Tuesday, December 3rd 2019

AWS Starts Designing 32-Core Arm Neoverse N1 CPU for Data Center

Amazon Web Services, a part of Amazon that is in charge of all things cloud, has announced plans to release 32 core CPU based on Arm Neoverse N1 microarchitecture that is designed to handle a diverse workload that today's cloud infrastructure needs. This new CPU should be the second iteration of AWS'es custom CPU based on the Arm architecture. First-generation AWS CPU was a processor called Graviton, which Amazon offered on-demand in the cloud.

The still-unnamed second-gen CPU will utilize a 7 nm manufacturing process if the Neoverce N1 core at its base is to be believed. Additionally, everything from the Neoverse line should translate to this next-gen CPU as well, meaning that there will be features like high frequency and high single-threaded performance, cache coherency, and interconnect fabric designed to connect special-purpose accelerators to the CPU complex. For reference, Arm's design of Neoverce N1 has a TDP of 105 W for the whole SoC and its packs 64 cores running at 3.1 GHz, delivering amazing power efficiency and high core count.
Source: AnandTech
Add your own comment

10 Comments on AWS Starts Designing 32-Core Arm Neoverse N1 CPU for Data Center

#1
Flanker
It's like everyone wants to design their own ASIC or something?
Posted on Reply
#2
ratirt
Even Arm can do better than Intel now. How about that :)
Posted on Reply
#3
R-T-B
ratirt
Even Arm can do better than Intel now. How about that :)
I mean, there have been ARM cpus past 14nm for literally more than two years now... that does not make them better.
Posted on Reply
#4
ratirt
R-T-B
I mean, there have been ARM cpus past 14nm for literally more than two years now... that does not make them better.
I was referring to core number not node.
Posted on Reply
#5
Vayra86
ratirt
I was referring to core number not node.
Same difference... core count on ARM is an entirely different beast than it is on x86.
Posted on Reply
#6
ratirt
Vayra86
Same difference... core count on ARM is an entirely different beast than it is on x86.
Well AMD made it happen and ARM as well. Intel still struggles and that was my point and irony at the same time.
Posted on Reply
#7
Deathy
ratirt
Well AMD made it happen and ARM as well. Intel still struggles and that was my point and irony at the same time.
Every heard of Xeon Phi? Those are basically tiny Atom cores thrown together with some other stuff. They are around 60+ cores per chip and are avaiable since 2012, with the earliest models being 32 core PCIe cards. Having many cores is not a feat and if it were, Intel would still be early to the party by your standards.
Posted on Reply
#9
kapone32
This makes me wonder what hardware Google Stadia is using.
Posted on Reply
#10
ratirt
Deathy
Every heard of Xeon Phi? Those are basically tiny Atom cores thrown together with some other stuff. They are around 60+ cores per chip and are avaiable since 2012, with the earliest models being 32 core PCIe cards. Having many cores is not a feat and if it were, Intel would still be early to the party by your standards.
No they are not. These are accelerators not processors and shouldn't be treated as such. It was based on an GPU design and uses these applications made for GPUs. It competes against GPGPU not processors. The only thing that is similar to a CPU is that the Xeon Phi cores are compatible with the x86 and can run windows without a processor. But it is not a processor like 9900k for instance or server market Xeons.
Use it instead of your 3700x and see what happens.
R0H1T
A big win for ARM in the enterprise space, especially if AWS can make it work!
I bet ARM will try to reach for the Laptop segment for sure or even desktop at some point. That would be something. Imagine ARM competing with x86 architecture.
Posted on Reply