RISC-V Servers

Servers

Server-class systems, boards, workstations and cloud instances, built from publicly checkable sources

Servers (systems and boards)2

Server systems, boards and workstations, built from publicly checkable sources, one entry per model.

Lanxin LX500 compute-AI convergence server CPU (complete system: Lenovo WenTian WR5219 G5)

Tested on real hardware by Project Jiachen In production

蓝芯算力(深圳)科技有限公司 · Lanxin Computing (Shenzhen) Technology Co., Ltd.

SoC / processor
LX500 single die: 48 heterogeneous cores = 32 performance cores + 16 efficiency cores (sharing the on-die memory space); an AI matrix engine integrated on-die, peak 75 TOPS INT8, supporting INT4/INT8/FP16/FP32; RVV 1.0 and RVA23 support; matrix-multiply extension instructions plus dedicated matrix-arithmetic hardware units.
Form factor
The chip is a single-die package (package type and dimensions not disclosed); the complete system is a 2U rack server (Lenovo WenTian WR5219 G5).
Memory
4-channel DDR5, up to 6400 MT/s; unified memory between the CPU and the on-die AI engine (one shared memory space, zero data movement); CXL 2.0 memory expansion supported.
Storage and I/O
64 PCIe Gen5 lanes (with bifurcation support) + 4 PCIe Gen3 lanes; CXL 2.0; the complete system (WR5219 G5) offers multiple storage/IO/memory options and supports hot-swap drives.
Official OS
Ubuntu, openEuler (the official bring-up demo booted on real hardware under both); China Mobile's BC-Linux has completed in-depth adaptation.
Where to get it
No pricing announced for either the chip or the system; the system ships through Lenovo's WenTian product line
First shown
2026-02

Official positioning: the world's first server-oriented RISC-V + AI “convergence” CPU (flagship product), putting general-purpose compute and on-die AI inference on the same chip in place of a “CPU + discrete accelerator card” pairing; the host system is the 2U rack server Lenovo WenTian WR5219 G5, with all-in-one and converged super-node form factors also offered.

Buy enquiry

SpacemiT V100 server-class CPU (systems: C2201 / G5201 / S5236)

Tested on real hardware by Project Jiachen Announced

进迭时空(北京)智能科技有限公司 · SpacemiT

SoC / processor
V100 single die: 46 heterogeneous cores = 6× XiangShan Kunminghu V2R2 (up to 2.1 GHz, single-core SPECint2006 ≥15.0/GHz) + 40× in-house SpacemiT X100 (up to 2.4 GHz, ≥9.0/GHz); RVA23; 64 MB SLC; dual-socket coherent interconnect scaling to 92 cores.
Form factor
FCBGA package, 58.5 × 75.4 mm, 0.94 mm pitch; the C2201 system is a 2U rack unit, 820×447.4×87.8 mm, ≤35 kg (G5201 4U 895×447.4×176.8 mm; S5236 4U).
Memory
8-channel DDR5, 1DPC, DDR5 RDIMM, 4800 MT/s, 307 GB/s peak bandwidth per chip; up to 256 GB per DIMM, up to 2 TB per chip; the C2201 system has 16 DIMM slots and supports up to 2 TB.
Storage and I/O
52 PCIe 5.0 lanes in total (flexible x16–x1 configuration, expandable with AI cards and SAS/SATA/NVMe); 1×GMAC (IEEE 1588); peripherals 1×QSPI, 1×LPC, 1×eSPI, 3×SPI master, 4×I3C, 12×I2C, 12×SMBUS, 5×UART. C2201 system: 88 PCIe 5.0 lanes across two sockets, 8×3.5" U.2 NVMe front-accessible, 1×M.2 internal, 2×2.5" SATA rear.
Power
Chip TDP 120W (official datasheet); the C2201 system ships with 1600W CRPS Platinum power supplies in 1+1 redundancy.
Official OS
openEuler, openRuyi, openKylin, Debian, Ubuntu (26.10), Red Hat Enterprise Linux 10.2 (developer preview), Fedora, Rocky Linux — all directly installable and bootable; firmware UEFI 2.1.1 + ACPI 6.6, EDK2, openBMC; virtualization KVM/QEMU, Docker.
Where to get it
No pricing or volume-production date announced; companion systems C2201 / G5201 / S5236
First shown
2026-10

Official positioning: the first compute chip / server-class CPU fully compliant with the RISC-V server specification, aimed at general-purpose data-center compute, agent infrastructure and domestically developed AI computing platforms; launched alongside the C2201 compute server (2U, dual-socket), G5201 AI server (4U, 8 cards) and S5236 storage server (4U, 36 bays).

Buy enquiry

Racks (rack-scale POD)1

Rack-scale products: delivered as a whole cabinet and organised as a POD or cluster, usually holding several server nodes and AI accelerators.

MetaStone RISMA® cross-architecture cloud-native POD (rack-scale)

In production

Wuhan MetaStone Technology Co., Ltd. · MetaStone

SoC / processor
Dual-socket 80-core RISC-V server node (the SoC model is not published) supporting multiple AI accelerators
Form factor
Rack-scale POD (cross-architecture, cloud-native)
Storage and I/O
Vendor-claimed 10Gbps line-rate networking; room for multiple AI accelerators
Official OS
Vendor says "multiple mainstream operating systems"; no specific distributions listed
Where to get it
Project-based delivery; the vendor says it is deployed in several AI computing centres, with no price or customer list published
First shown
2025-07

Incubated within the ESWIN group and founded in 2023-07. Its in-house RVOCs open heterogeneous compute system covers kernel drivers, compilers, inference frameworks and model serving; at the 6th RISC-V Summit China in 2026-10 it reported a dual-socket 80-core server with multiple AI accelerators reaching 95.8% of x86 throughput at 32-way concurrency and passing a 12-hour stress test (vendor figures, not independently reproduced)

Buy enquiry

Entries are ordered by how well they are confirmed; the first two have been checked against datasheets by hand. Further candidates are added once their sources are verified — an entry is created only for a product whose specifications have been confirmed, never on hearsay.