Merge pull request 'feat: add Dashboard tab with live GPU overview' (#5) from feat/dashboard-tab into main

Reviewed-on: #5
2026-09-02 14:57:54 +00:00
commit 0320cea895
@@ -1,8 +1,10 @@
 import { useGpu } from "./hooks/useGpu";
 import { useCurve } from "./hooks/useCurve";
 import { useMonitor } from "./hooks/useMonitor";
+import { useDashboard } from "./hooks/useDashboard";
 import { StatusBar } from "./components/Monitor/StatusBar";
 import { LiveMonitor } from "./components/Monitor/LiveMonitor";
+import { Dashboard } from "./components/Dashboard/Dashboard";
 import { CurveEditor } from "./components/CurveEditor/CurveEditor";
 import { PointTable } from "./components/PointTable/PointTable";
 import { PerformancePanel } from "./components/Limits/PerformancePanel";
@@ -89,12 +91,13 @@ function MainApp({
  const gpuInfo = useGpu();
  const { curve, wsStatus: curveWsStatus } = useCurve();
  const { monitor, monitorHistory, wsStatus: monitorWsStatus } = useMonitor();
+  const { dashboard, loading: dashboardLoading } = useDashboard();
  const { setCurve, activeProfile, setActiveProfile, selectedGpuIndex } =
    useCurveStore();

-  const [activeTab, setActiveTab] = useState<"curve" | "performance" | "fans">(
-    "curve",
-  );
+  const [activeTab, setActiveTab] = useState<
+    "dashboard" | "curve" | "performance" | "fans"
+  >("dashboard");
  const [fanState, setFanState] = useState<FanState | null>(null);
  const [activeDomain, setActiveDomain] = useState<"gpu" | "memory">("gpu");
  const [isProfileOpen, setIsProfileOpen] = useState(false);
@@ -168,6 +171,12 @@ function MainApp({
        {/* Tab Header and Profile Selector */}
        <div className="flex justify-between items-end border-b border-zinc-800 pb-2">
          <div className="flex gap-6">
+            <button
+              onClick={() => setActiveTab("dashboard")}
+              className={`text-lg font-medium pb-2 -mb-[9px] border-b-2 transition-colors ${activeTab === "dashboard" ? "border-pink-500 text-zinc-100" : "border-transparent text-zinc-500 hover:text-zinc-300"}`}
+            >
+              Dashboard
+            </button>
            <button
              onClick={() => setActiveTab("curve")}
              className={`text-lg font-medium pb-2 -mb-[9px] border-b-2 transition-colors ${activeTab === "curve" ? "border-pink-500 text-zinc-100" : "border-transparent text-zinc-500 hover:text-zinc-300"}`}
@@ -222,7 +231,14 @@ function MainApp({

        {/* Main content area */}
        <div className="flex flex-col gap-4 w-full">
-          {activeTab === "curve" ? (
+          {activeTab === "dashboard" ? (
+            <Dashboard
+              monitor={monitor}
+              history={monitorHistory}
+              dashboard={dashboard}
+              loading={dashboardLoading}
+            />
+          ) : activeTab === "curve" ? (
            <>
              <div className="flex gap-4 items-stretch w-full">
                <div className="flex-1 min-w-0 flex flex-col">
@@ -7,6 +7,7 @@ import type {
  ProfileData,
  FanState,
  FanPoint,
+  DashboardInfo,
 } from "../types";

