New 8-GPU AI Servers from Lenovo

The new Lenovo ThinkSystem SR680a V3, SR685a V3, and SR780a V3 GPU systems deliver massive computational performance for Artificial Intelligence (AI), High-Performance Computing (HPC), and graphical and simulation workloads across various industries. The family of servers supports eight high-performance GPUs, either from AMD or NVIDIA, with planned support for GPUs from Intel.

Figure 1. New Lenovo ThinkSystem servers that have 2 processors and 8 GPUs (left to right): SR680a V3, SR780a V3, and SR685a V3

The SR680a V3 and SR685a V3 servers feature two Intel® Xeon® and two AMD EPYC™ processors, respectively, and are 100% air-cooled servers suitable for most data centers. The SR780a V3 server features two Intel Xeon processors and uses a hybrid water-air cooling system to take advantage of the efficiencies of cooling the CPUs and GPUs using a direct water cooled infrastructure.

We also announced enhancements to our Lenovo AI Services Practice with new offerings such as AI Discover and new Fast Start services. We unveiled services to implement, adopt, and scale Generative AI solutions which customers can easily deploy and manage as a Service with Lenovo TruScale.

The ThinkSystem SR680a V3 offers a choice of acceleration platforms featuring NVIDIA H100 or H200 GPUs with planned support for AMD MI300X and other future GPUs. With high-speed interconnects between the GPUs, the system delivers unparalleled computational power for demanding AI and HPC workloads.

• 8U chassis with 100% air-cooling
• Two 5th Gen Intel® Xeon® Scalable processors
• 8 GPUs with high-speed interconnect, choice of:
  • NVIDIA H100 700W SXM5 GPUs with 80GB HBM3 GPU memory per GPU
  • NVIDIA H200 700W SXM5 GPUs with 141GB HBM3 GPU memory per GPU
  • AMD MI300X 750W OAM GPUs with 192GB HBM3 memory per GPU (planned)
• Up to 2TB memory with 32x 64GB RDIMMs
• PCIe 5.0 x16 interfaces to all GPUs and network adapters
• Support for high-speed networking up to 400 Gb/s, directly connected to the GPU complex
• Up to 16x high-speed 2.5” NVMe drives

Figure 2. ThinkSystem SR680a V3

For more information see the following resources:

• ThinkSystem SR680a V3 datasheet
• ThinkSystem SR680a V3 product web page
• ThinkSystem SR680a V3 product guide
• ThinkSystem SR680a V3 interactive 3D tour (coming in April)

The ThinkSystem SR685a V3 is an 8U2S rack server built for demanding AI and HPC workloads. It’s accelerated for intense computing using industry-leading 4th Generation AMD EPYC™ Processors, interconnects with the fastest transfer rate with AMD Infinity Fabric or NVIDIA NVLink, and supports 8x of the latest GPUs. It has the computing power to tackle modeling, training, rendering, financial tech, scientific research, and more.

• 8U chassis with 100% air-cooling
• Two 4th Gen AMD EPYC processors
• 8 GPUs with high-speed interconnect, choice of:
  • AMD MI300X 750W OAM GPUs with 192GB HBM3 memory per GPU
  • NVIDIA H100 700W SXM5 GPUs with 80GB HBM3 GPU memory per GPU
  • NVIDIA H200 700W SXM5 GPUs with 141GB HBM3 GPU memory per GPU (planned)
• Up to 2.25TB memory with 32x 96GB RDIMMs
• PCIe 5.0 x16 interfaces to all GPUs and network adapters
• Support for high-speed networking up to 400 Gb/s, directly connected to the GPU complex
• Up to 16x high-speed 2.5” NVMe drives

Figure 3. ThinkSystem SR685a V3

For more information see the following resources:

• ThinkSystem SR685a V3 datasheet
• ThinkSystem SR685a V3 product web page
• ThinkSystem SR685a V3 product guide
• ThinkSystem SR685a V3 interactive 3D tour (coming in April)

The ThinkSystem SR780a V3 harnesses the might of eight NVIDIA H200 Tensor Core GPUs, paired with two 5th Gen Intel® Xeon® Scalable processors and 32 DDR5 DIMMs. Through Lenovo Neptune™ liquid cooling, this system achieves the extensive computational prowess crucial for handling demanding AI and HPC workloads. Housed in a compact 5U design, it seamlessly fits into a standard 19-inch rack with a water manifold attached at the rear of the rack.

• 5U chassis with up 80% water-cooling (CPUs & GPUs), 20% air-cooling (other components)
• Two 5th Gen Intel® Xeon® Scalable processors
• 8 GPUs with high-speed interconnect, choice of:
  • NVIDIA H100 700W SXM5 GPUs with 80GB HBM3 GPU memory per GPU
  • NVIDIA H200 700W SXM5 GPUs with 141GB HBM3 GPU memory per GPU
  • NVIDIA B200 SXM6 GPUs with 175GB HBM3 GPU memory per GPU (planned)
  • AMD MI300X 750W OAM GPUs with 192GB HBM3 memory per GPU (planned)
• Up to 2TB memory with 32x 64GB RDIMMs
• PCIe 5.0 x16 interfaces to all GPUs and network adapters
• Support for high-speed networking up to 400 Gb/s, directly connected to the GPU complex
• Up to 8x high-speed 2.5” NVMe drives

Figure 4. ThinkSystem SR780a V3

For more information see the following resources:

• ThinkSystem SR780a V3 datasheet
• ThinkSystem SR780a V3 product web page
• Thinksystem SR780a V3 product guide (planned for later in 2024)
• Thinksystem SR780a V3 interactive 3D tour (planned for later in 2024)

For more information about these servers, contact your Lenovo sales representative or use the Contact Us button on one of the product web pages listed above. Preorders for the SR680a V3 and SR685a V3 are planned in March with general orders planned to open towards the end April. Orders for the SR780a V3 are planned for later in 2024.