Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

IBM analog AI chip could give the Nvidia H100 a run for its money

techradar.com

Post reply on HN
Print view View original

21–30 of 71 posts

Re: IBM analog AI chip could give the Nvidia H100 a run for its money

#21

Post by PeterisP »

JKCalhoun wrote: post #9
ilaksh wrote: post #2

I thinks this type of paradigm shift is what will be needed for the next level of performance for AI. But it also seems like for that to be competitive with current methods, they need to reduce the size by a factor of 2-3 and also make bigger chips. I just feel like IBM has a track record recently of doing great research but not actually manufacturing anything. Kind of like Google.

Just a laymen here, but if the organic brain is something we're trying to model I always ached for some kind of analog processor for things like neural networks. Op-amps, not logic gates. (Quick google found this fun hobbyist article of a perceptron built with op-amps: https://www.nutsvolts.com/magazine/article/the_perceptron_ci... )

We're not really trying to model the organic brain; while some structures are somewhat inspired by some structures in the brain, the current learning structures are fundamentally quite different from biological brains.

While there is a bunch of research for modeling organic brains, IMHO it is primarily driven by neuroscience trying to understand how humans work, and not directly applicable to making computation more efficient.

Re: IBM analog AI chip could give the Nvidia H100 a run for its money

#22

Post by wil421 »

ginko wrote: post #7
pjmlp wrote: post #5

Earlier quoted context omitted.

Usually IBM doesn't sell to the kind of HN crowd, rather people happy with Aix, IBM i and z/OS, DB2, cloud for mainframes, own JVM implementations, Java EE servers, offshoring services,... Some of the research does end in such products, that are outside HN radar, because IBM isn't cool, yet many HN startups won't achieve a legacy like it and still be pumping money like IBM, even for boring business.

>are outside HN radar, because IBM isn't cool More like outside people's radar because IBM only sells to big companies.

I would bet a large portion of software engineers work in Big Companies. FANG employees comment regularly here and they work for the biggest companies in Tech.

Re: IBM analog AI chip could give the Nvidia H100 a run for its money

#24

Post by pjmlp »

magic_man wrote: post #8
pjmlp wrote: post #5

Earlier quoted context omitted.

Usually IBM doesn't sell to the kind of HN crowd, rather people happy with Aix, IBM i and z/OS, DB2, cloud for mainframes, own JVM implementations, Java EE servers, offshoring services,... Some of the research does end in such products, that are outside HN radar, because IBM isn't cool, yet many HN startups won't achieve a legacy like it and still be pumping money like IBM, even for boring business.

IBM revenue has been shrinking over the last 10 years. They aren't selling well.

Many companies would dream of a revenue with that many digits.

Re: IBM analog AI chip could give the Nvidia H100 a run for its money

#25

Post by pjmlp »

wil421 wrote: post #22
ginko wrote: post #7

Earlier quoted context omitted.

>are outside HN radar, because IBM isn't cool More like outside people's radar because IBM only sells to big companies.

I would bet a large portion of software engineers work in Big Companies. FANG employees comment regularly here and they work for the biggest companies in Tech.

Those are the cool ones, the "do no evil" stuff.

Re: IBM analog AI chip could give the Nvidia H100 a run for its money

#26

Post by fidotron »

Gravityloss wrote: post #20
fidotron wrote: post #18

Earlier quoted context omitted.

Arguably that is what the Foveon sensor did. https://en.m.wikipedia.org/wiki/Foveon_X3_sensor There is a book called The Silicon Eye about the history, and not coincidentally Carver Mead also wrote a book on neuromorphic analog neural network implementation back in the 80s.

Hmm all I've seen about Foveon is just stacked RGB sensing. Ie there is no computation involving multiple pixels or anything like that.

Outside of the books I mention the Foveon information online is bad and deteriorating. I mean they don’t seem to mention https://en.m.wikipedia.org/wiki/Misha_Mahowald

From what I recall there are analog feedback stages between adjacent cells, and that was a ongoing theme throughout the work of the teams around Mead. His neuromorphic book is almost entirely about doing that to sound and images.

As an aside the Foveon cameras are worth experiencing. They are amazingly slow, the colours go wrong in less than perfect lighting (the stacked filter) but the edges on objects in resulting images have a definition that you only realize bayer filters completely destroy when they are gone.

Re: IBM analog AI chip could give the Nvidia H100 a run for its money

#27

Post by ginko »

wil421 wrote: post #22
ginko wrote: post #7

Earlier quoted context omitted.

>are outside HN radar, because IBM isn't cool More like outside people's radar because IBM only sells to big companies.

I would bet a large portion of software engineers work in Big Companies. FANG employees comment regularly here and they work for the biggest companies in Tech.

Even in a FANG you can't just walk to your manager and tell them to get a mainframe so you can play around with z/OS.

Re: IBM analog AI chip could give the Nvidia H100 a run for its money

#28

Post by qeternity »

dunefox wrote: post #19
bob1029 wrote: post #16

Earlier quoted context omitted.

Why can't we develop different architectures?

Well, show me one.

You do realize that Transformers were designed around the hardware that exists.

It's not as if we said wow, all you need is attention, and then Nvidia built some GPUs for it.

Re: IBM analog AI chip could give the Nvidia H100 a run for its money

#29

Post by hiddencost »

Rule of thumb: IBM tends to spoof results to get published, then use the published results to trick tech-illiterate clients into using their stack.

At conferences I used to go to, they'd just phack the hyperparameters of an ensemble model on the benchmark sets, and put out a press release saying they were state of the art. They were mostly ignored by the academics who went to the talks on the actual novel work.

Re: IBM analog AI chip could give the Nvidia H100 a run for its money

#30

Post by dleeftink »

PeterisP wrote: post #21
JKCalhoun wrote: post #9

Earlier quoted context omitted.

Just a laymen here, but if the organic brain is something we're trying to model I always ached for some kind of analog processor for things like neural networks. Op-amps, not logic gates. (Quick google found this fun hobbyist article of a perceptron built with op-amps: https://www.nutsvolts.com/magazine/article/the_perceptron_ci... )

We're not really trying to model the organic brain; while some structures are somewhat inspired by some structures in the brain, the current learning structures are fundamentally quite different from biological brains. While there is a bunch of research for modeling organic brains, IMHO it is primarily driven by neuroscience trying to understand how humans work, and not directly applicable to making computation more…

But wouldn't some computations be more efficient it they could run on brain-like synapses/clusters or hardware analogs thereof? Pattern recognition, multi-modal perception and such.
Post reply on HN