> Studies 1-5 showed that people are disproportionately likely to live in places whose names resemble their own first or last names (e.g., people named Louis are disproportionately likely to live in St. Louis).
When I lived in Austin, it seemed like a third of boys born were being named Austin. I presume many of them will end up living there as adults but not because of this particular bias, because they were raised there and have family’s there seems to be a more likely driver.
Seems more likely this falls under the replication crisis umbrella. My wife's favorite numbers are my birthday (mm-dd), which is a small reason she fell in love with me. Neither of those numbers are related to her birthday. My favorite number(s) do not overlap with my birthday. Maybe my mm-dd values just aren't low enough, like 02-02?
The name of this CPU is bordering on securities fraud. When people see the term "AGI" now, they are assuming "Artificial General Intelligence", not "Agentic AI Infrastructure".
Of course people don't realize that, and people will buy ARM stock thinking they've cracked AGI. The people running Arm absolutely know this, so this name is what we in the industry call a "lie".
This is just a Neoverse CPU that Arm will manufacture themselves at TSMC and then sell directly to customers.
It isn't an "AI" CPU. There is nothing AI about it. There is nothing about it that makes it more AI than Graviton, Epyc, Xeon, etc.
This was already revealed in the Qualcomm vs Arm lawsuit a few years ago. Qualcomm accused Arm of planning to sell their CPUs directly instead of just licensing. Arm's CEO at the time denied it. Qualcomm ends up being right.
Meta are heavily invested in building their own chips with ARM to reduce their reliance on Nvidia as everyone is going after their (Nvidia) data center revenues.
This is why Meta acquired a chip startup for this reason [0] months ago.
Call this an “AGI CPU” just feels like the most out of touch, terrible marketing possible. Maybe this is unfair but it makes me think ARM as a whole is incompetent just because it is so tasteless.
> Arm has additionally partnered with Supermicro on a liquid-cooled 200kW design capable of housing 336 Arm AGI CPUs for over 45,000 cores.
Also just bad timing on trying to brag about a partnership with Supermicro, after a founder was just indicted on charges of smuggling Nvidia GPUs. Just bizarre to mention them at all.
How fun would it be if due to improved chips handling more model state RAM needs are reduced and Sama cannot make all those RAM purchases he booked?
VC without a degree who has no grasp of hardware engineering failed up when all he had to do was noodle numbers in an Excel sheet.
He is so far behind the hardware scene he thinks its sitting still and RAM requirements will be a nice linear path to AGI. Not if new chips optimized for model streaming crater RAM needs.
Hilarious how last decades software geniuses are being revealed as incompetent finance engineers whose success was all due to ZIRP offering endless runway.
For those wanting to know more about software stack,
> Arm is actively collaborating with leading Linux distributions from Canonical, Red Hat, and SUSE to ensure certified support for the production systems.
If I try to cut through the hype, it seems the main features of this processor, or rather processor + memory controller + system architecture, is < 100 ns for accessing anything in system memory and 6 GB/sec for each of a large-ish number of cores, so a (much?) higher overall bandwidth than what we would see in a comparable Intel x86_64 machine.
It only took a quarter century, but I'm glad that somebody is finally adding a little multicore competition since Moore's law began failing in the mid-2000s.
I looked around a bit, and the going rate appears to be about $10,000 per 64 cores, or around $150 per core. Here is an Intel Xeon Platinum 8592+ 64 Core Processor with 61 billion transistors:
So that's about 500 million transistors per dollar, or 1 billion transistors for $2.
It looks like Arm's 136 core Neoverse V3 has between 150 and 200 billion transistors, so it should cost around $400. Each blade has 2 of those chips, so should be around $800-1000 for compute. It doesn't say how much memory the blades come with, but that's a secondary concern.
Note that this is way too many cores for 1 bus, since by Amdahl's law, more than about 4-8 cores per bus typically results in the remaining cores getting wasted. Real-world performance will be bandwidth-limited, so I would expect a blade to perform about the same as a 16-64 core computer. But that depends on mesh topology, so maybe I'm wrong (AI thinks I might be):
Intel Xeon Scalable: Switched from a Ring to a Mesh Architecture starting with Skylake-SP to handle higher core counts.
