Jim Keller was also a former colleague of mine, as long as we are listing my former colleagues.. (Rick Bergman should count twice, since I also worked with him at Exponential Technology
I meant 2013 for the a7, wires got crossed in my brain.
oh man, I dunno how I made that mistake. The video is from mlid, not rgt. See above about wires getting crossed!
APFS existed since Mojave (2018?) if my memory serves well. APFS did not existe before that.
. My daughter keeps them in a collection of chips I designed. 
I know you worked with Keller. However, as far as I am aware, he's not a source for Tom's video. At this point, you've worked with so many people, I'm assuming that @cmaier is an omnipotent immortal that has stories from your days working with the likes of Tesla, Edison, Galileo, and how you were the real driving force behind trigonometry, in ancient Babylon, a millennia before Pythagorus. You'll also surely complain about Newton being a grump.Jim Keller was also a former colleague of mine, as long as we are listing my former colleagues.
Sierra 10.12.4 had experimental support for it (I believe only by using Terminal commands), and then High Sierra and iOS 10.3 automatically converted devices with SSDs to APFS.
They have such things today, in fact!
Oh, absolutely. I didn't intend on diminishing the hard work that @cmaier and the team at AMD did by using that nomenclature. It just seems that x86-64 is the most common name today among most tech users, but it's entirely an AMD original technology, that Intel had no choice but to adopt. Today, Intel shamelessly refers to it as Intel® 64, because of course they do.
It was amazingly ad-hoc. We had a whole lot of people working on K8, which was going to be a very complicated architecture - I don’t know the details, but at the time I think we thought we had to go blow-for-blow against itanium. I was not involved - I was working on K6 revisions, a whole bunch of them at once.
I absolutely love stories like this, because those of us on the outside really had no idea what was going on. I remember reading an article on the Register by Mike Magee that stated that the first time that Opteron was demoed that Intel executies were "in shock". In a separate article, the Register claimed that Intel execs were "confident and not bothered" by your creation. On the outside, we didn't really know what to think, until of course it was hammer time with the new processors.
My favorite story was when we taped out, there was a small glitch somewhere. The fix was to move a little bit of metal around. We loaded up one of our tools to find the exact coordinates where the metal had to be removed and other metal added, then I loaded up a .def file in vim and manually edited the coordinates in the file, rather than run everything through our tools again, which would have cost another day. I took advantage of the fact that the next tool in the chain was something I wrote, and it was flexible with its .def parsing, so I didn’t have to go back and change many lines of the file to make it work. I don’t think Intel would ever dare something like that 
A simple answer to this is just we have to wait and see and judge by what we have now.I think you missed the forest for the trees here . The "disaggregation" and "building blocks" agenda most likely means that "4-way Hyperthreading" would not show up in the part they were throwing at the competitive to Apple products market. One of the primary point of doing thing in a chunks is that you don't have to use it where you don't need it. So they don't need to entangle as many high end server features into low-mid range consumer laptop products.
#4 above is muddled because that isn't going to be one leap in a single bound. +20% -> +20% -> +20% -> 40% probably is easier to +100% in one jump.
#6 DD5 7400 being "Buck Rogers" fantasy stuff of far off decades ....
Samsung Teases 512 GB DDR5-7200 Modules
www.anandtech.com
Machine Learning Accelerator would be off in a separate tile in the SoC. ( not necessarily going to show up in the combo pointed at by Apple. ML Training Accelerator may not may the same cut as a ML inference one.
Again with the desegregation Intel could bring Xe-HPG tiles on TMSC N3 to the Arrow lake about the same time Apple is moving to N3 . So yeah, the intel P and E cores may be "back" on Intel 7 or 4 but if like Apple removing dGPUs from laptops that would still be a key move to stay competitive. If Intel's GPU tiles work in laptops pretty well at TSMC N3 then most of the "pain points" that Apple's M-series at that point would be against the mobile versions of AMD and Nvidia. Intel's Xe-HPG is already on TSMC N6. Going to N3 in 2 years isn't a moonshot project. That should be pretty straightforward to do.
it "chop off" the iGPU then what is left on the other tile on Intel N4 (or N7) just got smaller and easier to make (yield wise).
Will Intel's 3D SoC package be more expensive to make than Apple's probably monolithic contemporary M-series. Probably. Is Apple going to sell their SoC for razor thin margins less than Intel? Probably not. TMSC just announced price increases. Qualcomm, Nvidia, AMD/Xlinix , Apple, Microsoft/Sony , HiSilcon, Intel , etc. etc. all buying up 100% of the capacity probably is going to keep prices higher 2-3 years out too. [ Likely this isn't going to be a temporary bubble. ]
Intel's probably isn't just with Apple though. There is real threat from Qualcomm and Samsung/AMD-GPU eating into a sizable volume of the Windows laptop market. And also AMD changing their mind for a laptop "almost last" in priority focus. I think Intel is using Apple as more of a visible focal point for the threat in general rather than getting all wound up in that they are being attacked from multiple directions all at once.
Also if Intel's graphics drivers for Xe-HPG have problems then having a N3 Xe-HPG tile isn't going to help them a whole lot.
The disaggregation strategy is in part a way for Intel to attack in 6 different directions all at the same time. (they are still in the "do everything for everybody" business). I don't think they have to technically "beat" Apple in that one specific , relatively narrow niches Apple is going to highly likely confine themselves.
Also on the software front. Windows 10 should be "de-supported" in 2025. Pragmatically, that means 32-bit Windows will be de-supported around at that point too. Having too few registers doesn't help getting to higher IPC. If Microsoft and Intel (and AMD ) also do a rational garbage collection on the grossly bloated SIMD opcodes .... that could be a big contributor to some IPC uplift.
Windows 11 Microsoft is cutting off some "old stuff". That could be precusor to a bigger cut in 2025. Four years for the slackers to get their stuff cleaned up. Freed from having to run code from 1997 ... yeah that could help Intel close the gap on Apple in 5 years or so. Get rid of the boat anchor and can actually start to hit some decent knots cruising speed.
Isn't it kind of a bad sign that they expect to need FOUR threads to fully utilize the resources of a single core? Or am I missing something here?
No, you didn’t miss anything. Unless they plan to add a lot of pipelines to each core, this is just further evidence of the inefficiency of the cores. And if they *are* adding a lot of pipelines, that’s a bad idea.
