Article Complete
AIM
GraphCore releases new 3D chip that speeds AI by 40%

GraphCore releases new 3D chip that speeds AI by 40%

Discover how GraphCore's Bow processor boosts AI by 40%. Click to learn more about its groundbreaking technology and energy efficiency!

GraphCore releases new 3D chip that speeds AI by 40%
poulomi.chatterjee·Published Updated

Key takeaways

  • GraphCore has introduced the Bow processor, the world's first Wafer-on-Wafer chip, enhancing AI processes significantly.
  • The Bow processor operates at a frequency of 1.85 GHz, an improvement over its predecessor's 1.35 GHz.
  • Bow is claimed to accelerate deep learning tasks by 40% while consuming 16% less energy than earlier models.
  • GraphCore's new Bow IPU is compatible with existing software, ensuring seamless integration for current users.
  • The updated IPU-POD systems using Bow are reportedly five times faster than NVIDIA's DGX machines at half the cost.
5 min

UK-based AI computer company GraphCore has announced a new combination chip called Bow, which is the world’s first Wafer-on-Wafer (WoW) processor. GraphCore claims that the processor will speed up processes like deep learning by 40 per cent and use 16 per cent less energy than previous generation processors. GraphCore has partnered closely with TSMC to make the Bow IPU. 

This is the latest version of an IPU or Intelligence Processing Unit from GraphCore. The firm had previously released two versions of the IPU. The Bow processor has a higher frequency of 1.85 GHz versus 1.35 GHz of its previous version, which came out in 2020. GraphCore has stated that its superscale Bow Pod 1024 offers up to 350 PetaFLOPS of AI compute. For users who are already on GraphCore systems, the new Bow IPU uses the same software minus any modifications. 

https://www.youtube.com/watch?v=maBKpM3oa2w

GraphCore also said it will release new, updated models of its multiprocessor computers called IPU-POD, which uses the Bow chip. The company has also stated that these systems are proven to be five times faster than NVIDIA’s DGX machines and cost half. 

TSMC’s wafer-on-wafer technology offers a high density of connections between multiple processors than fusing an individual chip with a wafer. GraphCore’s CTO Simon Knowles said that the process of fusing chips is known as ‘hybrid bonding’ and is similar to cold welding. 

Membership

Stay ahead with AI insights.

Unlimited access to premium articles, exclusive analysis, and the stories shaping India's AI economy.

Subscribe now

AIM Americas

1190 Miraloma Way Suite P, Sunnyvale, CA 94085

AIM India

1st Floor, Sakti Statesman, Bellandur, Bengaluru – 560103

© 2026 AIM. All rights reserved.