> on Apple Silicon, a WebAssembly module's linear memory can be shared directly with the GPU: no copies, no serialization, no intermediate buffers
enhance
> no copies, no serialization, no intermediate buffers
would it kill people to write their own stuff why are we doing this. out of all the things people immediately cede to AI they cede their human ability to communicate and convey/share ideas. this timeline is bonkers.
Huh, I’m 100% going to interview this way the next time I have to hire an engineer. I can’t think of a better way to get a sense of how a candidate reasons about things, and of their values - do they have a sense of responsibility, conscientiousness, team fit.
All other things that could be LLM-mediated have no more signal.
> I can’t think of a better way to get a sense of how a candidate reasons about things
Some ideas to help you: ask the candidate something underspecified and watch what they do first. Do they ask clarifying questions, make their assumptions explicit? After they answer ask what would change their mind, where does that break down? Pick a topic they know and ask them to explain it to a smart non-engineer. Make them estimate something they can’t look up (forces them to decompose, bound, and calibrate). Once they’ve proposed a solution to a question, change the constraints to see if they can adapt or whether they’re stuck.
What you want to evaluate is dynamic reasoning, adaptability.
Well, there is a long tradition of "testing" developer candidates by asking them to exhibit skills in tasks that they never, ever, do in their work. Like whiteboard coding.
It doesn't have a great success record.
I personally would rather they exhibited expert skills in using tools, and expressing their design insight as a part of that skillset.
I’ve become overly sensitive to it as well because it’s such a reliable indicator that there are other problems in the work.
I’ve wasted so much time looking at interesting repos this year before discovering that one of the main claims was a hallucination, or that when I got to the specific part of the codebase it just had a big note from the LLM that’s it’s a placeholder until it can figure out how to do the requested thing.
The people who have AI write their articles don’t care if it works or if it’s correct. They’re trying to get jobs and want something quick and interesting that will appeal to a lazy hiring manager. We’re just taking the bait too.
> The people who have AI write their articles don’t care if it works or if it’s correct.
I'd build on this: The people who have AI write their articles very likely don't know how their thing works or is correct. High chance they'll stumble when they are expected to speak about whatever it is they are presenting with some authority and demonstration of knowledge. Human to human, not being able to do that = obliterates trust. Places it somewhere near the realm of misinformation, which everyone unilaterally has no interest in consuming.
Good luck to people who want to fluff expertise and present as more-capable for job prospects, the world is shit and I know there's more people who need income than there are jobs that provide for our basic human needs, but this level of AI crutching is just going to bode poorly for those who think this is going to get them where they need to go.
I don't know, to me your sentiment sounds a lot like how back in the day they used to say "you can't just use a calculator all the time, use your brain and show the work on pen and paper".
humans have been using tools to communicate since pre-history. language itself is one tool of communication invented to supersede body-language and grunting and noises. the thought and idea is theirs, it was communicated. Would it be that much different if they used a spellchecker extensively to edit their work?
I get why you're annoyed but is it really such a big deal? random people aren't to blame for whatever other annoyances "AI slop" has created.
> language itself is one tool of communication invented to supersede body-language and grunting and noises
That's a pretty utilitarian view of language. How would it feel if everyone spoke and wrote like a PR representative? This is what an article written by an LLM is starting to sound like.
I'm even willing to argue that the way in which you convey your ideas is as important as the idea itself. Like we could all be eating soylent for our daily nutritional requirements but we don't. The taste of the food we eat is important. It's the same with writing for me
> the thought and idea is theirs, it was communicated
Are they? I don't know how much they used AI, the entire article could be written from a one sentence prompt and so I'd argue that the thoughts and ideas are not their own.
This isn't like using a spell checker, it's like using a ghost writer.
Calculators have never been the medium in which we communicate our human experience and knowledge transfer. Calculators aren't part of the social fabric or culture. Very 2d extrapolation that somehow resulted in an alleged parallel. Language is woven deeply into civilization and our histories & been a part of our species literal survival against the most unforgiving odds/environments. Using what is effectively a ghost writer nukes trust. You cannot ascertain anything about the person behind the blog if it's clear they used AI to write it. And without that there's no way to infer expertise, rule out hallucinations, falsehoods presented as matter of fact, and the whole broad set of things LLM's get wrong because of their limitations as a technology. I have literally nothing to go off of that would prove this person knows what they are talking about. Why would anyone want to consume that?
