vacancies advertise contact news tip The Vault
facebook rss twitter

Arm launches Neoverse N2 and V1 platforms for data centres

by Mark Tyson on 28 April 2021, 11:11

Tags: ARM

Quick Link: HEXUS.net/qaeqi3

Add to My Vault: x

Chip designer Arm has outlined a pair of new products which will be important in deepening its penetration into the data centre market. The firm added the new Neoverse N2 and Neoverse V1 platforms to its roadmap last September, and yesterday it fleshed out the new features and the main selling points of the new platforms in a special presentation.

As you might expect, the Neoverse N2 and Neoverse V1 platforms take different approaches to data centre computing, broadly forking to cover the two most important application areas for such processors.

Arm's Neoverse N2 is the first of its data centre processors to use the new Armv9 architecture. The highlight is that the N2 offers 40 per cent higher 1T performance (compared with the N1) at the same power/efficiency levels. Arm forecasts the Neoverse N2 being popular in edge applications to the cloud.

The Arm Neoverse V1 is the bigger change, the longer evolutionary step from the N1, hence the new 'V' product branch. The biggest change is the creation of a wide and deep architecture with the addition of scalable vector extensions (SVE). Arm says the V1 delivers "a massive 50 per cent uplift, 1.8x improvement for a range of vector workloads and 4x improvement for machine learning workloads over N1". This 'revolutionary' chip will be targeting HPC and and exascale computing server markets.

Arm couldn't release 'platforms' without a system IP so has also introduced (Coherent Mesh Network) CMN-700, which it described as "a key element for constructing high-performance Neoverse V1 and Neoverse N2-based SoCs". From the diagram you can see that CMN-700 enables next-generation use cases for multi-chip, memory expansions and accelerators. It supports up to 4x the number of cores as its predecessor (256 cores per socket), chiplet architectures, plus heterogeneous chiplet designs.



HEXUS Forums :: 2 Comments

Login with Forum Account

Don't have an account? Register today!
So they went with what Apple did with their M1 chip, just they are targeting server market?
darcotech
So they went with what Apple did with their M1 chip, just they are targeting server market?

Not really, ARM makes the IP so people can build SoCs around them. As at the bottom of the article, this is a set of IPs for people to build Neoverse based chips / systems. Apple took the ARM IP and made their own chip with it, just like Qualcomm does with the Snapdragon and how Samsung did with the Exynos. Also, the M1 didn't come out that long ago, ARM couldn't reverse engineer Apple's product and make their own, non-patent violating version, in such a short time.

Apple is using ARM's tech, not the other way round. Apple have done something very interesting with the M1 and their tight control of their software ecosystem means they have a far bigger chance of making it work than Microsoft's ARM stuff (although they did this a loooong time ago with Windows 2000 Mobile on the Jornada 720, couldn't make it work properly then and can't do it now because the appeal of Windows is the software availability). I'm very interested to see how this stuff works out. But it's not a miracle of Apple's genius. It's an application of ARM's IP and you can see the development path has been over several years in the iPads. They are right to ditch Intel, as everyone else is moving forward without them. Using ARM made a lot of sense.