fix(AI): vendor-aware AMD HSA override + benchmark discrete-GPU detection

Closes #810.

## Bug A: HSA_OVERRIDE_GFX_VERSION=11.0.0 was unconditional

PR #804 set HSA_OVERRIDE_GFX_VERSION=11.0.0 for any AMD GPU. The inline comment
claimed this was harmless on supported discrete cards (gfx1030 RX 6800, etc.) — empirically false. With the override, Ollama crashes during GPU discovery on
gfx1030 and falls back to CPU silently. Affects every NOMAD user with an
RX 6800 or other RDNA 2 discrete card.

The correct value depends on the gfx version:
- gfx1030, gfx1100, gfx1101, gfx1102: officially supported by ROCm — no override
- gfx1031..gfx1036 (RDNA 2 variants + iGPUs like Rembrandt 680M): 10.3.0
- gfx1103, gfx1150, gfx1151 (Phoenix 780M, Strix 890M, Strix Halo): 11.0.0

### Resolution chain in `_resolveAmdHsaOverride()`

1. KV `ai.amdHsaOverride` — manual override; accepts 'none' to disable, or a
   semver-style value to force.
2. Marker file `/app/storage/.nomad-amd-gfx` — written by install_nomad.sh
   based on lspci codename. Mapped to override via `_mapGfxToHsaOverride()`.
3. Default: `11.0.0` — preserves prior behavior so existing iGPU users
   (780M / 890M, the dominant AMD population today) don't regress on upgrade.

Discrete RDNA 2 users on existing installs can opt out via
`ai.amdHsaOverride='none'` and force-reinstall AI Assistant, OR re-run
install_nomad.sh to refresh the marker file.

The helper is used in both `createContainer` (initial install) and
`updateContainer` (image update) paths, replacing the unconditional push.

## Bug B: BenchmarkService had no AMD discrete detection path

`BenchmarkService.getHardwareInfo()` had three GPU detection fallbacks:
1. `si.graphics()` — empty inside Docker for AMD
2. nvidia-smi — NVIDIA only
3. AMD APU regex from CPU model — integrated only

Result: AMD discrete cards (RX 6800, RX 7900 XTX, etc.) showed up as
"GPU: Not detected" on the leaderboard despite ROCm working. Corrupts
leaderboard data quality for that population.

Fix: after the existing fallbacks, call `SystemService.getSystemInfo()` and
read `graphics.controllers[0].model`. That path already handles AMD via the
marker file + Ollama log probe added in PR #804, so we're reusing existing
plumbing rather than duplicating detection logic.

## install_nomad.sh changes

The existing AMD detection block already runs lspci. Added a codename parse
step that maps Navi 21/22/23/24, Rembrandt, Phoenix1/Phoenix2, Strix/Strix
Point/Strix Halo, and Navi 31/32/33 to gfx versions, then writes
`/opt/project-nomad/storage/.nomad-amd-gfx`. Unknown codenames write nothing
(admin handles missing-marker case via the backward-compat default).

## Validation

Both bugs were originally surfaced and validated empirically on RX 6800 /
gfx1030 / Ubuntu 24.04 + kernel 6.17 + ollama/ollama:rocm during the #810
filing. Validation grid from that report:

| Run                                           | NOMAD Score | tok/s | GPU detected            |
|-----------------------------------------------|-------------|-------|-------------------------|
| Pre-fix (Bug A active)                        | n/a         | 0     | yes, but library=cpu    |
| HSA_OVERRIDE removed, Bug B unfixed           | 73.8        | 221.6 | "Not detected"          |
| Both fixes hot-patched (this PR's behavior)   | 73.7        | 216.0 | AMD Radeon RX 6800      |

Local checks: `npm run typecheck` clean, `npm run build` clean.

admin/app/services/benchmark_service.ts

@@ -317,6 +317,23 @@ export class BenchmarkService {
         }
       }
 
+      // Fallback: AMD discrete cards. si.graphics() returns empty inside Docker for AMD,
+      // the nvidia-smi path doesn't apply, and the APU regex only catches integrated parts.
+      // SystemService.getSystemInfo() already handles AMD via the marker file + Ollama log
+      // probe added in PR #804, so reuse that plumbing rather than duplicating it here.
+      if (!gpuModel) {
+        try {
+          const systemService = new (await import('./system_service.js')).SystemService(this.dockerService)
+          const sysInfo = await systemService.getSystemInfo()
+          const sysGpuModel = sysInfo?.graphics?.controllers?.[0]?.model
+          if (sysGpuModel) {
+            gpuModel = sysGpuModel
+          }
+        } catch (sysError: any) {
+          logger.warn(`[BenchmarkService] system_service AMD fallback failed: ${sysError.message}`)
+        }
+      }
+
       return {
         cpu_model: `${cpu.manufacturer} ${cpu.brand}`,
         cpu_cores: cpu.physicalCores,

