Blackwell Ultra
NVIDIA GB300 NVL72
wins
4
The GB300 NVL72 is NVIDIA's Blackwell Ultra rack-scale system — 72 GPUs in a single rack delivering 7,500 TFLOPS FP16 per GPU.
Volta
NVIDIA V100 16GB
wins
1
The V100 PCIe 16GB was NVIDIA's flagship AI GPU before the A100. It delivers 112 TFLOPS FP16 (dense) and 900 GB/s HBM2 bandwidth.
Performance comparison
Visual bars — winner highlighted
NVIDIA GB300 NVL72
Metric
NVIDIA V100 16GB
FP16 Tensor TFLOPS (dense)
Bandwidth
VRAM
TDP (lower=better)
TFLOPS/watt
GB/s per watt
Full specification table
| Spec | NVIDIA GB300 NVL72 | NVIDIA V100 16GB |
|---|---|---|
| Architecture | Blackwell Ultra | Volta |
| VRAM | 288 GB HBM3e (per GPU) | 16 GB HBM2 |
| VRAM type | HBM3e (per GPU) | HBM2 |
| Memory bandwidth | 16,000 GB/s | 900 GB/s |
| FP16 Tensor TFLOPS (dense) | 7,500 TFLOPS | 112 TFLOPS |
| TDP | 1,200 W | 250 W |
| TFLOPS/watt | 6.25 T/W | 0.45 T/W |
| GB/s per watt | 13.33 GB/s/W | 3.60 GB/s/W |
| NVLink | Yes | Yes |
| Release year | 2026 | 2018 |
Live cloud pricing
On-demand pricing across providers — updated in real time
Power efficiency analysis
TFLOPS/watt and GB/s/watt — critical for data center TCO
NVIDIA GB300 NVL72
NVIDIA V100 16GB
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.
When to choose each GPU
Choose NVIDIA GB300 NVL72 for:
- Next-gen frontier training
- Trillion-param inference
- Rack-scale AI
Weaknesses / Limitations
- Rack-scale system — not a single GPU, requires full NVL72 rack deployment
- Extremely high cost — rack-scale pricing only
- Very limited availability, primarily hyperscaler deployments
Choose NVIDIA V100 16GB for:
- Legacy ML training
- Budget HPC
- Mature framework support
Weaknesses / Limitations
- Volta architecture — significantly lower TFLOPS than A100 or H100
- 16 GB VRAM ceiling limits modern LLM workloads
- No BF16 support — less efficient for modern transformer training
Popular GPU comparisons