GPU MODELS / A16

A16 cloud GPU pricing

14 listings across 3 providers. Cheapest published rate: $0.082/GPU/hr at MatPool. Sorted by per-GPU hourly price. Every row links to the provider's own published price.

WHAT IS THE A16

The A16 is four small Ampere GPUs on a single PCB, built for one job: virtual desktop density. Each 16 GB slice serves vGPU sessions, so one 250 W card hosts dozens of virtual workstations. On rental panels it shows up cheap because it is a VDI part, not a compute part - and buyers comparing it to an L4 or A10 are asking the wrong question.

MAKER
NVIDIA
ARCHITECTURE
Ampere
LAUNCHED
2021
MEMORY
4x 16 GB GDDR6 ECC (64 GB total)
MEMORY BANDWIDTH
4x 200 GB/s
TENSOR COMPUTE
Four Ampere GPUs on one board (third-gen Tensor Cores per GPU)
POWER DRAW
250 W
INTERCONNECT
PCIe 4.0 x16

WHAT IT IS GOOD AT

VDI and virtual workstations
The only real fit. Four 16 GB GPUs per card with NVIDIA vPC/vWS software is the highest-density graphics VDI play of its generation.
Small inference jobs on idle VDI capacity
NVIDIA explicitly positions spare A16 capacity for off-hours compute. Workable, but each slice is a small 16 GB Ampere.
LLM training or serious inference
No. Four isolated 16 GB GPUs with 200 GB/s each cannot hold or feed modern models, and there is no NVLink between the slices.

WHO SHOULD NOT RENT THE A16

Any single-model workload over 16 GB - the four GPUs cannot share memory.
Compute buyers: an L4 at similar panel prices outperforms the whole card on one 24 GB slice with 300 GB/s and newer tensor cores.
ProviderConfigVRAMLocation$/hr$/GPU/hrSource
MatPoolMatPoolA16 × 115 GBChina$0.082$0.082receipt
VultrVultrA16 × 116 GBNot published$0.471$0.471STALEreceipt
Hostnot GPUHostnot GPUA16 × 216 GBAsia-Pacific$1.250$0.625receipt
Hostnot GPUHostnot GPUA16 × 216 GBEU West$1.250$0.625receipt
Hostnot GPUHostnot GPUA16 × 216 GBIndia South$1.250$0.625receipt
Hostnot GPUHostnot GPUA16 × 216 GBUS Central$1.250$0.625receipt
Hostnot GPUHostnot GPUA16 × 216 GBUS East$1.250$0.625receipt
Hostnot GPUHostnot GPUA16 × 216 GBUS West$1.250$0.625receipt
Hostnot GPUHostnot GPUA16 × 116 GBAsia-Pacific$0.630$0.630receipt
Hostnot GPUHostnot GPUA16 × 116 GBEU West$0.630$0.630receipt
Hostnot GPUHostnot GPUA16 × 116 GBIndia South$0.630$0.630receipt
Hostnot GPUHostnot GPUA16 × 116 GBUS Central$0.630$0.630receipt
Hostnot GPUHostnot GPUA16 × 116 GBUS East$0.630$0.630receipt
Hostnot GPUHostnot GPUA16 × 116 GBUS West$0.630$0.630receipt

THE A16 MARKET, BY THE NUMBERS

PROVIDERS
3
CONFIGURATIONS
14
MEDIAN $/GPU/HR
$0.63
SPREAD
7.7x

3 providers publish 14 A16 configurations on the panel today. Published per-GPU hourly rates run from $0.082 (MatPool) to $0.63 (Hostnot GPU), a 8x spread between the cheapest and the most expensive published rate for the same chip. The median listing sits at $0.63/GPU/hr. 13 rows were reported in stock by the provider at last observation; the rest are published catalog rates. All rates are on-demand - committed-use and reserved pricing is excluded.

CHEAPEST A16 BY GEOGRAPHY
chinaMatPool$0.082/hrindiaHostnot GPU$0.625/hrunited statesHostnot GPU$0.625/hr
A16 PRICE TREND, DAILY PANEL SNAPSHOTS

Across every A16 row on the panel, the daily floor held flat at $0.082/GPU/hr over 46 days of tracking. Snapshots run daily; the full history is public in the repo.

A16 PROVIDER NOTES
MatPool · 1 config · 1 region · 1 in stock
Sets the panel floor for this model.
from $0.082/hr
Hostnot GPU · 12 configs · 6 regions · 12 in stock
Widest regional footprint here - 6 regions.
from $0.625/hr
BUYING A16: WHAT THE PANEL SAYS

At the current floor, one A16 running around the clock costs about $60 a month (730 hours of on-demand arithmetic). Per-GPU floors by config size: 1x $0.08 · 2x $0.63. Compare the alternatives below before committing - the same budget often buys more than one chip class.

A16 VS THE ALTERNATIVES

A16 vs A10
Same generation, different intent: the A10 is one 24 GB GPU for compute and graphics, the A16 is four 16 GB GPUs for VDI density. For AI workloads the A10 wins outright; for desktops-per-rack the A16 wins outright.

Compare live rates: A16 vs A10

A16 PRICING FAQ

What is the cheapest A16 cloud GPU right now?
The cheapest published A16 rate on the panel is $0.082 per GPU-hour at MatPool (A16 x 1, China). Every price links to the provider's own published rate.
How much does a A16 cost per hour?
Across 3 providers tracked today, published per-GPU hourly rates for A16 run from $0.08 to $0.63, with a median of $0.63. Committed-use and reserved rates are excluded; everything shown is on-demand.
How many providers rent A16 GPUs?
3 providers publish 14 distinct A16 configurations on Compute Cafe. 13 of those rows were reported in stock by the provider's live endpoint at last observation.
Where is A16 cheapest?
By published per-GPU hourly rate, the cheapest geography for A16 today is china at $0.08/hr (MatPool). The spread across 3 tracked geographies runs to $0.63/hr in united states.
Are A16 prices going up or down?
Over the 46 days Compute Cafe has tracked the panel, the A16 floor is holding flat: $0.082 to $0.082/GPU/hr (+0.5%). Daily snapshots are public in the repo, so the trend is auditable.
What does a A16 cost per month?
At the current floor of $0.082/GPU/hr, a A16 running 24/7 costs about $60 per month (730 hours of on-demand arithmetic on the cheapest published rate). Committed-use contracts can price lower; the panel tracks on-demand rates only.
Is the A16 good for AI workloads?
Only incidentally. It is a VDI card - four separate 16 GB GPUs that cannot pool memory. A model must fit in a single 16 GB slice, and 200 GB/s per slice is slow for inference. Rent it for virtual desktops, not for models.
Why is the A16 so cheap on GPU rental panels?
Supply and use case: VDI fleets refreshed out large numbers of them, and the four-slice design does not map onto LLM workloads, so AI demand does not bid the price up the way it does for 24+ GB cards.

See also alternatives to the A16 · cheapest GPU-hour per model · inference chips · training chips · by location