Google Announces Axion, Its First Custom Arm-based Data Center Processor (techcrunch.com) 22
Google Cloud on Tuesday joined AWS and Azure in announcing its first custom-built Arm processor, dubbed Axion. From a report: Based on Arm's Neoverse 2 designs, Google says its Axion instances offer 30% better performance than other Arm-based instances from competitors like AWS and Microsoft and up to 50% better performance and 60% better energy efficiency than comparable X86-based instances. [...] "Technical documentation, including benchmarking and architecture details, will be available later this year," Google spokesperson Amanda Lam said. Maybe the chips aren't even ready yet? After all, it took Google a while to announce Arm-chips in the cloud, especially considering that Google has long built its in-house TPU AI chips and, more recently, custom Arm-based mobile chips for its Pixel phones. AWS launched its Graviton chips back in 2018.
New Improved Death by PowerPoint (Score:2)
Re: (Score:3)
Based on Arm's CEO warning (Score:2)
Re:Based on Arm's CEO warning (Score:5, Insightful)
I would be surprised if higher efficiency was used to maintain compute at reduced energy consumption.
The much more likely result is increased compute at even higher energy consumption. The more efficient things get, the more of it is used, not less. Credits to Mr. Jevon.
Re: (Score:2)
Re: (Score:2)
about future AI data centers which would suck up 20-25% of the entire US power supply by 2030 - a 5x increase from today's 4%. If Google's Axion processor is 60% more efficient, this looks great.
Much of the energy for AI is for training. However, that extrapolation is based on the current wild west of AI. Most of these companies and organizations doing AI training will stop after either realizing that their models are either less than functional or are non-competitive in the marketplace. Furthermore, there is a huge amount of exploration for the fields that might benefit from AI. Those fields will be pruned over time. The combination of pruning both companies and fields will dramatically limit
how much lockin with that custom arm? (Score:3)
how much lockin with that custom arm?
Can it run arm windows?
Can it run any arm Linux?
Can it run full docker in Linux?
Can it run full K8?
Etc?
Re: (Score:2)
It could run "any" ARM Linux but you may have to install manually if it is even supported by the platform. You would have to use Google's loader, but that's not a deal breaker for the question as asked.
If it runs Linux it will support whatever can run on ARM Linux.
It might run ARM Windows, but that's really up to Microsoft.
Relevance (Score:2)
You can't buy one, so why would you care? They are going to use them in their own datacenters.
Re: (Score:2)
You can't buy one, so why would you care? They are going to use them in their own datacenters.
Running the internal Linux distribution that you cannot download, since it is used internally only.
Re: (Score:2)
Does it matter if the chips will be deployed only internally at Google?
How could Google possibly know about Graviton? (Score:2)
How could Google possibly know about Graviton's power consumption to make that kind of claim? Amazon doesn't disclose its numbers to that level of detail.
Re: (Score:2)
Google is also not telling the truth. Conveniently, they ignore the fact that the "first custom Arm-based CPU designed for the datacenter" is actually AWS' Graviton processors way back in 2018.
Re: (Score:2)
You read it wrong. Google is not announcing THE first custom Arm-based CPU. It's announcing ITS first custom Arm-based CPU.
Re: (Score:2)
No, Google's slide explicitly states that it is THE first custom Arm-based CPU.
They are lying or conveniently ignoring the facts.
Re: (Score:2)
That's one slide taken out of the context of a full presentation.
Netcraft confirms it (Score:2)
Intel is dying.
Or, at least, their processor division is.
Re: Netcraft confirms it (Score:2)
Oh no! Why are we US taxpayers giving them $8.5 billion dollars to build new chip fabs?
https://www.aztechcouncil.org/... [aztechcouncil.org]
Re: (Score:1)