feat: support GB10 shared memory and refine AI model UI (#14254)

agent/app/dto/dashboard.go

@@ -177,6 +177,7 @@ type DiskInfo struct {
 }
 
 type GPUInfo struct {
+	SharedMemory     bool   `json:"sharedMemory"`
 	Type             string `json:"type"`
 	Index            uint   `json:"index"`
 	NPUIndex         uint   `json:"npuIndex"`

agent/app/dto/monitor.go

@@ -56,17 +56,18 @@ type MonitorGPUOptions struct {
 	Options   []string       `json:"options"`
 }
 type GPUChartHide struct {
-	DeviceID    string `json:"deviceID"`
-	Legacy      bool   `json:"legacy"`
-	ProductName string `json:"productName"`
-	Type        string `json:"type"`
-	Process     bool   `json:"process"`
-	GPU         bool   `json:"gpu"`
-	Memory      bool   `json:"memory"`
-	Power       bool   `json:"power"`
-	PowerLimit  bool   `json:"powerLimit"`
-	Temperature bool   `json:"temperature"`
-	Speed       bool   `json:"speed"`
+	SharedMemory bool   `json:"sharedMemory"`
+	DeviceID     string `json:"deviceID"`
+	Legacy       bool   `json:"legacy"`
+	ProductName  string `json:"productName"`
+	Type         string `json:"type"`
+	Process      bool   `json:"process"`
+	GPU          bool   `json:"gpu"`
+	Memory       bool   `json:"memory"`
+	Power        bool   `json:"power"`
+	PowerLimit   bool   `json:"powerLimit"`
+	Temperature  bool   `json:"temperature"`
+	Speed        bool   `json:"speed"`
 }
 type MonitorGPUSearch struct {
 	Aggregation string    `json:"aggregation" validate:"omitempty,oneof=avg max"`

agent/app/service/monitor.go

@@ -8,6 +8,7 @@ import (
 	"path/filepath"
 	"sort"
 	"strconv"
+	"strings"
 	"sync"
 	"time"
 
@@ -19,6 +20,7 @@ import (
 	"github.com/1Panel-dev/1Panel/agent/app/model"
 	"github.com/1Panel-dev/1Panel/agent/global"
 	"github.com/1Panel-dev/1Panel/agent/utils/ai_tools/accelerator"
+	"github.com/1Panel-dev/1Panel/agent/utils/ai_tools/gpu"
 	"github.com/1Panel-dev/1Panel/agent/utils/common"
 	"github.com/1Panel-dev/1Panel/agent/utils/psutil"
 	"github.com/robfig/cron/v3"
@@ -205,7 +207,14 @@ func (m *MonitorService) LoadGPUOptions() dto.MonitorGPUOptions {
 			continue
 		}
 		seen[key] = true
-		data.ChartHide = append(data.ChartHide, dto.GPUChartHide{DeviceID: device.DeviceID, ProductName: device.ProductName, Type: device.DeviceType, Legacy: device.DeviceID == ""})
+		_, productName, _ := strings.Cut(device.ProductName, " - ")
+		data.ChartHide = append(data.ChartHide, dto.GPUChartHide{
+			DeviceID:     device.DeviceID,
+			ProductName:  device.ProductName,
+			Type:         device.DeviceType,
+			Legacy:       device.DeviceID == "",
+			SharedMemory: strings.HasPrefix(device.DeviceID, "nvidia:") && gpu.UsesSystemMemory(productName),
+		})
 		data.Options = append(data.Options, device.ProductName)
 	}
 	return data
@@ -800,9 +809,10 @@ func loadGPUOptions(snapshot *accelerator.Snapshot) dto.MonitorGPUOptions {
 	})
 	for _, item := range snapshot.Devices {
 		chartHide := dto.GPUChartHide{
-			DeviceID:    item.ID,
-			ProductName: item.Label,
-			Type:        string(item.Kind),
+			SharedMemory: item.GPU != nil && item.GPU.SharedMemory,
+			DeviceID:     item.ID,
+			ProductName:  item.Label,
+			Type:         string(item.Kind),
 		}
 		data.ChartHide = append(data.ChartHide, chartHide)
 		data.Options = append(data.Options, chartHide.ProductName)

agent/utils/ai_tools/gpu/nvidia.go

@@ -4,10 +4,12 @@ import (
 	"context"
 	"encoding/xml"
 	"fmt"
+	"slices"
 	"strings"
 	"time"
 
 	"github.com/1Panel-dev/1Panel/agent/utils/cmd"
+	"github.com/shirou/gopsutil/v4/mem"
 )
 
 const nvidiaSMICommand = "nvidia-smi"
@@ -20,7 +22,31 @@ func (n nvidiaSMI) LoadInfo(ctx context.Context) (*Info, error) {
 	if err != nil {
 		return nil, fmt.Errorf("calling %s failed: %w", nvidiaSMICommand, err)
 	}
-	return parseNvidiaSMI([]byte(itemData))
+	info, err := parseNvidiaSMI([]byte(itemData))
+	if err != nil {
+		return nil, err
+	}
+	if slices.ContainsFunc(info.Devices, func(device Device) bool { return device.SharedMemory }) {
+		memory, err := mem.VirtualMemoryWithContext(ctx)
+		if err != nil {
+			info.Warnings = append(info.Warnings, fmt.Sprintf("read system shared memory: %v", err))
+		} else {
+			for index := range info.Devices {
+				device := &info.Devices[index]
+				if !device.SharedMemory {
+					continue
+				}
+				device.MemUsed = fmt.Sprintf("%d B", memory.Used)
+				device.MemTotal = fmt.Sprintf("%d B", memory.Total)
+				device.MemoryFree = fmt.Sprintf("%.2f MiB", float64(memory.Available)/(1024*1024))
+			}
+		}
+	}
+	return info, nil
+}
+
+func UsesSystemMemory(productName string) bool {
+	return strings.EqualFold(strings.TrimSpace(productName), "NVIDIA GB10")
 }
 
 func parseNvidiaSMI(buf []byte) (*Info, error) {
@@ -38,6 +64,7 @@ func parseNvidiaSMI(buf []byte) (*Info, error) {
 	for i := range s.Gpu {
 		gpuItem := Device{
 			Type:              "nvidia",
+			SharedMemory:      UsesSystemMemory(s.Gpu[i].ProductName),
 			MemoryActivity:    s.Gpu[i].Utilization.MemoryUtil,
 			EncoderUtil:       s.Gpu[i].Utilization.EncoderUtil,
 			DecoderUtil:       s.Gpu[i].Utilization.DecoderUtil,
@@ -72,6 +99,12 @@ func parseNvidiaSMI(buf []byte) (*Info, error) {
 			ComputeMode:       s.Gpu[i].ComputeMode,
 			MigMode:           s.Gpu[i].MigMode.CurrentMig,
 		}
+		if gpuItem.SharedMemory {
+			gpuItem.MemUsed = "N/A"
+			gpuItem.MemTotal = "N/A"
+			gpuItem.MemoryFree = "N/A"
+			gpuItem.MemoryReserved = "N/A"
+		}
 		gpuItem.ECCPending = s.Gpu[i].EccMode.PendingEcc
 		gpuItem.ECCErrors = []ECCError{
 			{Scope: "Volatile Total", Correctable: s.Gpu[i].EccErrors.Volatile.SingleBit.Total, Uncorrectable: s.Gpu[i].EccErrors.Volatile.DoubleBit.Total},

