- Bolt Graphics is building a RISC-V-based GPU to challenge Nvidia
- CUDA support on RISC-V could lower software barriers for alternative accelerators
- Zeus targets path tracing, HPC and large memory workloads over traditional shaders
Bolt Graphics is pressing ahead with its plan to challenge Nvidia and AMD by building a graphics processor around a RISC-V controlled architecture rather than a conventional GPU design.
The Sunnyvale, California-based startup’s Zeus architecture is a reimagining of graphics, rendering and high-performance computing workloads.
Instead of relying on traditional shader-heavy designs, Zeus combines fixed-function hardware for rasterization, ray tracing, and path tracing with an internal SIMD engine.
A standalone Linux system
Command and scheduling tasks are handled by a RISC-V processor that also acts as a general-purpose CPU, allowing Zeus to run as a standalone Linux system instead of relying solely on a host processor.
We wrote about Bolt and Zeus in 2025 , and the company used CES 2026 to showcase its plans, which seem even more viable after Nvidia’s move to bring CUDA support to RISC-V systems.
Since CUDA is no longer tied exclusively to x86 or Arm hosts, a RISC-V-based accelerator stack becomes more practical for developers already invested in Nvidia’s software ecosystem.
Zeus cards support Vulkan and DirectX 12, along with engines such as Unreal and Unity, while also supporting common programming environments used in HPC, including Python, Fortran and OSL compiled through LLVM.
The prototype add-in card uses a PCIe 5.0 x16 interface and pairs LPDDR5X graphics memory with DDR5 SODIMM slots for the RISC-V processor.
Depending on the configuration, the total memory capacity can reach 384 GB on a single card.
Bolt plans several Zeus variants, including the Zeus 1c26-032, Zeus 2c26-064, Zeus 2c26-128, and Zeus 4c26-256, spanning single-chip PCIe card and multi-chip 2U server designs with combined memory capacities in excess of 2TB.
Networking is addressed through integrated 400 Gbps and 800 Gbps interfaces intended for render farms and clustered workloads. These interfaces are designed to allow direct GPU to GPU connections without separate network cards.
The board also includes BMC and IPMI hardware, features more commonly found on servers than consumer graphics cards.
Power consumption is limited for its class, with the card relying on a single 8-pin PCIe connector for up to 225W, while higher-end server configurations scale up to 500W.
Bolt has made some intriguing (pinch salt) performance claims, including path tracking throughput several times higher than Nvidia’s RTX 5090 and extreme gains in FP64 simulation workloads.
These numbers are naturally based on internal tests and simulations, with real hardware validation still pending.
If CUDA on RISC-V gains traction, Bolt’s approach may face fewer software barriers than similar efforts in the past.
Of course, this still presents a risk of execution, but the technical direction suggests that Bolt is betting on ecosystem change rather than brute force scale alone.
Via TechPowerUp
Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews and opinions in your feeds. Be sure to click the Follow button!
And of course you can too follow TechRadar on TikTok for news, reviews, video unboxings, and get regular updates from us on WhatsApp also.
