IBM says these back-office, network edge Power 10 servers would be sweet for – yes, you guessed it – AI

Not to be left out of the AI infrastructure game, on Tuesday IBM unveiled a pair of tiny Power 10 servers designed to preprocess data at the network edge. The Power S1012 systems are available in both a PC-style tower configuration and a more traditional 2U half-width rack mount chassis. Both can be equipped with IBM's …

[Androgynous Cow Herd]

Hope they put the marketing drones on danger money!

"We still have these same old architectures, built for a completely different workload - but CEO sez we have to say 'AI" ten times in all product discussions from now on"

Tech marketing drone glances at the copy currently displayed on his beige colored monitor...

"What, no more 'Blockchain'?"

[TVU]

Re: Hope they put the marketing drones on danger money!

Indeed, and no amount of trendy AI phrases will keep either IBM or Oracle relevant today. They are late to the party and PR hype now will not help them to catch up with the likes of Microsoft, Google and Amazon.

[Anonymous Coward]

I have an application which is entirely dependent on carrying out lots of large matrix inversions and divisions, in a manner which can’t be parallelised - the output of one slice has to be passed to the next for processing along the length of the model.

It is a little slow as a result. I’d be curious if this hardware genuinely alters performance. The time saving is unlikely to be worth the cost of entry, and probable redevelopment to exploit it.

[Bebu]

lots of large matrix inversions and divisions

I assume the divisions are inversions of the dividing matrix followed by the multiplication of the matrix to be divided (or vice versa.)

I suspect some fairly deep numerical analysis of the inversion of your matrices especially if they have some specific structure. might pay dividends in identifying specific algorithms with the machine architectures that implement them most efficiently.

With such a table you can perform a cost/benefit analysis.

Most vendors would allow you run a testsuite of your intended workloads on their offering as part of your "presales experience."

[Anthidote]

small large language models

This isn't really even a mistake, but stylistically I just don't like the phrase "small large language models"...