agent/utils/ai_tools/gpu/types.go

@@ -17,6 +17,7 @@ type ECCError struct {
 }
 
 type Device struct {
+	SharedMemory       bool       `json:"sharedMemory"`
 	ECCPending         string     `json:"eccPending"`
 	ECCErrors          []ECCError `json:"eccErrors"`
 	MemoryActivity     string     `json:"memoryActivity"`

frontend/src/api/interface/ai.ts

@@ -49,6 +49,7 @@ export namespace AI {
         xpu: XpuInfo['xpu'];
     }
     export interface GPU {
+        sharedMemory?: boolean;
         eccPending?: string;
         eccErrors?: { scope: string; correctable: string; uncorrectable: string }[];
         memoryActivity?: string;
@@ -139,6 +140,7 @@ export namespace AI {
         chartHide: Array<ChartHide>;
     }
     export interface ChartHide {
+        sharedMemory?: boolean;
         deviceID?: string;
         legacy?: boolean;
         productName: string;

frontend/src/api/interface/dashboard.ts

@@ -151,6 +151,7 @@ export namespace Dashboard {
         inodesUsedPercent: number;
     }
     export interface GPUInfo {
+        sharedMemory?: boolean;
         type: string;
         index: number;
         npuIndex: number;

frontend/src/lang/modules/en.ts

@@ -1071,11 +1071,11 @@ const message = {
             onlineModel: 'Online models',
             instanceDeployment: 'Deploy by instance',
             nvidiaDeployment: 'NVIDIA-optimized deployment',
-            engines: 'Inference engines',
+            engines: 'Local inference engines',
             modelManagement: 'Model management',
             serviceDetails: 'Service details',
             deployments: 'Local deployments',
-            files: 'Model files',
+            files: 'Model hub',
             connectExternal: 'Connect external service',
             deploymentTemplates: 'vLLM startup command templates',
             engineProvider: 'Engine / Provider',
@@ -1643,6 +1643,11 @@ const message = {
             timePerOutputTokenChart: 'Inter-token latency',
         },
         gpu: {
+            systemMemory: 'Shared system memory',
+            systemMemoryUsed: 'System memory used',
+            systemMemoryTotal: 'Total system memory',
+            systemMemoryAvailable: 'Available system memory',
+            systemMemoryHelper: 'Shared by the CPU and GPU. Shows system-wide memory usage, not GPU-only usage.',
             memoryActivity: 'Memory read/write activity',
             hotspotTemperature: 'Hotspot temperature',
             fanRPM: 'Fan speed (RPM)',
@@ -1678,7 +1683,6 @@ const message = {
             historyPeak: 'Peak',
             historyAggregated:
                 '{count} samples aggregated into {seconds}-second intervals; processes show counts only.',
-            historyRaw: '{count} original samples.',
             historyMetricEmpty: 'No data for this metric in the selected time range',
             historyLoadFailed: 'Failed to load history. Please retry.',
             collectedAt: 'Collected at',

frontend/src/lang/modules/es-es.ts

@@ -1073,11 +1073,11 @@ const message = {
             onlineModel: 'Modelos en línea',
             instanceDeployment: 'Despliegue por instancia',
             nvidiaDeployment: 'Despliegue optimizado para NVIDIA',
-            engines: 'Motores de inferencia',
+            engines: 'Motores de inferencia locales',
             modelManagement: 'Gestión de modelos',
             serviceDetails: 'Detalles del servicio',
             deployments: 'Despliegues locales',
-            files: 'Archivos de modelos',
+            files: 'Catálogo de modelos',
             connectExternal: 'Conectar servicio externo',
             deploymentTemplates: 'Plantillas de comandos de inicio de vLLM',
             engineProvider: 'Motor / Proveedor',
@@ -1653,6 +1653,12 @@ const message = {
             timePerOutputTokenChart: 'Latencia entre tokens',
         },
         gpu: {
+            systemMemory: 'Memoria compartida del sistema',
+            systemMemoryUsed: 'Memoria del sistema utilizada',
+            systemMemoryTotal: 'Memoria total del sistema',
+            systemMemoryAvailable: 'Memoria disponible del sistema',
+            systemMemoryHelper:
+                'Compartida por la CPU y la GPU. Muestra el uso de memoria de todo el sistema, no solo de la GPU.',
             memoryActivity: 'Actividad de lectura/escritura de memoria',
             hotspotTemperature: 'Temperatura del punto caliente',
             fanRPM: 'Velocidad del ventilador (RPM)',
@@ -1688,7 +1694,6 @@ const message = {
             historyPeak: 'Máximo',
             historyAggregated:
                 '{count} muestras agrupadas en intervalos de {seconds} segundos; los procesos solo muestran cantidades.',
-            historyRaw: '{count} muestras originales.',
             historyMetricEmpty: 'No hay datos de esta métrica en el período seleccionado',
             historyLoadFailed: 'No se pudo cargar el historial. Inténtelo de nuevo.',
             collectedAt: 'Fecha de recopilación',

frontend/src/lang/modules/fa.ts

@@ -1055,11 +1055,11 @@ const message = {
             onlineModel: 'مدل‌های آنلاین',
             instanceDeployment: 'استقرار بر اساس نمونه',
             nvidiaDeployment: 'استقرار بهینه‌شده برای NVIDIA',
-            engines: 'موتورهای استنتاج',
+            engines: 'موتورهای استنتاج محلی',
             modelManagement: 'مدیریت مدل‌ها',
             serviceDetails: 'جزئیات سرویس',
             deployments: 'استقرارهای محلی',
-            files: 'فایل‌های مدل',
+            files: 'مرکز مدل‌ها',
             connectExternal: 'اتصال به سرویس خارجی',
             deploymentTemplates: 'الگوهای فرمان راه‌اندازی vLLM',
             engineProvider: 'موتور / ارائه‌دهنده',
@@ -1625,6 +1625,11 @@ const message = {
             timePerOutputTokenChart: 'تأخیر بین توکن‌ها',
         },
         gpu: {
+            systemMemory: 'حافظهٔ مشترک سیستم',
+            systemMemoryUsed: 'حافظهٔ مصرف‌شدهٔ سیستم',
+            systemMemoryTotal: 'کل حافظهٔ سیستم',
+            systemMemoryAvailable: 'حافظهٔ در دسترس سیستم',
+            systemMemoryHelper: 'بین CPU و GPU مشترک است. مصرف حافظهٔ کل سیستم را نشان می‌دهد، نه فقط مصرف GPU.',
             memoryActivity: 'فعالیت خواندن/نوشتن حافظه',
             hotspotTemperature: 'دمای نقطه داغ',
             fanRPM: 'سرعت فن (RPM)',
@@ -1660,7 +1665,6 @@ const message = {
             historyPeak: 'بیشینه',
             historyAggregated:
                 '{count} نمونه در بازه‌های {seconds} ثانیه‌ای تجمیع شده‌اند؛ فقط تعداد فرایندها نمایش داده می‌شود.',
-            historyRaw: '{count} نمونه اصلی.',
             historyMetricEmpty: 'در بازه زمانی انتخاب‌شده داده‌ای برای این شاخص وجود ندارد',
             historyLoadFailed: 'بارگیری سابقه ناموفق بود. دوباره تلاش کنید.',
             collectedAt: 'زمان گردآوری',

