I'm quite skeptical; their baseline is a in-order CPU, instead of a modern out-of-order design, and they're testing a small amount of programs.
Also, I'm not an expert in CPUs, but the underlying idea doesn't seem entirely novel - WaveScalar explored dataflow-based architecture 20+ years ago.
Feels like the visualization is doing a lot of heavy lifting, because the explanation sounds quite mundane (randomly rewriting interactions between two points in a program graph?)
It is also written as if this is a new idea for programming languages, but modelling things as a graph of nodes is like 101 stuff?
Did they independently rediscover 80's style fine-grained Dynamic Dataflow architectures?
(RIP Arvind)
> Moreover, single-core CPU performance has largely stalled since the early 2000s, pushing people towards large multicore CPUs such as Intel Xeon, AMD Zen, and NVIDIA Vera.
Huh? Have you tried the Apple's A + M-series of processors? It has dramatically improved performance and tends to increase single core around 15% per year and it compounds.
The drive towards multi-core architectures is real but it isn't because single core performance has stalled, it is that we want to scale faster than single core performance increases.
These are basic facts that this article is getting wrong. How many more basic errors are in this?
the title is a bit misleading. humans built the checker and picked the workloads. the swarm did the grinding.
Why would you hold off evaluation criteria? This isn’t LLM training.
The frostbite tiles and their work-stealing is reminiscent of a 2d-grid of basic blocks from compiler design. Register renaming or something similar to it also seems to help building intuition about this.
My prediction:
In the AI/AGI/ASI era, the 4 most valuable things will be:
- Model
- AI hardware
- Manufacturers
- Materials
You can already see it beginning at the chip design stage. If AI makes it so that it takes 2 months to design a chip instead of 2 years, then there will way more chips designed to accelerate applications that previously weren't viable economically.
Therefore, it's manufacturers like TSMC that will become more valuable.
And in order to make things, you need materials.
In order to design the chip, you need the model and AI hardware.
Instruction nets sound intriguing, but it looks like execution is based on rewrite rules. Does it need a lot of copying and marshalling to get work done?
Wouldn't a SIMD/SIMT architecture achieve similar speedups? There are many ways to parallelise programs. What am I missing?
Is this a legit company? Looks like it was founded by 2 interns...
In my opinion, the benchmarks chosen look small (e.g. json parsing with 4096 chars, mergesort with 1024 elements, fib(20)).
Is the core RISC-V ISA not too alien to this µarchitecture?
Looks like AI doesn't have to fall for C hubris. Honestly, if you want to make it in this new AI age I recommend you to question every single thing that is considered set in stone, anything that is considered "the natural order of things" in software development. Think hard outside the box.
I don't know why nobody is seeing the problems with C. From my perspective it's not useful to think of C as a portable assembler or a thin layer on "top of the machine", because it makes C machines look eternal and timeless, even though we are deep into the post C machine era of multi core processors with cache coherence.
>Mainstream programming languages, when compiled for general-purpose CPUs, hide the dependencies between the parts of a program. Any piece of code may read or write any variable through a shared, mutable state, and so modern CPU hardware assumes the worst case and rediscovers, at runtime, which instructions actually depend on each other.
To this day I still wonder why people aren't seeing the immediate glaring red flag that the conventional sequential programming model presents. To this day people keep building inherently sequential languages and then wonder why parallelism is so hard. It's hard because you made a fatal assumption: There is only one global memory. How exactly is anyone supposed to reason about such an omnipotent and all powerful concept, where everything can potentially affect anything? If you have two cores, the entire idea of a singular memory screams like a bottleneck. Like, it's crazy to me that anyone would think in 2026 "yup, lets build another sequential language that makes our lives a living hell".
>Moreover, single-core CPU performance has largely stalled since the early 2000s, pushing people towards large multicore CPUs such as Intel Xeon, AMD Zen, and NVIDIA Vera. This forces CPUs to maintain the illusion that every core sees the same memory, and keeping that illusion consistent and performant requires deep cache hierarchies and coherence protocols that cost area and power, and scale poorly.
Correct. Because everything potentially depends on everything, you need to figure out at runtime when things don't depend on each other, which is kind of insane when you think about it.
The assumption of a single global memory is what doomed C. It's the 21th century. Computers don't have global memories anymore, they pretend they do. You can't fix it. No amount of nostalgia towards C is going to fix this mistake.
Also before someone accuses me of e.g. being a C hater, I'll tell you that Zig's IO Backend parameters are the same type of flaw in Rust that probably will never be fixed. Implicit IO is a different kind of global state that Rust inherited and from the 2026 agentic coding perspective it's a pretty egregious flaw because agents don't care about passing in a parameter, meanwhile from the perspective of PL design, using a parameter instead of a full on effects system has this certain charm / beauty to it, how you don't need to special case everything to get the same results. You just use what is already there.
[flagged]
[dead]
Absolutely breathtaking number of claims, and impossible to find the seam between human endeavour and spambot output. Would love a sober overview of this from a suitably competent human, history is littered with promising sounding architectural approaches (anyone remember the Mill?) that went nowhere.
Before even reading anything the word discovered on the title triggered me. It's was not there waiting for someone to find it - like electrostatic attraction/repulsion. It was either designed or "statistically and agentically arrived at it" if my high level mediocre knowledge about how people use AI to research those things is even a little relevant.
Why does it say "discovered"? What is this, a finding? This is an invention.
We will continue to invent Transputer-shaped things forever, it seems. Maybe AI actually makes it tractable/ergonomic though.
I'd love to see them run the experiment "Invent a general-purpose computer that exists entirely within a GPU." see if that turns up any potential.
A new architecture discovered, designed, verified, tested, used, enjoyed, glazed, and eventually forgotten solely by their LLM.
When they actually make a CPU and offer it for use, then we'll talk about their magic LLM-designed CPU.
hardware slopocalypse inc
Discovered? I don't think this means what you think it means
It's not like Intel dug x86 from some mine in Oregon
So far this thread is a little breathless about AI.
This looks cool and all.
But, can someone with CPU experience give an opinion? Does this blow your mind? Or is it trash?
Like how the new math solutions when read by a mathematician seem 'ugly', even if technically correct. Does this look like something actually alien and groundbreaking?
Once again, you are "not allowed" to invest in the frontier.
[dead]
The future is going to be fascinating
If we have enough fabs theres a world where every single problem has a completely custom chip to maximise efficiency of whatever problem needs to be solved. End to end AI - Ask to solve a problem, AI designs the chips, fab delivers it, AI writes the software in a completely custom language. Nobody will understand how anything works other than knowing that it does work
Future = AI + Robots
I am interested to have end products with really good UX driven by peak performance and optimizations which blows my mind because it’s so well perfected as humans weren’t part of its process.
As of now, I don’t see any of that being the goals even let alone the result. Instead what I see is that everyone is in a race to have as much product out as possible by the same standards that applied to products made by humans, but even lesser quality.
I the last two decades we have gotten so used to average or less than average product quality that we are willing to pay 2000$ for a device running an operating system like macOS or windows that ‘works most of the time’.
I hope we have products thanks to AI that compete on reliability of software and hardware to the point that the big tech will have to reinvent their processes just to be able to justify prices.