Blackwell Ultra
NVIDIA GB300 NVL72
wins
3
The GB300 NVL72 is NVIDIA's Blackwell Ultra rack-scale system — 72 GPUs in a single rack delivering 7,500 TFLOPS FP16 per GPU.
Blackwell
NVIDIA RTX 5090 32GB
wins
1
The RTX 5090 is NVIDIA's Blackwell consumer flagship with 32 GB GDDR7 and 838 TFLOPS FP16.
Performance comparison
Visual bars — winner highlighted
NVIDIA GB300 NVL72
Metric
NVIDIA RTX 5090 32GB
FP16 Tensor TFLOPS (dense)
Bandwidth
VRAM
TDP (lower=better)
TFLOPS/watt
GB/s per watt
Full specification table
| Spec | NVIDIA GB300 NVL72 | NVIDIA RTX 5090 32GB |
|---|---|---|
| Architecture | Blackwell Ultra | Blackwell |
| VRAM | 288 GB HBM3e (per GPU) | 32 GB GDDR7 |
| VRAM type | HBM3e (per GPU) | GDDR7 |
| Memory bandwidth | 16,000 GB/s | 1,792 GB/s |
| FP16 Tensor TFLOPS (dense) | 7,500 TFLOPS | n/a |
| TDP | 1,200 W | 575 W |
| TFLOPS/watt | 6.25 T/W | 0.00 T/W |
| GB/s per watt | 13.33 GB/s/W | 3.12 GB/s/W |
| NVLink | Yes | No |
| Release year | 2026 | 2025 |
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 RTX 5090 32GB
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 RTX 5090 32GB for:
- Consumer AI inference
- Fine-tuning 13B–30B models
- Cost-effective Blackwell
Weaknesses / Limitations
- 32 GB VRAM — cannot run 70B+ models unquantized
- No NVLink — cannot pool VRAM across cards
- 575W TDP — high power draw for a consumer card
Popular GPU comparisons