 export class ApiError extends Error {
@@ -108,6 +109,7 @@ export const api = {

  gpus: () => get<GpuInfo[]>("/gpus"),
  gpu: (gpuIndex: number) => get<GpuInfo>("/gpu", gpuIndex),
+  dashboard: (gpuIndex: number) => get<DashboardInfo>("/dashboard", gpuIndex),
  curve: (gpuIndex: number) => get<CurveState>("/curve", gpuIndex),
  ranges: (gpuIndex: number) =>
    get<Record<string, { min_khz: number; max_khz: number }>>(
@@ -0,0 +1,318 @@
+import { Loader } from "lucide-react";
+import { GaugeCard } from "../Monitor/GaugeCard";
+import { fmt } from "../../utils/units";
+import type { MonitoringSample, DashboardInfo } from "../../types";
+
+interface Props {
+  monitor: MonitoringSample | null;
+  history: MonitoringSample[];
+  dashboard: DashboardInfo | null;
+  loading: boolean;
+}
+
+function pluck<K extends keyof MonitoringSample>(
+  history: MonitoringSample[],
+  key: K,
+): number[] {
+  return history.map((s) => (s[key] as number | null) ?? 0);
+}
+
+function pcieLink(
+  width: number | null | undefined,
+  gen: number | null | undefined,
+): string | null {
+  if (width == null && gen == null) return null;
+  return `x${width ?? "?"} Gen${gen ?? "?"}`;
+}
+
+/** A static label/value pair. Renders nothing when the value is unavailable. */
+function InfoItem({
+  label,
+  value,
+  accent,
+}: {
+  label: string;
+  value: string | null | undefined;
+  accent?: string;
+}) {
+  if (value == null || value === "") return null;
+  return (
+    <div className="flex flex-col gap-0.5 min-w-0">
+      <span className="text-xs text-zinc-500">{label}</span>
+      <span
+        className={`text-sm font-mono truncate ${accent ?? "text-zinc-200"}`}
+        title={value}
+      >
+        {value}
+      </span>
+    </div>
+  );
+}
+
+function TempRow({ label, value }: { label: string; value: number | null }) {
+  if (value == null) return null;
+  return (
+    <div className="flex items-baseline justify-between gap-2">
+      <span className="text-xs text-zinc-500">{label}</span>
+      <span className="text-lg font-mono font-semibold text-orange-400">
+        {fmt.celsius(value)}
+      </span>
+    </div>
+  );
+}
+
+export function Dashboard({ monitor, history, dashboard: d, loading }: Props) {
+  if (loading && !d) {
+    return (
+      <div className="bg-zinc-900 rounded-lg p-10 text-center text-zinc-500 flex items-center justify-center gap-2">
+        <Loader size={18} className="animate-spin" /> Loading dashboard…
+      </div>
+    );
+  }
+
+  const throttleLabel = monitor?.throttle_reasons_label ?? null;
+  const isThrottled = throttleLabel != null && throttleLabel !== "No";
+
+  const memUsed = monitor?.mem_used_mib ?? null;
+  const memTotal = monitor?.mem_total_mib ?? null;
+  const memLabel =
+    memUsed != null && memTotal != null
+      ? `${memUsed.toFixed(0)} / ${memTotal.toFixed(0)} MiB`
+      : "—";
+
+  const powerLimit = d?.power_limit_w ?? null;
+  const powerRange =
+    d?.power_min_limit_w != null && d?.power_max_limit_w != null
+      ? `${d.power_min_limit_w}–${d.power_max_limit_w} W`
+      : null;
+
+  return (
+    <div className="flex flex-col gap-4 w-full">
+      {/* ── Critical row ─────────────────────────────────────────────── */}
+      <div className="grid grid-cols-1 md:grid-cols-3 gap-4">
+        {/* Throttling */}
+        <div
+          className={`rounded-lg p-4 border ${
+            isThrottled
+              ? "bg-amber-500/10 border-amber-500/30"
+              : "bg-emerald-500/10 border-emerald-500/30"
+          }`}
+        >
+          <div className="text-xs text-zinc-500 uppercase tracking-wider font-semibold">
+            Throttling
+          </div>
+          <div
+            className={`text-2xl font-semibold mt-1 break-words ${
+              isThrottled ? "text-amber-400" : "text-emerald-400"
+            }`}
+          >
+            {throttleLabel ?? "—"}
+          </div>
+        </div>
+
+        {/* GPU Voltage */}
+        <div className="bg-zinc-900 rounded-lg p-4 border border-zinc-800">
+          <div className="text-xs text-zinc-500 uppercase tracking-wider font-semibold">
+            GPU Voltage
+          </div>
+          <div className="text-2xl font-mono font-semibold mt-1 text-violet-400">
+            {fmt.mv(monitor?.voltage_mv)}
+          </div>
+        </div>
+
+        {/* Temperature */}
+        <div className="bg-zinc-900 rounded-lg p-4 border border-zinc-800">
+          <div className="text-xs text-zinc-500 uppercase tracking-wider font-semibold">
+            Temperature
+          </div>
+          <div className="flex flex-col gap-1 mt-1">
+            <TempRow label="GPU" value={monitor?.temp_c ?? null} />
+            <TempRow label="VRAM" value={monitor?.mem_temp_c ?? null} />
+          </div>
+        </div>
+      </div>
+
+      {/* ── Live Monitor ─────────────────────────────────────────────── */}
+      <div className="bg-zinc-900 rounded-lg p-4">
+        <div className="text-xs text-zinc-500 uppercase tracking-wider font-semibold pb-3 border-b border-zinc-800">
+          Live Monitor
+        </div>
+        <div className="grid grid-cols-2 md:grid-cols-3 lg:grid-cols-4 gap-3 mt-3">
+          <GaugeCard
+            label="Core Clock"
+            value={fmt.mhz(monitor?.clock_mhz)}
+            history={pluck(history, "clock_mhz")}
+            color="#34d399"
+            max={d?.max_graphics_clock_mhz ?? 3000}
+          />
+          <GaugeCard
+            label="Mem Clock"
+            value={fmt.mhz(monitor?.mem_clock_mhz)}
+            history={pluck(history, "mem_clock_mhz")}
+            color="#67e8f9"
+            max={d?.max_memory_clock_mhz ?? 20000}
+          />
+          <GaugeCard
+            label="Voltage"
+            value={fmt.mv(monitor?.voltage_mv)}
+            history={pluck(history, "voltage_mv")}
+            color="#a78bfa"
+            max={1100}
+          />
+          <GaugeCard
+            label="Power Draw"
+            value={fmt.watts(monitor?.power_w)}
+            history={pluck(history, "power_w")}
+            color="#f472b6"
+            max={powerLimit != null && powerLimit > 0 ? powerLimit : 600}
+          />
+          <GaugeCard
+            label="GPU Util"
+            value={fmt.pct(monitor?.gpu_util_pct)}
+            history={pluck(history, "gpu_util_pct")}
+            color="#facc15"
+            max={100}
+          />
+          <GaugeCard
+            label="Mem Util"
+            value={fmt.pct(monitor?.mem_util_pct)}
+            history={pluck(history, "mem_util_pct")}
+            color="#38bdf8"
+            max={100}
+          />
+          <GaugeCard
+            label="Fan Speed"
+            value={fmt.pct(monitor?.fan_pct)}
+            history={pluck(history, "fan_pct")}
+            color="#fb923c"
+            max={100}
+          />
+          <GaugeCard
+            label="P-State"
+            value={monitor?.pstate_label ?? "Unknown"}
+            history={pluck(history, "pstate")}
+            color="#a8a29e"
+            max={15}
+          />
+          <GaugeCard
+            label="VRAM Used"
+            value={memLabel}
+            history={pluck(history, "mem_used_mib")}
+            color="#a78bfa"
+            max={memTotal ?? 32768}
+          />
+        </div>
+      </div>
+
+      {/* ── GPU Information ──────────────────────────────────────────── */}
+      <div className="bg-zinc-900 rounded-lg p-4">
+        <div className="text-xs text-zinc-500 uppercase tracking-wider font-semibold pb-3 border-b border-zinc-800">
+          GPU Information
+        </div>
+        <div className="grid grid-cols-2 md:grid-cols-3 lg:grid-cols-4 gap-x-6 gap-y-3 mt-3">
+          <InfoItem label="Driver" value={d?.driver_version} />
+          <InfoItem label="VBIOS" value={d?.vbios_version} />
+          <InfoItem
+            label="CUDA Cores"
+            value={d?.cuda_cores != null ? String(d.cuda_cores) : null}
+          />
+          <InfoItem label="CUDA Compute" value={d?.cuda_compute_capability} />
+          <InfoItem
+            label="ROP Count"
+            value={d?.rop_count != null ? String(d.rop_count) : null}
+          />
+          <InfoItem
+            label="VRAM Type"
+            value={d?.vram_type}
+            accent="text-cyan-400"
+          />
+          <InfoItem label="VRAM Total" value={fmt.bytes(d?.vram_total_bytes)} />
+          <InfoItem
+            label="Bus Width"
+            value={
+              d?.memory_bus_width != null ? `${d.memory_bus_width}-bit` : null
+            }
+          />
+          <InfoItem
+            label="CPU Accessible VRAM"
+            value={fmt.bytes(d?.bar1_total_bytes)}
+          />
+          <InfoItem
+            label="Resizable BAR"
+            value={
+              d?.resizable_bar == null
+                ? null
+                : d.resizable_bar
+                  ? "Enabled"
+                  : "Disabled"
+            }
+            accent={
+              d?.resizable_bar == null
+                ? undefined
+                : d.resizable_bar
+                  ? "text-emerald-400"
+                  : "text-zinc-400"
+            }
+          />
+          <InfoItem
+            label="PCIe Link"
+            value={pcieLink(
+              monitor?.pcie_link_width,
+              monitor?.pcie_link_generation,
+            )}
+            accent="text-cyan-400"
+          />
+          <InfoItem
+            label="PCIe Max"
+            value={pcieLink(
+              d?.pcie_max_link_width,
+              d?.pcie_max_link_generation,
+            )}
+          />
+          <InfoItem
+            label="Max Core Clock"
+            value={fmt.mhz(d?.max_graphics_clock_mhz)}
+          />
+          <InfoItem
+            label="Max Mem Clock"
+            value={fmt.mhz(d?.max_memory_clock_mhz)}
+          />
+          <InfoItem
+            label="Power Limit"
+            value={powerLimit != null ? `${powerLimit} W` : null}
+          />
+          <InfoItem label="Power Range" value={powerRange} />
+          <InfoItem
+            label="Temp Slowdown"
+            value={
+              d?.temp_slowdown_c != null ? `${d.temp_slowdown_c} °C` : null
+            }
+          />
+          <InfoItem
+            label="Temp Shutdown"
+            value={
+              d?.temp_shutdown_c != null ? `${d.temp_shutdown_c} °C` : null
+            }
+          />
+          <InfoItem
+            label="Persistence Mode"
+            value={
+              d?.persistence_mode == null
+                ? null
+                : d.persistence_mode
+                  ? "Enabled"
+                  : "Disabled"
+            }
+          />
+          <InfoItem
+            label="Fans"
+            value={d?.num_fans != null ? String(d.num_fans) : null}
+          />
+          <InfoItem label="Serial" value={d?.serial} />
+          <InfoItem label="Board Part No." value={d?.board_part_number} />
+          <InfoItem label="UUID" value={d?.uuid} />
+        </div>
+      </div>
+    </div>
+  );
+}
@@ -0,0 +1,47 @@
+import { useEffect, useState } from "react";
+import { api } from "../api/client";
+import { useCurveStore } from "../store/curveStore";
+import type { DashboardInfo } from "../types";
+
+interface DashboardState {
+  gpuIndex: number;
+  data: DashboardInfo | null;
+  done: boolean;
+}
+
+/**
+ * Fetch static GPU info for the Dashboard tab.
+ * Re-fetches when the selected GPU changes.
+ */
+export function useDashboard() {
+  const { selectedGpuIndex } = useCurveStore();
+  const [state, setState] = useState<DashboardState>({
+    gpuIndex: -1,
+    data: null,
+    done: false,
+  });
+
+  useEffect(() => {
+    let cancelled = false;
+    api
+      .dashboard(selectedGpuIndex)
+      .then((data) => {
+        if (!cancelled)
+          setState({ gpuIndex: selectedGpuIndex, data, done: true });
+      })
+      .catch((err) => {
+        console.error("Failed to load dashboard info:", err);
+        if (!cancelled)
+          setState({ gpuIndex: selectedGpuIndex, data: null, done: true });
+      });
+    return () => {
+      cancelled = true;
+    };
+  }, [selectedGpuIndex]);
+
+  const isCurrent = state.gpuIndex === selectedGpuIndex;
+  return {
+    dashboard: isCurrent ? state.data : null,
+    loading: !isCurrent || !state.done,
+  };
+}
@@ -8,7 +8,7 @@ export interface VFPoint {
  delta_mhz: number;
  effective_freq_khz: number;
  effective_freq_mhz: number;
-  domain: 'gpu' | 'memory';
+  domain: "gpu" | "memory";
 }

