Lenovo HPC servers with Neptune
For research, engineering, life sciences or energy: simulation and AI on a cluster sized for your applications, with water-cooled SC777 V4 trays carrying two Grace CPUs and four B200 GPUs, eight-GPU servers and DSS-G storage with IBM Storage Scale.
ThinkSystem SC777 V4Choose by how your applications scale
Simulation and AI in the same tray
Two Grace CPUs and four B200 GPUs linked by NVLink, eight trays per enclosure and three enclosures per rack. For codes that already run on GPUs and need density.
Eight GPUs per server
The whole HGX in one server, water-cooled in the SR780a V3 or air-cooled in the SR680a V4. For training large models, or for a data centre with no water loop.
Lenovo DSS-G: storage for the cluster
Lenovo's parallel storage: two servers and one to eight enclosures of 90 drives, with IBM Storage Scale, formerly GPFS. It's all on hard drives, so it gives you capacity at a good cost per terabyte for simulation data and training datasets. Storage Scale is our home ground: we install it, run it and train people on it. More on this in IBM Storage Scale.
- Servers
- 2 ThinkSystem SR655 V3One AMD EPYC 9004 each
- Drives
- 1 to 8 D4390 enclosures90 × 3.5" NL-SAS drives per enclosure, 12 to 24 TB
- Protection
- Storage Scale RAIDIn software and spread across every drive, no RAID controllers
- Network
- Up to 400 Gb/sTwo ConnectX-7 per server, InfiniBand or Ethernet
- Licence
- By usable capacityNo per-core or per-client licences
- Delivery
- Rack-integratedFrom the factory, in a 42U or 48U rack
Before you ask for a price
Neptune trays or eight-GPU servers?
If your codes scale well across many nodes and your data centre has water, SC777 V4 trays put more GPUs in each rack. If each job fits in eight GPUs, or you need air cooling, an HGX server is simpler to run.
What your data centre needs
The SC777 V4 is cooled by water alone, with no fans, using inlet water from 2 to 45 °C through a CDU. Each 13U N1380 enclosure takes up to four 15 kW power conversion stations, and a rack with three enclosures reaches 162 kW. Before you order, we go through the water loop, the power per rack and the floor loading with you: a full enclosure weighs 470 kg.
Comparison
| Model | GPU | Processor | Max. memory | Form factor | Cooling |
|---|---|---|---|---|---|
| ThinkSystem SC777 V4 | 4 × NVIDIA B200 with 186 GB (GB200 NVL4) | 2 × NVIDIA Grace, 72 Arm cores each | 960 GB LPDDR5X, fixed | Tray, 8 per 13U N1380 enclosure | Liquid (Neptune), no fans |
| ThinkSystem SR780a V3 | NVIDIA HGX B200 (8 × 180 GB GPUs) or HGX H100 | 2 × 5th Gen Intel Xeon, up to 64 cores | 4 TB | 5U | Liquid (Neptune) |
| ThinkSystem SR680a V4 | NVIDIA HGX B300: 8 × 288 GB GPUs with NVLink | 2 Intel Xeon 6, up to 86 cores | 4 TB | 8U | Air |
Why buy it from us
The cluster arrives built and tested with your applications, and if it ever fails, you call the same team that put it into service.
We start from your applications: whether they scale on GPUs or CPUs, how much memory they need, how they communicate across nodes and how much data they move. That tells us how many trays, which network and how much storage, and whether your data centre can handle the water and the power.
Nodes, InfiniBand or Ethernet network, storage with IBM Storage Scale and the scheduler, Slurm or IBM LSF, with your applications compiled and tested before it opens to users.
When something breaks you talk to the engineers who commissioned that machine, and if you need it there's 24×7 on-call cover with a contractual SLA. It's the same independent support we provide for IBM Power, Linux and Ceph estates.
We're an IBM Business Partner and partners of Red Hat, SUSE, Canonical and CIQ. With CIQ we install Rocky Linux for AI, with the NVIDIA driver and CUDA already validated, on the same servers.
We work in all three languages, support included, so if your company has teams in several countries they deal with the same group of engineers.
How we build it
What runs on top of these servers, explained in detail:
Support and training for your team
Frequently asked questions
Does my data centre need a water loop?
For the SC777 V4 and the SR780a V3, yes: both are cooled with warm water through a CDU. If your data centre has no water loop, the SR680a V4 carries eight air-cooled B300 GPUs.
Will my applications run on Arm?
The SC777 V4 uses NVIDIA Grace CPUs, which are Arm. CUDA runs on Arm, but whatever runs on the CPU has to be compiled for that architecture, and not every commercial application has an Arm version. Before you order, we go through each application and its version with you.
What if my workload is CPU only?
The SC750 V4 goes in the same N1380 enclosure, with two nodes per tray, each with two Intel Xeon 6900 processors. Tell us about your applications and we'll tell you which mix of nodes fits.
What software do you install?
RHEL or Ubuntu, CUDA and MPI, Slurm or IBM LSF as the scheduler and IBM Storage Scale for the data. We explain it in HPC with Slurm, LSF and Storage Scale.
Do you have list prices?
No. A cluster depends on the number of nodes, the network, the storage and the warranty. We send you a quote with a confirmed lead time for your configuration.
What cluster do your applications need?
Tell us what applications you run, how many users and what your data centre looks like, and we'll come back with the architecture, the price and a confirmed lead time.
Blog!
Our mission
SIXE ensures that your infrastructures based on IBM®, Red Hat®, Canonical®, and SUSE® technologies remain stable, secure, and optimized for as long as you need them.
At SIXE, we are committed to sustainability. That’s why we help you design, implement, and maintain durable and efficient technological solutions.
