The holy grail of supercomputing chip design is an architecture that combines the versatility and programmability of CPUs with the sheer synchronicity of GPUs, and that's what InspireSemi strives to achieve. InspireSemi's Thunderbird 'supercomputer-cluster-on-a-chip' packs 1,536 RISC-V cores designed specifically for high-performance computing, but also supports the general-purpose CPU programming model. . It also has incredible scalability—four chips can be placed on a single accelerator card, which comes in a standard GPU-like form factor (AIC), bringing the total number of cores per card to 6,144, which Extended scalability up to 360,000 cores per cluster. .
InspireSemi's Thunderbird processor packs 1,536 custom 64-bit superscalar RISC-V cores with very high-performance SRAM, accelerators for several cryptography algorithms, and an on-chip low-latency mesh fabric for inter- and intra-chip connectivity. . The chip also supports LPDDR memory, NVMe storage, PCIe, and GbE connectivity. It has been tapped and will be fabbed at TSMC, then packaged at ASE.
InspireSemi aims to mount four Thunderbird chips on a single board to offer developers 6,144 RISC-V cores. The current Thunderbird architecture supports scale-out capability for up to 256 processors connected using high-speed serial transceivers.
When it comes to performance, InspireSemi says its solution offers up to 24 FP64 TFLOPS at 50 GFLOPS/W (at 480W), which is an impressive performance. To put that into context, Nvidia's A100 delivers 19.5 FP64 TFLOPS, while Nvidia's H100 reaches 67 FP64 TFLOPS. It's not clear if we're dealing with the performance of a single-chip Thunderbird card or a 4-way model. It's hardly possible to deliver 1920W in an add-in card, so we're likely dealing with four Thunderbird processors on one card that can deliver 24 FP64 TFLOPS per chip.
The superscalar cores support vector tensor operations and mixed-precision floating-point data formats, although there's no word on whether these cores are Linux-capable, which is why InspireSemi calls the Thunderbird an accelerator rather than a general-purpose processor. Is. Still, this processor can be programmed like a regular RISC-V CPU and supports a variety of workloads, such as AI, HPC, graph analytics, blockchain, and other compute-intensive applications. As a result, InspireSemi users won't need to use proprietary tools or software stacks like Nvidia's CUDA. The only question is whether the industry standard tools and software stack will be enough to get the most out of Thunderbird I in all kinds of workloads.
InspireSemi CEO Ron Van Dael said, “We are excited by our engineering and operations team to finalize the Thunderbird I design and deliver it to our world-class supply chain partners, TSMC, ASE, and imec for production. Proud of the achievement.” “We expect to begin customer deliveries in the fourth quarter.”
Speaking of customers and partners, InspireSemi has a long list of companies it works with, including Lenovo, Penguin Computing, 2CRSI, World Wide Computing, GigaIO, Cadence, and GUC, just to name a few.
“This is an important milestone for our company and an exciting time to bring this versatile accelerated computing solution to market,” said Alex Gray, founder, CTO, and president of InspireSemi. “Thunderbird accelerates many critical applications in key industries that other approaches do not, including life sciences, genomics, medical devices, climate change research, and applications that require deep simulation and modeling. “