Low-Profile GPU Upgrade for an Avigilon NVR

An Avigilon NVR4 Value can take a better GPU for Appearance Search and Face Recognition, but only one shape of card fits its 1U chassis: single-slot, low-profile, and powered entirely from the slot, with no 6- or 8-pin connector. The NVIDIA RTX A1000 8 GB is the card to buy. It fits, it has four times the memory of Avigilon’s own analytics kit, and Avigilon’s Unity Video 8 datasheet names it as the recommended analytics GPU.

Why the Analytics Kit card falls short

The NVR4 Value ships without a GPU; the STD and PRM models come with Avigilon’s HD-NVR-ANK2-1 Analytics Kit. Its card is an NVIDIA Quadro P620: Pascal, 2 GB, single-slot, low-profile, about 40 W.

That’s half of what Avigilon now asks for. The ACC 7 and Unity Video 8 requirements both call for at least 4 GB of GPU memory for Appearance Search and Face Recognition.

What Avigilon supports

From Avigilon’s Unity Video 8 datasheet (October 2025) and the ACC 7 system requirements:

	Unity Video 8	ACC 7
Architectures	Ada Lovelace, Ampere, Turing, Pascal	Ampere, Turing, Pascal (Maxwell for Appearance Search only)
GPU memory	4 GB or more	4 GB or more
Driver	520.06 or later	Ampere 473.81, Turing 462.31, Pascal 451.48 or later
Recommended card	RTX A1000 8 GB	T1000 or P1000

Unity’s GPU requirements apply from Unity 8.5.0. Ada Lovelace cards aren’t on the ACC 7 list.

What physically fits

The NVR4 Value’s chassis sets three hard rules for any card:

• Single slot. It’s a 1U server; a dual-slot card won’t go in, however short.
• Low-profile, to be safe. Avigilon’s own kit card is low-profile, and Avigilon’s guide only says it goes in “the appropriate expansion riser slot”. If the chassis is the Dell R340 it appears to be, its riser has one full-height slot and one low-profile slot. A low-profile card fits either, so check your riser before buying anything bigger.
• No power connector. Everything comes from the slot.

The power supply is 350 W, and it isn’t the limit: every candidate draws about 50 W, 10 W more than the P620.

Cards that fit

Card	Architecture	Memory	Power	Supported by
RTX A1000	Ampere	8 GB	50 W	ACC 7 and Unity 8. Unity’s recommended card
RTX 2000E Ada	Ada Lovelace	16 GB	50 W	Unity 8 only
T1000 8 GB	Turing	8 GB	50 W	ACC 7 and Unity 8
RTX A400	Ampere	4 GB	50 W	ACC 7 and Unity 8, at the minimum memory

All four are single-slot, low-profile and slot-powered.

• RTX A1000: the pick. 8 GB, Tensor cores, supported by both ACC and Unity, and easy to buy.
• RTX 2000E Ada: the most memory that fits, but Unity-only and scarce. Only the “E” version is single-slot.
• T1000 8 GB: the fallback if the A1000 is out of stock. Same memory, older and slower.
• RTX A400: meets the 4 GB minimum and nothing more.

Cards that look right but don’t fit

Several cards appear to match on paper and fail on one rule:

• RTX 2000 Ada (without the E): low-profile and slot-powered, but dual-slot and 70 W. This is the easy mistake, because it’s far more common than the 2000E.
• RTX A2000 and RTX 4000 SFF Ada: low-profile, but dual-slot.
• T400: fits, but its 2 GB version is below Avigilon’s 4 GB minimum.
• Anything with a power connector, whatever its size.

Fit the card

The order matters: the NVIDIA driver goes on before the card is switched to TCC mode.

1. Note the NVR’s configuration and make sure recordings are safe, as for any hardware change.
2. Shut down the NVR and unplug its power.
3. Fit the low-profile bracket to the card if it isn’t already on.
4. Install the card in the expansion riser and screw the bracket in. If both riser slots are free, the full-height slot is the one with more PCIe lanes on an R340 (x8 rather than x4).
5. Power on and let Windows start.
6. Install the NVIDIA driver for the card, at or above the version Avigilon lists for your ACC or Unity release.
7. Switch the card to TCC mode, as described below. Avigilon’s GPU Configuration Tool on the official kit’s USB drive does the same job.

Switch the card to TCC mode

Windows runs an NVIDIA card in WDDM mode by default, as a display adapter. For analytics it should run in TCC (Tesla Compute Cluster) mode, which is compute only. Until it does, Avigilon’s Site Health report shows a warning for the GPU.

In TCC mode the card can’t drive a monitor. That’s fine on the NVR4 Value, because the monitor goes on the server’s own VGA port, not the GPU.

Open Command Prompt or PowerShell as administrator, then:

1. Find the card’s index and current mode:

   nvidia-smi --query-gpu=index,name,driver_model.current,driver_model.pending --format=csv

   As the only NVIDIA card in the NVR, it’s index 0, showing WDDM.

2. Switch it to TCC:

   nvidia-smi -i 0 -dm TCC

   nvidia-smi confirms the change and says a reboot is needed. If it refuses because a display is attached, unplug the monitor from the GPU and run it again. Don’t use -fdm to force it. If the card doesn’t support TCC, nvidia-smi says so and nothing changes.

3. Restart the NVR.

4. Check the change took:

   nvidia-smi --query-gpu=index,name,driver_model.current --format=csv

   It should now say TCC. Plain nvidia-smi shows the mode too, in the TCC/WDDM column.

Check the mode again after any NVIDIA driver update. To undo it, run nvidia-smi -i 0 -dm WDDM and restart.

Check it worked

• In Windows Device Manager, the card appears under Display adapters with no warning icon.
• nvidia-smi shows the card in TCC mode with its 8 GB of memory, and GPU load rises once analytics are running.
• Site Health no longer warns about the GPU.
• In the Avigilon client, Appearance Search returns results for cameras recorded on that NVR.
• GPU temperature stays steady under a full analytics load. A small card in a 1U chassis depends on the server’s own airflow, so keep the front clear and the filters clean.