Supercomputers/Systems/Databricks DBRX Training Cluster
USA · operational · ai training
Databricks DBRX Training Cluster
Operated by Databricks at Databricks Mosaic AI (undisclosed US site) .
Phase 1 seed dataset, compiled by hand. These rows were built from public operator, laboratory and vendor sources. A mechanical second-reader pass has since fetched every cited source: 1092 of 1153 systems have a readable citation that names them, and 40 are genuinely weakly sourced. Every claim carries its source and a confidence tier. Treat anything below verified as a lead, not a citation.
Measured performance
- Rmax
- -
- Rpeak
- -
- Rmax ÷ Rpeak
- -
- Cores
- -
- Power
- -
- Per watt
- - GF/W
Figures are as last publicly reported for the configuration described below, not a live measurement. Where a system was upgraded in place, the post-upgrade configuration is shown and the earlier one appears in the timeline.
What this machine is made of
One row per supplier relationship. “Supplier at build” is the company that shipped the part at the time; where that company has since been acquired, the parent it rolls up to today is shown beside it. That distinction is what makes ticker-level aggregation possible across a thirty-year dataset.
| Role | Part | Supplier at build | Quantity | Confidence | Source |
|---|---|---|---|---|---|
| Accelerator | NVIDIA H100 SXM5 Accelerators CUDA (Hopper GH100) · 16,896C · 1.98 GHz · TSMC 4N · 700 W 3,072 NVIDIA H100s used to pretrain DBRX | NVIDIA | 3,072 reported | Reported | databricks.comdatabricks.com |
| Interconnect | NVIDIA InfiniBand NDR (Quantum-2) Interconnect Fat tree or dragonfly+ · 400 Gb/s per port · Under 0.6 microseconds 3.2 Tbps aggregate InfiniBand fabric connecting the 3,072 H100s | NVIDIA | - | Reported | databricks.com |
AI datacenter metrics
Figures beyond an accelerator count, each with who stated it. What can and cannot be compared across AI datacenters.
| Metric | Value | As of | Basis | Note | Source |
|---|---|---|---|---|---|
| Accelerators | 3,072 accelerators | 2024-03 | Operator stated | DBRX was trained on 3,072 NVIDIA H100 on 3.2 Tbps InfiniBand; the post gives no MFU or power figure | databricks.com |
The read
The cluster Databricks used to pretrain its own DBRX open model on its Mosaic AI Training platform: 3,072 NVIDIA H100 GPUs connected by an aggregate 3.2 Tbps InfiniBand fabric, run over roughly three months and processing more than 12 trillion tokens, announced in Databricks' own 27 March 2024 blog post. A second Databricks post on Mosaic AI Training's capabilities repeats the same GPU count and token figure but is the same organization, not an independent source, so the tier stays reported. Databricks names no single cloud provider for this run, saying only that Mosaic AI Training spans AWS, Azure, OCI and CoreWeave capacity generally, so no site location or integrator is recorded beyond an undisclosed US site. This is a training-run deployment on Databricks' own multi-tenant training stack, not a permanently branded machine, and no FLOPS or HPL figure has been published for it.
Timeline
- 2024-03 Announced source
Site & facility
Change history
- 2026-10-04 AI datacenter metrics: Databricks DBRX Training Cluster
- 2026-09-24 Added Databricks' own 3,072-H100 DBRX training cluster on its Mosaic AI Training platform
Source check
We fetched this system's own citations and recorded whether each page actually mentions it. This is a corroboration signal, not a fact check, and it is published so you can see how well the sourcing holds up rather than take it on trust.
| Cited source | Result | Found on the page |
|---|---|---|
| databricks.com | names system, part and count | NVIDIA H100 SXM5, NVIDIA InfiniBand NDR (Quantum-2) · counts: 3072 |
| databricks.com | names system, part and count | NVIDIA H100 SXM5 · counts: 3072 |
Checked 2026-10-09 by pnpm verify. Re-run it and the table changes with the web.