Arm Neoverse V3 / AGI: Uses the Arm CMN-700 (Coherent Mesh Network), which is a high-bandwidth 2D mesh designed specifically to link over 100 cores and multiple memory controllers.
I find all of this to be somewhat exhausting. We're long overdue for modular transputers. I'm envisioning small boards with 4-16 cores between 1-4 GHz and 1-16 GB of memory approaching $100 or less with economies of scale. They would be stackable horizontally and vertically, to easily create clusters with as many cores as one desires. The cluster could appear to the user as an array of separate computers, a single multicore computer running in a unified address space, or various custom configurations. Then libraries could provide APIs to run existing 3D, AI, tensor and similar SIMD code, since it's trivial to run SIMD on MIMD but very challenging to run MIMD on SIMD. This is similar to how we often see Lisp runtimes written in C/C++, but never C/C++ runtimes written in Lisp.
It would have been unthinkable to design such a thing even a year ago, but with the arrival of AI, that seems straightforward, even pedestrian. If this design ever manifests, I do wonder how hard it would be to get into a fab. It's a chicken and egg problem, because people can't imagine a world that isn't compute-bound, just like they couldn't imagine a world after the arrival of AI.
Edit: https://news.ycombinator.com/item?id=47506641 has Arm AGI specs. Looks like it has DDR5-8800 (12x DDR5 channels) so that's just under 12 cores per bus, which actually aligns well with Amdahl's law. Maybe Arm is building the transputer I always wanted. I just wish prices were an order of magnitude lower so that we could actually play around with this stuff.
Poor TSMC (and ASML)! They were already struggling with capacity to fulfill orders from their established customers. With ARM now joining the party, I don't know how they're going to cope.
Edit: The new CPU will be built with the soon-to-be-former leading edge process of 3nm lithography.
Many of these words are unexplained. "Memory and I/O on the same die". Oh? What does this mean? All of the DRAM in the photo/render is still on sticks. Do they mean the memory controller? Or is there an embedded DRAM component?
What is “agentic AI cloud era” referring to? I honestly don’t know what this buzz-speak is targeting. Running models locally on the server, for cloud workloads? Agentic, that is just a LLM pattern.
Hmm all my experience with using AI has been mostly VRAM. I haven't experienced any bottleneck on the CPU side. What does this chip offer over Intel or Apple Silicon? Anyone expert here know whatit is?
is this a cpu that's meant for AI training or is it more for serving inference? I don't quite get why I would want to buy an arm CPU over a nvidia GPU for ai applications.
313 comments
My realtor's last name is House
> Studies 1-5 showed that people are disproportionately likely to live in places whose names resemble their own first or last names (e.g., people named Louis are disproportionately likely to live in St. Louis).
When I lived in Austin, it seemed like a third of boys born were being named Austin. I presume many of them will end up living there as adults but not because of this particular bias, because they were raised there and have family’s there seems to be a more likely driver.
> Studies 1-5 showed that people are disproportionately likely to live in places whose names resemble their own first or last names
There are several cities in the US that share my last name. I don't live near any of them.
> Study 6 extended this finding to birthday number preferences.
D'oh!
https://publishing.rcseng.ac.uk/doi/10.1308/147363515X141345...
The TDP to memory bandwidth& capacity ratio form these blades is in a class of its own, yes?
I no longer believe this is like the dotcom. Now it feels like the 1983 video game crash.
So sad.
Of course people don't realize that, and people will buy ARM stock thinking they've cracked AGI. The people running Arm absolutely know this, so this name is what we in the industry call a "lie".
It isn't an "AI" CPU. There is nothing AI about it. There is nothing about it that makes it more AI than Graviton, Epyc, Xeon, etc.
This was already revealed in the Qualcomm vs Arm lawsuit a few years ago. Qualcomm accused Arm of planning to sell their CPUs directly instead of just licensing. Arm's CEO at the time denied it. Qualcomm ends up being right.
I wrote a post here on why Arm is doing this and why now: https://news.ycombinator.com/item?id=47032932
For the first time in our more than 35-year history, Arm is delivering its own silicon products
In case you were thinking about some other abbreviation...
This is why Meta acquired a chip startup for this reason [0] months ago.
[0] https://www.reuters.com/business/meta-buy-chip-startup-rivos...