Would it kill anyone at all to add a preamble that is forthcoming about using AI to write something? A chance to say these are my ideas and I've used claude to help me state it eloquently because <english is not my first language / i dont write well / claude said it better than i ever could> etc ? Not doing that, presenting as more capable/knowing than one probably is, is what destroys trust immediately the moment it's sniffed out that AI was used to write something.
It's irresponsible, a self-nerf, and it's annoying. Venn diagram there is basically a circle. We're all familiar with how vibe coding appears to weaken your ability to write code, like skipping the gym and expecting good muscle density. All I'm saying is people shouldn't be skipping the gym for literally communicating with each other because there's gonna be a lot of times in life where you're not gonna be able to whip out chat jippity to continue a real conversation with another person. Ceding that turf means you're willingly trading your ability to deal with real life scenarios with other human beings for short term gain. It's funny how the universe tends to find balance. Yeah, being well read and expressing ideas well is a skill, it takes work.
It's not that deep man, it's just a blog post about some software library. There's no civilisational communication going on here, relax. This whole thing will become irrelevant in a few decades before the end of our lifespans. It's just never that deep.
Why does it matter if it's their thought or not. If you currently care about GPU inference from webassembly on apple silicon, you can use this article. That's really about it.
Now if you care about GPU inference from wasm on apple silicon, and you found problems with this articles content, then great, comment about it. If you say that the problem with the content is due to the usual surface level slop LLMs belt out, then great complain about LLMs. But your comment didn't say anything about gpu inference from wasm on apple silicon.
Apple Silicon uses unified memory where the CPU and GPU use the exact same memory and no copies from RAM to VRAM are needed. The article opens with mentioning just that and indeed it is the whole point of the article.
I am always a bit baffled why Apple gets credited with this. Unified memory has been a thing for decades. I can still load the biggest models on my 10th gen Intel Core CPU and the integrated GPU can run inference.
The difference being that modern integrated GPU are just that much faster and can run inference at tolerable speeds.
(Plus NPUs being a thing now, but that also started much earlier. Thr 10th gen Intel Core architecture already had instructions to deal with "AI" workloads... just very preliminary)
That’s shared, not unified, it’s partitioned where cpu and gpu copies are managed by driver. Lunar lake (2024) is getting closer but still not as tightly integrated as apple and capped to 32GB only (Apple has up to 512GB). AMD ryzen ai max is closer to Apple but still 3 times slower memory.
Shared vs unified is merely a driver implementation detail. Regardless, in practice (IIUC) data is still going to be copied if you perform a transfer using a graphics API because the driver has no way of knowing what the host might do with the pointed-to memory after the transfer.
If you make use of host pointers and run on an iGPU no copy will take place.
My last serious GPU programming was with OpenCL. And if my memory does not fail me the API was quite specific about copying and/or sharing memory on a shared memory system.
I am pretty sure that my old 10th gen CPU/GPU combo has the ability to use the "unified"/zero-copy access mode for the GPU.
I don't think people are crediting Apple with inventing unified memory - I certainly did not. There have been similar systems for decades. What Apple did is popularize this with widely available hardware with GPUs that don't totally suck for inference in combination with RAM that has decent speed at an affordable price. You either had iGPUs which were slow (plus not exactly the fastest DDR memory) but at least sitting on the same die or you had fast dGPUs which had their own limited amount of VRAM. So the choice was between direct memory access but not powerfull or powerfull but strangled by having to go through the PCIE subsystem to access RAM.
The article is talking about one particular optimization that one can implement with Apple Silicon and I at least wasn't aware that it is now possible to do so from WebAssembly - so to completely dismiss it as if it had nothing to do with Apple Silicon is imho not fair.
