Intel CEO Pat Gelsinger Retires

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts
  • avis
    Senior Member
    • Dec 2022
    • 2273

    Originally posted by duby229 View Post

    How much fixed function hardware does that product have exactly? Benchmark results and computer performance are unfortunately two very different things. Apple CAN claim that they have optimized fixed function hardware which reduces power and compute loads. What they CANNOT claim is that their compute performance is faster or lower power than anyone else's.

    Fixed function hardware has some pros and cons. The pros include that it can be very compact which in turn can mean lower power consumption or faster performance. But the cons can be severe, like it's fixed and as such it just is what it is. Newer APIs often drop support for older fixed functions. And this is apple here, so yeah, your fixed function hardware isn't likely to be supported past the the next major version. If apple even releases it for your hardware at all. Not to mention it's apple so it's gonna be more expensive than any other brand....

    EDIT: As is always the case, fixed function hardware will always be superceded by programmable hardware as soon as process node and microarchitecture gets to the point that running it in software has more features and runs faster.

    You can cherish your benchmarks and fixed function hardware all you want.
    You OK bro?

    CPUs are based on speculative execution and advanced AI to speed up execution. What you've just said is basically complete nonsense - go watch a couple of videos with Jim Keller and never say this again.

    Fixed function hardware is mainly 1) hardware video encoding/decoding for GPUs and 2) AES encryption/SHA hashing for CPUs. The latter is basically not used at all by modern benchmarks.

    Here, the most advanced Zen 5 CPU with over 155W TDP vs 40W M4 Pro:



    Fixed-function HW, FFS. No AES, no hashing in the benchmark in any shape or form. LMAO.

    Intel's best 285K



    Yeah, at 235W it's faster in MT while having almost twice as many cores. Cringe.

    Maybe you could show benchmarks where Zen 5/ARL beat M4 Pro. "Non-fixed function benchmarks" because you're an expert the world has never noticed.

    Maybe you should join the Flat Earth society. And never get any vaccines. They are evil, lead to autism, heart inflammation and other long-term issues.
    Last edited by avis; 09 December 2024, 11:44 AM.

    Comment

    • fotomar
      Phoronix Member
      • Jun 2024
      • 103

      Originally posted by avis View Post

      And never get any vaccines. They are evil, lead to autism, heart inflammation and other long-term issues.
      Finally the spook accounts are coming around

      Comment

      • GermanyChris
        Junior Member
        • May 2014
        • 21

        FWIW normal people buy integrated systems which means more than which processor in this case is faster. I've been buying Apples and Mac's since the mid 90's some have been faster than the competition and some slower but I like the OS and the community. I think integrated system is what's being ignored here in our collective effort to build up or tear down Apple.

        Comment

        • mSparks
          Senior Member
          • Oct 2007
          • 2110

          Originally posted by duby229 View Post

          EDIT: As is always the case, fixed function hardware will always be superceded by programmable hardware as soon as process node and microarchitecture gets to the point that running it in software has more features and runs faster.

          You can cherish your benchmarks and fixed function hardware all you want.
          Apple silicons advantage is its entirely different instruction architecture, specifically its 8 word wide instruction decoder iirc. Thats not "fixed function" its just an order of magnitude faster for the same power consumption at the expense of zero backward compatibility (hence rosetta).

          it may very well end up being the architecture that "officially" kills off X86 (4 word decoders) - and winblows "by association". I suspect if apple and nvidia were on friendlier terms it already would have.
          Last edited by mSparks; 09 December 2024, 09:14 PM.

          Comment

          • duby229
            Senior Member
            • Nov 2007
            • 7783

            Originally posted by mSparks View Post

            Apple silicons advantage is its entirely different instruction architecture, specifically its 8 word wide instruction decoder iirc. Thats not "fixed function" its just an order of magnitude faster for the same power consumption at the expense of zero backward compatibility (hence rosetta).

            it may very well end up being the architecture that "officially" kills off X86 (4 word decoders) - and winblows "by association". I suspect if apple and nvidia were on friendlier terms it already would have.
            I hope AMD agrees with me, but the actual truth is that a CMT architecture could scale easily to 8 wide frontend on just 2 threads per module, or even 16 on 4. If an 8 wide CMT microarchitecture was produced on a new TSMC node that is.....

