Nvidia Groq 3 LPX
The Nvidia Groq 3 LPX is an inference rack built from 256 Groq 3 LPUs, sold as part of the Vera Rubin platform.
- Vendor
- Nvidia
- Architecture
- Groq 3 LPU
- Memory
- 128 GB SRAM per rack (256 LPUs)
- Memory bandwidth (TB/s)
- 40,000 TB/s
- Form factor
- Rack-scale system: 256 LPUs
- Status
- In production
Nvidia sells Groq 3 LPX as part of the Vera Rubin platform, paired with Vera Rubin NVL72 to speed up token generation. Nvidia lists 128 GB of SRAM, 40 PB/s of memory bandwidth and 640 TB/s of scale-up bandwidth per rack. Nvidia announced full production on August 24, 2026, with Nebius the first cloud to adopt it.
- Memory (GB)
- 128 GB
- Precision note (sparse or dense)
- Rack totals of on-chip SRAM: 128 GB and 40 PB/s memory bandwidth, plus 640 TB/s scale-up bandwidth per rack.
- Launch
- Full production announced August 24, 2026
- Vendor source
- nvidianews.nvidia.com
- Checked
Questions
What is the Nvidia Groq 3 LPX?
It is an inference rack of 256 Groq 3 LPUs, paired with Vera Rubin NVL72 to speed up token generation.
How much memory does a Groq 3 LPX rack have?
Nvidia lists 128 GB of SRAM and 40 PB/s of memory bandwidth per rack, plus 640 TB/s of scale-up bandwidth.
When did Groq 3 LPX enter production?
Nvidia announced full production on August 24, 2026, with Nebius the first cloud to adopt it.
Source
Every figure is taken from the vendor’s own datasheet, product page, technical documentation or announcement, linked on each chip with the date it was last checked. How we pick and check sources: editorial standards.