frontend/src/lang/modules/ja.ts

@@ -1059,11 +1059,11 @@ const message = {
             onlineModel: 'オンラインモデル',
             instanceDeployment: 'インスタンス単位でデプロイ',
             nvidiaDeployment: 'NVIDIA 向け最適化デプロイ',
-            engines: '推論エンジン',
+            engines: 'ローカル推論エンジン',
             modelManagement: 'モデル管理',
             serviceDetails: 'サービス詳細',
             deployments: 'ローカルデプロイ',
-            files: 'モデルファイル',
+            files: 'モデル広場',
             connectExternal: '外部サービスに接続',
             deploymentTemplates: 'vLLM 起動コマンドテンプレート',
             engineProvider: 'エンジン / プロバイダー',
@@ -1629,6 +1629,11 @@ const message = {
             timePerOutputTokenChart: 'トークン間遅延',
         },
         gpu: {
+            systemMemory: '共有システムメモリ',
+            systemMemoryUsed: 'システムメモリ使用量',
+            systemMemoryTotal: 'システムメモリ合計',
+            systemMemoryAvailable: '利用可能なシステムメモリ',
+            systemMemoryHelper: 'CPU と GPU で共有します。GPU 単体ではなく、システム全体のメモリ使用量を表示します。',
             memoryActivity: 'メモリ読み書き稼働率',
             hotspotTemperature: 'ホットスポット温度',
             fanRPM: 'ファン回転数（RPM）',
@@ -1663,7 +1668,6 @@ const message = {
             historyAverage: '平均値',
             historyPeak: '最大値',
             historyAggregated: '{count} 件のサンプルを {seconds} 秒間隔で集計。プロセスは件数のみ表示します。',
-            historyRaw: '元のサンプル {count} 件。',
             historyMetricEmpty: '選択した期間にこの指標のデータはありません',
             historyLoadFailed: '履歴の読み込みに失敗しました。再試行してください。',
             collectedAt: '収集日時',

frontend/src/lang/modules/ko.ts

@@ -1049,11 +1049,11 @@ const message = {
             onlineModel: '온라인 모델',
             instanceDeployment: '인스턴스별 배포',
             nvidiaDeployment: 'NVIDIA 최적화 배포',
-            engines: '추론 엔진',
+            engines: '로컬 추론 엔진',
             modelManagement: '모델 관리',
             serviceDetails: '서비스 상세 정보',
             deployments: '로컬 배포',
-            files: '모델 파일',
+            files: '모델 허브',
             connectExternal: '외부 서비스 연결',
             deploymentTemplates: 'vLLM 시작 명령 템플릿',
             engineProvider: '엔진 / 공급자',
@@ -1614,6 +1614,11 @@ const message = {
             timePerOutputTokenChart: '토큰 간 지연',
         },
         gpu: {
+            systemMemory: '공유 시스템 메모리',
+            systemMemoryUsed: '사용 중인 시스템 메모리',
+            systemMemoryTotal: '전체 시스템 메모리',
+            systemMemoryAvailable: '사용 가능한 시스템 메모리',
+            systemMemoryHelper: 'CPU와 GPU가 공유합니다. GPU만의 사용량이 아닌 시스템 전체 메모리 사용량을 표시합니다.',
             memoryActivity: '메모리 읽기/쓰기 활성률',
             hotspotTemperature: '핫스폿 온도',
             fanRPM: '팬 회전 속도 (RPM)',
@@ -1648,7 +1653,6 @@ const message = {
             historyAverage: '평균',
             historyPeak: '최댓값',
             historyAggregated: '샘플 {count}개를 {seconds}초 간격으로 집계합니다. 프로세스는 개수만 표시합니다.',
-            historyRaw: '원본 샘플 {count}개.',
             historyMetricEmpty: '선택한 기간에 이 지표의 데이터가 없습니다',
             historyLoadFailed: '기록을 불러오지 못했습니다. 다시 시도하세요.',
             collectedAt: '수집 시간',

frontend/src/lang/modules/lo.ts

@@ -1046,11 +1046,11 @@ const message = {
             onlineModel: 'ໂມເດວອອນລາຍ',
             instanceDeployment: 'ນຳໃຊ້ຕາມອິນສະແຕນ',
             nvidiaDeployment: 'ການນຳໃຊ້ທີ່ປັບໃຫ້ເໝາະກັບ NVIDIA',
-            engines: 'ເຄື່ອງຈັກອະນຸມານ',
+            engines: 'ເຄື່ອງຈັກອະນຸມານພາຍໃນ',
             modelManagement: 'ຈັດການໂມເດວ',
             serviceDetails: 'ລາຍລະອຽດບໍລິການ',
             deployments: 'ການນຳໃຊ້ພາຍໃນເຄື່ອງ',
-            files: 'ໄຟລ໌ໂມເດວ',
+            files: 'ສູນລວມໂມເດວ',
             connectExternal: 'ເຊື່ອມຕໍ່ບໍລິການພາຍນອກ',
             deploymentTemplates: 'ແມ່ແບບຄຳສັ່ງເລີ່ມຕົ້ນ vLLM',
             engineProvider: 'ເຄື່ອງຈັກ / ຜູ້ໃຫ້ບໍລິການ',
@@ -1606,6 +1606,11 @@ const message = {
             timePerOutputTokenChart: 'ເວລາລໍຖ້າລະຫວ່າງ Token',
         },
         gpu: {
+            systemMemory: 'ໜ່ວຍຄວາມຈຳລະບົບທີ່ໃຊ້ຮ່ວມກັນ',
+            systemMemoryUsed: 'ໜ່ວຍຄວາມຈຳລະບົບທີ່ໃຊ້ແລ້ວ',
+            systemMemoryTotal: 'ໜ່ວຍຄວາມຈຳລະບົບທັງໝົດ',
+            systemMemoryAvailable: 'ໜ່ວຍຄວາມຈຳລະບົບທີ່ພ້ອມໃຊ້',
+            systemMemoryHelper: 'CPU ແລະ GPU ໃຊ້ຮ່ວມກັນ. ສະແດງການໃຊ້ໜ່ວຍຄວາມຈຳຂອງທັງລະບົບ ບໍ່ແມ່ນສະເພາະ GPU.',
             memoryActivity: 'ກິດຈະກຳອ່ານ/ຂຽນໜ່ວຍຄວາມຈຳ',
             hotspotTemperature: 'ອຸນຫະພູມຈຸດຮ້ອນ',
             fanRPM: 'ຄວາມໄວພັດລົມ (RPM)',
@@ -1640,7 +1645,6 @@ const message = {
             historyAverage: 'ຄ່າສະເລ່ຍ',
             historyPeak: 'ຄ່າສູງສຸດ',
             historyAggregated: 'ລວມ {count} ຕົວຢ່າງເປັນຊ່ວງ {seconds} ວິນາທີ; ສະແດງພຽງຈຳນວນໂປຣເຊສ.',
-            historyRaw: '{count} ຕົວຢ່າງຕົ້ນສະບັບ.',
             historyMetricEmpty: 'ບໍ່ມີຂໍ້ມູນຕົວຊີ້ວັດນີ້ໃນຊ່ວງເວລາທີ່ເລືອກ',
             historyLoadFailed: 'ໂຫຼດປະຫວັດບໍ່ສຳເລັດ. ກະລຸນາລອງໃໝ່.',
             collectedAt: 'ເວລາເກັບຂໍ້ມູນ',

