Air-Cooled vs Liquid-Cooled HGX H100 Servers
TL;DR
Both our HGX H100 builds carry the same eight H100 80 GB SXM GPUs (640 GB HBM3 per node) — the choice is cooling and density. The air-cooled 8U server drops into a standard air-cooled facility with no liquid loop, so it deploys fast with no retrofit. The liquid-cooled 4U server packs the same eight GPUs into half the rack units for higher density but needs a coolant distribution unit (CDU) and a facility liquid loop. The GPU is identical; the decision is your facility and density target.
On this page
The short answer
The GPU is the same in both — eight H100 80 GB SXM5 GPUs, 640 GB of HBM3 per node — so this is not a performance comparison. It is a facility and density decision.
The air-cooled 8U server is the drop-in option: it runs in a standard air-cooled data center with no liquid loop, so there is no cooling retrofit and it deploys fast. The liquid-cooled 4U server puts the same eight GPUs into half the rack units for higher rack density, but it needs a coolant distribution unit (CDU) and a facility liquid loop. If you have (or are building) liquid cooling and want maximum density, take the 4U; if you need to deploy into an existing air-cooled hall without a retrofit, take the 8U.
Air vs liquid at a glance
Both builds are the Supermicro H13 platform with the same 8× H100 80 GB SXM configuration; the difference is form factor, rack density, and what the facility has to provide.
| Air-cooled | Liquid-cooled | |
|---|---|---|
| Form factor | 8U | 4U |
| GPUs | 8× H100 80 GB SXM | 8× H100 80 GB SXM |
| Rack density | Standard | Higher (half the U) |
| Facility requirement | Standard air / no loop | CDU / liquid loop |
| Best for | Fast drop-in, no retrofit | Max density where a loop exists |
When to choose air-cooled (8U)
Reach for the air-cooled 8U H100 when the facility is the constraint:
- No liquid loop, no retrofit. It runs in a standard air-cooled data center as-is. There is no CDU to plumb and no facility water to arrange, so nothing about the building has to change first.
- Fastest path to deployed compute. With no cooling retrofit on the critical path, the 8U node is the quicker way to stand up eight H100s in an existing hall.
- Standard rack density is fine. If you are not trying to maximize GPUs per rack, the extra rack units cost you little and buy you a simpler deployment.
When to choose liquid-cooled (4U)
Reach for the liquid-cooled 4U H100 when density is the goal and the plumbing exists:
- Maximum rack density. The same eight GPUs occupy half the rack units, so you fit more compute per rack — the point of liquid cooling.
- You have (or are building) a liquid loop. The 4U build cools through per-GPU cold plates and needs a CDU and a facility liquid loop. Where that infrastructure is already in place, the 4U is the efficient choice.
- Dense, efficient buildouts. For new high-density halls designed around liquid cooling, the 4U is the natural fit.
Procurement note
Both are new, factory-integrated Supermicro H13 nodes with the same eight H100 80 GB SXM GPUs — the liquid-cooled 4U and the air-cooled 8U. Configuration (CPU, networking, cooling) and pricing are quoted per configuration on request, since the figure depends on the build and the facility fit. Tell us your density target and what your facility can cool, and we will return pricing and availability. Browse the GPU catalog to compare integrations.
Frequently asked questions
Should I buy air-cooled or liquid-cooled H100 servers?
It comes down to your facility and density target, not GPU performance — both carry the same eight H100 80 GB SXM GPUs. Choose the air-cooled 8U to drop into a standard air-cooled facility with no liquid loop and no retrofit. Choose the liquid-cooled 4U for higher rack density where you have (or are building) a CDU and a facility liquid loop.
Do I need a liquid loop for the 4U H100?
Yes. The liquid-cooled 4U H100 cools through per-GPU cold plates and needs a coolant distribution unit (CDU) and a facility liquid loop. If your facility is air-cooled with no loop, the air-cooled 8U server is the drop-in option that requires no cooling retrofit.
Is the GPU the same in both?
Yes — both builds carry eight H100 80 GB SXM GPUs (640 GB of HBM3 per node) on the Supermicro H13 platform. The compute is identical; the difference is cooling, form factor (8U air vs 4U liquid), and rack density.
How much does an HGX H100 server cost?
GPU systems are quoted per configuration on request — the figure depends on cooling, CPU, and networking. Tell us the build you need and we will return pricing and availability.
Related
Last updated