 export interface CurveState {
@@ -34,6 +34,46 @@ export interface MonitoringSample {
  gpu_util_pct: number | null;
  mem_util_pct: number | null;
  mem_clock_mhz: number | null;
+  throttle_reasons: number | null;
+  throttle_reasons_label: string | null;
+  pcie_link_width: number | null;
+  pcie_link_generation: number | null;
+  mem_temp_c: number | null;
+}
+
+export interface DashboardInfo {
+  name: string | null;
+  index: number;
+  driver_version: string | null;
+  vbios_version: string | null;
+  serial: string | null;
+  board_part_number: string | null;
+  uuid: string | null;
+  cuda_cores: number | null;
+  cuda_compute_capability: string | null;
+  rop_count: number | null;
+  vram_total_bytes: number | null;
+  vram_type: string | null;
+  memory_bus_width: number | null;
+  bar1_total_bytes: number | null;
+  bar1_used_bytes: number | null;
+  resizable_bar: boolean | null;
+  pcie_link_width: number | null;
+  pcie_link_generation: number | null;
+  pcie_max_link_width: number | null;
+  pcie_max_link_generation: number | null;
+  max_graphics_clock_mhz: number | null;
+  max_memory_clock_mhz: number | null;
+  max_video_clock_mhz: number | null;
+  power_limit_w: number | null;
+  power_min_limit_w: number | null;
+  power_max_limit_w: number | null;
+  persistence_mode: boolean | null;
+  num_fans: number | null;
+  supported_throttle_reasons: number | null;
+  temp_slowdown_c: number | null;
+  temp_shutdown_c: number | null;
+  temp_gpu_max_c: number | null;
 }

