Some workloads need both a large coherent system and room for accelerators or extensive I/O, and that combination is hard to deliver in a dense 2U. The ThinkSystem SR860 is Lenovo's four-socket 4U server built for exactly that: mission-critical scale-up with the physical space for GPUs, more drives and more PCIe than its 2U sibling can offer. This guide explains where the SR860 fits against the dense SR850 V3, how to spec it for in-memory databases or accelerated workloads, and when a 4U scale-up box is the right call.
What the SR860 is for
The SR860 is a four-socket 4U server. The extra rack height over a 2U four-socket platform buys two things buyers often need together: physical room for GPUs and full-height, full-length cards, and generous drive and PCIe capacity. That makes it suited to mission-critical scale-up workloads that also want accelerators, large local storage, or simply more expansion than a dense 2U can physically house.
Think of it as the scale-up platform for jobs that are not only memory-hungry but also I/O-hungry or accelerator-adjacent: large in-memory databases with demanding storage, consolidation hosts that need extensive networking, or analytics workloads that pair big memory with GPUs.
Memory and four-socket sizing
As with any four-socket server, memory is the primary reason for the architecture. The SR860 supports a very large DDR5 footprint across its sockets, and the design goal is to hold a big working set in RAM. Size to the application's certified requirement plus headroom, balance every channel, and plan the layout up front. Use our memory and RAM guidance to keep a high-capacity configuration balanced rather than bandwidth-starved.
Four sockets mean four NUMA domains. Keep the workload NUMA-local where possible, and remember that on per-core-licensed software the temptation to max core count is usually a false economy. Choose processors for the role with our processors guidance.
The GPU and expansion dimension
The SR860's defining advantage over the 2U SR850 V3 is space. Where the SR850 V3 prioritises rack density, the SR860 gives you the slots and thermal envelope to add accelerators and high-speed cards alongside a four-socket compute and memory base. For workloads that pair a large in-memory dataset with GPU-accelerated analytics or inference, that physical headroom is the deciding factor.
If you do add accelerators, size power and cooling for them deliberately rather than assuming the chassis will absorb the load. Match the accelerator to the workload with our GPU accelerators guidance, and remember that a few large GPUs change the server's power profile substantially.
SR860 or SR850 V3?
The choice between the two four-socket Lenovo platforms is largely physical. Choose the dense SR850 V3 when rack space is tight and you do not need GPUs or extensive expansion: it delivers four-socket scale-up at close to 2U density. Choose the SR860 when you need GPUs, more drives, or more PCIe than a 2U can physically provide, and you have the rack units to spend.
Both share Lenovo's XClarity management and operational model, so the decision is about expansion and density rather than how you run the box. For cross-vendor context on four-socket scale-up, our Dell vs HPE vs Lenovo comparison frames the platform landscape.
Resilience and management
Mission-critical scale-up demands the full RAS toolkit: ECC memory with patrol scrub and advanced protection, redundant hot-plug power supplies on independent feeds, redundant fans, and licensed XClarity for remote console and lifecycle management. Consolidating important workloads onto one large server only makes sense if that server is built to keep running through component faults.
Decide the high-availability approach early, since it determines system count and storage design. Build and price an SR860 to your memory, core and accelerator targets in our Lenovo configurator.