frontend/src/lang/modules/ms.ts

@@ -1081,11 +1081,11 @@ const message = {
             onlineModel: 'Model dalam talian',
             instanceDeployment: 'Pengerahan mengikut instans',
             nvidiaDeployment: 'Pengerahan dioptimumkan untuk NVIDIA',
-            engines: 'Enjin inferens',
+            engines: 'Enjin inferens setempat',
             modelManagement: 'Pengurusan model',
             serviceDetails: 'Butiran perkhidmatan',
             deployments: 'Pengerahan setempat',
-            files: 'Fail model',
+            files: 'Hab model',
             connectExternal: 'Sambungkan perkhidmatan luaran',
             deploymentTemplates: 'Templat arahan permulaan vLLM',
             engineProvider: 'Enjin / Pembekal',
@@ -1655,6 +1655,12 @@ const message = {
             timePerOutputTokenChart: 'Kependaman antara token',
         },
         gpu: {
+            systemMemory: 'Memori sistem dikongsi',
+            systemMemoryUsed: 'Memori sistem digunakan',
+            systemMemoryTotal: 'Jumlah memori sistem',
+            systemMemoryAvailable: 'Memori sistem tersedia',
+            systemMemoryHelper:
+                'Dikongsi oleh CPU dan GPU. Memaparkan penggunaan memori seluruh sistem, bukan GPU sahaja.',
             memoryActivity: 'Aktiviti baca/tulis memori',
             hotspotTemperature: 'Suhu titik panas',
             fanRPM: 'Kelajuan kipas (RPM)',
@@ -1690,7 +1696,6 @@ const message = {
             historyPeak: 'Puncak',
             historyAggregated:
                 '{count} sampel diagregatkan dalam sela {seconds} saat; proses hanya memaparkan bilangan.',
-            historyRaw: '{count} sampel asal.',
             historyMetricEmpty: 'Tiada data untuk metrik ini dalam julat masa yang dipilih',
             historyLoadFailed: 'Gagal memuatkan sejarah. Sila cuba lagi.',
             collectedAt: 'Masa pengumpulan',

frontend/src/lang/modules/pt-br.ts

@@ -1077,11 +1077,11 @@ const message = {
             onlineModel: 'Modelos online',
             instanceDeployment: 'Implantação por instância',
             nvidiaDeployment: 'Implantação otimizada para NVIDIA',
-            engines: 'Mecanismos de inferência',
+            engines: 'Mecanismos de inferência locais',
             modelManagement: 'Gerenciamento de modelos',
             serviceDetails: 'Detalhes do serviço',
             deployments: 'Implantações locais',
-            files: 'Arquivos de modelos',
+            files: 'Central de modelos',
             connectExternal: 'Conectar serviço externo',
             deploymentTemplates: 'Modelos de comandos de inicialização do vLLM',
             engineProvider: 'Mecanismo / Provedor',
@@ -1654,6 +1654,12 @@ const message = {
             timePerOutputTokenChart: 'Latência entre tokens',
         },
         gpu: {
+            systemMemory: 'Memória compartilhada do sistema',
+            systemMemoryUsed: 'Memória do sistema em uso',
+            systemMemoryTotal: 'Memória total do sistema',
+            systemMemoryAvailable: 'Memória disponível do sistema',
+            systemMemoryHelper:
+                'Compartilhada pela CPU e GPU. Exibe o uso de memória de todo o sistema, não apenas da GPU.',
             memoryActivity: 'Atividade de leitura/escrita da memória',
             hotspotTemperature: 'Temperatura do ponto quente',
             fanRPM: 'Rotação da ventoinha (RPM)',
@@ -1689,7 +1695,6 @@ const message = {
             historyPeak: 'Pico',
             historyAggregated:
                 '{count} amostras agregadas em intervalos de {seconds} segundos; processos exibem apenas contagens.',
-            historyRaw: '{count} amostras originais.',
             historyMetricEmpty: 'Sem dados desta métrica no período selecionado',
             historyLoadFailed: 'Falha ao carregar o histórico. Tente novamente.',
             collectedAt: 'Data da coleta',

frontend/src/lang/modules/ru.ts

@@ -1070,11 +1070,11 @@ const message = {
             onlineModel: 'Онлайн-модели',
             instanceDeployment: 'Развертывание по экземплярам',
             nvidiaDeployment: 'Развертывание с оптимизацией для NVIDIA',
-            engines: 'Движки инференса',
+            engines: 'Локальные движки инференса',
             modelManagement: 'Управление моделями',
             serviceDetails: 'Сведения о сервисе',
             deployments: 'Локальные развертывания',
-            files: 'Файлы моделей',
+            files: 'Каталог моделей',
             connectExternal: 'Подключить внешний сервис',
             deploymentTemplates: 'Шаблоны команд запуска vLLM',
             engineProvider: 'Движок / Поставщик',
@@ -1644,6 +1644,12 @@ const message = {
             timePerOutputTokenChart: 'Задержка между токенами',
         },
         gpu: {
+            systemMemory: 'Общая системная память',
+            systemMemoryUsed: 'Использовано системной памяти',
+            systemMemoryTotal: 'Всего системной памяти',
+            systemMemoryAvailable: 'Доступная системная память',
+            systemMemoryHelper:
+                'Память совместно используется CPU и GPU. Показано потребление всей системы, а не только GPU.',
             memoryActivity: 'Активность чтения/записи памяти',
             hotspotTemperature: 'Температура горячей точки',
             fanRPM: 'Скорость вентилятора (об/мин)',
@@ -1679,7 +1685,6 @@ const message = {
             historyPeak: 'Максимум',
             historyAggregated:
                 '{count} измерений сгруппированы по {seconds} с; для процессов показано только количество.',
-            historyRaw: 'Исходных измерений: {count}.',
             historyMetricEmpty: 'Нет данных этой метрики за выбранный период',
             historyLoadFailed: 'Не удалось загрузить историю. Повторите попытку.',
             collectedAt: 'Время сбора',

