NVIDIA A100 GPU: Specs, VRAM, Price & Benchmarks (2026)
The NVIDIA A100 (80GB HBM2e) launched in 2020. This guide covers its full specs, VRAM, SXM/PCIe differences where they apply, AI performance, and live per-GPU rental pricing on Aquanode.
Short answer on cost: the A100 rents from $0.991 per GPU per hour on Aquanode.
How much VRAM does the A100 have?
The A100 has 80GB HBM2e, with 2,039 GB/s of peak memory bandwidth.
- VRAM: A100 80GB HBM2e vs H100 80GB HBM3
- VRAM: A100 80GB HBM2e vs V100 16GB or 32GB HBM2
What fits in 80GB HBM2e of VRAM
| Model | Precision | Fits? |
|---|---|---|
| Llama 3 8B | BF16 | ~16GB. Fits with room to spare. |
| Llama 3.1 70B | BF16 | ~140GB. Needs 2 GPUs; a single 80GB card can't hold it. |
| Llama 3.1 70B | INT4 | ~35-40GB. Fits on a single card. |
| Mistral 7B / similar | BF16 | ~14GB. Fits easily, leaves room for several MIG partitions. |
Approximate, based on published parameter counts and standard bytes-per-parameter rules of thumb (FP16 ≈ 2 bytes/param, INT4 ≈ 0.5-0.6 bytes/param). Real footprint also depends on KV-cache size and framework overhead.
What can the A100 run?
Popular open models from small to frontier scale, with the memory each needs and how many A100 cards (80GB HBM2e each) that takes.
| Model | As published | FP8 | INT4 |
|---|---|---|---|
| Qwen/Qwen3-8B 8.2B | BF16: ~18.3 GB, 1 GPU | FP8: not supported | INT4: ~4.6 GB, 1 GPU |
| Qwen/Qwen2.5-14B-Instruct 14.8B | BF16: ~33 GB, 1 GPU | FP8: not supported | INT4: ~8.3 GB, 1 GPU |
| Qwen/Qwen3-32B 32.8B | BF16: ~73.2 GB, 1 GPU | FP8: not supported | INT4: ~18.3 GB, 1 GPU |
| Qwen/Qwen-72B 72.3B | BF16: ~162 GB, 3 GPUs | FP8: not supported | INT4: ~40.4 GB, 1 GPU |
| MiniMaxAI/MiniMax-M2.7 228.7B | FP8: not supported | – | INT4: ~128 GB, 2 GPUs |
| deepseek-ai/DeepSeek-R1 684.5B | FP8: not supported | – | INT4: ~383 GB, 5 GPUs |
Estimates: weights at the stated precision plus a flat 20% for KV cache and overhead, at a moderate context length. A dash means the precision is not offered for that model (it is already published at that size). INT4 needs a published quantized checkpoint. Open any model for a per-GPU breakdown, or use the A100 VRAM calculator.
A100 VRAM calculator: check which models fit in its memory at each precision.
All models that fit in 80 GB: the open models whose weights and overhead fit, at native, FP8 and INT4 precision.
A100 specs
A100 SXM4
| VRAM | 80GB HBM2e |
| Memory bandwidth | 2,039 GB/s |
| TDP | 400W |
| Form factor | SXM (DGX A100 / HGX A100 platform only) |
| Interconnect | NVLink 3, 600 GB/s |
A100 PCIe
| VRAM | 80GB HBM2e |
| Memory bandwidth | 1,935 GB/s |
| TDP | 300W |
| Form factor | Standard PCIe, dual-slot |
Specs sourced from the vendor's public datasheet/product page. See the source.
Related reading: A100 vs H100, A100 vs V100, and The best GPUs for AI, ranked.
GPU Glossary: What is VRAM?, HBM, Tensor Cores, CUDA Cores, TFLOPS, NVLink vs PCIe
A100 SXM4 vs PCIe: which should you use?
