GPUBeacon
Language
NVIDIA

Cloud GPU · verified 2026-09-05

NVIDIA A10

Ampere24 GB

Ampere 24 GB GDDR6, 150 W, the quiet inference card hyperscalers still have in quota. Not a trainer. Not a 4090 killer. A cheap, cool, boring GPU, which is sometimes the point.

Author Valentin FourierLast modified

VRAM24 GB
From /GPU/hr$0.45
ArchitectureAmpere
VendorNVIDIA

In brief

What to know before you rent

Catalog listings

A10 Pricing and Availability

Published on-demand rates first. Spot and waitlists are labeled. Confirm the live rate before you provision.

ProvidersRegionBilling typeInterconnect/GPU/hr
RunPodIn stockUS-EastOn-demandPCIe$0.45Open provider
Vast.aiIn stockUS-EastOn-demandPCIe$0.49Open provider
AWSIn stockUS-EastOn-demandPCIe$1.01Open provider
Google CloudLimitedUS-WestOn-demandPCIe$1.10Open provider

Our reading

Who this GPU is for

I rent A10 when the workload is inference or light fine-tunes that fit 24 GB and the cloud already has quota. I do not rent A10 to race a 4090 on diffusion. The 4090 will win and then crash. The A10 will finish the batch.

Rent it if Always-on endpoints, T4/A10-shaped serving, hyperscaler shops with existing A10 quota.

Skip it if Training that wants Tensor cores at 4090/L40S levels, 48 GB+ models, NVLink jobs.

How we compare hours

Datasheet

Technical specifications

NVIDIA A10 · Per GPU · PCIe

Compute

FP32
31.2 TFLOPS
FP16 Tensor
125 TFLOPS250 with sparsity
CUDA cores
9,216
Precision support
FP16, BF16, TF32, FP32, INT8

Memory

Capacity
24 GB GDDR6
Bandwidth
600 GB/s
ECC
Yes

Silicon

Architecture
Ampere
Class
Datacenter PCIe / graphics+inference

Fabric and host

GPU interconnect
None
Host interface
PCIe 4.0 x16

Power

Board power
150 W

A10 is not A10G (AWS branding) automatically, and it is not A100. If the listing says A10G, read AWS’s SKU. This review is the 24 GB A10 class clouds put next to T4/L4.

Trade-offs

Pros and cons of A10

What this GPU does well, and when the hour is the wrong buy.

Strengths

  • 150 W means dense packing and fewer thermal novels
  • 24 GB is enough for a lot of production 7B serving
  • Enterprise PCIe, not a GeForce lottery
  • Often sitting in AWS/GCP/Azure already

Limits

  • Ampere without A100’s HBM
  • Slow versus Ada/Blackwell consumer cards on raw throughput
  • Easy to keep running because it is cheap, then forget you should have moved to L4/L40S

A lane, not a destination

A10 is what you rent when the endpoint must stay up and the model fits. The moment you are waiting on samples, price a 4090 or L40S. Leaving production on A10 “because it works” is valid. Leaving training on A10 is how epochs become weekends.

How I actually rent it

  1. 01

    Confirm 24 GB is enough

    Including KV cache at your real context.

  2. 02

    Compare L4

    Newer Ada, 72 W, often better efficiency. A10 is the incumbent.

  3. 03

    Do not train here by accident

    If the job looks like a trainer, leave.

FAQ

A10 questions

Who has the cheapest A10?+

As of September 5, 2026, the cheapest verified in-stock on-demand A10 tracked by GPUBeacon is $0.45 per GPU per hour from RunPod.

How much does the A10 cost per month?+

As of September 5, 2026, at 720 hours per month, one A10 costs an estimated $540 at the median on-demand price in this sample. Cheapest verified in stock: $324 per month on-demand.

Where can I rent an A10?+

GPUBeacon currently tracks A10 configs from 4 providers. The cheapest verified in-stock on-demand configs come from RunPod, Vast.ai and AWS. See the full price comparison above for every provider and config.

Are A10 prices going up or down?+

GPUBeacon publishes a catalog sample, not a 90-day price index. As of September 5, 2026, the median on-demand A10 in this sample is $0.75 per GPU per hour. The cheapest verified in-stock hour is $0.45.

Why choose the A10?+

Always-on endpoints, T4/A10-shaped serving, hyperscaler shops with existing A10 quota.

When is the A10 not a good fit?+

Training that wants Tensor cores at 4090/L40S levels, 48 GB+ models, NVLink jobs.

A10 or T4?+

A10 is the upgrade: 24 GB vs 16 GB, Ampere vs Turing. If you are still on T4 for a 7B, move.

A10 or RTX 4090?+

4090 is faster and riskier. A10 is calmer and slower. Serving vs experimenting.

Keep comparing

Alternatives to NVIDIA A10

Same job, different envelope. Compare VRAM, bandwidth and the hour before you lock a badge.

NVIDIA L4

24 GB GDDR6 · 50% less bandwidth than the A10

From$0.39/ GPU / hr

Compare

NVIDIA RTX 4090

24 GB GDDR6X · 68% more bandwidth than the A10

From$0.29/ GPU / hr

Compare

NVIDIA L40S

48 GB GDDR6 · 44% more bandwidth than the A10

From$0.79/ GPU / hr

Compare

NVIDIA A100

80 GB HBM2e · 3.4x the bandwidth of the A10

From$1.19/ GPU / hr

Compare

Browse all GPUs

Sources

Where these numbers come from

TFLOPS and TDP are from the vendor datasheet. Hourly prices are from the GPUBeacon catalog sample. Open the originals, then confirm the live SKU with the provider before you rent.