t`VRAM usage of ${gpu.n}`}
+										description={t`Precise utilization at the recorded time`}
 									>
 										<AreaChartDefault
 											chartData={chartData}