> Apple Silicon changes the physics. The CPU and GPU share the same physical memory (Apple's Unified Memory Architecture) ... no bus!
Beware the reality distortion field: This is of course how it's worked on most x86 machines for a long time. And also on most Macs when they were using Intel chips.
Why did all my x86 onboard iGPU reserve a fixed amount of RAM on boot, inaccessible to the OS? Why do dGPU bring their own VRAM and how to directly manipulate it from the CPU without copying?
To the first question: blame Windows I guess. But even on older chips, GPU code could access memory allocated on the CPU side so this didn't cap the amount of data your GPGPU code could crunch.
I remember this was mostly a BIOS setting how much memory to allocate for iGPU - and once set in the BIOS, that memory was not accessible to the underlying OS (besides GPU I/O).
Correct me if I'm wrong, but that reserved memory is for the framebuffer? The iBoot bootloader also reserves some memory for the framebuffer.
dGPUs bring their own VRAM because it's a different type of memory, allowing them to get higher performance than they could with DDR. The M4 Max requires 128GB of LPDDR5X to reach its ~500GB/s bandwidth. The RX Vega 64 had that same bandwidth in 2017 with just 8GB of HBM2.
Nope, the reserved memory is what's available to use from the various APIs (VK, GL, etc). More recently there's OS support for flexible on demand allocation by the GPU driver.
Of course the APIs have allowed you to make direct use of pointers to CPU memory for something like a decade. However that requires maintaining two separate code paths because doing so while running on a dGPU is _extremely_ expensive.
As someone that's worked on GPU drivers for shared memory systems for over 15 years, supporting hardware that was put on the market over 20 years ago, and they've "always" (in my experience) been able to dynamically assign memory pages to the GPU.
The "reserved" memory is more about the guaranteed minimum to allow the thing to actually light up, and sometimes specific hardware blocks had more limited requirements (e.g. the display block might require contiguous physical addresses, or the MMU data/page tables themselves) so we would reserve a chunk to ensure they can actually be allocated with those requirements. But they tended to be a small proportion of the total "GPU Memory used".
Sure, sharing the virtual address space is less well supported, but the total amount of memory the GPU can use is flexible at runtime.
Agree, maybe "changes the physics" was too strong, shared cpu/gpu memory is not new.
What is different then is the combination of
1. UMA memory (and yes, iGPU had this, pre-M1)
2. enough bandwidth / GPU throughput for local inference
3. straightforward `makeBuffer(bytesNoCopy:)` path
So, the novelty isn't the shared memory itself, but the whole chain lining up to make the Wasm linear memory -> Metal-buffer approach practical + performant enough.
(and not saying there's some Apple Silicon magic here either ... it'd work anywhere there was UMA and no-copy host-pointer path)
On one side it sounds promising to exploit shared memory properties to speed up inference. But on the other hand, the well established inference engines are perhaps already well optimized to overlap compute and communication efficiently. In this case the host-device copies are likely not a problem to tackle.
loading third party agents in a sandbox with full custom model support. right now you need to either run that code directly (super dangerous) use a vm/container (slow and complicated) or a interpreter like lua (language bound, slow and weak security). wasm is perfect for this, its almost native speed, built for security and language neutral. onnx and coreml are secure but they can only do the actual model not all the code around it.
wmf | a day ago
thrill | a day ago
m00dy | a day ago
junon | 23 hours ago
koolala | 7 hours ago
trueno | a day ago
enhance
> no copies, no serialization, no intermediate buffers
would it kill people to write their own stuff why are we doing this. out of all the things people immediately cede to AI they cede their human ability to communicate and convey/share ideas. this timeline is bonkers.
rvz | a day ago
There will be a time where it will be problematic for those who over-rely on AI and will struggle on on-site interviews with whiteboard tests.
bensyverson | a day ago
m00dy | a day ago
dylan604 | 21 hours ago
JSR_FDED | 23 hours ago
All other things that could be LLM-mediated have no more signal.
andsoitis | 22 hours ago
Some ideas to help you: ask the candidate something underspecified and watch what they do first. Do they ask clarifying questions, make their assumptions explicit? After they answer ask what would change their mind, where does that break down? Pick a topic they know and ask them to explain it to a smart non-engineer. Make them estimate something they can’t look up (forces them to decompose, bound, and calibrate). Once they’ve proposed a solution to a question, change the constraints to see if they can adapt or whether they’re stuck.
