HAHAHAHAHA Return to office, like that's a solution as opposed to RETURN YOUR BONUS, BUYBACKS, and BUYOUTS. Such hilarity. No wonder you're a laughing stock. Oh wait, that's another wordplay joke. Did you need billions in computing power to figure that out?
To me, the main part is Panther Lake in high volume production on 18 this year. After all, there's never been anything wrong with Intel CPU designs. Sometimes oddities in the chiplet composition of the package, sometimes bios/driver issues. But I'm running on a bunch of Xeon6 ("3" nm) which are just fine in comparison to AMD/TSMC's offerings.
So far, this CEO and his predecessor don't seem to have managed to really clarify the message: HT or not, LPE or not, outsourcing or not, dGPU or not, AI or not. Those are just the high-profile messages that end-users would see. Not even getting into the weeds with things like AVX512-type features or followons, ambivalence about providing a lot of PCIe lanes, how seriously they take the NIC business, the whole 3dxpoint debacle, whether they're actually trying at the very high end (supercomputing/large-AI)...
The workforce things are mainly performative for Wall Street's benefit. At least assuming they're not getting rid of the people who could address the live-or-die features (those messaging topics).
Intel may be too far gone... What concerns me in this letter is the doubling down on x86 and SMT.
This is like trying to teach a (very) old dog new tricks.
The "Core"/Centrino architecture was needed revolution at the time, and to keep x86 alive something similar is needed. SMT, done over 20yeara ago sounds too little.
The old "x86 survivea bc the world has a too big code base" is no longer... Web apps, emulators, and Apple with Rosetta showed you are not architecture dependent
New data center use cases, are re designed from the ground up, where efficiency is key in reducing cost. (Less power = less heat = less everything = much lower cost)
I would be interested in seeing how SMT helps improve performance per watt when competing with Arm based instruction sets
Thanks for passing this along early!
HAHAHAHAHA Return to office, like that's a solution as opposed to RETURN YOUR BONUS, BUYBACKS, and BUYOUTS. Such hilarity. No wonder you're a laughing stock. Oh wait, that's another wordplay joke. Did you need billions in computing power to figure that out?
Weird that Intel-haters find validation here.
To me, the main part is Panther Lake in high volume production on 18 this year. After all, there's never been anything wrong with Intel CPU designs. Sometimes oddities in the chiplet composition of the package, sometimes bios/driver issues. But I'm running on a bunch of Xeon6 ("3" nm) which are just fine in comparison to AMD/TSMC's offerings.
So far, this CEO and his predecessor don't seem to have managed to really clarify the message: HT or not, LPE or not, outsourcing or not, dGPU or not, AI or not. Those are just the high-profile messages that end-users would see. Not even getting into the weeds with things like AVX512-type features or followons, ambivalence about providing a lot of PCIe lanes, how seriously they take the NIC business, the whole 3dxpoint debacle, whether they're actually trying at the very high end (supercomputing/large-AI)...
The workforce things are mainly performative for Wall Street's benefit. At least assuming they're not getting rid of the people who could address the live-or-die features (those messaging topics).
Intel may be too far gone... What concerns me in this letter is the doubling down on x86 and SMT.
This is like trying to teach a (very) old dog new tricks.
The "Core"/Centrino architecture was needed revolution at the time, and to keep x86 alive something similar is needed. SMT, done over 20yeara ago sounds too little.
The old "x86 survivea bc the world has a too big code base" is no longer... Web apps, emulators, and Apple with Rosetta showed you are not architecture dependent
New data center use cases, are re designed from the ground up, where efficiency is key in reducing cost. (Less power = less heat = less everything = much lower cost)
I would be interested in seeing how SMT helps improve performance per watt when competing with Arm based instruction sets
I hope they turn around to profitability soon
🤯