Hi SimonF,

First off, thanks a lot for all the work on this plugin — and especially the recent i915 + qmassa backend additions. The NVIDIA side of my setup works beautifully.

I'm running into one stubborn little thing and was hoping you could point me in the right direction. I have a mixed setup (an NVIDIA RTX PRO 6000 plus an Intel Arc A310),

and while the NVIDIA panel renders perfectly, the A310 panel just won't appear on the dashboard — even with an active Plex transcode on the Arc. I've done a fair bit of

digging before posting, so here's everything I've found:

Setup:

- Unraid 7.3.1, gpustat 2026.05.19

- Intel Backend (i915) = "intel_gpu_top + qmassa"; qmassa is present at /usr/local/sbin/qmassa

- GPUs: NVIDIA RTX PRO 6000 (renders fine) + Intel Arc A310 (DG2, 8086:56a6), i915,

/dev/dri/card2, renderD129, PCI 0000:91:00.0. Both are selected in

"Unit ID for Dashboard (Multiple)".

What I've checked so far:

- intel_gpu_top -L and -J both give correct output.

- qmassa reads the A310 fine. Idle sample (key fields): "pci_id":"8086:56A6",

"drv_name":"i915", "dev_nodes":"card2, renderD129", vram_total ~4GB, freq cur 2450,

power ~14.7W, temp 36C, fan ~2900 RPM — but "eng_names":[], "eng_usage":{} and

"clis_stats":[] are empty while idle.

- /tmp has A310 cache files (gpustat_power_reading_cache_0000_91_00.0.json and

gpustat_fan_cache_0000_91_00.0.json), so the plugin does seem to be gathering the

A310 data.

- DEBUGLOG = 1 in gpustat.cfg, but I can't find a debug log anywhere

(no /var/log/gpustats.debug, nothing matching gpustat in /var/log).

- php .../gpustatusmulti.php from the CLI returns [].

So the data is clearly being collected, but the panel never renders. Two questions:

where should the debug log live in 2026.05.19 (so I can grab something useful for you),

and is there a known fix or an updated Intel.php for A310/DG2 cards not appearing

next to an NVIDIA card?

Happy to run any tests or pull whatever logs would help — just let me know.

Thanks again for maintaining this, it's much appreciated!

i have a couple containers whose icons render in this plugin with dead image urls. is there anyway i can force clearing or regenerating the cached icon images this plugin uses?

I have a A380. I just added qmassa, it works great. But I see that I can set it to only use qmassa. Is there a plan to stop requiiring intel_gpu_top in the future? For now the plugin doesn't work without it even if qmassa should supply the same info. I'm planning for a future where I'm using the Xe driver.

Edited September 12Sep 12 by kaares

Hello.

Bug: dashboard shows nothing for GPUs with “&” in the model name (Intel Atom Z36xxx/Z37xxx)

Unraid 7.3, GPU Statistics version: 2026.05.19
iGPU: “Atom Processor Z36xxx/Z37xxx Series Graphics & Display” (i915)

intel_gpu_top works fine, but the dashboard tile shows no data.

Cause: the request to gpustatusmulti.php sends the model name unencoded in the gpus parameter, so the & in “Graphics & Display” splits the query string. The JSON is truncated, json_decode fails, and the endpoint returns []. Calling the same URL with %26 instead of & returns the correct data.

Suggested fix: encode the JSON with encodeURIComponent(JSON.stringify(...)) when building the URL in gpustatmulti.js.

Workaround: remove the & from the model name in MULTIGPUJSON in /boot/config/plugins/gpustat/gpustat.cfg (Graphics+%26+Display → Graphics+Display).

Hi @SimonF,

I've noticed an issue with the GPU Statistics plugin on my Unraid server.

System:

• Unraid 7.3.4

• NVIDIA GeForce RTX 4060

• NVIDIA driver 615.78.08

• GPU Statistics (Multi-GPU)

Problem:

The dashboard always displays 0W GPU power consumption, even while the GPU is actively encoding and decoding video.