What you want to evaluate is dynamic reasoning, adaptability.
z0r | 22 hours ago
jhayward | 12 hours ago
It doesn't have a great success record.
I personally would rather they exhibited expert skills in using tools, and expressing their design insight as a part of that skillset.
Aurornis | 23 hours ago
I’ve wasted so much time looking at interesting repos this year before discovering that one of the main claims was a hallucination, or that when I got to the specific part of the codebase it just had a big note from the LLM that’s it’s a placeholder until it can figure out how to do the requested thing.
The people who have AI write their articles don’t care if it works or if it’s correct. They’re trying to get jobs and want something quick and interesting that will appeal to a lazy hiring manager. We’re just taking the bait too.
trueno | 20 hours ago
I'd build on this: The people who have AI write their articles very likely don't know how their thing works or is correct. High chance they'll stumble when they are expected to speak about whatever it is they are presenting with some authority and demonstration of knowledge. Human to human, not being able to do that = obliterates trust. Places it somewhere near the realm of misinformation, which everyone unilaterally has no interest in consuming.
Good luck to people who want to fluff expertise and present as more-capable for job prospects, the world is shit and I know there's more people who need income than there are jobs that provide for our basic human needs, but this level of AI crutching is just going to bode poorly for those who think this is going to get them where they need to go.
notepad0x90 | 22 hours ago
humans have been using tools to communicate since pre-history. language itself is one tool of communication invented to supersede body-language and grunting and noises. the thought and idea is theirs, it was communicated. Would it be that much different if they used a spellchecker extensively to edit their work?
I get why you're annoyed but is it really such a big deal? random people aren't to blame for whatever other annoyances "AI slop" has created.
rdedev | 21 hours ago
That's a pretty utilitarian view of language. How would it feel if everyone spoke and wrote like a PR representative? This is what an article written by an LLM is starting to sound like.
I'm even willing to argue that the way in which you convey your ideas is as important as the idea itself. Like we could all be eating soylent for our daily nutritional requirements but we don't. The taste of the food we eat is important. It's the same with writing for me
ben-schaaf | 21 hours ago
Are they? I don't know how much they used AI, the entire article could be written from a one sentence prompt and so I'd argue that the thoughts and ideas are not their own.
This isn't like using a spell checker, it's like using a ghost writer.
trueno | 20 hours ago
Would it kill anyone at all to add a preamble that is forthcoming about using AI to write something? A chance to say these are my ideas and I've used claude to help me state it eloquently because <english is not my first language / i dont write well / claude said it better than i ever could> etc ? Not doing that, presenting as more capable/knowing than one probably is, is what destroys trust immediately the moment it's sniffed out that AI was used to write something.
It's irresponsible, a self-nerf, and it's annoying. Venn diagram there is basically a circle. We're all familiar with how vibe coding appears to weaken your ability to write code, like skipping the gym and expecting good muscle density. All I'm saying is people shouldn't be skipping the gym for literally communicating with each other because there's gonna be a lot of times in life where you're not gonna be able to whip out chat jippity to continue a real conversation with another person. Ceding that turf means you're willingly trading your ability to deal with real life scenarios with other human beings for short term gain. It's funny how the universe tends to find balance. Yeah, being well read and expressing ideas well is a skill, it takes work.
porridgeraisin | 17 hours ago
Why does it matter if it's their thought or not. If you currently care about GPU inference from webassembly on apple silicon, you can use this article. That's really about it.
Now if you care about GPU inference from wasm on apple silicon, and you found problems with this articles content, then great, comment about it. If you say that the problem with the content is due to the usual surface level slop LLMs belt out, then great complain about LLMs. But your comment didn't say anything about gpu inference from wasm on apple silicon.
saagarjha | 23 hours ago
jsomedon | 21 hours ago
swiftcoder | 16 hours ago
saagarjha | 15 hours ago
swiftcoder | 10 hours ago
[OP] agambrahma | an hour ago
The value would be in actor processes, where you can delegate inference without paying the 'copy tax' for crossing the sandbox boundary.