 export interface GpuInfo {
@@ -73,7 +113,7 @@ export interface FanPoint {

 export interface FanState {
  fan_pct: number | null;
-  fan_mode: 'auto' | 'curve' | null;
+  fan_mode: "auto" | "curve" | null;
  min_fan_pct: number | null;
  max_fan_pct: number | null;
  curve: FanPoint[] | null;
@@ -1,16 +1,24 @@
 export const fmt = {
  mhz: (v: number | null | undefined, decimals = 0) =>
-    v == null ? '—' : `${v.toFixed(decimals)} MHz`,
+    v == null ? "—" : `${v.toFixed(decimals)} MHz`,
  mv: (v: number | null | undefined, decimals = 0) =>
-    v == null ? '—' : `${v.toFixed(decimals)} mV`,
+    v == null ? "—" : `${v.toFixed(decimals)} mV`,
  celsius: (v: number | null | undefined, decimals = 0) =>
-    v == null ? '—' : `${v.toFixed(decimals)} °C`,
+    v == null ? "—" : `${v.toFixed(decimals)} °C`,
  watts: (v: number | null | undefined, decimals = 0) =>
-    v == null ? '—' : `${v.toFixed(decimals)} W`,
+    v == null ? "—" : `${v.toFixed(decimals)} W`,
  pct: (v: number | null | undefined, decimals = 0) =>
-    v == null ? '—' : `${v.toFixed(decimals)}%`,
+    v == null ? "—" : `${v.toFixed(decimals)}%`,
  mib: (v: number | null | undefined) =>
-    v == null ? '—' : `${v.toFixed(0)} MiB`,
+    v == null ? "—" : `${v.toFixed(0)} MiB`,
+  // Human-readable byte size (auto-scales to MiB / GiB).
+  bytes: (v: number | null | undefined) => {
+    if (v == null) return "—";
+    const gib = v / 1024 ** 3;
+    if (gib >= 1) return `${gib.toFixed(2)} GiB`;
+    const mib = v / 1024 ** 2;
+    return `${mib.toFixed(0)} MiB`;
+  },
 };

 export const conv = {
@@ -1,13 +1,24 @@
-from .gpu import get_gpu, discover_gpus
-from .vfcurve import read_curve, read_clock_offsets, write_offsets, reset_offsets
+from .dashboard import get_dashboard_info
+from .gpu import discover_gpus, get_gpu
 from .monitoring import poll, read_voltage
 from .ranges import get_clock_ranges
-from .snapshot import save as snapshot_save, restore as snapshot_restore, list_snapshots
+from .snapshot import list_snapshots
+from .snapshot import restore as snapshot_restore
+from .snapshot import save as snapshot_save
+from .vfcurve import read_clock_offsets, read_curve, reset_offsets, write_offsets

