Harpia node architecture
Harpia is operated by Petróleo Brasileiro S.A in Brazil. This profile covers Ranked system / partition. Specifications are listed below; anything the operator has not published is itemised in the documented-fields list.
Published benchmark (TOP500 · June 2026)
- HPL rank
- #37
- HPL Rmax
- 75.2 PFlop/s
- Theoretical peak Rpeak
- 141 PFlop/s
- Ranked core count
- 303,104
- Measured power
- 2008 kW
- Energy efficiency
- 37.45 GFlop/s per watt
Cluster · ThinkSystem SR665 V3 · public profile
- Operator / site
- Petróleo Brasileiro S.A
- Country
- Brazil
- Profile scope
- Ranked system / partition
- Data checked
- 14 September 2026
- Manufacturer
- Lenovo
- Ranked core count
- 303104
- Theoretical peak (PFlop/s)
- 141
- Measured power (kW)
- 2008
- Energy efficiency (GFlop/s per watt)
- 37.45019920318725
- Operating system
- RedHat 8.10
ThinkSystem SR665 V3 · public profile
- System / node model
- ThinkSystem SR665 V3
- CPU model
- AMD EPYC 9354 32C 3.25GHz
- Cores per socket
- 32
- CPU note
- The submitted server model changed between TOP500 lists (ThinkSystem SR675 V3 with 284,160 cores in November 2025, SR665 V3 with 303,104 cores in June 2026). Petrobras states roughly 146 PFlop/s peak against the 141.00 PFlop/s on the TOP500 record. The submission is also internally inconsistent on the accelerator: Lenovo documents the SR665 V3 as a 2U PCIe server taking at most three double-wide GPUs and lists no SXM or HGX baseboard support, while the TOP500 string names an SXM5 part. No per-node accelerator count is recorded for this system.
- Accelerator model
- NVIDIA H100 SXM5 80GB
- Accelerator memory
- 80 GB per accelerator
Scope and uncertainty
Partial public profile: processor and fabric follow the TOP500 submission. Node count, socket count, memory and on-node wiring are not assumed from the server model or benchmark core count. This entry describes the ranked system or partition, not every resource operated at this site.
Documented fields: 12 / 27
Every field this catalog checks. Fields listed above have a value; the remainder are shown here. A field marked absent is not publicly available in the sources we could verify, which does not mean the hardware is missing. Fields that cannot apply, such as benchmark results for an unranked system, are excluded from the total.
Cluster — 6 / 7
- ✓Manufacturer
- ✓Ranked core count
- ✓Theoretical peak (PFlop/s)
- ✓Measured power (kW)
- ✓Energy efficiency (GFlop/s per watt)
- ✓Operating system
- ✕Network topology
ThinkSystem SR665 V3 · public profile — 6 / 20
- ✓System / node model
- ✕Compute nodes of this type
- ✓CPU model
- ✕CPU sockets per node
- ✓Cores per socket
- ✕Physical CPU cores
- ✓CPU note
- ✓Accelerator model
- ✕Physical accelerators per node
- ✕Host / unified memory
- ✓Accelerator memory
- –Accelerator ↔ accelerator link
- –On-node fabric
- ✕CPU ↔ accelerator link
- ✕Network interfaces per node
- ✕Interface model
- ✕Interface speed
- ✕Interface-to-accelerator topology
- –Aggregate scale-out per node
- ✕GPUDirect RDMA
- ✕Management / storage networking
- ✕Local storage
- –NVMe data disks
- –Max remote data disks
- –Uncached disk IOPS
- –Uncached disk throughput (MBps)
- ✕Form factor
Sources
Browse all cluster architecture pages →
CLUSTERATLAS