6 ms·

I feel like we're just trading one bottleneck for another here. So instead of slow storage, we now have a system that's hyper-sensitive to any interruption and

I feel like we're just trading one bottleneck for another here. So instead of slow storage, we now have a system that's hyper-sensitive to any interruption and probably requires a dedicated power plant to run. Cool experiment, but is this actually a practical path forward or just a dead end with a great headline? Someone convince me I'm wrong...

JumpCrisscross 1y ago
> we're just trading one bottleneck for another If you have two systems with opposite bottlenecks you can build a composite system with the bottlenecks reduced.
- moralestapia 1y ago
  Usually, you get a state with two bottlenecks ...
  - rzzzt 1y ago
    Think of L2 cache (small access time, small capacity) vs. memory modules (larger access time, large capacity) on a motherboard. You get the large capacity and an access time somewhere in between, depending on the hit rate.
    - moralestapia 1y ago
      >depending on the hit rate Yes, a different usage pattern could give you the worse of both worlds. That's a very deep insight from queue theory and similar topics, I don't know if you wrote what you wrote with that in mind (I'm not being an ass, I just want to highlight this observation). A system's behavior is highly dependent, to the point of it being fundamental, on how it is going to be used.
  - DonHopkins 1y ago
    Sounds like you need a massively parallel hardware regexp accelerator (a RePU), so you can have two millions of problems!
    - rzzzt 1y ago
      Hardware accelerated regex engine (HARE): https://ieeexplore.ieee.org/document/7783747 https://ieeexplore.ieee.org/document/7783747
tokyolights2 1y ago
Sandia National Labs is one of the few places in the country (on the planet?) doing blue-sky research. My first thought was similar to yours--If it doesn't have storage, what can I realistically even do with it!? But sometimes you just have to let the academics cook for a few decades and then something fantastical pops out the other end. If we ever make something that is truely AGI, its architecture is probably going to look more like this SpiNNaker machine than anything we are currently using.
- rbanffy 1y ago
  > what can I realistically even do with it!? It doesn't have built-in storage, but that doesn't mean it can't connect to external storage, or that its memory cannot be retrieved from a front-end computer.
mipsISA69 1y ago
This smells like a VC derived sentiment - the only value is from identifying the be all end all solution. There's plenty to learn from endeavors like this, even if this particular approach isn't the one that e.g. achieves AGI.
fc417fc802 1y ago
HPC jobs generally don't stream data to disk in the first place. They write out (huge) snapshots periodically. So mount a network filesystem and be done with it. I don't see the issue.

JumpCrisscross 1y ago

> we're just trading one bottleneck for another If you have two systems with opposite bottlenecks you can build a composite system with the bottlenecks reduced.

moralestapia 1y ago
Usually, you get a state with two bottlenecks ...
- rzzzt 1y ago
  Think of L2 cache (small access time, small capacity) vs. memory modules (larger access time, large capacity) on a motherboard. You get the large capacity and an access time somewhere in between, depending on the hit rate.
  - moralestapia 1y ago
    >depending on the hit rate Yes, a different usage pattern could give you the worse of both worlds. That's a very deep insight from queue theory and similar topics, I don't know if you wrote what you wrote with that in mind (I'm not being an ass, I just want to highlight this observation). A system's behavior is highly dependent, to the point of it being fundamental, on how it is going to be used.
- DonHopkins 1y ago
  Sounds like you need a massively parallel hardware regexp accelerator (a RePU), so you can have two millions of problems!
  - rzzzt 1y ago
    Hardware accelerated regex engine (HARE): https://ieeexplore.ieee.org/document/7783747 https://ieeexplore.ieee.org/document/7783747

tokyolights2 1y ago

Sandia National Labs is one of the few places in the country (on the planet?) doing blue-sky research. My first thought was similar to yours--If it doesn't have storage, what can I realistically even do with it!? But sometimes you just have to let the academics cook for a few decades and then something fantastical pops out the other end. If we ever make something that is truely AGI, its architecture is probably going to look more like this SpiNNaker machine than anything we are currently using.

rbanffy 1y ago
> what can I realistically even do with it!? It doesn't have built-in storage, but that doesn't mean it can't connect to external storage, or that its memory cannot be retrieved from a front-end computer.

mipsISA69 1y ago

This smells like a VC derived sentiment - the only value is from identifying the be all end all solution. There's plenty to learn from endeavors like this, even if this particular approach isn't the one that e.g. achieves AGI.

fc417fc802 1y ago

HPC jobs generally don't stream data to disk in the first place. They write out (huge) snapshots periodically. So mount a network filesystem and be done with it. I don't see the issue.