Both power.draw and power.draw.instant return [N/A].

However, NVIDIA still provides valid power measurements using:

nvidia-smi -q -d POWER -i 0

Output:

Power Samples
    Duration          : 2.36 sec
    Number of Samples : 119
    Max               : 74.53 W
    Min               : 57.59 W
    Avg               : 61.05 W

Looking at lib/Nvidia.php, around lines 262–279, it seems that N/A values are converted to floats, resulting in 0W.

Would it be possible to add a fallback that uses the average value from Power Samples when normal power readings are unavailable?

Ideally, the plugin would also display N/A instead of 0W when neither source provides a valid measurement.

This could improve compatibility with newer NVIDIA drivers while avoiding unnecessary polling overhead.

Thanks for your work on the plugin!

Hi,

I wanted to report a telemetry/parsing regression with the GPU Statistics plugin on Unraid 7.3.4 when running an RTX 50-series (Blackwell) GPU with NVIDIA's open-source kernel modules (nvos).

Environment

• Unraid Version: 7.3.4

• Plugin: GPU Statistics (gpustat)

• GPU: NVIDIA GeForce RTX 5060 Ti 16 GB (GB206 / Blackwell)

• Driver: v580.196.07 (NVIDIA Open Kernel Module / nvos)

Observed Behaviors

1. Dashboard XML Parsing Error (Code 302)

   • Symptom: The dashboard widget fails to populate and displays:

     Vendor command returned unparseable data.

   • Root Cause: In driver branch 580+ under the open kernel module on Blackwell silicon, several XML nodes/attributes returned by nvidia-smi -q -x have either shifted or altered types, causing the PHP XML deserializer to fail validation.

2. Core Clock Domain Glitch (GDDR7 Data Rate Misread)

   • Symptom: When metrics do briefly render, the GPU Graphics Clock reads an absurd value (e.g., 17,317 MHz).

   • Root Cause: The parser appears to be picking up the effective GDDR7 memory data rate/clock multiplier or an unshifted memory frequency field and mapping it into the core graphics clock property.

3. GSP Power State Wake Thrashing

   • Symptom: Periodic polling causes the GPU to rapidly oscillate into P0 (14W, ~49% fan, sw_power_cap) when the card is otherwise idle with zero client processes.

   • Note: Without an active CUDA handle or nvidia-persistenced / nvidia-smi -pm 1 explicitly active, frequent nvidia-smi -q -x invocations trigger the GSP firmware's fail-safe clock state.

Workaround / Reproducibility

• Disabling polling or switching the monitored GPU to None stops the error logs and allows the GPU to settle back into P8 idle (~7W, 0% fan).

• Manual terminal invocation of nvidia-smi -q -x -i 0 runs cleanly, confirming the hardware and kernel driver are alive, but the resulting XML schema diverges from what gpustat expects for Ada/Ampere cards.

Let me know if you would like a raw dump of nvidia-smi -q -x from this card/driver combination to help update the parser schemas for Blackwell.

Thanks for maintaining the plugin!

I started getting this error after update to 7.3.4:

Oct 29 13:34:17 unRAID kernel: NVRM: nvCheckOkFailedNoLog: Check failed: Out of memory [NV_ERR_NO_MEMORY] (0x00000051) returned from memdescAllocInternal(pMemDesc) @ memdesc.c:1359

Oct 29 13:34:19 unRAID kernel: ACPI BIOS Error (bug): Failure creating named object [\_SB.PC00.RP01.PXSX._DSM.USRG], AE_ALREADY_EXISTS (20250807/dsfield-184)

Oct 29 13:34:19 unRAID kernel: ACPI Error: AE_ALREADY_EXISTS, CreateBufferField failure (20250807/dswload2-477)

Oct 29 13:34:19 unRAID kernel: ACPI Error: Aborting method \_SB.PC00.RP01.PXSX._DSM due to previous error (AE_ALREADY_EXISTS) (20250807/psparse-543)