 __all__ = [
-    "get_gpu", "discover_gpus",
-    "read_curve", "read_clock_offsets", "write_offsets", "reset_offsets",
-    "poll", "read_voltage",
+    "get_gpu",
+    "discover_gpus",
+    "get_dashboard_info",
+    "read_curve",
+    "read_clock_offsets",
+    "write_offsets",
+    "reset_offsets",
+    "poll",
+    "read_voltage",
    "get_clock_ranges",
-    "snapshot_save", "snapshot_restore", "list_snapshots",
+    "snapshot_save",
+    "snapshot_restore",
+    "list_snapshots",
 ]
@@ -0,0 +1,285 @@
+"""Static GPU information for the Dashboard tab.
+
+Collects hardware-identifying and capability data via NVML (pynvml).
+Everything here is a one-shot read (no polling) — live values (clocks, temps,
+power, throttle) come from the monitoring WebSocket instead.
+
+Fields that NVML cannot provide (e.g. ROP count, VRAM type) are filled from a
+best-effort per-model table keyed on the GPU name. When a value cannot be
+determined, the field is ``None`` and the frontend omits it.
+"""
+
+import logging
+from typing import Any
+
+try:
+    import pynvml as _pynvml_import
+
+    _NVML_AVAILABLE = True
+except ImportError:
+    _pynvml_import = None
+    _NVML_AVAILABLE = False
+
+# Aliased as Any so attribute access is not flagged when the import failed.
+pynvml: Any = _pynvml_import
+
+log = logging.getLogger("nvcurve.hal.dashboard")
+
+# ── Best-effort per-model specs ───────────────────────────────────────────────
+# NVML does not expose the ROP count or the VRAM type, so we infer them from
+# the GPU model name. Keyed on the model token (e.g. "RTX 5090"). Values are
+# (vram_type, rop_count). Only well-known cards are listed; unknown models
+# simply yield None for both.
+_GPU_SPECS: dict[str, tuple[str, int]] = {
+    # Blackwell (RTX 50) — GDDR7
+    "RTX 5090": ("GDDR7", 192),
+    "RTX 5080": ("GDDR7", 112),
+    "RTX 5070 Ti": ("GDDR7", 96),
+    "RTX 5070": ("GDDR7", 64),
+    "RTX 5060 Ti": ("GDDR7", 48),
+    "RTX 5060": ("GDDR7", 48),
+    # Ada (RTX 40)
+    "RTX 4090": ("GDDR6X", 128),
+    "RTX 4080 Super": ("GDDR6X", 112),
+    "RTX 4080": ("GDDR6X", 112),
+    "RTX 4070 Ti Super": ("GDDR6X", 96),
+    "RTX 4070 Super": ("GDDR6X", 64),
+    "RTX 4070 Ti": ("GDDR6X", 80),
+    "RTX 4070": ("GDDR6", 64),
+    "RTX 4060 Ti": ("GDDR6", 48),
+    "RTX 4060": ("GDDR6", 32),
+    # Ampere (RTX 30)
+    "RTX 3090 Ti": ("GDDR6X", 88),
+    "RTX 3090": ("GDDR6X", 88),
+    "RTX 3080 Ti": ("GDDR6X", 88),
+    "RTX 3080": ("GDDR6X", 88),
+    "RTX 3070 Ti": ("GDDR6", 80),
+    "RTX 3070": ("GDDR6", 80),
+    "RTX 3060 Ti": ("GDDR6", 64),
+    "RTX 3060": ("GDDR6", 48),
+    # Turing (RTX 20)
+    "RTX 2080 Ti": ("GDDR6", 64),
+    "RTX 2080 Super": ("GDDR6", 48),
+    "RTX 2080": ("GDDR6", 48),
+    "RTX 2070 Super": ("GDDR6", 48),
+    "RTX 2070": ("GDDR6", 48),
+    "RTX 2060 Super": ("GDDR6", 48),
+    "RTX 2060": ("GDDR6", 36),
+    # Pascal / Maxwell (GTX 10 / GTX 9)
+    "GTX 1080 Ti": ("GDDR5X", 64),
+    "GTX 1080": ("GDDR5X", 64),
+    "GTX 1070": ("GDDR5", 64),
+    "GTX 1060": ("GDDR5", 96),
+    "GTX 980 Ti": ("GDDR5", 64),
+    "GTX 980": ("GDDR5", 64),
+}
+
+
+def _lookup_spec(gpu_name: str) -> tuple[str | None, int | None]:
+    """Return (vram_type, rop_count) inferred from the GPU name, or (None, None)."""
+    if not gpu_name:
+        return None, None
+    name = gpu_name.upper()
+    # Longest keys first so "RTX 4070 Ti Super" wins over "RTX 4070 Ti".
+    for key in sorted(_GPU_SPECS, key=len, reverse=True):
+        if key.upper() in name:
+            vram_type, rop = _GPU_SPECS[key]
+            return vram_type, rop
+    return None, None
+
+
+def _safe(fn, *args, **kwargs):
+    """Call an NVML function, returning None on any error."""
+    try:
+        return fn(*args, **kwargs)
+    except Exception:
+        return None
+
+
+def _to_int(val) -> int | None:
+    try:
+        return int(val)
+    except (TypeError, ValueError):
+        return None
+
+
+def _to_float(val) -> float | None:
+    try:
+        return float(val)
+    except (TypeError, ValueError):
+        return None
+
+
+def _get_handle(gpu_index: int):
+    if not _NVML_AVAILABLE:
+        return None
+    try:
+        return pynvml.nvmlDeviceGetHandleByIndex(gpu_index)
+    except Exception:
+        return None
+
+
+def _read_temp_thresholds(handle) -> dict:
+    out: dict[str, int | None] = {
+        "temp_slowdown_c": None,
+        "temp_shutdown_c": None,
+        "temp_gpu_max_c": None,
+    }
+    if handle is None:
+        return out
+    mapping = {
+        "temp_slowdown_c": "NVML_TEMPERATURE_THRESHOLD_SLOWDOWN",
+        "temp_shutdown_c": "NVML_TEMPERATURE_THRESHOLD_SHUTDOWN",
+        "temp_gpu_max_c": "NVML_TEMPERATURE_THRESHOLD_GPU_MAX",
+    }
+    for key, const in mapping.items():
+        val = _safe(
+            pynvml.nvmlDeviceGetTemperatureThreshold,
+            handle,
+            getattr(pynvml, const, None),
+        )
+        if isinstance(val, (int, float)):
+            out[key] = _to_int(val)
+    return out
+
+
+def get_dashboard_info(gpu_index: int = 0, gpu_name: str = "") -> dict:
+    """Collect static GPU info for the dashboard.
+
+    ``gpu_name`` is passed in (already known by the server) so the spec lookup
+    works even if NVML name retrieval fails.
+    """
+    handle = _get_handle(gpu_index)
+
+    out: dict[str, Any] = {
+        "name": gpu_name or None,
+        "index": gpu_index,
+        "driver_version": None,
+        "vbios_version": None,
+        "serial": None,
+        "board_part_number": None,
+        "uuid": None,
+        "cuda_cores": None,
+        "cuda_compute_capability": None,
+        "rop_count": None,
+        "vram_total_bytes": None,
+        "vram_type": None,
+        "memory_bus_width": None,
+        "bar1_total_bytes": None,
+        "bar1_used_bytes": None,
+        "resizable_bar": None,
+        "pcie_link_width": None,
+        "pcie_link_generation": None,
+        "pcie_max_link_width": None,
+        "pcie_max_link_generation": None,
+        "max_graphics_clock_mhz": None,
+        "max_memory_clock_mhz": None,
+        "max_video_clock_mhz": None,
+        "power_limit_w": None,
+        "power_min_limit_w": None,
+        "power_max_limit_w": None,
+        "persistence_mode": None,
+        "num_fans": None,
+        "supported_throttle_reasons": None,
+    }
+
+    if not _NVML_AVAILABLE:
+        return out
+
+    # Driver version (system-wide)
+    out["driver_version"] = _safe(pynvml.nvmlSystemGetDriverVersion)
+
+    if handle is None:
+        return out
+
+    # Identity
+    out["vbios_version"] = _safe(pynvml.nvmlDeviceGetVbiosVersion, handle)
+    serial = _safe(pynvml.nvmlDeviceGetSerial, handle)
+    if isinstance(serial, bytes):
+        serial = serial.decode(errors="replace")
+    out["serial"] = serial or None
+    bpn = _safe(pynvml.nvmlDeviceGetBoardPartNumber, handle)
+    if isinstance(bpn, bytes):
+        bpn = bpn.decode(errors="replace")
+    out["board_part_number"] = bpn or None
+    uuid = _safe(pynvml.nvmlDeviceGetUUID, handle)
+    if isinstance(uuid, bytes):
+        uuid = uuid.decode(errors="replace")
+    out["uuid"] = uuid or None
+
+    # Compute
+    cores = _safe(pynvml.nvmlDeviceGetNumGpuCores, handle)
+    if isinstance(cores, (int, float)):
+        out["cuda_cores"] = _to_int(cores)
+    cc = _safe(pynvml.nvmlDeviceGetCudaComputeCapability, handle)
+    if isinstance(cc, (list, tuple)) and len(cc) >= 2:
+        out["cuda_compute_capability"] = f"{cc[0]}.{cc[1]}"
+
+    # Memory
+    mem = _safe(pynvml.nvmlDeviceGetMemoryInfo, handle)
+    if mem is not None:
+        out["vram_total_bytes"] = _to_int(mem.total)
+    bus_width = _safe(pynvml.nvmlDeviceGetMemoryBusWidth, handle)
+    if isinstance(bus_width, (int, float)):
+        out["memory_bus_width"] = _to_int(bus_width)
+    bar1 = _safe(pynvml.nvmlDeviceGetBAR1MemoryInfo, handle)
+    if bar1 is not None:
+        out["bar1_total_bytes"] = _to_int(bar1.bar1Total)
+        out["bar1_used_bytes"] = _to_int(bar1.bar1Used)
+
+    # Best-effort specs (VRAM type + ROP count) from the GPU name.
+    vram_type, rop = _lookup_spec(gpu_name or (out.get("name") or ""))
+    out["vram_type"] = vram_type
+    out["rop_count"] = rop
+
+    # Resizable BAR: enabled when the BAR1 aperture is a substantial fraction
+    # of total VRAM (legacy BAR1 is a fixed 256 MB).
+    if out["bar1_total_bytes"] is not None and out["vram_total_bytes"]:
+        out["resizable_bar"] = out["bar1_total_bytes"] >= out["vram_total_bytes"] * 0.5
+
+    # PCIe
+    out["pcie_link_width"] = _safe(pynvml.nvmlDeviceGetCurrPcieLinkWidth, handle)
+    out["pcie_link_generation"] = _safe(
+        pynvml.nvmlDeviceGetCurrPcieLinkGeneration, handle
+    )
+    out["pcie_max_link_width"] = _safe(pynvml.nvmlDeviceGetMaxPcieLinkWidth, handle)
+    out["pcie_max_link_generation"] = _safe(
+        pynvml.nvmlDeviceGetMaxPcieLinkGeneration, handle
+    )
+
+    # Max clocks
+    out["max_graphics_clock_mhz"] = _safe(
+        pynvml.nvmlDeviceGetMaxClockInfo, handle, pynvml.NVML_CLOCK_GRAPHICS
+    )
+    out["max_memory_clock_mhz"] = _safe(
+        pynvml.nvmlDeviceGetMaxClockInfo, handle, pynvml.NVML_CLOCK_MEM
+    )
+    out["max_video_clock_mhz"] = _safe(
+        pynvml.nvmlDeviceGetMaxClockInfo, handle, pynvml.NVML_CLOCK_VIDEO
+    )
+
+    # Power limits (mW → W)
+    limit = _safe(pynvml.nvmlDeviceGetPowerManagementLimit, handle)
+    if isinstance(limit, (int, float)):
+        out["power_limit_w"] = (_to_int(limit) or 0) // 1000
+    constrs = _safe(pynvml.nvmlDeviceGetPowerManagementLimitConstraints, handle)
+    if isinstance(constrs, (list, tuple)) and len(constrs) >= 2:
+        out["power_min_limit_w"] = (_to_int(constrs[0]) or 0) // 1000
+        out["power_max_limit_w"] = (_to_int(constrs[1]) or 0) // 1000
+
+    # Misc
+    persist = _safe(pynvml.nvmlDeviceGetPersistenceMode, handle)
+    if isinstance(persist, (bool, int)):
+        out["persistence_mode"] = bool(persist)
+    fans = _safe(pynvml.nvmlDeviceGetNumFans, handle)
+    if isinstance(fans, (int, float)):
+        out["num_fans"] = _to_int(fans)
+    out["supported_throttle_reasons"] = _safe(
+        pynvml.nvmlDeviceGetSupportedClocksThrottleReasons, handle
+    )
+
+    # Temperature thresholds (static). Current temps come from the monitor
+    # WebSocket (live), not this one-shot read.
+    out.update(_read_temp_thresholds(handle))
+
+    return out
@@ -4,21 +4,26 @@ Voltage is read via NvAPI GetCurrentVoltage.
 Clock, temperature, power draw, and fan speed are read via NVML (nvidia-ml-py).
 """

