NVIDIA A16 64GB
One board, four GPUs, 64GB of GDDR6 with ECC: the A16 exists to put graphics-rich virtual desktops on cheap compute. Add NVIDIA vPC licensing and a single 250W dual-slot card carries up to 64 concurrent users, with 4x NVENC and 8x NVDEC handling the video. It is a VDI and virtual-workstation card first — not an AI accelerator. Graded ex-fleet stock, tested and shipped from Hong Kong.
Fully Tested & Documented · Hong Kong Warehouse Stock · Worldwide Shipping 3–7 Days · 1-Year Warranty
Hardware Gallery
The A16, Up Close
Four GPUs on One PCB
The A16 is a quad-GPU board: four 16GB GPUs behind a single PCIe Gen4 x16 interface. That architecture is the whole point — one slot, one 250W budget and up to 64 virtual desktops — and it is also why we test all four GPUs independently rather than treating the card as one device.
Passive, Full-Height, Full-Length
Angle view showing the full-length passive heatsink and 8-pin power inlet. Like every passive data center card this one lives or dies by chassis airflow, and it occupies a full-height full-length double slot — check your VDI host's riser and airflow before ordering.
4x16
GB GDDR6 with ECC (64GB total)
64
concurrent VDI users per board
250
W TDP, dual-slot passive
4
NVENC engines (8x NVDEC, AV1 decode)
Key Specifications
| Component | Specification |
|---|---|
| Model | NVIDIA A16 64GB PCIe (quad-GPU board; Dell and other OEM-branded boards are the same GPU) |
| Architecture | Ampere, 4x 1,280 CUDA cores (5,120 total), 4x 40 third-gen Tensor cores, 4x 10 second-gen RT cores |
| Memory | 4x 16GB GDDR6 with ECC (64GB total) |
| Interface | PCIe Gen4 x16, full-height full-length dual-slot, passive |
| Power | 250W TDP, 8-pin auxiliary power connector |
| Media | 4x NVENC / 8x NVDEC — includes AV1 decode |
| Virtualization | vGPU support: NVIDIA vPC, vApps, RTX vWS and vCS |
| User Density | Up to 64 concurrent VDI users per board |
| Compliance | Secure and measured boot with hardware root of trust; NEBS Level 3 |
The A16 is bought for user density, so buy it with the licensing decision already made: NVIDIA vPC, vApps, RTX vWS and vCS entitlements are sold separately, and the software is what turns four 16GB GPUs into 64 virtual desktops. If your goal is actually inference rather than desktops, compare the single-GPU L4 for dense serving or the A100 40GB for real tensor throughput — both put more usable compute per dollar behind a model. A16 cards typically come out of 2U VDI hosts such as the PowerEdge R750.
Transparent Pricing
2026 Reference Pricing by Configuration
Tracked Sep 20, 2026 — nine asking prices for the A16 64GB, every one of them a Pre-Owned listing. There is no new-retail channel in this dataset, and the spread is unusually tight: the whole board trades inside a $780 band.
Pre-Owned — Entry Band
$2,190-$2,800
Four tracked listings: $2,189.99 (Dell), $2,450, $2,469 and $2,800 (band shown rounded to whole dollars, as the index reports it). Dell-branded boards appear at both ends of this band, so branding alone does not explain the spread — condition grading and what is included do.
Pre-Owned — Upper Band
$2,899-$2,970
Five tracked listings: $2,899 twice, $2,900 (Dell), $2,929.99 and $2,970. At the top of the band you are paying about $800 more than the entry listing for an identical spec — worth asking what test report and warranty come with it.
Asking prices, not transaction prices — volume deals typically close 5-15% below asking. Sources: eBay seller listings (NVIDIA and Dell variants, all Pre-Owned). Collected Sep 20, 2026. Full dataset: Used Server Price Index.