frontend/src/lang/modules/tr.ts

@@ -1080,11 +1080,11 @@ const message = {
             onlineModel: 'Çevrimiçi modeller',
             instanceDeployment: 'Örnek bazında dağıtım',
             nvidiaDeployment: 'NVIDIA için optimize edilmiş dağıtım',
-            engines: 'Çıkarım motorları',
+            engines: 'Yerel çıkarım motorları',
             modelManagement: 'Model yönetimi',
             serviceDetails: 'Hizmet ayrıntıları',
             deployments: 'Yerel dağıtımlar',
-            files: 'Model dosyaları',
+            files: 'Model merkezi',
             connectExternal: 'Harici hizmete bağlan',
             deploymentTemplates: 'vLLM başlatma komutu şablonları',
             engineProvider: 'Motor / Sağlayıcı',
@@ -1652,6 +1652,12 @@ const message = {
             timePerOutputTokenChart: 'Token arası gecikme',
         },
         gpu: {
+            systemMemory: 'Paylaşılan sistem belleği',
+            systemMemoryUsed: 'Kullanılan sistem belleği',
+            systemMemoryTotal: 'Toplam sistem belleği',
+            systemMemoryAvailable: 'Kullanılabilir sistem belleği',
+            systemMemoryHelper:
+                'CPU ve GPU tarafından paylaşılır. Yalnızca GPU kullanımını değil, tüm sistemin bellek kullanımını gösterir.',
             memoryActivity: 'Bellek okuma/yazma etkinliği',
             hotspotTemperature: 'Sıcak nokta sıcaklığı',
             fanRPM: 'Fan hızı (RPM)',
@@ -1687,7 +1693,6 @@ const message = {
             historyPeak: 'En yüksek',
             historyAggregated:
                 '{count} örnek {seconds} saniyelik aralıklarla toplandı; işlemler için yalnızca sayılar gösterilir.',
-            historyRaw: '{count} ham örnek.',
             historyMetricEmpty: 'Seçilen zaman aralığında bu metriğe ait veri yok',
             historyLoadFailed: 'Geçmiş yüklenemedi. Lütfen tekrar deneyin.',
             collectedAt: 'Toplanma zamanı',

frontend/src/lang/modules/zh-Hant.ts

@@ -1009,11 +1009,11 @@ const message = {
             onlineModel: '線上模型',
             instanceDeployment: '按執行個體部署',
             nvidiaDeployment: 'NVIDIA 最佳化部署',
-            engines: '推理引擎',
+            engines: '本機推理引擎',
             modelManagement: '模型管理',
             serviceDetails: '服務詳情',
             deployments: '本機部署',
-            files: '模型檔案',
+            files: '模型廣場',
             connectExternal: '接入外部服務',
             deploymentTemplates: 'vLLM 啟動命令範本',
             engineProvider: '引擎 / 供應商',
@@ -1555,6 +1555,11 @@ const message = {
             timePerOutputTokenChart: 'Token 間延遲',
         },
         gpu: {
+            systemMemory: '系統共享記憶體',
+            systemMemoryUsed: '系統已用記憶體',
+            systemMemoryTotal: '系統總記憶體',
+            systemMemoryAvailable: '系統可用記憶體',
+            systemMemoryHelper: 'CPU 與 GPU 共用，顯示整機記憶體使用情況，不代表 GPU 獨占用量。',
             memoryActivity: '顯示記憶體讀寫活躍率',
             hotspotTemperature: '熱點溫度',
             fanRPM: '風扇轉速（RPM）',
@@ -1589,7 +1594,6 @@ const message = {
             historyAverage: '平均值',
             historyPeak: '峰值',
             historyAggregated: '共 {count} 筆取樣，按 {seconds} 秒彙整；程序僅顯示統計數量。',
-            historyRaw: '共 {count} 筆原始取樣。',
             historyMetricEmpty: '所選時間範圍內沒有此指標的資料',
             historyLoadFailed: '歷史記錄載入失敗，請重試',
             collectedAt: '採集時間',

frontend/src/lang/modules/zh.ts

@@ -1022,11 +1022,11 @@ const message = {
             onlineModel: '在线模型',
             instanceDeployment: '按实例部署',
             nvidiaDeployment: 'NVIDIA 优化部署',
-            engines: '推理引擎',
+            engines: '本地推理引擎',
             modelManagement: '模型管理',
             serviceDetails: '服务详情',
             deployments: '本地部署',
-            files: '模型文件',
+            files: '模型广场',
             connectExternal: '接入外部服务',
             deploymentTemplates: 'vLLM 启动命令模板',
             engineProvider: '引擎 / 供应商',
@@ -1568,6 +1568,11 @@ const message = {
             timePerOutputTokenChart: 'Token 间延迟',
         },
         gpu: {
+            systemMemory: '系统共享内存',
+            systemMemoryUsed: '系统已用内存',
+            systemMemoryTotal: '系统总内存',
+            systemMemoryAvailable: '系统可用内存',
+            systemMemoryHelper: 'CPU 与 GPU 共用，显示整机内存使用情况，不代表 GPU 独占用量。',
             memoryActivity: '显存读写活跃率',
             hotspotTemperature: '热点温度',
             fanRPM: '风扇转速（RPM）',
@@ -1602,7 +1607,6 @@ const message = {
             historyAverage: '平均值',
             historyPeak: '峰值',
             historyAggregated: '共 {count} 条采样，按 {seconds} 秒聚合；进程仅展示统计数量。',
-            historyRaw: '共 {count} 条原始采样。',
             historyMetricEmpty: '所选时间范围内没有该指标的数据',
             historyLoadFailed: '历史记录加载失败，请重试',
             collectedAt: '采集时间',

