If it will deliver and match Apples Mx designs in terms of energy eff and perf, then ARMs expansion on the client side will slow down, I guess.
No point in porting stuff for marginal gains since ISA does not matter as much as ppl think
https://chipsandcheese.com/2021/07/13/arm-or-x86-isa-doesnt-...
Theres also talk about their shift to GPU, maybe Battle Mage will deliver in this year
Considering the stock tanking and the announced layoffs, I fear Battle Mage will be unfortunate collateral damage along the line.
Really unfortunate timing as that was the only competition we could hope for to the Nvidia and AMD GPU duopoly who jacked up their prices, and Intel was actually committed to the new GPUs constantly pushing updates and fixes.
There's no point in rooting for Intel's downfall here as some people seem to take pleasure in, as then AMD will just jack up their prices as they've done in the past, since they're a business not a charity, and only acted nice to consumers while they were the underdog but will not hesitate to milk us dry the moment they own the market like Intel and Nvidia did.
Too risky to go all in on AI now when Nvidia already has a stronghold monopoly that even AMD can't break.
But the "AI bubble"? Are we going to discover that none of the LLM-based applications that are being developed now actually works (despite many having already been working for years, e.g. copilot) AND at the same time that further improvements, both in the LLMs themselves and in the code surrounding them, are impossible- in other words, that LLMs, after four or five years of development at most are already a mature technology? Are we not going to find any new use cases for agents that have almost the skills of specialists, minus some kind of cognitive impairment? Honest question, I don't understand.
Large corps and FAANGS are spending billions on Nvidia HW and on ML training but all those LLMs aren't earning them any money, in fact they're still loosing money just to capture the market share and consumer mindset, but there's doesn't seem to be a profit light at the end of the tunnel.
Once they've bled enough money with no chance of RoI, they'll put a halt on ML expansion and stop buying Nvidia HW. Intel entering at that point would be suicide.
The spend seems to be based on faith that there will be some kind of major breakthrough...but that does appear to be entirely faith based. We don't have any evidence to indicate that it's likely.
I think that rather than being a matter of having or not faith in a future breakthrough, the difference is between those who believe that there's already been a major breakthrough in the last few years and those who deny it.
Because that would eat into the market of hyper expensive dedicated AI accelerator cards. Wasn't it obvious?
At TSMC you'll pay 20 per square inch of silicon(number I made up for exemplification), would you rather a) print a GPU on it and sell it to consumers for $800 or b) print an AI accelerator on it and sell it to datacenters for $8000?
It's really that simple. What company you know that says no to making more money?
Right now everyone's still busy shoving the not-very-efficient N100 into every possible product that exists and making up some new ones too, while half of them end up underperforming from improper cooling.