3 ms·

Is it just me or are we past the stage where being at the absolute apex of chip development is strategically important? For at least a decade, chips have been f

Is it just me or are we past the stage where being at the absolute apex of chip development is strategically important? For at least a decade, chips have been fine and even pretty often overpowered for the vast majority of applications. You want to have access to chips, yes, but who cares if they're five years out of date?

nothercastle 2y ago
Power consumption is a big deal. You can power modem ai chips with very small battery packs. That leaves more room for explosives
- m4rtink 2y ago
  Or longer range with the same warhead. Or better feature detection & terminal targetting, etc.
- andrewflnr 2y ago
  True, but how much has power consumption changed in the last few years? Enough to be a serious obstacle in the face of, say, a determined national defense effort? Serious question, I guess. I don't really know know how power consumption varies across the last few generations of chips, or among the chips you'd plausibly choose between for a warhead or something. But I'm guessing it's not a huge effect.
  - nothercastle 2y ago
    I mean just look at modern cpus and gpus. Power consumption gets cut in half between generations. So it can be a big deal but you are right about diminishing returns at some point it’s low enough.
Jensson 2y ago
> For at least a decade, chips have been fine and even pretty often overpowered for the vast majority of applications True, but that changed the last few years, next gen AI require vast amounts of compute, that isn't strategically important today but could be in a few years.
3eb7988a1663 2y ago
I am of a similar mind. 95%+ of work done today is not limited by compute capacity. The remaining frontiers are already heavily distributed (datacenters for processing weather, spying on citizens, early universe simulations, AI model du jour, etc) so that you can buy a few more racks if you really need to get the process done sooner. I already enable ECO mode (AMD power limitation) because the chip is fast enough without burning extra watts for a marginal performance improvement (which in my testing was incredibly minor relative to the additional draw).
re-thc 2y ago
> You want to have access to chips, yes, but who cares if they're five years out of date? Everyone cares? Faster chips = less cost. If you can run more with the same power then it'll be cheaper. Large businesses run it at scale and so...
- andrewflnr 2y ago
  Ok, fair, "who cares" was too broad. But "older chips would cut into our profit margin" is not the same as "strategically important". Would any of those large businesses go bankrupt if they had to use slightly older chips, or would they grumble and adapt (especially if their competitors were in the same situation)?
  - re-thc 2y ago
    > Would any of those large businesses go bankrupt if they had to use slightly older chips, or would they grumble and adapt (especially if their competitors were in the same situation)? It all depends on the business. If you compare AMD EPYC growth in 5 years, there are huge savings, maybe >200% (in extra cores, less power and faster cores). Is that profit margin not significant? I'm sure a lot of businesses relying on a fleet of servers would struggle. If you're selling phones for example and Samsung/Apple have the latest chips whilst your local company doesn't it will soon matter. Today it might be close, but in 5 years? There's the you don't "need" it as a consumer but there's also a why would a consumer buy a vastly inferior product? So who are you competing against and hence how do you adapt? You assume the competitors are in the same situation, but are they? A lot of companies compete globally.

nothercastle 2y ago

Power consumption is a big deal. You can power modem ai chips with very small battery packs. That leaves more room for explosives

m4rtink 2y ago
Or longer range with the same warhead. Or better feature detection & terminal targetting, etc.
andrewflnr 2y ago
True, but how much has power consumption changed in the last few years? Enough to be a serious obstacle in the face of, say, a determined national defense effort? Serious question, I guess. I don't really know know how power consumption varies across the last few generations of chips, or among the chips you'd plausibly choose between for a warhead or something. But I'm guessing it's not a huge effect.
- nothercastle 2y ago
  I mean just look at modern cpus and gpus. Power consumption gets cut in half between generations. So it can be a big deal but you are right about diminishing returns at some point it’s low enough.

Jensson 2y ago

> For at least a decade, chips have been fine and even pretty often overpowered for the vast majority of applications True, but that changed the last few years, next gen AI require vast amounts of compute, that isn't strategically important today but could be in a few years.

3eb7988a1663 2y ago

I am of a similar mind. 95%+ of work done today is not limited by compute capacity. The remaining frontiers are already heavily distributed (datacenters for processing weather, spying on citizens, early universe simulations, AI model du jour, etc) so that you can buy a few more racks if you really need to get the process done sooner. I already enable ECO mode (AMD power limitation) because the chip is fast enough without burning extra watts for a marginal performance improvement (which in my testing was incredibly minor relative to the additional draw).

re-thc 2y ago

> You want to have access to chips, yes, but who cares if they're five years out of date? Everyone cares? Faster chips = less cost. If you can run more with the same power then it'll be cheaper. Large businesses run it at scale and so...

andrewflnr 2y ago
Ok, fair, "who cares" was too broad. But "older chips would cut into our profit margin" is not the same as "strategically important". Would any of those large businesses go bankrupt if they had to use slightly older chips, or would they grumble and adapt (especially if their competitors were in the same situation)?
- re-thc 2y ago
  > Would any of those large businesses go bankrupt if they had to use slightly older chips, or would they grumble and adapt (especially if their competitors were in the same situation)? It all depends on the business. If you compare AMD EPYC growth in 5 years, there are huge savings, maybe >200% (in extra cores, less power and faster cores). Is that profit margin not significant? I'm sure a lot of businesses relying on a fleet of servers would struggle. If you're selling phones for example and Samsung/Apple have the latest chips whilst your local company doesn't it will soon matter. Today it might be close, but in 5 years? There's the you don't "need" it as a consumer but there's also a why would a consumer buy a vastly inferior product? So who are you competing against and hence how do you adapt? You assume the competitors are in the same situation, but are they? A lot of companies compete globally.