frontend/src/views/ai/gpu/current/index.vue

@@ -79,6 +79,9 @@
                                                 >
                                                     {{ $t('commons.button.view') }}
                                                 </el-button>
+                                                <el-button type="primary" link @click="openHistory(item)">
+                                                    {{ $t('aiTools.gpu.history') }}
+                                                </el-button>
                                             </div>
                                         </div>
                                     </template>
@@ -97,7 +100,21 @@
                                             />
                                         </div>
                                         <div class="metric-item metric-primary metric-memory">
-                                            <span>{{ $t('aiTools.gpu.memoryUsed') }}</span>
+                                            <span>
+                                                {{
+                                                    $t(
+                                                        item.type !== 'ascend' && item.sharedMemory
+                                                            ? 'aiTools.gpu.systemMemory'
+                                                            : 'aiTools.gpu.memoryUsed',
+                                                    )
+                                                }}
+                                                <el-tooltip
+                                                    v-if="item.type !== 'ascend' && item.sharedMemory"
+                                                    :content="$t('aiTools.gpu.systemMemoryHelper')"
+                                                >
+                                                    <el-icon><InfoFilled /></el-icon>
+                                                </el-tooltip>
+                                            </span>
                                             <strong>{{ formatMemory(item.memUsed, item.memTotal) }}</strong>
                                             <el-progress
                                                 v-if="memoryPercentage(item) !== null"
@@ -191,6 +208,9 @@
                                                 >
                                                     {{ $t('commons.button.view') }}
                                                 </el-button>
+                                                <el-button type="primary" link @click="openHistory(item)">
+                                                    {{ $t('aiTools.gpu.history') }}
+                                                </el-button>
                                             </div>
                                         </div>
                                     </template>
@@ -321,6 +341,8 @@
 
 <script lang="ts" setup>
 import { computed, onMounted, ref } from 'vue';
+import { useRouter } from 'vue-router';
+import { InfoFilled } from '@element-plus/icons-vue';
 import { loadGPUInfo } from '@/api/modules/ai';
 import RouterMenu from '@/views/ai/gpu/index.vue';
 import { AI } from '@/api/interface/ai';
@@ -328,6 +350,7 @@ import i18n from '@/lang';
 import { MsgWarning } from '@/utils/message';
 import DeviceDetails, { type DetailSection } from './components/DeviceDetails.vue';
 
+const router = useRouter();
 const loading = ref();
 const processDrawerVisible = ref(false);
 const processDeviceKey = ref('');
@@ -574,10 +597,14 @@ const deviceDetailSections = (item: AcceleratorDevice): DetailSection[] => {
         },
         {
             id: 'memory',
-            title: t('aiTools.gpu.memory'),
+            title: t(item.sharedMemory ? 'aiTools.gpu.systemMemory' : 'aiTools.gpu.memory'),
             items: [
                 { label: t('aiTools.gpu.memoryActivity'), value: item.memoryActivity },
-                { label: t('aiTools.gpu.freeMemory'), value: item.memoryFree },
+                {
+                    label: t(item.sharedMemory ? 'aiTools.gpu.systemMemoryAvailable' : 'aiTools.gpu.freeMemory'),
+                    value: item.memoryFree,
+                    help: item.sharedMemory ? t('aiTools.gpu.systemMemoryHelper') : undefined,
+                },
                 { label: t('aiTools.gpu.memoryReserved'), value: item.memoryReserved },
             ],
         },
@@ -733,6 +760,26 @@ const openXPUProcesses = (deviceID: number) => {
     processDrawerVisible.value = true;
 };
 
+const stableDeviceID = (vendor: string, identifier: string | undefined, fallback: string) => {
+    return `${vendor}:${identifier && identifier.toUpperCase() !== 'N/A' ? identifier : fallback}`;
+};
+
+const openHistory = (item: AcceleratorDevice | AI.XpuInfo['xpu'][number]) => {
+    let deviceID: string;
+    if ('basic' in item) {
+        deviceID = stableDeviceID(
+            'xpu',
+            item.basic.uuid,
+            stableDeviceID('pci', item.basic.pciBdfAddress, String(item.basic.deviceID)),
+        );
+    } else if (item.type === 'ascend') {
+        deviceID = `${stableDeviceID('ascend', item.busID, String(item.npuIndex))}:${item.chipIndex}`;
+    } else {
+        deviceID = stableDeviceID(item.type, item.uuid, stableDeviceID('pci', item.busID, String(item.index)));
+    }
+    void router.push({ path: '/ai/gpu/history', query: { deviceID } });
+};
+
 const deviceTitle = (item: AcceleratorDevice) => {
     return item.type === 'ascend' ? `Chip ${item.chipIndex}` : `GPU ${item.index}`;
 };

frontend/src/views/ai/gpu/history/index.vue

@@ -4,45 +4,40 @@
         <template v-if="supported">
             <el-card class="history-toolbar">
                 <div class="flex flex-wrap gap-3 items-center">
-                    <el-date-picker
-                        v-model="timeRangeGlobal"
-                        type="datetimerange"
-                        style="max-width: 100%; width: 360px; flex-grow: 0"
-                        range-separator="-"
-                        :start-placeholder="$t('commons.search.timeStart')"
-                        :end-placeholder="$t('commons.search.timeEnd')"
-                        :shortcuts="shortcuts"
-                        :clearable="false"
-                        :size="isMobile ? 'small' : 'default'"
-                        @change="changeTimeRange"
-                    />
-                    <el-select style="max-width: 100%" class="p-w-300" v-model="selectedDevice" @change="search">
+                    <el-select style="max-width: 100%" class="p-w-400" v-model="selectedDevice" @change="search">
+                        <template #prefix>{{ $t('aiTools.gpu.deviceInfo') }}</template>
                         <el-option v-for="item in options" :key="item.value" :label="item.label" :value="item.value" />
                     </el-select>
                     <el-radio-group v-if="history && history.bucketSeconds > 0" v-model="aggregation" @change="search">
                         <el-radio-button value="avg">{{ $t('aiTools.gpu.historyAverage') }}</el-radio-button>
                         <el-radio-button value="max">{{ $t('aiTools.gpu.historyPeak') }}</el-radio-button>
                     </el-radio-group>
-                    <TableRefresh @search="refresh" />
+                    <div class="ml-auto flex flex-wrap gap-3 items-center">
+                        <el-date-picker
+                            v-model="timeRangeGlobal"
+                            type="datetimerange"
+                            class="history-time-range"
+                            style="max-width: 100%; width: 400px; flex-grow: 0"
+                            :prefix-icon="timeRangePrefix"
+                            range-separator="-"
+                            :start-placeholder="$t('commons.search.timeStart')"
+                            :end-placeholder="$t('commons.search.timeEnd')"
+                            :shortcuts="shortcuts"
+                            :clearable="false"
+                            @change="changeTimeRange"
+                        />
+                        <TableRefresh @search="refresh" />
+                    </div>
                 </div>
-                <div v-if="history?.sampleCount" class="history-summary">
+                <div v-if="history?.sampleCount && history.bucketSeconds > 0" class="history-summary">
                     {{
-                        history.bucketSeconds > 0
-                            ? $t('aiTools.gpu.historyAggregated', {
-                                  seconds: history.bucketSeconds,
-                                  count: history.sampleCount,
-                              })
-                            : $t('aiTools.gpu.historyRaw', { count: history.sampleCount })
+                        $t('aiTools.gpu.historyAggregated', {
+                            seconds: history.bucketSeconds,
+                            count: history.sampleCount,
+                        })
                     }}
                 </div>
             </el-card>
-            <el-alert
-                v-if="currentDevice?.legacy"
-                :title="$t('aiTools.gpu.legacyHistory')"
-                type="info"
-                :closable="false"
-                show-icon
-            />
             <el-alert
                 v-if="failed"
                 :title="$t('aiTools.gpu.historyLoadFailed')"
@@ -58,7 +53,15 @@
             <el-row v-else-if="history?.sampleCount" :gutter="12">
                 <el-col v-for="chart in charts" :key="chart.id" :xs="24" :md="12">
                     <el-card class="card-interval">
-                        <template #header>{{ chart.title }}</template>
+                        <template #header>
+                            {{ chart.title }}
+                            <el-tooltip
+                                v-if="chart.id === 'gpu-memory' && currentDevice?.sharedMemory"
+                                :content="$t('aiTools.gpu.systemMemoryHelper')"
+                            >
+                                <el-icon><InfoFilled /></el-icon>
+                            </el-tooltip>
+                        </template>
                         <v-charts :id="chart.id" height="320px" type="line" :option="chart.option" :dataZoom="true" />
                     </el-card>
                 </el-col>
@@ -90,7 +93,9 @@
 </template>
 
