Ampere
NVIDIA A100 40GB
wins
6
The A100 40GB is the more affordable sibling of the A100 80GB.
Ada Lovelace
NVIDIA RTX 4090
wins
0
The RTX 4090 is the most cost-effective consumer GPU for AI workloads. At $0.40–$0.80/hr on GPU clouds, it's ideal for fine-tuning and small-model inference.
Performance comparison
Visual bars — winner highlighted
NVIDIA A100 40GB
Metric
NVIDIA RTX 4090
FP16 TFLOPS
Bandwidth
VRAM
TDP (lower=better)
TFLOPS/watt
GB/s per watt
TFLOPS/$
FP16 TFLOPS ÷ best on-demand $/hr
GB/s per $
Memory bandwidth ÷ best on-demand $/hr
Full specification table
| Spec | NVIDIA A100 40GB | NVIDIA RTX 4090 |
|---|---|---|
| Architecture | Ampere | Ada Lovelace |
| VRAM | 40 GB HBM2e | 24 GB GDDR6X |
| VRAM type | HBM2e | GDDR6X |
| Memory bandwidth | 1,555 GB/s | 1,008 GB/s |
| FP16 TFLOPS | 312 TFLOPS | 165 TFLOPS |
| TDP | 400 W | 450 W |
| TFLOPS/watt | 0.78 T/W | 0.37 T/W |
| GB/s per watt | 3.89 GB/s/W | 2.24 GB/s/W |
| NVLink | Yes | No |
| Release year | 2020 | 2022 |
| Best cloud price | $0.29/hr | $0.16/hr |
| TFLOPS/$ | 1075.9 | 1031.3 |
| GB/s per $ | 5362 | 6300 |
Live cloud pricing
On-demand pricing across providers — updated in real time
NVIDIA A100 40GB
NVIDIA RTX 4090
Salad: Lowest priority tier — preemptible
Power efficiency analysis
TFLOPS/watt and GB/s/watt — critical for data center TCO
NVIDIA A100 40GB
NVIDIA RTX 4090
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 40GB
NVIDIA RTX 4090
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
NVIDIA A100 40GB costs 1.8× more at current best prices. NVIDIA A100 40GB has 402GB VRAM vs 246GB — larger models fit without offloading; 1.5× higher memory bandwidth (1555 vs 1008 GB/s) improves token throughput for memory-bound inference.
| Workload | Scale | NVIDIA A100 40GB | NVIDIA RTX 4090 |
|---|---|---|---|
| 7B LLM fine-tune | 8× GPU, 24 h | $56 | $31▼ |
| 70B LLM fine-tune | 64× GPU, 72 h | $1.3k | $737▼ |
| Inference server | 1× GPU, 730 h | $212 | $117▼ |
| Image gen batch | 4× GPU, 8 h | $9 | $5▼ |
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 40GB for:
- ML training
- Inference
- Cost-effective HPC
Weaknesses / Limitations
- 40 GB VRAM is a hard ceiling — cannot run models that require 80 GB
- Lower bandwidth than A100 80GB (1,555 vs 2,039 GB/s)
- Older Ampere architecture — lower TFLOPS than H100
Choose NVIDIA RTX 4090 for:
- Fine-tuning
- Small model inference
- Cost-effective compute
Weaknesses / Limitations
- 24 GB VRAM limits models to ~34B at INT4 — cannot run 70B+ unquantized
- No NVLink — cannot pool VRAM across cards
- Consumer GPU — no ECC memory, lower reliability for 24/7 production
Popular GPU comparisons