Solutions
Deterministic + Predictable PerformanceMore Applications, High Utilization, and Low Latency
Thunderbird designed to deliver real world application benefits
-
Software friendly, all CPU architecture (double precision FP64 RISC-V cores) will work with all HPC & AI software
-
High speed, low latency core-to-core communications for predictable performance
-
MIMD architecture (vs. high latency GPU SIMD)
-
Large memory – can address larger problems than fit in many GPUs
-
Distributed memory – each core has its own 64KB local fast memory
Result = Greater application performance with less power consumption
Deterministic + Predictable Performance addresses even more applications GPUs cannot
Example – Thunderbird vs. leading GPU latency
Thunderbird is Deterministic, GPUs are not
Reproducibility of results is a must for many key applications, disqualifying GPUs
-
e.g.- financial trading, cryptography, robotics, self-driving cars, smart weapons, healthcare imaging…
Definition:
-
A deterministic system is one that always produces the same output for a given input
-
There is no randomness or chance involved in the system’s behavior
-
Instead, the system follows predictable rules that determine how the inputs will be transformed into outputs
GPU non-determinism problems
-
GPU’s use very complex hardware/software task schedulers to try to hide their latency problems (switching between tasks while waiting for others)
-
These schedulers are not deterministic, meaning users cannot know when their tasks will finish
-
They’re also proprietary, obscure, and change often, frustrating any attempt to predict their behavior (as with many aspects of GPU architecture)
-
Output results can also vary, due to varying order of sub-task completion and rounding errors
Thunderbird solves this
-
Each core is controlled independently by its own program thread
-
Real-time operating systems provide deterministic software scheduling
-
Easy bare-metal programming gives developers absolute control down to single clock cycles
-
Atomic instruction set provides rich, straightforward task synchronization options
-
Delivering the same results, the same time, every time
MISSION
Setting new standards of performance, energy efficiency, versatility, and ease of programming
InspireSemi provides revolutionary high-performance, energy-efficient accelerated computing solutions for High-Performance Computing (HPC), AI, graph analytics, and other compute-intensive workloads. The Thunderbird “supercomputer-cluster-on-a-chip” is a disruptive, next-generation datacenter accelerator designed to address multiple underserved and diversified industries, including financial services, computer-aided engineering, energy, climate modeling, and life sciences & drug discovery. Based on the open standard RISC-V instruction set architecture, InspireSemi’s solutions leverage an established and thriving software ecosystem.
Download More Information
Solution Overview
Executive Summary
Company Overview Presentation