Oct 29 13:34:20 unRAID kernel: NVRM: nvCheckOkFailedNoLog: Check failed: Out of memory [NV_ERR_NO_MEMORY] (0x00000051) returned from memdescAllocInternal(pMemDesc) @ memdesc.c:1359

Oct 29 13:34:22 unRAID kernel: ACPI BIOS Error (bug): Failure creating named object [\_SB.PC00.RP01.PXSX._DSM.USRG], AE_ALREADY_EXISTS (20250807/dsfield-184)

Oct 29 13:34:22 unRAID kernel: ACPI Error: AE_ALREADY_EXISTS, CreateBufferField failure (20250807/dswload2-477)

Oct 29 13:34:22 unRAID kernel: ACPI Error: Aborting method \_SB.PC00.RP01.PXSX._DSM due to previous error (AE_ALREADY_EXISTS) (20250807/psparse-543)

Oct 29 13:34:23 unRAID kernel: NVRM: nvCheckOkFailedNoLog: Check failed: Out of memory [NV_ERR_NO_MEMORY] (0x00000051) returned from memdescAllocInternal(pMemDesc) @ memdesc.c:1359

Oct 29 13:34:25 unRAID kernel: ACPI BIOS Error (bug): Failure creating named object [\_SB.PC00.RP01.PXSX._DSM.USRG], AE_ALREADY_EXISTS (20250807/dsfield-184)

Oct 29 13:34:25 unRAID kernel: ACPI Error: AE_ALREADY_EXISTS, CreateBufferField failure (20250807/dswload2-477)

Oct 29 13:34:25 unRAID kernel: ACPI Error: Aborting method \_SB.PC00.RP01.PXSX._DSM due to previous error (AE_ALREADY_EXISTS) (20250807/psparse-543)

Oct 29 13:34:28 unRAID kernel: ACPI BIOS Error (bug): Failure creating named object [\_SB.PC00.RP01.PXSX._DSM.USRG], AE_ALREADY_EXISTS (20250807/dsfield-184)

Oct 29 13:34:28 unRAID kernel: ACPI Error: AE_ALREADY_EXISTS, CreateBufferField failure (20250807/dswload2-477)

Oct 29 13:34:28 unRAID kernel: ACPI Error: Aborting method \_SB.PC00.RP01.PXSX._DSM due to previous error (AE_ALREADY_EXISTS) (20250807/psparse-543)

Oct 29 13:34:31 unRAID kernel: ACPI BIOS Error (bug): Failure creating named object [\_SB.PC00.RP01.PXSX._DSM.USRG], AE_ALREADY_EXISTS (20250807/dsfield-184)

Oct 29 13:34:31 unRAID kernel: ACPI Error: AE_ALREADY_EXISTS, CreateBufferField failure (20250807/dswload2-477)

Oct 29 13:34:31 unRAID kernel: ACPI Error: Aborting method \_SB.PC00.RP01.PXSX._DSM due to previous error (AE_ALREADY_EXISTS) (20250807/psparse-543)

Oct 29 13:34:34 unRAID kernel: ACPI BIOS Error (bug): Failure creating named object [\_SB.PC00.RP01.PXSX._DSM.USRG], AE_ALREADY_EXISTS (20250807/dsfield-184)

Oct 29 13:34:34 unRAID kernel: ACPI Error: AE_ALREADY_EXISTS, CreateBufferField failure (20250807/dswload2-477)

Oct 29 13:34:34 unRAID kernel: ACPI Error: Aborting method \_SB.PC00.RP01.PXSX._DSM due to previous error (AE_ALREADY_EXISTS) (20250807/psparse-54

I am running a RTX PRO 4000 Blackwell with the 580 driver. It worked fine on 7.3.3. This error only shows up on when viewing the dashboard

Is it fair to say the plugin or its development is dead?