GPU servers and AI racks for San Francisco teams

Three places to run GPUs from San Francisco

For the South Bay option, the Santa Clara page compares published per-rack figures by facility.

Running GPUs in an SF office: the circuit is the limit

A GPU server that looks like a workstation can draw more than an office circuit is meant to carry. NVIDIA rates the RTX 5090 at 575 W total graphics power (NVIDIA), so a server with two of them, like the TOR-AI-Edge-2G, draws about 1,150 W for the GPUs alone, before the CPU, memory and drives.

A standard 120 V, 15 A circuit is rated at 1,800 W, and electrical codes generally limit a continuous load to 80% of the rating: 1,440 W. Sustained inference is a continuous load. In practice:

• Give the server its own circuit. Don’t share it with desks or kitchen equipment; a 20 A or 208/240 V circuit leaves headroom.
• Plan for the heat. Nearly every watt the server draws ends up as heat in the room.
• Check the building HVAC schedule. If cooling shuts off at night or on weekends, a server running a long job will heat the room with it.
• Consider noise. Under sustained load, GPU fans are loud enough to need a closet or server room rather than an open office.

TORmem quotes the measured draw of your exact configuration so the circuit can be specified before delivery.

San Francisco colocation

The city’s commercial colocation is concentrated in a few buildings, chiefly Digital Realty’s 365 Main Street and 200 Paul Avenue (Data Center Knowledge). 200 Paul is described as the best-connected building in the city (Baxtel), and 365 Main is a carrier-neutral facility near the Financial District (DataCenterMap).

Built in San Jose, installed in San Francisco

The hardware is configured, burned in and tested at TORmem’s San Jose facility. TORmem engineers then install it at your SF office, lab or data center and run the acceptance tests agreed in writing. If you want to see the hardware first, you can arrange an evaluation visit in San Jose.

Frequently asked questions

Does TORmem install GPU servers in San Francisco?

Yes. TORmem builds and tests the hardware in San Jose, and its engineers install it at your San Francisco office, lab or data center, then run the acceptance tests agreed in writing.

Can we run a GPU server in our San Francisco office?

Often, if the circuit and cooling allow it. A dual RTX 5090 server draws about 1,150 W for the GPUs alone, which is most of the 1,440 W continuous capacity of a standard 15 A office circuit. Plan a dedicated circuit, account for the heat and after-hours HVAC, and keep it out of open workspace for noise.

How much power does a dual RTX 5090 server need?

NVIDIA rates each RTX 5090 at 575 W total graphics power, so two GPUs draw about 1,150 W before the CPU, memory and storage. The total depends on the configuration and workload; TORmem provides the measured draw for your exact build so the circuit can be specified first.

Which San Francisco data centers can host GPU hardware?

Commercial colocation in the city is concentrated in a few buildings, notably Digital Realty’s 365 Main Street and 200 Paul Avenue. Whether a GPU rack fits depends on the power and cooling per cabinet you can contract there, so confirm those first. TORmem is independent of the operators.

Should an SF team put its GPUs in the city or the South Bay?

It depends on the power per rack you need, how often your team needs hands-on access and your network design. Santa Clara facilities publish higher per-rack densities; a city colo or your own office keeps hardware close. TORmem can install at any of them.