Latency Processing Unit (LPU)


Latency-optimized and highly scalable architecture that accelerates hyperscale models for Generative AI (e.g., GPT-3, LLaMA),

a multi-billion parameter, small-batch,

and memory-intensive workload

Streamlined memory access (SMA) and streamlined execution engine (SXE) are designed to maximize the memory bandwidth usage by continuously streaming and processing the data with minimal interference for high-speed inference

Expandable

Synchronization Link (ESL)


Innovative peer-to-peer communication technology that performs low-latency synchronization to achieve high scalability when supporting hyperscale models with multi-devices

Expandable Synchronization Link (ESL) is a 

lightweight full-duplex  peer-to-peer communication technology that performs data synchronization with low latency and latency hiding to achieve 1.8x speedup when doubling the number of devices

HyperDex Framework


A new paradigm of compiler technology

that bridges datacenter applications,

hyperscale models, and LPU-based hardware

via high-level easy-to-use API

and internal optimization tools 

to ultimately enable the transition

from narrow AI to general AI

HyperAccel Orion


LPU-based datacenter server that outperforms state-of-the art DGX A100 in text generation workloads, such as ChatGPT,

in terms of performance(>30%), cost-effectiveness (>2x), and power efficiency (>2x), with a superior accelerator scalability


At HyperAccel, we create hyper-accelerated hardware solutions for emerging AI applicaitons.

Our team consists of experienced researchers and engineers with high expertise and amazing character to be at the forefront of providing innovative and practical AI hardware.


We Are Hiring!

Join Our HyperAcceleration


We are looking for brilliant and passionate people to join us and play a major role

in building the next big thing in data analytics.

If you enjoy working on cutting edge technologies, solving complex problems,

have a team spirit, grit and a

can-do-attitude – your place is with us!

At HyperAccel, we create hyper-accelerated hardware solutions for emerging AI applications.

Our team consists of experienced research scientists and engineers with high expertise and amazing character

to be at the forefront of providing innovative and practical AI hardware.

Address

B-3, 119, 160 Dongtanyeok-ro

Hwaseong-si, Gyeonggi-do, 

South Korea

E - mail

contact@hyperaccel.ai


© Copyright 2023 HyperAccel. All Rights Reserved.

Address

B3, 119, 160 Dongtanyeok-ro

Hwaseong-si, Gyeonggi-do, 

South Korea 

E - mail

contact@hyperaccel.ai


© Copyright 2023 HyperAccel (하이퍼엑셀). All Rights Reserved.