Hardware / GPU servers / HGX A100 80GB
Spec sheet · Ampere · 2020
HGX A100 SXM4 80GB: 8-GPU Server System
The proven Ampere workhorse: eight A100 80GB GPUs on NVLink 3. End of life at the OEM level, and the lowest-power HGX node.
Full specifications
Compute
- GPUs
- 8× NVIDIA A100 SXM4 80GB
- FP16 (system)
- ~2.5 PFLOPS dense
Memory
- GPU memory
- 80 GB HBM2e per GPU · 640 GB total
- Memory bandwidth
- 2.0 TB/s per GPU · 16 TB/s total
Interconnect
- GPU interconnect
- NVLink 3 · 600 GB/s per GPU · NVSwitch
- Network fabric
- 8× ConnectX-6 200G HDR InfiniBand
Power and cooling
- GPU TDP
- 400 W per GPU
- System power draw
- 6.5 kW max (DGX A100)
Platform
- Host platform
- 2× Xeon SP or EPYC · 1 TB DDR4 · PCIe Gen 4
- Common chassis
- Supermicro SYS-420GP · DGX A100 · Dell XE8545
- Form factor
- 4U–6U air-cooled
- Architecture
- Ampere
- First shipped
- 2020
Power, cooling and rack density
| Cooling | Per node | Per rack | Rack power |
|---|---|---|---|
| Air | 6.5 kW | 4 nodes | 26 kW |
A 32-node cluster (256 GPUs) needs about 0.3 MW of facility power: a colocation cage or suite. Size your own cluster in the site planner →
What one node can serve
Largest open models that fit on eight GPUs at FP8 with an 8K context, with estimated output throughput.
- gpt-oss-120b13K tok/s
- Qwen3.6 27B18K tok/s
- gpt-oss-20b34K tok/s
Cost per million tokens in the inference calculator →
How it compares
| System | Architecture | Memory / GPU | Bandwidth | FP8 dense | Max power | Cooling |
|---|---|---|---|---|---|---|
| HGX B300 | Blackwell Ultra | 288 GB | 8 TB/s | 4.5 PF | 15 kW | Liquid / Air |
| HGX B200 | Blackwell | 180 GB | 8 TB/s | 4.5 PF | 14.3 kW | Air / Liquid |
| HGX H200 | Hopper | 141 GB | 4.8 TB/s | 2.0 PF | 10.2 kW | Air / Liquid |
| HGX H100 | Hopper | 80 GB | 3.35 TB/s | 2.0 PF | 10.2 kW | Air / Liquid |
| HGX A100 80GB | Ampere | 80 GB | 2 TB/s | — | 6.5 kW | Air |
| RTX Pro 6000 Blackwell | Blackwell | 96 GB | 1.8 TB/s | 1.0 PF | 6.5 kW | Air |
How it connects to the cluster
Each GPU gets its own network card and its own port on the leaf switch, one rail per GPU. These are the parts on that path.
On this path: ConnectX-6 · Quantum QM8700
Alternatives to consider
What to verify before you buy
- A100 systems have significant field hours; ECC error history is critical
- HBM2e degradation: confirm vendor ran full HBM memory test
- NVLink 3 fabric health: older NVSwitch boards can develop connector issues
- Check firmware: A100 systems may be on outdated BIOS/BMC, confirm update path
- EOL status: no new OEM support; factor extended warranty cost from third parties
Export controls
A100 80GB is export-controlled under EAR. While older than H100, the classification still applies. Verify end-user for all cross-border transactions.
Frequently asked questions
How much power does an HGX A100 80GB draw?
Up to 6.5 kW per 8-GPU HGX node at maximum rated power. Plan site power for that peak: training runs close to it. Inference typically averages 55–65% of it.
Is the HGX A100 80GB air-cooled or liquid-cooled?
The HGX A100 80GB is air-cooled.
How much GPU memory does an HGX A100 80GB have?
80 GB per GPU, 640 GB across eight, at 2 TB/s per GPU.
What models can one HGX A100 80GB node serve?
At FP8 with an 8K context, the largest open models that fit on eight GPUs include gpt-oss-120b, Qwen3.6 27B, gpt-oss-20b. The inference calculator shows throughput and cost for each.
What kind of site does an HGX A100 80GB cluster need?
A 32-node cluster (256 GPUs) needs roughly 0.3 MW of facility power with air cooling: a colocation cage or suite.
What comes after the HGX A100 80GB?
The HGX H100. The HGX A100 80GB is Ampere architecture, first shipped in 2020.
Sources: NVIDIA A100. Power is the maximum rated system draw; throughput and site figures are planning estimates from the same models as the Rack Rate tools.
Building with HGX A100 80GB?
Send us the cluster spec. We source the full bill of materials, and can finance it.