 
|
MQM8790-HS2R 200G InfiniBand Smart Switch – Unmanaged, C2P Airflow, 40x QSFP56
Λεπτομέρειες:
| Μάρκα: | Mellanox |
| Αριθμό μοντέλου: | MQM8790-HS2R (920-9B110-00RH-0D0) |
| Έγγραφο: | MQM8700 series.pdf |
Πληρωμής & Αποστολής Όροι:
| Ποσότητα παραγγελίας min: | 1 ΤΕΜ |
|---|---|
| Τιμή: | Negotiate |
| Συσκευασία λεπτομέρειες: | Εξωτερικό κουτί |
| Χρόνος παράδοσης: | Βασισμένο στο αποθεματικό |
| Όροι πληρωμής: | T/T |
| Δυνατότητα προσφοράς: | Προμήθεια ανά έργο/παρτίδα |
|
Λεπτομερής ενημέρωση |
|||
| Μοντέλο ΑΡΙΘ.: | MQM8790-HS2R | ρυθμό μετάδοσης: | 10/100/1000Mbps |
|---|---|---|---|
| λιμάνια: | ≧ 48 | Τεχνολογία: | Infiniband |
| Πακέτο μεταφοράς: | Συσκευασία | Εμπορικό σήμα: | Mellanox |
| Επισημαίνω: | Nvidia InfiniBand smart switch,200G unmanaged network switch,QSFP56 InfiniBand switch C2P airflow |
||
Περιγραφή προϊόντων
MQM8790-HS2R 200G InfiniBand Smart Switch
Unmanaged, C2P Airflow, 40x QSFP56 Ports
Product Overview
The NVIDIA Quantum MQM8790-HS2R is a 200Gb/s InfiniBand smart switch engineered for high-performance computing, AI clouds, and hyperscale storage environments. This unmanaged switch delivers 40 non-blocking 200Gb/s ports in a compact 1U form factor with 16Tb/s aggregate throughput and features in-network computing acceleration via SHARP technology.
Designed as an externally managed switch with C2P (Connector-to-Power) airflow direction, it leverages advanced fabric management platforms to dramatically boost cluster efficiency while reducing operational expenses.
Key Features
- 40 x 200Gb/s Non-blocking Ports: Aggregate data throughput up to 16Tb/s with zero packet loss and ultra-low latency
- Port Splitting Flexibility: Each QSFP56 port can operate as 2x100Gb/s, enabling 80 ports of 100G in a single 1U
- In-Network Computing (SHARP): Hardware-embedded aggregation reduces data movement and accelerates MPI/NCCL collectives
- Advanced Fabric Management: External management pairs with NVIDIA UFM for proactive telemetry and congestion control
- Redundant & Hot-swap Power: 1+1 redundant AC power supplies with 80 Plus Gold+ certification
- C2P Airflow Direction: Connector-to-Power airflow ideal for specific data center cooling layouts
Technology: SHARP & In-Network Computing
NVIDIA Quantum switches embed silicon-based aggregation engines that accelerate collective communication patterns. With SHARP technology, data is processed as it traverses the switch, reducing hops and latency. This is critical for large language model training, molecular dynamics, and weather simulation applications.
Typical Deployments
- High-Performance Computing (HPC): Large-scale clusters for research, energy, and scientific computing
- AI & Deep Learning Clusters: GPU-accelerated environments using NCCL optimized communication
- Hyperscale Cloud & Storage: NVMe over Fabrics (NVMe-oF) and distributed storage backends with RDMA
- Data Center Fat-Tree Topologies: Leaf-spine architectures needing dense 100G/200G radix
- DragonFly+ & Torus Networks: Externally managed switches for customized high-radix designs
- Reverse Airflow Racks: Data centers requiring C2P (connector-to-power) cooling pattern
Technical Specifications
| Parameter | Specification |
|---|---|
| Ports | 40 QSFP56 ports (200Gb/s each) or 80 ports at 100Gb/s via splitting |
| Switching Capacity | 16 Tb/s non-blocking aggregate throughput |
| Latency | Extremely low, sub-130ns typical |
| Management Type | Unmanaged (external controller required) |
| Power Supplies | 1+1 redundant, hot-swappable AC power supplies |
| Airflow | C2P (Connector-to-Power) |
| Dimensions (HxWxD) | 1.7 in (43.6mm) x 17 in (433.2mm) x 23.2 in (590.6mm) |
| Weight | With 1 PSU: 11.4 kg; with 2 PSUs: 12.48 kg |
| Operating Conditions | 0°C to 40°C, 10%-85% non-condensing humidity |
| Certifications | CE, FCC, VCCI, ICES, RCMS, RoHS compliant, ENERGY STAR |
| Warranty | 1 year (extended available upon request) |
Compatibility
Backward compatible with previous InfiniBand generations (EDR, HDR). Seamless interoperability with NVIDIA ConnectX-6, ConnectX-7 adapters, and Quantum-2 switches. Supported software includes MLNX-OS for management and compatibility with standard InfiniBand fabrics.
| Component | Compatible Models / Versions |
|---|---|
| Host Channel Adapters | NVIDIA ConnectX-6 (HDR/200G), ConnectX-7, ConnectX-5/EDR |
| Optical/Cables | QSFP56 passive DAC, active optical cables, SR4/LR4 optical modules |
| Subnet Management | NVIDIA UFM, OpenSM, native InfiniBand SM |
| Operating Systems | RHEL, Ubuntu, SLES, Windows Server with OFED |
Buyer Checklist
- Confirm AC power connector type (C14) and redundancy requirement
- Choose port count: 40x200G or split to 80x100G
- Ensure external Subnet Manager license/availability
- Check rack depth minimum 600mm for 590mm switch depth
- Verify airflow compatibility: C2P requires cold aisle on QSFP connector side
- Request volume quotation & lead time
Frequently Asked Questions
What is the difference between MQM8790-HS2R and MQM8790-HS2F?
The only difference is airflow direction: HS2R has C2P (Connector-to-Power), HS2F has P2C (Port-to-Power). Choose based on your rack's cold aisle location.
Can I split ports to 100Gb/s?
Yes. Each QSFP56 port can be configured as two independent 100Gb/s ports, providing double-density 100G radix.
Does this switch support RDMA?
InfiniBand natively supports RDMA with hardware transport acceleration; it is the native protocol for NVIDIA InfiniBand fabrics.
Related Products
Service & Support
Hong Kong Starsurge Group provides professional pre-sales consulting, configuration design, and post-sales technical assistance for NVIDIA Quantum switches. Services include free architecture consultation, RMA support, firmware updates, and custom rack integration with global shipping options.
Θέλετε να μάθετε περισσότερες λεπτομέρειες σχετικά με αυτό το προϊόν
