Ampere
NVIDIA A100 80GB
wins
1
The A100 remains one of the most widely available data center GPUs. Its broad provider support and mature software ecosystem make it a reliable choice for ML training and inference.
Blackwell Ultra
NVIDIA GB300 SXM6 288GB
wins
4
The GB300 SXM6 (Grace Blackwell Ultra) is a CPU+GPU superchip with 288 GB HBM3e and 8 TB/s memory bandwidth.
Performance comparison
Visual bars — winner highlighted
NVIDIA A100 80GB
Metric
NVIDIA GB300 SXM6 288GB
FP16 Tensor TFLOPS (dense)
Bandwidth
VRAM
TDP (lower=better)
TFLOPS/watt
GB/s per watt
Full specification table
| Spec | NVIDIA A100 80GB | NVIDIA GB300 SXM6 288GB |
|---|---|---|
| Architecture | Ampere | Blackwell Ultra |
| VRAM | 80 GB HBM2e | 288 GB HBM3e |
| VRAM type | HBM2e | HBM3e |
| Memory bandwidth | 2,039 GB/s | 8,000 GB/s |
| FP16 Tensor TFLOPS (dense) | 312 TFLOPS | 2,800 TFLOPS |
| TDP | 400 W | 1,200 W |
| TFLOPS/watt | 0.78 T/W | 2.33 T/W |
| GB/s per watt | 5.10 GB/s/W | 6.67 GB/s/W |
| NVLink | Yes | Yes |
| Release year | 2020 | 2025 |
Live cloud pricing
On-demand pricing across providers — updated in real time
NVIDIA A100 80GB
NVIDIA GB300 SXM6 288GB
No live pricing data available.
Salad: Lowest priority tier — preemptible
Power efficiency analysis
TFLOPS/watt and GB/s/watt — critical for data center TCO
NVIDIA A100 80GB
NVIDIA GB300 SXM6 288GB
TFLOPS/watt measures compute efficiency — how much AI throughput you get per watt of power consumed. For data centers with PUE of 1.2–1.5, a 10% improvement in TFLOPS/watt translates directly to lower electricity costs and cooling requirements. GB/s/watt measures memory bandwidth efficiency, which is the binding constraint for memory-bound LLM inference workloads.
Performance-per-dollar
TFLOPS/$ and GB/s/$ at current best cloud prices
NVIDIA A100 80GB
NVIDIA GB300 SXM6 288GB
TFLOPS/$ = FP16 TFLOPS ÷ best on-demand $/hr. Higher is better — more compute per dollar. GB/s/$ = memory bandwidth ÷ best on-demand $/hr. For memory-bound inference workloads (most LLM serving), GB/s/$ is often a better predictor of cost-efficiency than TFLOPS/$. Prices are live from the Compute Comparison API and update in real time.
Workload Cost Comparison
Total job cost = best on-demand $/hr × GPU count × hours
| Workload | Scale | NVIDIA A100 80GB | NVIDIA GB300 SXM6 288GB |
|---|---|---|---|
| 7B LLM fine-tune | 8× GPU, 24 h | $92 | — |
| 70B LLM fine-tune | 64× GPU, 72 h | $2.2k | — |
| Inference server | 1× GPU, 730 h | $350 | — |
| Image gen batch | 4× GPU, 8 h | $15 | — |
Based on current best on-demand prices. Spot pricing typically 40–70% lower. Actual cost depends on provider, region, and availability.
When to choose each GPU
Choose NVIDIA A100 80GB for:
- ML training
- Inference at scale
- HPC
Weaknesses / Limitations
- Older Ampere architecture — lower FP16 TFLOPS than H100 at similar price
- No Ada Lovelace efficiency improvements
- Being phased out by H100 on many platforms
Choose NVIDIA GB300 SXM6 288GB for:
- Frontier AI training
- Inference at scale
- Memory-bound LLM serving
Weaknesses / Limitations
- Very limited cloud availability — still ramping supply
- Extremely high cost — enterprise/hyperscaler pricing only
- Requires specialized NVLink-C2C infrastructure
Popular GPU comparisons