OCP Certified · R2v3 · OCP Solution Provider
Find your fit : what describes your workload?
Compute
Leopard V2
General-purpose hyperscale compute
Originally designed by Meta’s engineering teams to run 24/7 in demanding data centers. Dual-socket Xeon architecture optimized for maximum density, efficient cooling, and OCP compliance.
|
Basic Single Intel® Xeon® 2680v4 · 64 GB DDR4 · Single SFP NIC |
Entry |
|
Plus Dual Intel® Xeon® 2680v4 · 256 GB DDR4 · 1 TB SATA · Single SFP NIC |
Popular |
|
Advanced Dual Intel® Xeon® 2680v4 · 256 GB DDR4 · Up to 4 TB SATA · SFP NIC |
Max density |
Enterprise ITKubernetes / K8sCloud hostingDatabase workloadsCDN edge nodes
Compute: High Memory
Tioga Pass
High-memory ·| 100G networking ·| OpenBMC
Meta’s successor to the Leopard platform. Doubled PCIe bandwidth, 100G integrated NICs, and OpenBMC management. Built for memory-intensive workloads, ML training nodes, and high-density virtualization.
|
Basic Single Intel® Xeon® Gold 6138 · 64 GB DDR4 · Single SFP NIC |
|
|
Plus Dual Intel® Xeon® Gold 6138 · 256 GB DDR4 · 1 TB SATA · SFP NIC |
Popular |
|
Advanced Dual Intel® Xeon® Gold 6138 · 768 GB DDR4 · 4 TB SATA · SFP28 100G NIC |
Max RAM |
In-memory databasesML training nodesHigh-density VMsOpenBMC managed
Storage
Bryce Canyon JBOD
High-density SAS-3 storage ·| OCP Open Rack v2
Originally built for Meta’s photo and video storage at hyperscale. 20% higher HDD density than Open Vault, fully toolless servicing, and OpenBMC management, all in an OCP Open Rack v2 chassis.
|
Single node 72× 3.5″ SAS-3 drives · Up to 1.44 PB raw · Dual storage server or JBOD config |
|
Full rack 8× Bryce Canyon nodes · 576 drives total · Up to 11.5 PB per ORv2 rack |
Data lakesBackup & archivingAI dataset storageMedia archives
Coming Q2 2026
Grand Canyon SAS-4 & Lightning NVMe JBOF
Up to 11.5 PB SAS-4 · Up to 2.7 PB NVMe per rack (360× u.2 drives)
AI & HPC
Big Basin GPU Server
NVIDIA V100 / P100 ·| PCIe GPU clusters | OCP Open Rack v2
Meta’s high-performance AI and machine learning platform. Validated for inferencing, small-scale training, and cost-effective HPC clusters. Delivering TFLOPS per dollar that new hardware can’t match at this price point.
|
Inference Basic 16× NVIDIA P100 · FP-32: 4× 10.6 TFLOPS · 4× 10G connectivity |
|
Tensor Basic 16× NVIDIA V100 · FP-32: 4× 15 TFLOPS · Tensor Core: 125 TFLOPS Recommended |
|
Tensor Plus 32× NVIDIA V100 · FP-32: 8× 15 TFLOPS · Tensor Core: 8× 125 TFLOPS Max TFLOPS |
LLM inferencingSmall-scale trainingScientific simulationCFD / modellingResearch clusters
Not just refurbished. Remanufactured to OEM spec.
Every system that leaves our facility has been through full hardware diagnostics, component-level testing, firmware updates, and thermal validation, backed by ISO 9001, ISO 14001, and R2v3 certifications.
Italy · 2024HPC · District Heating
Brescia’s Data Center Revolution: Turning Waste Heat into Urban Value
A2A, Qarnot, and RackRenew partnered to transform Brescia’s new data center into a citywide heating source. RackRenew’s remanufactured OCP servers power liquid-cooled pods that recover thermal energy at up to 65°C, fed directly into the city’s district heating grid.
“This first deployment in Italy is a strategic milestone for Qarnot. Made possible through our collaboration with A2A and RackRenew, it gives us the opportunity to address the Italian market’s growing needs for sovereign, energy-efficient cloud computing.”
Paul Benoit, CEO & Co-founder, Qarnot
Impact at a glance
Partners on this project:
A2A Energy · Qarnot Computing · RackRenew
Remanufactured OCP compute nodes operating inside Qarnot’s liquid-cooled pod infrastructure, validating that second-life servers can power mission-critical workloads while simultaneously contributing to urban heating networks.