Origin
Hong Kong Warehouse
Free port — full export documentation
Transit
3–7 Days Air Freight
Tracked, insured, worldwide
Warranty
1-Year Warranty
Diagnostic report in every box
Quality Process
What We Test Before an A16 Ships
All Four GPUs Individually
Each of the board's four GPUs is enumerated and stress-tested on its own. A quad-GPU card that reports healthy while one device is dead is the single most common A16 failure mode, and it is not shippable.
Per-GPU Memory and ECC
All four 16GB memory partitions are scrubbed with ECC counters logged per device, so you see which GPU — not just which board — has history.
vGPU Profile Sweep
vGPU profiles are created and destroyed across the board, including mixed profiles on one card, because heterogeneous user profiles are the reason buyers choose an A16 over four separate cards.
Media Engine Test
All four NVENC and eight NVDEC engines are exercised with H.264, H.265 and AV1 decode pipelines — in VDI the video engines carry the user experience, so a dead encoder is a real defect.
Ex-Fleet Provenance and Airflow Check
Serial traced to its source fleet, OEM branding declared on the invoice, and the passive heatsink inspected for fin damage. We confirm your chassis provides the front-to-back airflow a 250W passive board needs.
Report in the Box
Every unit ships with its per-machine diagnostic report and warranty. Full process: inspection guide.
Frequently Asked Questions
How much is a used NVIDIA A16 64GB in 2026?
The tracked range is $2,190-$2,970 across nine Pre-Owned eBay asking prices collected Sep 20, 2026 — no new-retail listings appeared in this dataset. Four listings sit in the $2,190-$2,800 entry band and five in the $2,899-$2,970 upper band, so the whole market is inside a $780 spread and condition grading matters more than price shopping.
NVIDIA A16 vs A100 — can the A16 run AI workloads?
They are built for different jobs, and the A16 loses badly if you buy it for the wrong one. The A16 splits 5,120 CUDA cores and 64GB of GDDR6 across four separate GPUs, each with its own 16GB memory partition tuned for desktop graphics. An A100 40GB puts 6,912 CUDA cores, 40GB of HBM2 at roughly 1,555GB/s and 312 TFLOPS of FP16 tensor throughput behind a single device. Small quantized models and light inference can run on A16 slices, but if your workload is model serving or fine-tuning, the A100-class card is the correct purchase — see our A100 40GB page for current tracking. The A16 wins when the deliverable is virtual desktops, not tokens.
How many VDI users can one A16 support?
NVIDIA rates the board at up to 64 concurrent users, and that ceiling assumes NVIDIA vPC licensing and a light, graphics-rich office workload. Real density depends on the profile you choose: knowledge workers fit many per GPU, while CAD, medical imaging and multi-monitor power users consume far more of a 16GB partition each. Tell us your user profile mix and we will help you size the board count rather than the marketing number.
Is the A16 four GPUs or one?
Four. The A16 is a single PCIe Gen4 x16 card carrying four independent Ampere GPUs with 16GB of GDDR6 each, and vGPU software sees four devices that can run different profiles at the same time. That is what allows mixed user types — a virtual PC for one team, a vWS workstation for another — on one board. It also means one failed GPU degrades a quarter of the card, which is why every unit we sell ships with a per-GPU test result.
Does the A16 come with vGPU licensing?
No. We sell the hardware; NVIDIA vPC, vApps, RTX vWS and vCS entitlements are licensed separately through NVIDIA or its channel, and NVIDIA AI Enterprise is a separate subscription too. Budget for licensing alongside the card — an A16 without vGPU entitlement is a very expensive four-GPU paperweight. Every card we ship carries a 1-year warranty, and we will confirm the profile and licensing path for your hypervisor before quoting.
Get a Current NVIDIA A16 Quote
Tell us how many virtual desktops you are sizing for, which hypervisor and vGPU edition you run, and the chassis the card is going into — we confirm live graded stock, quote volume pricing below the tracked band, and ship worldwide from Hong Kong, usually within one business day.
Request a Quote