> Arm has additionally partnered with Supermicro on a liquid-cooled 200kW design capable of housing 336 Arm AGI CPUs for over 45,000 cores.
Also just bad timing on trying to brag about a partnership with Supermicro, after a founder was just indicted on charges of smuggling Nvidia GPUs. Just bizarre to mention them at all.
Yesterday everything was Agentic.
Everything was AI last week.
Waiting for AGI Agentic AI Crypto toilet paper to be in the supermarket shelves , next to the superseded Object oriented UML Rational Rose tuna.
VC without a degree who has no grasp of hardware engineering failed up when all he had to do was noodle numbers in an Excel sheet.
He is so far behind the hardware scene he thinks its sitting still and RAM requirements will be a nice linear path to AGI. Not if new chips optimized for model streaming crater RAM needs.
Hilarious how last decades software geniuses are being revealed as incompetent finance engineers whose success was all due to ZIRP offering endless runway.
> Arm is actively collaborating with leading Linux distributions from Canonical, Red Hat, and SUSE to ensure certified support for the production systems.
Taken from
https://developer.arm.com/community/arm-community-blogs/b/se...
That's...not much right? Maybe it's a lot times N-cores? But I really hope each individual core isn't limited to that.
Edit: 17 minutes to sum RAM?
Am I right or am I misunderstanding?
Seeing "Arm AGI" spelled out on a page with an "arm" logo looks slightly cheesy.
But maybe it's actually a good fit for the societal revolution driven by AGI, comparable to the one driven by the DOT.com RevoLut.Ion. (dot com).
Anyways, it sounds like an A.R.M. branded version of the AppleSilicon revolution?
But maybe that's just my shallow categorization.
I looked around a bit, and the going rate appears to be about $10,000 per 64 cores, or around $150 per core. Here is an Intel Xeon Platinum 8592+ 64 Core Processor with 61 billion transistors:
https://www.itcreations.com/product/144410
So that's about 500 million transistors per dollar, or 1 billion transistors for $2.
It looks like Arm's 136 core Neoverse V3 has between 150 and 200 billion transistors, so it should cost around $400. Each blade has 2 of those chips, so should be around $800-1000 for compute. It doesn't say how much memory the blades come with, but that's a secondary concern.
Note that this is way too many cores for 1 bus, since by Amdahl's law, more than about 4-8 cores per bus typically results in the remaining cores getting wasted. Real-world performance will be bandwidth-limited, so I would expect a blade to perform about the same as a 16-64 core computer. But that depends on mesh topology, so maybe I'm wrong (AI thinks I might be):
I find all of this to be somewhat exhausting. We're long overdue for modular transputers. I'm envisioning small boards with 4-16 cores between 1-4 GHz and 1-16 GB of memory approaching $100 or less with economies of scale. They would be stackable horizontally and vertically, to easily create clusters with as many cores as one desires. The cluster could appear to the user as an array of separate computers, a single multicore computer running in a unified address space, or various custom configurations. Then libraries could provide APIs to run existing 3D, AI, tensor and similar SIMD code, since it's trivial to run SIMD on MIMD but very challenging to run MIMD on SIMD. This is similar to how we often see Lisp runtimes written in C/C++, but never C/C++ runtimes written in Lisp.It would have been unthinkable to design such a thing even a year ago, but with the arrival of AI, that seems straightforward, even pedestrian. If this design ever manifests, I do wonder how hard it would be to get into a fab. It's a chicken and egg problem, because people can't imagine a world that isn't compute-bound, just like they couldn't imagine a world after the arrival of AI.
Edit: https://news.ycombinator.com/item?id=47506641 has Arm AGI specs. Looks like it has DDR5-8800 (12x DDR5 channels) so that's just under 12 cores per bus, which actually aligns well with Amdahl's law. Maybe Arm is building the transputer I always wanted. I just wish prices were an order of magnitude lower so that we could actually play around with this stuff.
Edit: The new CPU will be built with the soon-to-be-former leading edge process of 3nm lithography.
idk ... so ARM itself is now jumping on the AGI *) "train"... !?
how will they do this!? during system-bootup / splash-screens during execution of whatever we will run on these SOCs ... ;)
*) ads generated income
bruhahaha ... ;^)
just my 0.02€
So we will see AI Toilet Paper launching in the next months.