admin/app/services/docker_service.ts

@@ -592,10 +592,12 @@ export class DockerService {
           ollamaEnv.push('OLLAMA_FLASH_ATTENTION=1')
         }
         if (amdGpuConfigured) {
-          // RDNA3 iGPUs (gfx1103: 780M, 880M, 890M, ...) aren't on AMD's official ROCm
+          // gfx-aware HSA override — only set for cards that actually need it. See
-          // allowlist but work when forced to identify as gfx1100 via HSA_OVERRIDE_GFX_VERSION.
+          // _resolveAmdHsaOverride() for the resolution order and gfx → version mapping.
-          // Harmless on supported discrete cards (gfx1030 RX 6800, etc.) — they ignore the override.
+          const hsaOverride = await this._resolveAmdHsaOverride()
-          ollamaEnv.push('HSA_OVERRIDE_GFX_VERSION=11.0.0')
+          if (hsaOverride) {
+            ollamaEnv.push(`HSA_OVERRIDE_GFX_VERSION=${hsaOverride}`)
+          }
         }
       }
 
@@ -999,6 +1001,67 @@ export class DockerService {
     }
   }
 
+  /**
+   * Resolve the HSA_OVERRIDE_GFX_VERSION value for the host's AMD GPU.
+   *
+   * gfx1030 (RX 6800/6700/etc.), gfx1100/1101/1102 (RX 7900/7800/7600) are on AMD's
+   * official ROCm allowlist — forcing an override on these breaks GPU discovery.
+   * gfx1035 / gfx1036 (RDNA 2 iGPUs like 680M) need 10.3.0 to coerce to gfx1030.
+   * gfx1103 / gfx1150 / gfx1151 (RDNA 3/3.5 iGPUs like 780M / 890M / Strix Halo) need 11.0.0.
+   *
+   * Resolution order:
+   *   1. KV `ai.amdHsaOverride` — manual user override; accepts 'none' (disable) or a semver-style value.
+   *   2. Marker file `/app/storage/.nomad-amd-gfx` written by install_nomad.sh.
+   *   3. Default: '11.0.0' — preserves prior behavior so existing iGPU users don't regress on
+   *      upgrade. Discrete-card users on existing installs can opt out via the KV.
+   *
+   * Returns null when no override should be applied.
+   */
+  private async _resolveAmdHsaOverride(): Promise<string | null> {
+    const manualRaw = await KVStore.getValue('ai.amdHsaOverride')
+    if (manualRaw !== null && manualRaw !== undefined && String(manualRaw).trim() !== '') {
+      const manual = String(manualRaw).trim().toLowerCase()
+      if (manual === 'none' || manual === 'off' || manual === 'false') {
+        logger.info('[DockerService] HSA override disabled via ai.amdHsaOverride')
+        return null
+      }
+      if (/^\d+\.\d+\.\d+$/.test(manual)) {
+        logger.info(`[DockerService] HSA override forced to ${manual} via ai.amdHsaOverride`)
+        return manual
+      }
+      logger.warn(`[DockerService] Ignoring invalid ai.amdHsaOverride value: ${manualRaw}`)
+    }
+
+    try {
+      const gfx = (await readFile('/app/storage/.nomad-amd-gfx', 'utf8')).trim()
+      const mapped = this._mapGfxToHsaOverride(gfx)
+      logger.info(`[DockerService] AMD gfx marker '${gfx}' → HSA override ${mapped ?? 'none'}`)
+      return mapped
+    } catch {
+      // Marker absent — most likely an existing install upgraded without re-running
+      // install_nomad.sh. Fall through to the default.
+    }
+
+    logger.info('[DockerService] No AMD gfx marker; defaulting HSA override to 11.0.0 for backward compatibility')
+    return '11.0.0'
+  }
+
+  private _mapGfxToHsaOverride(gfx: string): string | null {
+    // Officially supported by ROCm — no override needed
+    if (gfx === 'gfx1030' || gfx === 'gfx1100' || gfx === 'gfx1101' || gfx === 'gfx1102') {
+      return null
+    }
+    // RDNA 2 variants + iGPUs (gfx1031..gfx1036, e.g. Rembrandt 680M)
+    if (/^gfx103[1-6]$/.test(gfx)) {
+      return '10.3.0'
+    }
+    // RDNA 3 / 3.5 mobile parts (Phoenix 780M = gfx1103, Strix 890M = gfx1150, Strix Halo = gfx1151)
+    if (gfx === 'gfx1103' || gfx === 'gfx1150' || gfx === 'gfx1151') {
+      return '11.0.0'
+    }
+    return '11.0.0'
+  }
+
   /**
    * Build the Docker Devices array for AMD GPU passthrough.
    *
@@ -1132,12 +1195,14 @@ export class DockerService {
       // and whether HSA_OVERRIDE needs injection. For AMD, replace any prior HSA_OVERRIDE in
       // the inspect-captured env so updates from older containers pick up the current value.
       const baseEnv = inspectData.Config?.Env || []
-      const finalEnv = updatedAmdGpuConfigured
+      let finalEnv = baseEnv
-        ? [
+      if (updatedAmdGpuConfigured) {
-            ...baseEnv.filter((e: string) => !e.startsWith('HSA_OVERRIDE_GFX_VERSION=')),
+        const hsaOverride = await this._resolveAmdHsaOverride()
-            'HSA_OVERRIDE_GFX_VERSION=11.0.0',
+        finalEnv = baseEnv.filter((e: string) => !e.startsWith('HSA_OVERRIDE_GFX_VERSION='))
-          ]
+        if (hsaOverride) {
-        : baseEnv
+          finalEnv.push(`HSA_OVERRIDE_GFX_VERSION=${hsaOverride}`)
+        }
+      }
 
