- Bolt Graphics is constructing a RISC-V primarily based GPU to problem Nvidia
- CUDA assist on RISC-V may decrease software program limitations for different accelerators
- Zeus targets path tracing, HPC, and enormous reminiscence workloads over conventional shaders
Bolt Graphics is urgent forward with its plan to problem Nvidia and AMD by constructing a graphics processor round a RISC-V managed structure slightly than a standard GPU design.
The Sunnyvale, California primarily based startup’s Zeus structure is a floor up rethink of graphics, rendering, and excessive efficiency compute workloads.
As a substitute of counting on conventional shader heavy designs, Zeus combines fastened operate {hardware} for rasterization, ray tracing, and path tracing with an in home SIMD engine.
A standalone Linux system
Command and scheduling duties are dealt with by a RISC-V processor that additionally features as a normal function CPU, permitting Zeus to run as a standalone Linux system slightly than relying fully on a number processor.
With CUDA now not tied solely to x86 or Arm hosts, a RISC-V primarily based accelerator stack turns into extra sensible for builders already invested in Nvidia’s software program ecosystem.
Zeus playing cards assist Vulkan and DirectX 12, alongside engines corresponding to Unreal and Unity, whereas additionally supporting frequent programming environments utilized in HPC, together with Python, Fortran, and OSL compiled via LLVM.
The prototype add in card makes use of a PCIe 5.0 x16 interface and pairs LPDDR5X graphics reminiscence with DDR5 SODIMM slots for the RISC-V processor.
Relying on configuration, whole reminiscence capability can attain 384GB on a single board.
Bolt plans a number of Zeus variants, together with the Zeus 1c26-032, Zeus 2c26-064, Zeus 2c26-128, and Zeus 4c26-256, spanning single chip PCIe playing cards and multi chip 2U server designs with mixed reminiscence capability exceeding 2TB.
Networking is addressed via built-in 400Gbps and 800Gbps interfaces meant for render farms and clustered workloads. These interfaces are designed to permit direct GPU to GPU connections with out separate community interface playing cards.
The board additionally consists of BMC and IPMI {hardware}, options extra generally present in servers than client graphics playing cards.
Energy draw is restrained for its class, with the cardboard counting on a single 8 pin PCIe connector for as much as 225W, whereas greater finish server configurations scale as much as 500W.
Bolt has made some intriguing (pinch of salt) efficiency claims, together with path tracing throughput a number of instances greater than Nvidia’s RTX 5090 and excessive positive aspects in FP64 simulation workloads.
These figures are naturally sufficient primarily based on inner testing and simulations, with actual {hardware} validation nonetheless pending.
If CUDA on RISC-V positive aspects traction, Bolt’s strategy may face fewer software program limitations than comparable efforts up to now.
That also leaves execution threat in fact, however the technical course suggests Bolt is betting on ecosystem shifts slightly than brute pressure scale alone.
Through TechPowerUp
Observe TechRadar on Google Information and add us as a most popular supply to get our knowledgeable information, critiques, and opinion in your feeds. Be sure to click on the Observe button!
And naturally it’s also possible to observe TechRadar on TikTok for information, critiques, unboxings in video type, and get common updates from us on WhatsApp too.
