Skip to content
Updated minutes ago
nvidia

A16

nvidia · ampere · 64 GB GDDR6 · 250W TDP

VRAM

64 GB

BF16 TFLOPS

16.8

Bandwidth

232 GB/s

From

$0.72/hr

Calculate ROI with this GPU →

Spec Sheet

VRAM64 GB GDDR6
Memory Bandwidth232 GB/s
BF16 TFLOPS16.8
FP16 TFLOPS16.8
FP8 TFLOPS16.8
INT8 TOPS33.6
TDP250W
InterconnectPCIE
Max per Node8
PCIe Gen4
CUDA Compute Capability8.6
Tensor CoresYes

Pricing by Provider

ProviderOn-DemandReservedSpotBadge
aws$1.10/hr$0.72/hr-Cheapest

Compatible Models (277)

Training Capabilities

Estimated GPU count for full fine-tuning (AdamW, BF16) and QLoRA

Model SizeFull Fine-TuneQLoRA
7B model3 GPUs1 GPU
13B model4 GPUs1 GPU
70B model21 GPUs1 GPU

Energy Efficiency

Estimated tokens/second per Watt for popular models

Mistral 7B
0.13 t/s/WFP8
Qwen 2.5 7B
0.12 t/s/WFP8
Llama 3.1 8B
0.12 t/s/WFP8
Llama 3.1 70B
0.01 t/s/WFP8
Qwen 2.5 72B
0.01 t/s/WFP8

Similar GPUs

GPUVRAMBF16 TFLOPSBW (GB/s)From
A100 80GB SXM80 GB3122039$1.19/hr
A100 80GB PCIe80 GB3122039$1.05/hr
RTX A600048 GB38.7768$0.49/hr
A4048 GB37.4696$0.42/hr
A100 40GB SXM40 GB3121555$0.85/hr

Methodology Note

Performance estimates for the A16are based on InferenceBench's roofline performance model with CUDA kernel-level optimization including FlashAttention v2 and PagedAttention. Memory calculations account for model weights (64 GB GDDR6 available), KV-cache allocation, and activation memory. Throughput predictions use the A16's rated 232 GB/s memory bandwidth and 16.8 BF16 TFLOPS compute capacity as roofline ceilings, with empirical correction factors per GPU architecture (ampere). See our full methodology.

Frequently Asked Questions

How many AI models can run on A16?

The A16 can run 277 AI models from our database within a single node. Compatible models range across various parameter sizes depending on the quantization precision (BF16, FP8, INT4). Smaller models fit on a single GPU while larger models may require multi-GPU setups up to 8x A16.

What is the A16 inference throughput?

The A16 delivers 16.8 BF16 TFLOPS and 16.8 FP8 TFLOPS with 232 GB/s memory bandwidth. Actual inference throughput (tokens/sec) depends on the model size, precision, and batch size. Use our calculator for model-specific throughput estimates.

How much does A16 cost per hour?

The A16 is available starting from $0.72/hour via aws. Prices vary by provider and pricing tier (on-demand, reserved, spot). Compare pricing across all providers in the table above.