- Memory bandwidth: SXM4 delivers 2,039 GB/s vs 1,935 GB/s on PCIe.
- Interconnect: SXM4 has NVLink 3, 600 GB/s. PCIe has no dedicated GPU-to-GPU interconnect.
- Power and form factor: SXM4 is 400W in a SXM (DGX A100 / HGX A100 platform only) form factor. PCIe is 300W in a Standard PCIe, dual-slot form factor.
For large-scale distributed training, the variant with NVLink and the highest memory bandwidth is usually the right choice, since those advantages compound across a multi-GPU cluster. For single-GPU inference or fine-tuning, the lower-power variant often gives the same usable VRAM at a lower hourly rate.
A100 AI performance
Still one of the most cost-effective cards for BF16 training and fine-tuning at the 7-30B scale, and its Multi-Instance GPU (MIG) support lets one 80GB card be partitioned into up to seven isolated instances for many small inference workloads at once.
- Dense FP16/BF16 tensor throughput: 312 TFLOPS
- Memory bandwidth: 2,039 GB/s
No FP8 tensor cores at all (FP8 shipped starting with the next generation, Ada/Hopper), so it can't hit the throughput or memory savings an H100/H200 gets from FP8 inference. Expect roughly half the tensor throughput of an H100 on comparable BF16 workloads.
See how it stacks up against other cards in the GPU benchmarks and specs table.
The cheapest A100 offer right now ($0.991/GPU/hr) is about 50% below the market median of $1.97/GPU/hr.
A100 price: what does it cost?
Buying. $9,500-$15,000 (commonly quoted street price (no fixed retail; OEM channel)). Older 40GB SXM4 cards trade for less; 80GB PCIe and SXM4 both fall in this range depending on new-vs-refurbished condition.
Renting. The cheapest current on-demand rate for the A100 on Aquanode is $1.49/GPU/hr. Live rates range from $0.991 to $1.97 per GPU per hour, with a median of $1.97/GPU/hr. Billed by the offer's own terms; the table below shows every live rate.
| Region | $/GPU/hr | Available | VRAM | vCPU | RAM |
|---|---|---|---|---|---|
| South Korea | $0.991 | 2 | 80 GB | 20 | 126 GB |
| – | $1.31 | 0 | 80 GB | – | – |
| Des Moines, Us | $1.49 | 1 | 80 GB | 16 | 96 GB |
| Hou, Us | $1.72 | 15 | 80 GB | 43.5 | 240.3 GB |
| United States | $1.78 | 8 | 80 GB | 24 | 120 GB |
| Finland | $2.08 | 1 | 80 GB | 22 | 120 GB |
A100 price history
Aquanode stores one snapshot of its GPU price index per UTC day. For the A100 that is 5 days so far, 2026-10-06 to 2026-10-10, so this is a short history, not a long-run trend. The lowest per-GPU rate went from $1.17 on 2026-10-06 to $1.23 on 2026-10-10.
| Day (UTC) | Lowest per GPU hour | Median per GPU hour | Data-center lowest per GPU hour | Offers |
|---|---|---|---|---|
| 2026-10-10 | $1.23 | $1.97 | $1.49 | 60 |
| 2026-10-09 | $1.21 | $1.97 | $1.49 | 53 |
| 2026-10-08 | $1.31 | $1.72 | $1.49 | 41 |
| 2026-10-07 | $1.21 | $1.75 | $1.49 | 55 |
| 2026-10-06 | $1.17 | $1.75 | $1.49 | 53 |
Each row is the stored daily snapshot of the live index, copied as recorded. A dash means that day stored no figure. The same series is available as JSON and summarised in the monthly GPU price report.
How this price is calculated
All prices on this page are normalized to a per-GPU hourly rate using each offer's authoritative GPU count, so that raw price is divided by the number of GPUs it actually covers; some offers report price as already per-GPU, so those are used as-listed. An offer with a missing, zero, or invalid GPU count is excluded entirely rather than published at a guessed rate.
