[AMPERE]
NVIDIA A100
The cost-effective legacy workhorse — still strong for fine-tuning.
The NVIDIA A100 (Ampere) remains widely deployed and cost-effective for fine-tuning, LoRA training, and mid-tier inference. No native FP8 — use BF16/FP16 paths.
VRAM
80 GB HBM2e
FP8 TFLOPS
—
Mem BW
2.0 TB/s
TDP
400 W
Best for
- LoRA & QLoRA fine-tuning
- Mid-tier inference
- Cost-sensitive workloads
Benchmarks
| WORKLOAD | METRIC | VALUE |
|---|---|---|
| Llama-2 13B LoRA | samples/sec | ~32 |
| BF16 dense matmul | TFLOPS | 312 |
NVIDIA A100 availability from $9.60/hr
NVIDIA Ampere
8x NVIDIA A100 80GB
Memory640GB HBM
RegionEU-North
Availability93%
ProviderAzure
$14.80/hr
Deploy →NVIDIA Ampere
8x NVIDIA A100 40GB
Memory320GB HBM
RegionAsia-Pacific-SE
Availability99%
ProviderRunPod
$9.60/hr
Deploy →Definition
What is Servers.Computer?
Servers.Computer is an AI compute routing and procurement layer that benchmarks, compares, and deploys GPU clusters (NVIDIA H100, H200, B200 and AMD MI300) across global cloud providers in real time.