 <script setup lang="ts">
-import { ref, computed, onMounted, onBeforeUnmount } from 'vue';
+import { ref, computed, watch, onBeforeUnmount, h } from 'vue';
+import { useRoute } from 'vue-router';
+import { InfoFilled } from '@element-plus/icons-vue';
 import { loadGPUMonitor, getGPUOptions } from '@/api/modules/ai';
 import RouterMenu from '@/views/ai/gpu/index.vue';
 import { shortcuts } from '@/utils/shortcuts';
@@ -98,6 +103,7 @@ import i18n from '@/lang';
 import { AI } from '@/api/interface/ai';
 import { useGlobalStore } from '@/composables/useGlobalStore';
 
+const route = useRoute();
 const { isMobile } = useGlobalStore();
 const loading = ref(false);
 const failed = ref(false);
@@ -108,6 +114,7 @@ const aggregation = ref<'avg' | 'max'>('avg');
 const history = ref<AI.MonitorGPUData>();
 const currentDevice = computed(() => options.value.find((item) => item.value === selectedDevice.value));
 const timeRangeGlobal = ref<[Date, Date]>([new Date(new Date().setHours(0, 0, 0, 0)), new Date()]);
+const timeRangePrefix = () => h('span', i18n.global.t('commons.table.date'));
 let followNow = true;
 let requestID = 0;
 
@@ -124,10 +131,18 @@ const charts = computed(() => {
         },
         {
             id: 'gpu-memory',
-            title: 'memory',
+            title: currentDevice.value?.sharedMemory ? 'systemMemory' : 'memory',
             series: [
-                { key: 'memoryUsed', label: 'memoryUsed', unit: 'MiB' },
-                { key: 'memoryTotal', label: 'memoryTotal', unit: 'MiB' },
+                {
+                    key: 'memoryUsed',
+                    label: currentDevice.value?.sharedMemory ? 'systemMemoryUsed' : 'memoryUsed',
+                    unit: 'MiB',
+                },
+                {
+                    key: 'memoryTotal',
+                    label: currentDevice.value?.sharedMemory ? 'systemMemoryTotal' : 'memoryTotal',
+                    unit: 'MiB',
+                },
                 { key: 'memoryPercent', label: 'percent', unit: '%', percent: true },
             ],
         },
@@ -338,12 +353,16 @@ const loadOptions = async () => {
         const response = await getGPUOptions();
         if (id !== requestID) return;
         supported.value = response.data.supported;
-        options.value = (supported.value ? response.data.chartHide || [] : []).map((item) => ({
-            ...item,
-            value: item.deviceID || `legacy:${item.productName}`,
-            label: `${item.productName} · ${item.deviceID || i18n.global.t('aiTools.gpu.legacyDevice')}`,
-        }));
-        if (!currentDevice.value) selectedDevice.value = options.value[0]?.value || '';
+        options.value = (supported.value ? response.data.chartHide || [] : [])
+            .filter((item) => item.deviceID && !item.legacy)
+            .map((item) => ({
+                ...item,
+                value: item.deviceID!,
+                label: item.productName
+                    .replace(/^(\d+) - /, `${item.type === 'xpu' ? 'XPU' : 'GPU'} $1 · `)
+                    .replace(' - ', ' · '),
+            }));
+        if (!currentDevice.value && !route.query.deviceID) selectedDevice.value = options.value[0]?.value || '';
         await search();
     } catch {
         if (id === requestID) failed.value = true;
@@ -401,7 +420,15 @@ const loadProcessType = (val: string) => {
     return val;
 };
 
-onMounted(loadOptions);
+watch(
+    () => [route.path, route.query.deviceID],
+    () => {
+        if (route.path !== '/ai/gpu/history') return;
+        if (typeof route.query.deviceID === 'string') selectedDevice.value = route.query.deviceID;
+        void loadOptions();
+    },
+    { immediate: true },
+);
 onBeforeUnmount(() => {
     requestID++;
 });
@@ -412,6 +439,14 @@ onBeforeUnmount(() => {
     margin: 7px 0 12px;
     --el-card-padding: 12px;
 }
+.history-toolbar :deep(.history-time-range .el-range__icon) {
+    width: auto;
+    flex-shrink: 0;
+    margin-left: 0;
+    margin-right: 6px;
+    font-style: normal;
+    white-space: nowrap;
+}
 .history-summary {
     margin-top: 12px;
     color: var(--el-text-color-secondary);

frontend/src/views/ai/model/components/deployment-manage/index.vue

@@ -65,7 +65,7 @@
                     ? { highlightDiff: 210 }
                     : view === TaskLog
                       ? { width: '70%' }
-                      : actionName === 'config'
+                      : ['config', 'monitor'].includes(actionName)
                         ? {}
                         : { dialog: true }
         "

frontend/src/views/ai/model/index.vue

@@ -357,8 +357,8 @@ const isManagement = computed(
     () => isCurrentPage.value && isLocal && ['vllm', 'tensorrt'].includes(String(route.query.tab)),
 );
 const tabs = computed(() => [
-    { label: i18n.global.t('aiTools.model.services'), path: '/ai/model/account' },
     { label: i18n.global.t('aiTools.model.files'), path: '/ai/model/files' },
+    { label: i18n.global.t('aiTools.model.services'), path: '/ai/model/account' },
     { label: i18n.global.t('aiTools.model.engines'), path: '/ai/model/local' },
 ]);
 const downloaderModules = import.meta.glob('/src/xpack/views/model-downloader/index.vue');

