Fix: VRAM/Memory-Leiste nutzt echtes GTT-Budget (~124GB) statt 16GB-Fallback

Cockpit VRAM-HUD las einen hardcodierten 16GB-APU-Fallback als Pool-Kapazitaet -> Leiste zeigte z.B. "11.3/16GB". Jetzt: gtt_total aus /api/system/status (unified memory, ~124GB). Label zeigt zusaetzlich die reale GTT-Belegung (inkl. KV) ehrlich an. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-27 12:43:26 +02:00
parent 6758bbfbd9
commit 31d5e5d727
5 changed files with 410 additions and 405 deletions
@@ -7,8 +7,8 @@
    <link rel="manifest" href="/manifest.webmanifest" />
    <link rel="icon" type="image/svg+xml" href="/favicon.svg" />
    <title>Mission Control 2.0</title>
-    <script type="module" crossorigin src="/assets/index-Bb8BpQr4.js"></script>
-    <link rel="stylesheet" crossorigin href="/assets/index-bw6S4bF_.css">
+    <script type="module" crossorigin src="/assets/index-OBh42VZw.js"></script>
+    <link rel="stylesheet" crossorigin href="/assets/index-BA8F-FbW.css">
  </head>
  <body>
    <div id="root"></div>
@@ -1,7 +1,7 @@
 import { useState, useRef, useCallback } from "react"
 import { Download, Trash2, Edit3, Activity, HardDrive, X, Check, Copy, Zap, Bot } from "lucide-react"
 import { api, type ModelInfo } from "@/lib/api"
-import { useModels, useRouting, useConnect, useUpdates, useHermesBrain, useQueryClient, qk } from "@/lib/queries"
+import { useModels, useRouting, useConnect, useUpdates, useHermesBrain, useSystemStatus, useQueryClient, qk } from "@/lib/queries"
 import { useDialog } from "@/lib/useDialog"
 import { CapsChips } from "@/components/CapsChips"
 import { cn } from "@/lib/utils"
@@ -16,6 +16,7 @@ export function Cockpit() {
  const { data: connectData } = useConnect()
  const { data: updates } = useUpdates(4_000)
  const { data: brain } = useHermesBrain()
+  const { data: sysStatus } = useSystemStatus()
  const { showAlert, showConfirm, showPrompt, dialogElement } = useDialog()
  const models = modelsResp?.models ?? []
  const running = modelsResp?.running ?? []
@@ -233,11 +234,15 @@ export function Cockpit() {
    )
  }

-  // VRAM calculation: APU fallback 16GB if sysfs returns 0
+  // Speicher-Pool = echtes GTT-Budget (unified memory, ~124 GB), nicht der alte 16-GB-Fallback.
  const runningWithInfo = models.filter((m) => running.includes(m.name))
  const totalRunningSize = runningWithInfo.reduce((acc, m) => acc + (m.size_bytes || 0), 0)
-  const virtualMax = 16 * 1024 ** 3 // 16 GB Default
-  const capacity = totalRunningSize > virtualMax ? totalRunningSize * 1.2 : virtualMax
+  const gttTotal = sysStatus?.gpu?.gtt_total || sysStatus?.gpu?.vram_total || 0
+  const gttUsed = sysStatus?.gpu?.gtt_used || 0
+  const virtualMax = 16 * 1024 ** 3 // Fallback, falls sysfs nichts liefert
+  const capacity = gttTotal > 2 * 1024 ** 3
+    ? gttTotal
+    : (totalRunningSize > virtualMax ? totalRunningSize * 1.2 : virtualMax)

  // Find model by role
  const getModelForRole = (role: string) => models.find((m) => m.role === role)
@@ -270,7 +275,7 @@ export function Cockpit() {
          </div>
          <div className="flex items-center gap-3">
            <span className="text-[10px] font-mono text-muted-foreground">
-              Llama Swap VRAM-Pool: {fmtSize(totalRunningSize)} / {fmtSize(capacity)} geladen
+              Speicher-Pool: {fmtSize(totalRunningSize)} Gewichte{gttUsed > 0 ? ` · ${fmtSize(gttUsed)} real belegt (inkl. KV)` : ""} / {fmtSize(capacity)}
            </span>
            {running.length > 0 && (
              <button