+import contextlib
 import struct
 import time
-from typing import Optional
+from typing import Any

 from ..nvapi.bootstrap import nvcall
 from ..nvapi.constants import FUNC, VOLT_SIZE
 from ..nvapi.types import MonitoringSample

 try:
-    import pynvml as _pynvml
+    import pynvml as _pynvml_import
+
    _NVML_AVAILABLE = True
 except ImportError:
-    _pynvml = None
+    _pynvml_import = None
    _NVML_AVAILABLE = False

+# Aliased as Any so attribute access is not flagged when the import failed.
+_pynvml: Any = _pynvml_import
+
 _nvml_initialized = False


@@ -39,14 +44,12 @@ def shutdown_nvml() -> None:
    """Shut down NVML. Call at process exit."""
    global _nvml_initialized
    if _NVML_AVAILABLE and _nvml_initialized:
-        try:
+        with contextlib.suppress(_pynvml.NVMLError):
            _pynvml.nvmlShutdown()
-        except _pynvml.NVMLError:
-            pass
        _nvml_initialized = False


-def get_driver_version() -> Optional[str]:
+def get_driver_version() -> str | None:
    """Return the NVIDIA driver version string, or None if unavailable."""
    if not (_NVML_AVAILABLE and _nvml_initialized):
        return None
@@ -56,7 +59,42 @@ def get_driver_version() -> Optional[str]:
        return None


-def get_vram_total(gpu_index: int = 0) -> Optional[int]:
+# NVML clock-throttle reason bits (from nvmlClocksThrottleReason* constants).
+# Mapped to short human-readable labels for the dashboard.
+_THROTTLE_REASONS: dict[int, str] = {
+    0x1: "GPU Idle",
+    0x2: "Application Clocks",
+    0x4: "SW Power Cap",
+    0x8: "HW Slowdown",
+    0x10: "Sync Boost",
+    0x20: "SW Thermal Slowdown",
+    0x40: "HW Thermal Slowdown",
+    0x80: "HW Power Brake",
+    0x100: "Display Clock Setting",
+}
+
+
+def throttle_reasons_label(mask: int | None) -> str | None:
+    """Convert a throttle-reasons bitmask to a comma-joined label.
+
+    Returns None when the mask is unknown, or "No" when nothing is active.
+    """
+    if mask is None:
+        return None
+    if mask == 0:
+        return "No"
+    active = [label for bit, label in _THROTTLE_REASONS.items() if mask & bit]
+    # Include any unknown high bits so nothing is silently dropped.
+    known = 0
+    for bit in _THROTTLE_REASONS:
+        known |= bit
+    extra = mask & ~known
+    if extra:
+        active.append(f"0x{extra:x}")
+    return ", ".join(active) if active else "No"
+
+
+def get_vram_total(gpu_index: int = 0) -> int | None:
    """Return total VRAM in bytes, or None if unavailable."""
    if not (_NVML_AVAILABLE and _nvml_initialized):
        return None
@@ -67,7 +105,7 @@ def get_vram_total(gpu_index: int = 0) -> Optional[int]:
        return None


-def read_voltage(gpu) -> tuple[Optional[int], str]:
+def read_voltage(gpu) -> tuple[int | None, str]:
    """Read current GPU core voltage in µV via NvAPI GetCurrentVoltage.

