Help me choose
Rung 2 vs Rung 3 — same GPUs, different deal
Both tiers run on the same datacenter GPU (the H200), so the raw computing power is the same either way. The difference is how you buy and build with them. Rung 2 — buy the cards, build it yourself. Purchase exactly the number of H200 cards your model needs and cluster them. Cost-effective — you pay only for the cards you need — but you handle setup: installing the cards, wiring them, supporting the system. Rung 3 — buy the finished machine. The same H200 GPUs come pre-installed, pre-wired, and factory-tested inside one server, ready to run. You pay more than the loose cards would cost, but you're paying for assembly, integration, and support — it arrives working, no cluster-building on your end. Which to choose: if you have the engineering expertise to build and maintain a cluster and want to save money, buy the cards (Rung 2). If you'd rather have a tested, supported machine that just works, buy the server (Rung 3). Same GPUs — you're choosing between saving money and saving effort.
| Rung 2 — H200 NVL cards | Rung 3 — DGX H200 server | |
|---|---|---|
| What you're buying | Individual datacenter cards | A factory-built server |
| Assembly | You build and wire the cluster | Arrives pre-built and tested |
| Expertise required | Yes — ops team needed | No — plug in and run |
| Cost vs equivalent | Lower — pay per card | Higher — paying for integration |
| Support | Self-managed | Manufacturer-backed |
Why not just buy a lot of desktop cards?
Datacenter GPUs pool memory via NVLink so several act as one large machine. Desktop cards can't — a pile of cards isn't a cluster.