       const newContainerConfig: any = {
         Image: newImage,

admin/types/kv_store.ts

@@ -13,6 +13,7 @@ export const KV_STORE_SCHEMA = {
   'ai.remoteOllamaUrl':         'string',
   'ai.ollamaFlashAttention':    'boolean',
   'ai.amdGpuAcceleration':      'boolean',
+  'ai.amdHsaOverride':          'string',
 } as const
 
 type KVTagToType<T extends string> = T extends 'boolean' ? boolean : string

install/install_nomad.sh

@@ -520,10 +520,40 @@ verify_gpu_setup() {
   # Check for AMD GPU — restrict to display controller classes to avoid false positives
   # from AMD CPU host bridges, PCI bridges, and chipset devices.
   local has_amd_gpu='false'
+  local amd_gfx_version=''
   if command -v lspci &> /dev/null; then
     if lspci 2>/dev/null | grep -iE "VGA|3D controller|Display" | grep -iE "amd|radeon" &> /dev/null; then
       has_amd_gpu='true'
       echo -e "${GREEN}✓${RESET} AMD GPU detected — ROCm acceleration will be configured automatically when AI Assistant is installed.\\n"
+
+      # Map AMD codename → gfx version so the admin can pick the right HSA_OVERRIDE_GFX_VERSION.
+      # gfx1030/1100/1101/1102 are on AMD's official ROCm allowlist and need NO override —
+      # forcing one (e.g. 11.0.0) breaks GPU discovery on these. Other variants do need it.
+      local amd_devices
+      amd_devices=$(lspci -vmm 2>/dev/null | awk -F'\t' '/^Class:.*(VGA|3D|Display)/{c=1} c && /^Device:/{print $2; c=0}')
+      if echo "${amd_devices}" | grep -iq 'Navi 21'; then
+        amd_gfx_version='gfx1030'
+      elif echo "${amd_devices}" | grep -iq 'Navi 22'; then
+        amd_gfx_version='gfx1031'
+      elif echo "${amd_devices}" | grep -iq 'Navi 23'; then
+        amd_gfx_version='gfx1032'
+      elif echo "${amd_devices}" | grep -iq 'Navi 24'; then
+        amd_gfx_version='gfx1034'
+      elif echo "${amd_devices}" | grep -iq 'Rembrandt'; then
+        amd_gfx_version='gfx1035'
+      elif echo "${amd_devices}" | grep -iEq 'Phoenix1?|Phoenix2'; then
+        amd_gfx_version='gfx1103'
+      elif echo "${amd_devices}" | grep -iEq 'Strix Halo'; then
+        amd_gfx_version='gfx1151'
+      elif echo "${amd_devices}" | grep -iEq 'Strix( Point)?'; then
+        amd_gfx_version='gfx1150'
+      elif echo "${amd_devices}" | grep -iq 'Navi 31'; then
+        amd_gfx_version='gfx1100'
+      elif echo "${amd_devices}" | grep -iq 'Navi 32'; then
+        amd_gfx_version='gfx1101'
+      elif echo "${amd_devices}" | grep -iq 'Navi 33'; then
+        amd_gfx_version='gfx1102'
+      fi
     fi
   fi
 
@@ -539,6 +569,16 @@ verify_gpu_setup() {
     sudo rm -f "${gpu_marker_path}" 2>/dev/null || true
   fi
 
+  # Companion marker used by the admin to pick the right HSA_OVERRIDE_GFX_VERSION for
+  # the detected card. Absence of this file means "unknown gfx" — the admin falls back
+  # to its built-in default. Always rewrite (or remove) on install to keep state fresh.
+  local amd_gfx_marker_path="${NOMAD_DIR}/storage/.nomad-amd-gfx"
+  if [[ -n "${amd_gfx_version}" ]]; then
+    echo "${amd_gfx_version}" | sudo tee "${amd_gfx_marker_path}" > /dev/null 2>&1 || true
+  else
+    sudo rm -f "${amd_gfx_marker_path}" 2>/dev/null || true
+  fi
+
   echo -e "${YELLOW}===========================================${RESET}\\n"
 
   # Summary