No offers were excluded from this snapshot for a missing or invalid price. No offers were dropped as price outliers in this snapshot.
Only the cheapest qualifying offer per provider is shown in the table above. This page regenerates at most once per hour.
A100 vs H100: how do they compare?
- VRAM: A100 80GB HBM2e vs H100 80GB HBM3
- Memory bandwidth: 2,039 GB/s vs 3.35 TB/s
- Dense FP16 tensor throughput: 312 TFLOPS vs 989 TFLOPS
On Aquanode right now, A100 starts at $0.991/GPU/hr against H100's $2.19/GPU/hr, about 55% less.
A100 vs V100: how do they compare?
- VRAM: A100 80GB HBM2e vs V100 16GB or 32GB HBM2
- Memory bandwidth: 2,039 GB/s vs 900 GB/s
- Dense FP16 tensor throughput: 312 TFLOPS vs 125 TFLOPS (SXM2), 112 TFLOPS (PCIe)
On Aquanode right now, V100 starts at $0.088/GPU/hr against A100's $0.991/GPU/hr, about 91% less.
Compare the A100 with other GPUs
Compare A100 with
- A100 vs B200
- A100 vs H200
- A100 vs H100
- A100 vs DGX A100
- A100 vs V100
- A100 vs AMD MI300X
- A100 vs L40S
- A100 vs L40
- A100 vs A40
- A100 vs RTX A6000
- A100 vs RTX A5000
- A100 vs RTX A4000
- A100 vs L4
- A100 vs T4
- A100 vs RTX 6000 Ada
- A100 vs RTX PRO 6000
- A100 vs RTX PRO 6000 WS
- A100 vs RTX PRO 6000 SE
- A100 vs RTX PRO 5000
- A100 vs RTX 5090
- A100 vs RTX 5080
- A100 vs RTX 5070 Ti
- A100 vs RTX 5070
- A100 vs RTX 5060 Ti
- A100 vs RTX 4090
- A100 vs RTX 4080 Super
- A100 vs RTX 4080
- A100 vs RTX 4070 Ti
- A100 vs RTX 4070 Super
- A100 vs RTX 4070
- A100 vs RTX 4060 Ti
- A100 vs RTX 3090
- A100 vs RTX 3080
- A100 vs RTX 3070
- A100 vs RTX 3060
- A100 vs B300
- A100 vs AMD MI355X
- A100 vs AMD MI325X
- A100 vs H100 NVL
- A100 vs RTX PRO 4500
- A100 vs RTX PRO 4500 SE
- A100 vs RTX PRO 4000
- A100 vs RTX 5880 Ada
- A100 vs RTX 5000 Ada
- A100 vs RTX 4000 Ada
- A100 vs RTX 4000 SFF Ada
- A100 vs RTX 2000 Ada
- A100 vs RTX 4070 Ti Super
- A100 vs RTX A4500
- A100 vs RTX 2080 Ti
- A100 vs RTX 3080 Ti
- A100 vs RTX 3070 Ti
- A100 vs RTX 3060 Ti
- A100 vs GTX 1080 Ti
- A100 vs RTX 4060
- A100 vs RTX 5060
- A100 vs GTX 1660 Super
- A100 vs RTX 2060
- A100 vs Radeon RX 9070 XT
- A100 vs Radeon RX 6700 XT
- A100 vs RTX 6000
- A100 vs A16
- A100 vs P40
- A100 vs P4
- A100 vs Quadro P2000
- A100 vs Quadro M4000
Get notified when the price drops
GPU supply moves hourly. Tell us what you're waiting for and we'll email you when a matching offer appears across any provider we track.
Rent a A100 on Aquanode
- On-demand instances from $0.991/GPU/hr, billed by the provider's own terms, with no hardware procurement or long-term commitment.
- 65 live offers across 6 regions today.
