Blackwell
NVIDIA B300 192GB
wins
4
The NVIDIA B300 is the next step in the Blackwell lineup.
Ada Lovelace
NVIDIA L40S
wins
1
The L40S is NVIDIA's inference-optimized Ada Lovelace GPU. 48GB GDDR6 and 362 TFLOPS FP16 make it a strong choice for production inference at lower cost than H100.
Performance comparison
Visual bars — winner highlighted
NVIDIA B300 192GB
Metric
NVIDIA L40S
FP16 TFLOPS
Bandwidth
VRAM
TDP (lower=better)
TFLOPS/watt
GB/s per watt
Full specification table
| Spec | NVIDIA B300 192GB | NVIDIA L40S |
|---|---|---|
| Architecture | Blackwell | Ada Lovelace |
| VRAM | 192 GB HBM3e | 48 GB GDDR6 |
| VRAM type | HBM3e | GDDR6 |
| Memory bandwidth | 8,000 GB/s | 864 GB/s |
| FP16 TFLOPS | 5,000 TFLOPS | 362 TFLOPS |
| TDP | 1,000 W | 350 W |
| TFLOPS/watt | 5.00 T/W | 1.03 T/W |
| GB/s per watt | 8.00 GB/s/W | 2.47 GB/s/W |
| NVLink | Yes | No |
| Release year | 2025 | 2023 |
Live cloud pricing
On-demand pricing across providers — updated in real time
NVIDIA B300 192GB
No live pricing data available.
NVIDIA L40S
Salad: Lowest priority tier — preemptible
Power efficiency analysis
TFLOPS/watt and GB/s/watt — critical for data center TCO
NVIDIA B300 192GB
NVIDIA L40S
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 B300 192GB
NVIDIA L40S
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 B300 192GB | NVIDIA L40S |
|---|---|---|---|
| 7B LLM fine-tune | 8× GPU, 24 h | — | $73 |
| 70B LLM fine-tune | 64× GPU, 72 h | — | $1.8k |
| Inference server | 1× GPU, 730 h | — | $277 |
| Image gen batch | 4× GPU, 8 h | — | $12 |
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 B300 192GB for:
- Next-gen LLM training
- Agent inference at scale
- Memory-intensive AI
Weaknesses / Limitations
- Very limited availability — not yet widely deployed on cloud platforms
- 1,000W TDP requires specialized power and cooling
- Premium pricing over B200 with incremental gains
Choose NVIDIA L40S for:
- Inference
- Video AI
- Multi-modal workloads
Weaknesses / Limitations
- GDDR6 bandwidth (864 GB/s) is lower than HBM-based GPUs — bottleneck for memory-bound LLM serving
- No NVLink — cannot scale across cards for very large models
- Less suitable for large-scale training than H100 or A100
Popular GPU comparisons