Awesome advertising for the Pinebunds Pro. No chance the Fairbuds can do this? I don't know much about them.
Also, with DOOM running on all these things now, is it still impossible to get it to run well on a 386?
Doom has been ported to an earbud
101–110 of 143 posts
Re: Doom has been ported to an earbud
#102Earlier quoted context omitted.
so 1mm2 peppered by those cores at 300MHz will give you 4 Tflops. And whole 200mm wafer - 100 Petaflops, like 10 B200s, and just at less than $3K/wafer. Giving half area to memory we'll get 50 PFlops with 300Gb RAM. Power draw is like 10-20KW. So, giving these numbers i'd guess Cerebras has tremendous margin and is just printing money :)
Yes, assuming you don't need to connect anything together and that RAM is tinier than it really is, sure. At 28nm, 3megabits/square millimeter is what you get of SRAM, so an entire wafer only gets you ~12 gigabytes of memory. And, of course, most of Cerebras' costs are NRE and the stuff like getting heat out of that wafer and power in.
Why not ddram?
Re: Doom has been ported to an earbud
#103Whenever I see another supposedly menial device including enough general purpose hardware to run Doom, I wonder whether I should think of that as a triumph of software over hardware or an economic failure to build cheaper purpose-built hardware for things like sending audio over a radio.
> Whenever I see another supposedly menial device including enough general purpose hardware The PineBuds are designed and sold as an open firmware platform to allow software experimentation, so there’s nothing bad nor any economic failures going on here. Having a powerful general purpose microcontroller to experiment with is a design goal of the product. That said, ANC Bluetooth earbuds are not menial products. Doing…
The other aspect to consider is changing requirements. Maybe a device capable of transmitting PSTN-level audio quality wirelessly would have been popular twenty years ago, but nowadays most people wouldn't settle for anything with less than 44.1kHz bandwidth. A faster processor means that there's some room for software upgrades later on, future-proofing the device and potentially reducing electronic waste. Unfortunately, that advantage is almost always squandered in practice by planned obsolescence and an industry obsession with locked-down, proprietary firmware.
Re: Doom has been ported to an earbud
#104Earlier quoted context omitted.
Yes, assuming you don't need to connect anything together and that RAM is tinier than it really is, sure. At 28nm, 3megabits/square millimeter is what you get of SRAM, so an entire wafer only gets you ~12 gigabytes of memory. And, of course, most of Cerebras' costs are NRE and the stuff like getting heat out of that wafer and power in.
Why not ddram?
Same reason why Cerebras doesn't use DRAM. The whole point of putting memory close is to increase performance and bandwidth, and DRAM is fundamentally latent.
Also, process that is good at making logic isn't necessarily good for making DRAM. Yes, eDRAM exists, but most designs don't put DRAM on the same die as logic and instead stack it or put it off-chip.
Almost all these microcontrollers that are single-die have flash+SRAM. Almost all microprocessor cache designs are SRAM for these reasons (with some designs using off-die L3 DRAM)-- for these reasons.
Re: Doom has been ported to an earbud
#105Earlier quoted context omitted.
Why not ddram?
Same reason why Cerebras doesn't use DRAM. The whole point of putting memory close is to increase performance and bandwidth, and DRAM is fundamentally latent. Also, process that is good at making logic isn't necessarily good for making DRAM. Yes, eDRAM exists, but most designs don't put DRAM on the same die as logic and instead stack it or put it off-chip. Almost all these microcontrollers that are single-die have fl…
CPU cache is understandably SRAM.
>The whole point of putting memory close is to increase performance and bandwidth, and DRAM is fundamentally latent.
When the access patterns are well established and understood, like in the case of transformers, you can mitigate latency by prefetch (we can even have very beefed up prefetch pipeline knowing that we target transformers), while putting memory on the same chip gives you huge number of data lines thus resulting in huge bandwidth.
Re: Doom has been ported to an earbud
#106Earlier quoted context omitted.
I can sort of see one angle for it, and the parent story kind of supports it. Bad software is a forcing function for good hardware - the worse that software has gotten in the past few decades the better hardware has had to get to support it. Such that if you actually tried like OP did, you can do some pretty crazy things on tiny hardware these days. Imagine what we could do on computers if they weren't so bottlenecke…
That wasn't the GP's claim. Their implication was that it's an economic failure that we don't produce less powerful hardware.
Yeah, that's more or less what I'm getting at.
Re: Doom has been ported to an earbud
#107Whenever I see another supposedly menial device including enough general purpose hardware to run Doom, I wonder whether I should think of that as a triumph of software over hardware or an economic failure to build cheaper purpose-built hardware for things like sending audio over a radio.
Earbuds often have features like mic beam forming and noise cancellation which require a substantial degree of processing power. It's hardly unjustified compared to your Teams instance making fans spin or Home Assistant bringing down an RPi to its knees.
These sorts of things feel like they would be quite inefficient on a general-purpose CPU so you would want to do them on some sort of dedicated DSP hardware instead. So I would expect an earbud to use some sort of specialized microcontroller with a slow-ish CPU core but extra peripherals to do all the signal processing and bluetooth-related stuff.
Re: Doom has been ported to an earbud
#108Earlier quoted context omitted.
Same reason why Cerebras doesn't use DRAM. The whole point of putting memory close is to increase performance and bandwidth, and DRAM is fundamentally latent. Also, process that is good at making logic isn't necessarily good for making DRAM. Yes, eDRAM exists, but most designs don't put DRAM on the same die as logic and instead stack it or put it off-chip. Almost all these microcontrollers that are single-die have fl…
CPU cache is understandably SRAM. >The whole point of putting memory close is to increase performance and bandwidth, and DRAM is fundamentally latent. When the access patterns are well established and understood, like in the case of transformers, you can mitigate latency by prefetch (we can even have very beefed up prefetch pipeline knowing that we target transformers), while putting memory on the same chip gives you…
With embedded SRAM close, you get startling amounts of bandwidth -- Cerebras claims to attain >2 bytes/FLOP in practice -- vs H200 attaining more like 0.001-0.002 to the external DRAM. So we're talking about a 3 order of magnitude difference.
Would it be a little better with on-wafer distributed DRAM and sophisticated prefetch? Sure, but it wouldn't match SRAM, and you'd end up with a lot more interconnect and associated logic. And, of course, there's no clear path to run on a leading logic process and embed DRAM cells.
In turn, you batch for inference on H200, where Cerebras can get full performance with very small batch sizes.
Re: Doom has been ported to an earbud
#109There's gotta be a Moore's law corollary for "Doom ported to [blank]" milestones. I wonder where this all ends? Doom ported to a mechanical pencil! Doom ported to a clipper card! To a lightbulb??
Re: Doom has been ported to an earbud
#110someone made a crude web server out of a vape pen a few weeks ago, we can’t be too far from running doom on one