Built and tested in San Jose
Integration happens at TORmem’s San Jose facility, not at a distant contract site. What gets tested depends on the level you buy:
| Level | Done at the San Jose facility |
|---|---|
| L10 server | Configure CPU, memory, GPUs, drives and NICs to the agreed BOM, match firmware, and run burn-in and sustained-load tests. |
| L11 rack | Rack, cable and power the servers, switches and memory appliances, then validate the fabric and run the agreed workload tests, typically thermal profiling, latency under load and synthetic AI training. |
See what L10 integration includes and what L11 integration includes. The pass criteria are agreed before assembly starts.
Evaluate hardware in person
Because the facility is local, Bay Area teams can see TORmem hardware before committing. That includes the M1000 memory appliance, which shares DDR5 memory across servers over RDMA/RoCEv2, and TORswitch, the FPGA-accelerated switch.
For a memory appliance, an evaluation is most useful with your own workload. The shared AI memory guide lists what a proof of concept should measure: job time or request latency, tail latency, pool occupancy and recovery from a link fault.
San Jose data centers
San Jose hosts Equinix’s Silicon Valley campus, including SV1 and SV11, the AI-focused building that houses an NVIDIA DGX SuperPOD (ServeTheHome; Equinix). Per-rack power is set by the space you contract, and operators do not publish one figure for a whole campus.
Before the BOM is final
Get the contracted circuit, voltage, PDU outlets and cooling limit for your cage in writing, then size the rack to them. The GPU rack power guide lists the questions to ask.
Delivery and on-site installation
TORmem delivers the tested hardware itself and installs it with the same team that built it. On site, engineers rack, cable and power up the hardware, bring up the fabric and run the acceptance tests you agreed. You receive the test records and as-built maps at handover.
TORmem also delivers to Santa Clara colocation and other Bay Area data centers.
Frequently asked questions
- Where in San Jose is TORmem?
- TORmem’s headquarters, engineering team and integration facility are in San Jose, California. Visits are arranged in advance through the contact page.
- Are GPU servers and racks tested before they leave San Jose?
- Yes. L10 servers are burned in and load-tested, and L11 racks are validated as a unit, against a test plan agreed with you before assembly starts. The plan sets the tests, durations and pass criteria.
- What can we evaluate at the San Jose facility?
- You can see and evaluate TORmem hardware such as the M1000 memory appliance and TORswitch. For the M1000, bring a representative workload and dataset so the evaluation measures your application rather than a synthetic benchmark.
- Does the team that builds the rack also install it?
- Yes. TORmem delivers the hardware and its engineers install it on site, then run the acceptance tests agreed in writing.
- How long does integration take in San Jose?
- It depends on the platform, GPU and component availability, and the length of the agreed test plan. TORmem quotes the timeline per project once the configuration is fixed.
Built in San Jose
Scope a Bay Area deployment
Send the GPU, quantity, target data center and contracted power per rack. TORmem can scope the build, test plan, delivery and on-site installation, or arrange a visit to evaluate the hardware.
Related
Sources
- ServeTheHome: touring Equinix SV1 and SV11 and an NVIDIA DGX SuperPOD
- Equinix: Silicon Valley data centers
Facility names and figures come from the published sources above and describe those facilities. TORmem is independent of the colocation operators named, and their mention doesn't imply any partnership or endorsement.