A40 vs A100: Specs, VRAM & Price
Short answer: pick the A100 when you're training or running bandwidth-bound large-batch inference. Pick the A40 when your workload is lighter-weight (rendering, virtualization, small-batch inference) and the lower cost matters more.
The A100 has 80 GB of VRAM; the A40 has 48 GB.
You can rent both by the hour, no purchase required. A100 from $0.991/GPU/hr; A40 from $0.649/GPU/hr. The A40 is 35% cheaper at the entry rate.
A100
Good for training and inference on models that don't need FP8 kernels
A40
Good for training and inference on models that don't need FP8 kernels
Last updated: 2026-10-10 21:21:46 UTCRefreshes hourly
Specs side by side
| Spec | A100 | A40 |
|---|---|---|
| VRAM | 80 GB | 48 GB |
| Memory bandwidth | 2,039 GB/s | – |
| Architecture | Ampere | Ampere |
| Compute capability | 8.0 | 8.6 |
| Precisions in hardware | FP32, FP16, BF16, INT4 | FP32, FP16, BF16, INT4 |
| Peak FP16/BF16 tensor TFLOPS | 312 TFLOPS | – |
| NVLink / multi-GPU interconnect | NVLink 3, 600 GB/s | – |
| Max board power (TDP) | 400W (SXM) | – |
| Form factor | SXM4 | – |
| Regions | 6 | 1 |
On-demand price
| Rate | A100 | A40 |
|---|---|---|
| Lowest per GPU/hr | $1.52/GPU/hr | $0.649/GPU/hr |
Which is better for AI: A40 or A100?
The A100 has just under 3x the memory bandwidth of the A40, 2,039 GB/s of HBM2e against 696 GB/s of GDDR6 (NVIDIA product pages, Aug 2026), which matters for training and large-batch inference, while the A40 is positioned as a workstation/virtualization card with lower cost. Pick the A100 when you're training or running bandwidth-bound large-batch inference; pick the A40 when your workload is lighter-weight (rendering, virtualization, small-batch inference) and the lower cost matters more. On price the A40 undercuts the A100 by 35% at the entry rate ($0.649 vs $0.991/GPU/hr). Worth weighing if the deciding factor above isn't a hard requirement for your job.
Decision criteria
| Memory bandwidth | A100 80GB SXM: 2,039 GB/s HBM2e. A40: 696 GB/s GDDR6 (NVIDIA product pages, Aug 2026). |
| Positioning | A100 is a data-center training/inference accelerator; A40 is a workstation/virtualization card (NVIDIA, Aug 2026). |
| VRAM | 80 GB (A100) vs 48 GB (A40). A40 memory can scale to 96GB via NVLink per NVIDIA's page. |
| Availability | 6 region(s) list the A100 vs 1 for the A40. |
| Entry price | $0.991/GPU/hr (A100) vs $0.649/GPU/hr (A40). |
Pick the A100 when
- You're training or fine-tuning and need the bandwidth
- Your job is large-batch and bandwidth-bound
- You need the highest-throughput option for the job regardless of cost
Pick the A40 when
- Your workload is lighter: rendering, virtualization, small-batch inference
- You want a lower hourly rate for a job that isn't bandwidth-bound
- A40 availability is better in your region
Price-performance
At list rates, 1,000 GPU-hours costs $649 on the A40 against $991 on the A100. Neither NVIDIA nor Aquanode publishes a workload-normalized $/token or $/epoch figure for this pair, so $/GPU-hour below is the only apples-to-apples number. A faster card can still cost less per finished job even at a higher hourly rate.
A40 vs A100: common questions
Is the A100 or the A40 better for AI?
Short answer: pick the A100 when you're training or running bandwidth-bound large-batch inference. Pick the A40 when your workload is lighter-weight (rendering, virtualization, small-batch inference) and the lower cost matters more.
Is the A100 or the A40 cheaper to rent?
On Aquanode's live marketplace the A100 starts at $0.991/GPU/hr (median $1.97/GPU/hr) and the A40 starts at $0.649/GPU/hr (median $0.649/GPU/hr). The A40 is the cheaper of the two at the entry rate, by 35%.
What is the difference between the A100 and the A40?
The A100 has 80 GB of VRAM against the A40's 48 GB; both are Ampere parts; both run BF16, INT4 workloads in hardware. On price, the A100 lists from $0.991/GPU/hr and the A40 from $0.649/GPU/hr.
How widely available are the A100 and the A40?
The A100 is listed in 6 regions and the A40 in 1 region right now.
How much does 1,000 GPU-hours cost on the A100 vs the A40?
At the lowest rates listed today, 1,000 GPU-hours costs $991 on the A100 and $649 on the A40, a difference of $342 for the same runtime. Rates are per GPU per hour and update hourly.
How much VRAM does the A100 have compared to the A40?
The A100 has 80 GB of VRAM; the A40 has 48 GB.
Can the A40 substitute for an A100 to save money?
For lighter workloads, yes. But the A100's 2,039 GB/s of bandwidth against the A40's 696 GB/s (NVIDIA, Aug 2026) means a bandwidth-bound training job will be meaningfully slower on an A40, which can erase the savings. The A40 lists from $0.649/GPU/hr against the A100's $0.991/GPU/hr on Aquanode today.
How this comparison is calculated
Every price is normalized to a per-GPU hourly rate using the same pipeline as every other pricing surface on Aquanode, and only the cheapest qualifying offer per provider is shown. A dash means that provider does not currently list that GPU.
Architecture, compute capability and supported precisions come from each vendor's published datasheet for that generation, not from the marketplace feed. This page regenerates at most once per hour.
Memory bandwidth, tensor TFLOPS, NVLink/interconnect and max board power are the same verified-from-datasheet figures shown on each model's own A100 page and A40 page, which link the primary vendor source for that model. A dash here means Aquanode hasn't verified that figure from a primary source yet, never a guess.
Go further
Check what fits in each card's memory with the A100 VRAM calculator and the A40 VRAM calculator, see vendor-published throughput in the GPU benchmarks and specs table, or browse every model by generation in the GPU index.
More A100 comparisons
More A40 comparisons
Related reading: A100 pricing and specs, A100 vs H100, A100 vs V100, and The best GPUs for AI, ranked.
