SiFive is attempting to move the RISC-V architecture from a theoretical concept to a production-grade reality within the enterprise datacenter. By introducing the BigSky SF-2U870, the company is providing a tangible hardware target for hyperscalers and ecosystem partners to perform critical software porting and workload tuning. This development marks a strategic shift toward commercializing open-standard silicon for large-scale infrastructure. The 2U rackable server, which SiFive claims is the first of its kind for the enterprise, is designed to bridge the gap between custom SoC development and standardized datacenter deployment. This move targets the total cost of ownership (TCO) optimization that custom silicon offers, while attempting to mitigate the traditional software compatibility risks associated with non-x86 or non-ARM architectures.
SiFive BigSky SF-2U870 Hardware Deployment
The BigSky SF-2U870 platform serves as a turnkey development environment intended to accelerate the deployment of RISC-V solutions. SiFive is positioning this 2U server as a way for customers to validate systems and optimize workloads before moving to full-scale production. According to the company, lead customers—including hyperscalers, software providers, and major SoC hardware leaders—are already running workloads on deployed BigSky systems. To facilitate immediate enterprise utility, the platform adheres to the RVA23 standard, which SiFive says eliminates the need for complex, custom software porting. This standardization is intended to allow for the "out-of-the-box" deployment of established enterprise software environments.
The hardware configuration includes 32 P870-D cores operating at 2.0GHz and 256GB of DDR5-5600 memory. For data-intensive enterprise tasks, the server provides 4x PCIe Gen5 x16 lanes (64 lanes total), support for x16 double-wide GPUs, and a 3.84 TB Data SSD U.2 NVME. Connectivity is handled via 10/25Gb OCP 3.0 NICs. SiFive notes that the platform is currently being produced in limited quantities to support strategic development, with current demand reportedly outpacing available supply. This limited availability suggests the company is focusing on high-value partnerships to stabilize the ecosystem before a broader commercial rollout.
Integrating CUDA and Enterprise Linux Ecosystems
A critical component of SiFive’s strategy involves securing the software layer through partnerships with major Linux distributors and NVIDIA. The BigSky platform provides out-of-the-box support for Canonical Ubuntu 26.04 LTS and Red Hat RHEL10, which SiFive identifies as essential for accelerating AI and cloud-native workloads. By ensuring compatibility with these standard distributions, the company aims to provide the stability that enterprise IT leaders require when transitioning away from proprietary architectures. Red Hat has characterized this integration as proof that the industry is moving into a "production-grade reality" via the RISE project.
Furthermore, SiFive is working with NVIDIA to port CUDA to RISC-V hardware and plans to integrate NVLink Fusion into future platforms. The company has already demonstrated CUDA running on the P870-D-powered BigSky platform. In this specific configuration, the BigSky server acts as the head node for modern Large Language Model (LLM) workloads running on NVIDIA GPUs. This capability is presented as a significant milestone for RISC-V, as it attempts to link the open-standard CPU architecture directly to the dominant hardware used for modern AI acceleration, potentially reducing the friction of adopting RISC-V in AI-centric datacenters.
Key Takeaways
- SiFive has released the BigSky SF-2U870, described as the first enterprise-grade, rackable 2U RISC-V server.
- The platform supports standard enterprise Linux environments, specifically Canonical Ubuntu 26.04 LTS and Red Hat RHEL10.
- SiFive is collaborating with NVIDIA to run CUDA on its P870-D-powered hardware, utilizing the server as a head node for LLM workloads.
TechInsyte's Take
In our view, the BigSky launch is less about selling individual servers and more about de-risking the RISC-V architecture for the world's largest cloud providers. The primary barrier to RISC-V adoption has never been the silicon itself, but rather the "software tax"—the immense effort required to port legacy enterprise workloads to a new instruction set. By shipping a RVA23-compliant 2U server that supports RHEL10 and Ubuntu 26.04 LTS, SiFive is attempting to bypass this hurdle. The most significant strategic signal, however, is the CUDA integration. If SiFive can successfully position RISC-V as a viable head node for NVIDIA-driven AI workloads, they move from being a niche IP provider to a central player in the AI infrastructure stack. This is a calculated attempt to prove that open standards can compete with the established dominance of x86 and ARM in the high-stakes datacenter market.
Questions & Answers
How does the BigSky platform address the software compatibility challenges of RISC-V?
The platform utilizes the RVA23 standard to enable "out-of-the-box" deployment of enterprise software. It specifically provides support for industry-standard Linux distributions, including Canonical Ubuntu 26.04 LTS and Red Hat RHEL10, which reduces the need for custom, complex software porting.
What role does NVIDIA play in the SiFive RISC-V ecosystem?
SiFive is working with NVIDIA to port CUDA to RISC-V hardware and intends to integrate NVLink Fusion into future platforms. Currently, CUDA is running on the P870-D-powered BigSky platform, allowing it to function as a head node for LLM workloads on NVIDIA GPUs.
What are the core technical specifications of the BigSky SF-2U870?
The server features 32 P870-D cores at 2.0GHz, 256GB of DDR5-5600 memory, 4x PCIe Gen5 x16 lanes, support for x16 double-wide GPUs, a 3.84 TB Data SSD U.2 NVME, and 10/25Gb OCP 3.0 NICs.
Is the BigSky platform available for general enterprise purchase?
The platform is currently being produced in limited quantities to drive strategic development opportunities. SiFive has stated that demand is currently outpacing supply, suggesting it is primarily targeted at strategic partners and hyperscalers for now.
Source: Businesswire
