Service · scoping included
The right machine for the workload, specced by the people who run it.
Buying GPU hardware for computational biology is easy to get wrong in expensive, quiet ways. We scope what the science actually needs. Catalytic AI builds it, racks it, and keeps it running.
Scoping is included in an audit or build engagement. The hardware itself is quoted by Catalytic AI.
Who this is for
Teams whose cloud GPU bill has started to look like a salary, or whose data cannot leave the building. Also teams who tried to buy a machine already and got quoted something specced to a sales target rather than to their workload.
What we scope
- Memory ceiling first. The question that actually decides the machine is what has to fit in VRAM at once. Getting this wrong is how a brand-new box turns out to be unable to run the job it was bought for.
- Throughput against real jobs — your batch sizes and your structures, not a benchmark suite.
- Where the bottleneck will move to. Often storage or PCIe lanes rather than the GPU, which is not obvious until the machine is assembled.
- Buy versus rent. Sometimes the honest answer is a few managed campaigns instead of a capital purchase.
Who builds it
Catalytic AI, in Troy, New York. They specced, bought and built the nodes we run on, which is why they can tell you exactly what is inside one:
- TROY-1 — a custom Blackwell workstation: RTX PRO 6000, 96 GB GDDR7 with ECC, 1,792 GB/s memory bandwidth, Ryzen 9 7900X, 64 GB DDR5.
- A DGX Spark — 128 GB unified memory, for work that needs one large address space more than it needs raw bandwidth.
Both sit in a 20U open-frame rack on a UPS. If your build ends up looking like one of those, it is because that is what the workload asked for.
How the two fit together
We are the science requirement; they are the machine. You can start at either end — if you already know you need hardware, talk to them directly. If you are not sure what the science needs yet, start with us.
Tell us the workload, not the budget.
We will work out what it needs, and say so if the honest answer is that you should rent instead of buy.