    Returns (voltage_uV, "OK") or (None, error).
@@ -80,19 +118,36 @@ def read_voltage(gpu) -> tuple[Optional[int], str]:

 def _nvml_read(gpu_index: int) -> dict:
    """Read all NVML fields. Returns a dict with keys matching MonitoringSample fields."""
-    out = {
-        "clock_mhz": None, "temp_c": None, "power_w": None, "fan_pct": None,
-        "pstate": None, "mem_used_bytes": None, "mem_total_bytes": None,
-        "gpu_util_pct": None, "mem_util_pct": None, "mem_clock_mhz": None,
+    out: dict[str, Any] = {
+        "clock_mhz": None,
+        "temp_c": None,
+        "power_w": None,
+        "fan_pct": None,
+        "pstate": None,
+        "mem_used_bytes": None,
+        "mem_total_bytes": None,
+        "gpu_util_pct": None,
+        "mem_util_pct": None,
+        "mem_clock_mhz": None,
+        "throttle_reasons": None,
+        "pcie_link_width": None,
+        "pcie_link_generation": None,
+        "mem_temp_c": None,
    }
    if not (_NVML_AVAILABLE and _nvml_initialized):
        return out
    try:
        handle = _pynvml.nvmlDeviceGetHandleByIndex(gpu_index)

-        out["clock_mhz"] = float(_pynvml.nvmlDeviceGetClockInfo(handle, _pynvml.NVML_CLOCK_GRAPHICS))
-        out["mem_clock_mhz"] = float(_pynvml.nvmlDeviceGetClockInfo(handle, _pynvml.NVML_CLOCK_MEM))
-        out["temp_c"] = float(_pynvml.nvmlDeviceGetTemperature(handle, _pynvml.NVML_TEMPERATURE_GPU))
+        out["clock_mhz"] = float(
+            _pynvml.nvmlDeviceGetClockInfo(handle, _pynvml.NVML_CLOCK_GRAPHICS)
+        )
+        out["mem_clock_mhz"] = float(
+            _pynvml.nvmlDeviceGetClockInfo(handle, _pynvml.NVML_CLOCK_MEM)
+        )
+        out["temp_c"] = float(
+            _pynvml.nvmlDeviceGetTemperature(handle, _pynvml.NVML_TEMPERATURE_GPU)
+        )
        out["power_w"] = _pynvml.nvmlDeviceGetPowerUsage(handle) / 1000.0  # mW → W
        out["pstate"] = int(_pynvml.nvmlDeviceGetPerformanceState(handle))

@@ -104,10 +159,33 @@ def _nvml_read(gpu_index: int) -> dict:
        out["gpu_util_pct"] = float(util.gpu)
        out["mem_util_pct"] = float(util.memory)

-        try:
+        with contextlib.suppress(_pynvml.NVMLError):
            out["fan_pct"] = float(_pynvml.nvmlDeviceGetFanSpeed(handle))
-        except _pynvml.NVMLError:
-            pass
+
+        with contextlib.suppress(_pynvml.NVMLError):
+            out["throttle_reasons"] = int(
+                _pynvml.nvmlDeviceGetCurrentClocksThrottleReasons(handle)
+            )
+
+        # PCIe link downclocks when idle, so read it live each poll.
+        with contextlib.suppress(_pynvml.NVMLError):
+            out["pcie_link_width"] = int(_pynvml.nvmlDeviceGetCurrPcieLinkWidth(handle))
+        with contextlib.suppress(_pynvml.NVMLError):
+            out["pcie_link_generation"] = int(
+                _pynvml.nvmlDeviceGetCurrPcieLinkGeneration(handle)
+            )
+
+        # VRAM temp (if the GPU exposes a MEMORY thermal sensor).
+        with contextlib.suppress(_pynvml.NVMLError):
+            sensors = _pynvml.nvmlDeviceGetThermalSettings(handle, 0)
+            for s in sensors:
+                if (
+                    s.target == _pynvml.NVML_THERMAL_TARGET_MEMORY
+                    and s.currentTemp is not None
+                    and s.currentTemp > 0
+                ):
+                    out["mem_temp_c"] = float(s.currentTemp)
+                    break
    except _pynvml.NVMLError:
        pass
    return out
@@ -4,7 +4,7 @@ Raw struct manipulation stays in the call layer (bootstrap.py + hal/).
 Everything above HAL works with these types.
 """

-from dataclasses import dataclass, field
+from dataclasses import dataclass


@dataclass
@@ -57,6 +57,10 @@ class MonitoringSample:
    gpu_util_pct: float | None  # GPU core utilization (0–100)
    mem_util_pct: float | None  # Memory bus utilization (0–100)
    mem_clock_mhz: float | None = None  # Current memory clock (NVML_CLOCK_MEM)
+    throttle_reasons: int | None = None  # Active clock-throttle bitmask (NVML)
+    pcie_link_width: int | None = None  # Current PCIe link width (x1..x16)
+    pcie_link_generation: int | None = None  # Current PCIe link generation (1..5)
+    mem_temp_c: float | None = None  # VRAM temperature (if the GPU exposes it)


@dataclass
@@ -9,7 +9,7 @@ Requires root (NvAPI needs it).
 import asyncio
 import logging
 import os
-from contextlib import asynccontextmanager
+from contextlib import asynccontextmanager, suppress
 from pathlib import Path
 from typing import Any

@@ -21,6 +21,7 @@ from pydantic import BaseModel

 from . import auth
 from .config import Config, default_config
+from .hal.dashboard import get_dashboard_info
 from .hal.fans import (
    get_fan_info,
    get_temp,
@@ -43,6 +44,7 @@ from .hal.monitoring import (
    init_nvml,
    poll,
    shutdown_nvml,
+    throttle_reasons_label,
 )
 from .hal.ranges import get_clock_ranges
 from .hal.snapshot import (
@@ -159,9 +161,29 @@ def _sample_dict(s) -> dict:
        else None,
        "gpu_util_pct": s.gpu_util_pct,
        "mem_util_pct": s.mem_util_pct,
+        "throttle_reasons": s.throttle_reasons,
+        "throttle_reasons_label": throttle_reasons_label(s.throttle_reasons),
+        "pcie_link_width": s.pcie_link_width,
+        "pcie_link_generation": s.pcie_link_generation,
+        "mem_temp_c": s.mem_temp_c,
    }


+def _int_key_deltas(deltas: dict) -> dict[int, int]:
+    """Convert string-keyed deltas (from JSON) to int-keyed.
+
+    Raises ValueError if any key is not a valid integer (corrupted profile),
+    so a bad profile fails closed rather than partially applying to hardware.
+    """
+    out: dict[int, int] = {}
+    for k, v in deltas.items():
+        try:
+            out[int(k)] = v
+        except (TypeError, ValueError) as exc:
+            raise ValueError(f"Invalid curve point index in profile: {k!r}") from exc
+    return out
+
+
 # ── WebSocket broadcast ───────────────────────────────────────────────────────