frontend/src/views/ai/model/tensorrt/operate/index.vue

@@ -55,10 +55,16 @@
                     <el-form-item :label="$t('aiTools.tensorRT.modelDir')" prop="modelDir">
                         <el-input v-model="tensorRTLLM.modelDir">
                             <template #prepend>
-                                <el-button icon="Folder" @click="modelDirRef.acceptParams({ dir: true })" />
+                                <el-button icon="Folder" @click="selectModelDir" />
                             </template>
                         </el-input>
                     </el-form-item>
+                </div>
+            </details>
+            <GPUConfig ref="gpuRef" v-model="tensorRTLLM.gpu" />
+            <details open class="form-section">
+                <summary class="form-section-header">{{ $t('runtime.runScript') }}</summary>
+                <div class="form-section-body">
                     <el-form-item :label="$t('runtime.runScript')" prop="command">
                         <el-input v-model="tensorRTLLM.command"></el-input>
                         <span class="input-help">
@@ -67,7 +73,6 @@
                     </el-form-item>
                 </div>
             </details>
-            <GPUConfig ref="gpuRef" v-model="tensorRTLLM.gpu" />
             <details open class="form-section">
                 <summary class="form-section-header">{{ $t('app.advanced') }}</summary>
                 <div class="form-section-body">
@@ -103,7 +108,7 @@ import { ElForm, FormInstance } from 'element-plus';
 import { createTensorRTLLM, updateTensorRTLLM } from '@/api/modules/ai';
 import { MsgSuccess } from '@/utils/message';
 import { useGlobalStore } from '@/composables/useGlobalStore';
-const { isFxplay, isMobile } = useGlobalStore();
+const { isFxplay, isMobile, isProductPro } = useGlobalStore();
 
 const loading = ref(false);
 const gpuRef = ref<InstanceType<typeof GPUConfig>>();
@@ -189,6 +194,24 @@ const handleClose = () => {
     emit('close');
 };
 
+const downloaderModules = import.meta.glob<{
+    getModelDownloaderSettings: () => Promise<{ data: { modelDir: string } }>;
+}>('/src/xpack/api/modules/model-downloader.ts');
+const selectModelDir = async () => {
+    let path = tensorRTLLM.value.modelDir;
+    const loadDownloader = Object.values(downloaderModules)[0];
+    if (isProductPro.value && loadDownloader) {
+        try {
+            const { getModelDownloaderSettings } = await loadDownloader();
+            const { data } = await getModelDownloaderSettings();
+            path = data.modelDir;
+        } catch {
+            path = '/';
+        }
+    }
+    modelDirRef.value.acceptParams({ dir: true, path });
+};
+
 const getModelDir = (path: string) => {
     tensorRTLLM.value.modelDir = path;
 };

frontend/src/views/home/status/gpu.vue

@@ -18,11 +18,18 @@
                     <span v-for="model in modelGroups" :key="model.name">{{ model.name }} × {{ model.count }}</span>
                 </div>
             </div>
-            <div class="summary-memory">
+            <div v-if="dedicatedMemoryDevices.length" class="summary-memory">
                 <span class="metric-label">{{ $t('aiTools.gpu.memoryTotal') }}</span>
                 <strong>{{ formatMemory(totalMemory) }}</strong>
                 <span class="metric-label">{{ $t('home.used') }} {{ formatMemory(usedMemory) }}</span>
             </div>
+            <div v-if="sharedMemoryDevice" class="summary-memory" :title="$t('aiTools.gpu.systemMemoryHelper')">
+                <span class="metric-label">{{ $t('aiTools.gpu.systemMemory') }}</span>
+                <strong>{{ formatMemory(sharedMemoryDevice.memoryTotal) }}</strong>
+                <span class="metric-label">
+                    {{ $t('home.used') }} {{ formatMemory(sharedMemoryDevice.memoryUsed) }}
+                </span>
+            </div>
         </div>
 
         <div class="gpu-grid">
@@ -59,7 +66,12 @@
                 </div>
                 <div class="device-metric">
                     <div class="metric-heading">
-                        <span class="metric-label">{{ $t('aiTools.gpu.memoryUsed') }}</span>
+                        <span
+                            class="metric-label"
+                            :title="device.sharedMemory ? $t('aiTools.gpu.systemMemoryHelper') : undefined"
+                        >
+                            {{ $t(device.sharedMemory ? 'aiTools.gpu.systemMemory' : 'aiTools.gpu.memoryUsed') }}
+                        </span>
                         <strong>
                             {{ formatMemory(device.memoryUsed) }}
                             <span class="metric-total">/ {{ formatMemory(device.memoryTotal) }}</span>
@@ -139,6 +151,7 @@ const devices = computed(() =>
         ...(props.current.gpuData || []).map((item) => ({
             key: `gpu-${item.type}-${item.busID || item.index}`,
             type: 'GPU',
+            sharedMemory: !!item.sharedMemory,
             card: `${item.type}-${item.busID || item.index}`,
             id: `GPU ${item.index}`,
             name: item.productName || 'GPU',
@@ -157,6 +170,7 @@ const devices = computed(() =>
         ...(props.current.npuData || []).map((item) => ({
             key: `npu-${item.npuIndex}-${item.chipIndex}`,
             type: 'NPU',
+            sharedMemory: false,
             card: item.npuIndex,
             id: `NPU ${item.npuIndex} · Chip ${item.chipIndex}`,
             name: item.productName || 'NPU',
@@ -176,6 +190,7 @@ const devices = computed(() =>
         ...(props.current.xpuData || []).map((item) => ({
             key: `xpu-${item.deviceID}`,
             type: 'XPU',
+            sharedMemory: false,
             card: item.deviceID,
             id: `XPU ${item.deviceID}`,
             name: item.deviceName || 'XPU',
@@ -222,15 +237,17 @@ const modelGroups = computed(() => {
     }
     return Array.from(models, ([name, cards]) => ({ name, count: cards.size }));
 });
+const sharedMemoryDevice = computed(() => devices.value.find((item) => item.sharedMemory));
+const dedicatedMemoryDevices = computed(() => devices.value.filter((item) => !item.sharedMemory));
 const totalMemory = computed(() =>
-    devices.value.some((item) => item.memoryTotal === null)
+    dedicatedMemoryDevices.value.some((item) => item.memoryTotal === null)
         ? null
-        : devices.value.reduce((sum, item) => sum + item.memoryTotal!, 0),
+        : dedicatedMemoryDevices.value.reduce((sum, item) => sum + item.memoryTotal!, 0),
 );
 const usedMemory = computed(() =>
-    devices.value.some((item) => item.memoryUsed === null)
+    dedicatedMemoryDevices.value.some((item) => item.memoryUsed === null)
         ? null
-        : devices.value.reduce((sum, item) => sum + item.memoryUsed!, 0),
+        : dedicatedMemoryDevices.value.reduce((sum, item) => sum + item.memoryUsed!, 0),
 );
 </script>