- Set a price/availability alert above to hear the moment a cheaper or newly-available A100 offer appears.
- Compare every A100 offer side by side, or browse the full multi-provider GPU marketplace.
Good for
Still one of the most cost-effective cards for BF16 training and fine-tuning at the 7-30B scale, and its Multi-Instance GPU (MIG) support lets one 80GB card be partitioned into up to seven isolated instances for many small inference workloads at once.
Not good for
No FP8 tensor cores at all (FP8 shipped starting with the next generation, Ada/Hopper), so it can't hit the throughput or memory savings an H100/H200 gets from FP8 inference. Expect roughly half the tensor throughput of an H100 on comparable BF16 workloads.
A100 FAQs
How much VRAM does the A100 have?
The A100 has 80GB HBM2e, with 2,039 GB/s of peak memory bandwidth.
What is the NVIDIA A100?
The NVIDIA A100 is a GPU released in 2020, with 80GB HBM2e of memory and a 400W (SXM) power envelope. See the full spec table above for interconnect, form factor and tensor-throughput details.
How much does it cost to rent a A100?
Live A100 rental prices currently range from $0.991 to $1.97 per GPU per hour, with a median of $1.97 per GPU per hour.
What's the cheapest A100 rate?
The lowest current A100 rate on Aquanode is $0.991 per GPU per hour in South Korea.
What is the difference between A100 SXM4 and PCIe?
SXM4 has 2,039 GB/s of memory bandwidth and NVLink 3, 600 GB/s. PCIe has 1,935 GB/s, no dedicated GPU-to-GPU interconnect. Both carry 80GB HBM2e.
How does the A100 compare to the H100?
VRAM: A100 80GB HBM2e vs H100 80GB HBM3 Memory bandwidth: 2,039 GB/s vs 3.35 TB/s Dense FP16 tensor throughput: 312 TFLOPS vs 989 TFLOPS On Aquanode right now, A100 starts at $0.991/GPU/hr against H100's $2.19/GPU/hr, about 55% less. See the full A100 vs H100 comparison for a shared-provider price breakdown.
How does the A100 compare to the V100?
VRAM: A100 80GB HBM2e vs V100 16GB or 32GB HBM2 Memory bandwidth: 2,039 GB/s vs 900 GB/s Dense FP16 tensor throughput: 312 TFLOPS vs 125 TFLOPS (SXM2), 112 TFLOPS (PCIe) On Aquanode right now, V100 starts at $0.088/GPU/hr against A100's $0.991/GPU/hr, about 91% less. See the full A100 vs V100 comparison for a shared-provider price breakdown.
What is the A100 good for?
Still one of the most cost-effective cards for BF16 training and fine-tuning at the 7-30B scale, and its Multi-Instance GPU (MIG) support lets one 80GB card be partitioned into up to seven isolated instances for many small inference workloads at once.
What are the A100's limitations?
No FP8 tensor cores at all (FP8 shipped starting with the next generation, Ada/Hopper), so it can't hit the throughput or memory savings an H100/H200 gets from FP8 inference. Expect roughly half the tensor throughput of an H100 on comparable BF16 workloads.
Is renting cheaper than buying?
Renting avoids the upfront hardware cost and lets you match spend to actual usage. A rented A100 at $0.991/hr only costs money while it's running, whereas buying ties up capital in hardware that keeps depreciating whether it's in use or not. Buying outright runs $9,500-$15,000 (commonly quoted street price (no fixed retail; OEM channel)). Which is cheaper depends on how continuously you'd run it; short or bursty workloads usually favor renting.
How is the A100 price calculated?
All prices are normalized to a per-GPU hourly rate using each offer's authoritative GPU count, which the raw price is divided by; some offers report price as already per-GPU. Offers whose price can't be safely normalized, or whose rate is an extreme outlier against the rest of the market, are excluded.
A100 price by region
Related guides
Other models in the same generation, then the rest of the GPU index.