So, less "inference engine" and more "Tmux for AI agents"
Think pausing, moving, resuming, swapping model backend.
I scoped the post to memory architecture, since it was the least obvious part ... will follow up with one about the actor model aspect.
saagarjha | an hour ago
pjmlp | 20 hours ago
Also, these folks should be amazed by 8 and 16 bit games development, or games consoles in general.
nl | 18 hours ago
The whole Apple Silicon thing is (in this case) just added details that don't actually matter.
[1] https://github.com/WebAssembly/memory-control/blob/main/prop...
eis | 17 hours ago
fho | 15 hours ago
The difference being that modern integrated GPU are just that much faster and can run inference at tolerable speeds.
(Plus NPUs being a thing now, but that also started much earlier. Thr 10th gen Intel Core architecture already had instructions to deal with "AI" workloads... just very preliminary)
mirekrusin | 15 hours ago
fc417fc802 | 14 hours ago
If you make use of host pointers and run on an iGPU no copy will take place.
fho | 14 hours ago
I am pretty sure that my old 10th gen CPU/GPU combo has the ability to use the "unified"/zero-copy access mode for the GPU.
eis | 11 hours ago
The article is talking about one particular optimization that one can implement with Apple Silicon and I at least wasn't aware that it is now possible to do so from WebAssembly - so to completely dismiss it as if it had nothing to do with Apple Silicon is imho not fair.
pjmlp | 10 hours ago
And yes things like the Amiga Blitter, arcade or console graphics units were already baby GPUs.
nl | 2 hours ago
That's the same no matter the physical memory system architecture.
fulafel | 18 hours ago
Beware the reality distortion field: This is of course how it's worked on most x86 machines for a long time. And also on most Macs when they were using Intel chips.
littlecranky67 | 18 hours ago
fulafel | 18 hours ago
littlecranky67 | 7 hours ago
fulafel | 5 hours ago
ben-schaaf | 15 hours ago
dGPUs bring their own VRAM because it's a different type of memory, allowing them to get higher performance than they could with DDR. The M4 Max requires 128GB of LPDDR5X to reach its ~500GB/s bandwidth. The RX Vega 64 had that same bandwidth in 2017 with just 8GB of HBM2.
fc417fc802 | 14 hours ago
Of course the APIs have allowed you to make direct use of pointers to CPU memory for something like a decade. However that requires maintaining two separate code paths because doing so while running on a dGPU is _extremely_ expensive.
kimixa | 7 hours ago
The "reserved" memory is more about the guaranteed minimum to allow the thing to actually light up, and sometimes specific hardware blocks had more limited requirements (e.g. the display block might require contiguous physical addresses, or the MMU data/page tables themselves) so we would reserve a chunk to ensure they can actually be allocated with those requirements. But they tended to be a small proportion of the total "GPU Memory used".
Sure, sharing the virtual address space is less well supported, but the total amount of memory the GPU can use is flexible at runtime.
[OP] agambrahma | an hour ago
What is different then is the combination of
1. UMA memory (and yes, iGPU had this, pre-M1) 2. enough bandwidth / GPU throughput for local inference 3. straightforward `makeBuffer(bytesNoCopy:)` path
So, the novelty isn't the shared memory itself, but the whole chain lining up to make the Wasm linear memory -> Metal-buffer approach practical + performant enough.
(and not saying there's some Apple Silicon magic here either ... it'd work anywhere there was UMA and no-copy host-pointer path)
itamos | 15 hours ago
jedisct1 | 10 hours ago
tancop | 6 hours ago
[OP] agambrahma | an hour ago
It's less about browsers, and more about server/edge/local-agent runtimes.
Wasm lets you have
- sandboxing (untrusted actor code)
- clean snapshot/restore
- portability of actor across machines
If you don’t need those properties, then yes ... native is obviously the better choice