@@ -309,18 +331,14 @@ async def lifespan(app: FastAPI):
    for task in poller_tasks:
        task.cancel()
    for task in poller_tasks:
-        try:
+        with suppress(asyncio.CancelledError):
            await task
-        except asyncio.CancelledError:
-            pass

    for gpu_index, g_state in _state["gpus"].items():
        if g_state.get("fan_poller_task"):
            g_state["fan_poller_task"].cancel()
-            try:
+            with suppress(asyncio.CancelledError):
                await g_state["fan_poller_task"]
-            except asyncio.CancelledError:
-                pass
        if g_state.get("fan_curve_active"):
            g_state["fan_curve_active"] = False
            g_state["fan_curve"] = None
@@ -345,7 +363,10 @@ app = FastAPI(title="nvcurve", version="0.5.0", lifespan=lifespan)

 app.add_middleware(
    CORSMiddleware,
-    allow_origins=["*"],
+    # The SPA is served same-origin by this server, so CORS only matters for
+    # local development (e.g. the Vite dev server). Restrict to localhost
+    # origins rather than a wildcard.
+    allow_origin_regex=r"https?://(localhost|127\.0\.0\.1)(:\d+)?$",
    allow_methods=["*"],
    allow_headers=["*"],
 )
@@ -589,6 +610,17 @@ async def api_gpu(gpu_index: int = 0):
    }


+@app.get("/api/dashboard")
+async def api_dashboard(gpu_index: int = 0):
+    """Static GPU info for the Dashboard tab (VBIOS, CUDA cores, PCIe, BAR1, etc.).
+
+    Live values (clocks, temps, power, throttle) come from the monitor WebSocket.
+    """
+    _, g_state = _require_gpu(gpu_index)
+    info = await _run(get_dashboard_info, gpu_index, g_state["gpu_name"])
+    return info
+
+
@app.get("/api/curve")
 async def api_curve(gpu_index: int = 0):
    """Full CurveState: all V/F points with base freq, voltage, delta, effective freq."""
@@ -780,9 +812,7 @@ async def _auto_apply_profile_with_retry(
    profile = await _run(load_profile, filepath)  # raises FileNotFoundError if missing

    expected: dict[int, int] = (
-        {int(k): v for k, v in profile.curve_deltas.items()}
-        if profile.curve_deltas
-        else {}
+        _int_key_deltas(profile.curve_deltas) if profile.curve_deltas else {}
    )

    for attempt in range(max_retries):
@@ -880,7 +910,7 @@ async def _apply_profile(name: str, gpu_index: int = 0) -> list[str]:
    # Apply curve deltas (after mem offset which may have wiped them).
    async with g_state["write_lock"]:
        if profile.curve_deltas:
-            deltas = {int(k): v for k, v in profile.curve_deltas.items()}
+            deltas = _int_key_deltas(profile.curve_deltas)
            errors = validate_write(deltas, cfg.max_delta_khz)
            if errors:
                errs.append("Curve: " + "; ".join(errors))
@@ -910,10 +940,8 @@ async def _apply_profile(name: str, gpu_index: int = 0) -> list[str]:
            # Stop existing fan poller if running
            if g_state.get("fan_poller_task"):
                g_state["fan_poller_task"].cancel()
-                try:
+                with suppress(asyncio.CancelledError):
                    await g_state["fan_poller_task"]
-                except asyncio.CancelledError:
-                    pass

            g_state["fan_curve"] = profile.fan_curve
            g_state["fan_curve_active"] = True
@@ -922,10 +950,8 @@ async def _apply_profile(name: str, gpu_index: int = 0) -> list[str]:
        # Profile has no fan curve, deactivate any active fan curve
        if g_state.get("fan_poller_task"):
            g_state["fan_poller_task"].cancel()
-            try:
+            with suppress(asyncio.CancelledError):
                await g_state["fan_poller_task"]
-            except asyncio.CancelledError:
-                pass
            g_state["fan_poller_task"] = None
        g_state["fan_curve"] = None
        g_state["fan_curve_active"] = False
@@ -942,10 +968,14 @@ async def api_profile_apply(name: str, gpu_index: int = 0):
    _require_gpu(gpu_index)
    try:
        errs = await _apply_profile(name, gpu_index)
-    except FileNotFoundError:
-        raise HTTPException(status_code=404, detail=f"Profile '{name}' not found")
+    except FileNotFoundError as err:
+        raise HTTPException(
+            status_code=404, detail=f"Profile '{name}' not found"
+        ) from err
    except Exception as e:
-        raise HTTPException(status_code=500, detail=f"Failed to load profile: {e}")
+        raise HTTPException(
+            status_code=500, detail=f"Failed to load profile: {e}"
+        ) from e
    if errs:
        raise HTTPException(status_code=500, detail="; ".join(errs))
    return {"ok": True}
@@ -1169,10 +1199,8 @@ async def api_fans_update(req: FanCurveRequest, gpu_index: int = 0):
    # Stop existing poller if running
    if g_state.get("fan_poller_task"):
        g_state["fan_poller_task"].cancel()
-        try:
+        with suppress(asyncio.CancelledError):
            await g_state["fan_poller_task"]
-        except asyncio.CancelledError:
-            pass

    g_state["fan_curve"] = curve_data
    g_state["fan_curve_active"] = True
@@ -1189,10 +1217,8 @@ async def api_fans_reset(gpu_index: int = 0):
    # Stop poller
    if g_state.get("fan_poller_task"):
        g_state["fan_poller_task"].cancel()
-        try:
+        with suppress(asyncio.CancelledError):
            await g_state["fan_poller_task"]
-        except asyncio.CancelledError:
-            pass
        g_state["fan_poller_task"] = None

    g_state["fan_curve"] = None
@@ -1234,7 +1260,7 @@ async def _reconcile_check(gpu_index: int) -> dict | None:
    if current is None:
        return None  # Can't read — let the write attempt proceed

-    changed = [i for i, (a, b) in enumerate(zip(last, current)) if a != b]
+    changed = [i for i, (a, b) in enumerate(zip(last, current, strict=False)) if a != b]
    if not changed:
        return None