GPU Model Comparison
Select up to 4 GPUs to compare specs and live pricing side-by-side
Select GPUs to Compare (3/4)
H100 80GB
GH100 · TSMC 4N
VRAM
80 GB HBM3
Bandwidth
3350 GB/s
TDP
700 W
Released
2022
Interconnect
NVLink 4.0 / PCIe 5.0
Flagship data center GPU. Transformer Engine with FP8 support. SXM5 form factor for max bandwidth.
A100 80GB
GA100 · TSMC 7nm
VRAM
80 GB HBM2e
Bandwidth
2000 GB/s
TDP
400 W
Released
2020
Interconnect
NVLink 3.0 / PCIe 4.0
Workhorse of the AI era. Widely available, mature software support, excellent price/performance.
RTX 4090
AD102 · TSMC 4N
VRAM
24 GB GDDR6X
Bandwidth
1008 GB/s
TDP
450 W
Released
2022
Interconnect
PCIe 4.0
Consumer GPU with surprisingly strong FP32. No NVLink. Best $/TFLOP for budget workloads.
Performance Comparison
FP16 PerformanceTFLOPS
H100 80GB
2.0K
A100 80GB
312
RTX 4090
165.2
BF16 PerformanceTFLOPS
H100 80GB
2.0K
A100 80GB
312
RTX 4090
165.2
INT8 PerformanceTOPS
H100 80GB
4.0K
A100 80GB
624
RTX 4090
330.3
VRAMGB
H100 80GB
80
A100 80GB
80
RTX 4090
24
Memory BandwidthGB/s
H100 80GB
3.4K
A100 80GB
2.0K
RTX 4090
1.0K
FP32 PerformanceTFLOPS
H100 80GB
67
A100 80GB
19.5
RTX 4090
82.6
Power Draw (TDP)W
H100 80GB
700
A100 80GB
400
RTX 4090
450
Full Spec Sheet
| Spec | H100 80GB | A100 80GB | RTX 4090 |
|---|---|---|---|
| Architecture | GH100 | GA100 | AD102 |
| Generation | Hopper | Ampere | Ada Lovelace |
| Process Node | TSMC 4N | TSMC 7nm | TSMC 4N |
| Transistors | 80B | 54.2B | 76.3B |
| VRAM | 80 GB | 80 GB | 24 GB |
| VRAM Type | HBM3 | HBM2e | GDDR6X |
| Mem Bandwidth | 3350 GB/s | 2000 GB/s | 1008 GB/s |
| FP32 | 67 TFLOPS | 19.5 TFLOPS | 82.6 TFLOPS |
| FP16 | 1979 TFLOPS | 312 TFLOPS | 165.2 TFLOPS |
| BF16 | 1979 TFLOPS | 312 TFLOPS | 165.2 TFLOPS |
| FP8 | 3958 TFLOPS | — | — |
| INT8 | 3958 TOPS | 624 TOPS | 330.3 TOPS |
| NVLink BW | 900 GB/s | 600 GB/s | — |
| Interconnect | NVLink 4.0 / PCIe 5.0 | NVLink 3.0 / PCIe 4.0 | PCIe 4.0 |
| TDP | 700 W | 400 W | 450 W |
| Released | 2022 | 2020 | 2022 |
Best Use Cases
H100 80GB
- ▸LLM training
- ▸Large-scale inference
- ▸Scientific HPC
A100 80GB
- ▸ML training
- ▸Large model inference
- ▸HPC workloads
RTX 4090
- ▸Fine-tuning small models
- ▸Inference up to 13B
- ▸Cost-sensitive workloads