            It's just wishful thinking on my part. But make no mistake that x86 is most definitely capable of scaling to anything as wide as can possibly be fabricated.

            EDIT: How far x86 can scale depends entirely on what a particular process node is capable of. Bulldozer and it's derivatives all only had a 4 wide frontend and AMD made the mistake of incorrectly defining what a core is, imo the core -is- the module as a whole. If counting the module as the core, then all Bulldozer derivatives actually had only half the cores advertised. Fucked up but true. They were all stuck at 32-28nm for over a decade and that hurt them the most because it limited them to their 2 wide integer units. A modern CMT microarchitecture on a new TSMC node with an 8 wide frontend and 2 4 wide integer pipelines would be truly awesome, it would make Ryzen look like a baby.

            EDIT: I just wanted to mention that also cache coherency is really important for a CMT architecture and the old SOI process AMD used to use had relatively poor cache performance. But not so for the new TSMC nodes, cache performance seems amazing on Ryzen. I really think a modern take on CMT is due.
            Last edited by duby229; 10 December 2024, 03:36 AM.

            Comment

            • mSparks
              Senior Member
              • Oct 2007
              • 2110

              Originally posted by duby229 View Post

              I hope AMD agrees with me, but the actual truth is that a CMT architecture could scale easily to 8 wide frontend on just 2 threads per module, or even 16 on 4. If an 8 wide CMT microarchitecture was produced on a new TSMC node that is.....
              I'm sure the architecture could, the issue is Winblows cannot. Windows, at its heart is an 8 bit, single core operating system (wchar, NTFS.... "nuff said") with everything after that "hacked on", they had to be dragged into the 64bit era kicking and screaming and that was a disaster for years.

              Apple had the extreme advantage of being oriented around POSIX standards, that meant everything compiled for apple hardware was already coded to basically recompile with little or no changes for the new architecture (heck, even recompiling apple stuff for Linux and vice versa generally just needs OS folder and library references changing), for windows to do the same the entire operating system and every bit of software written for it will basically need rewriting from scratch, and that is never going to happen.

              It took silicon manufacturing reaching the point that Apple could make their own and free themselves from those chains, now that genie is out of the bottle there is no going back. Windows devices are already in a very distant last place. Windows 11 will not save them.
              Last edited by mSparks; 10 December 2024, 09:30 AM.

              Comment

              • duby229
                Senior Member
                • Nov 2007
                • 7783

                Originally posted by mSparks View Post

                I'm sure the architecture could, the issue is Winblows cannot. Windows, at its heart is an 8 bit, single core operating system (wchar, NTFS.... "nuff said") with everything after that "hacked on", they had to be dragged into the 64bit era kicking and screaming and that was a disaster for years.

                Apple had the extreme advantage of being oriented around POSIX standards, that meant everything compiled for apple hardware was already coded to basically recompile with little or no changes for the new architecture (heck, even recompiling apple stuff for Linux and vice versa generally just needs OS folder and library references changing), for windows to do the same the entire operating system and every bit of software written for it will basically need rewriting from scratch, and that is never going to happen.

                It took silicon manufacturing reaching the point that Apple could make their own and free themselves from those chains, now that genie is out of the bottle there is no going back. Windows devices are already in a very distant last place. Windows 11 will not save them.
                I think I mostly agree with you about Windows. But there is no alternative really. You mention NTFS is a problem for windows, but Linux has BTRFS as a problem and that is a way bigger problem. NTFS might not eat your data, but BTRFS most definitely will. If you are using it right now, then you have silent corruption right now. It's seriously bad. I don't think the ecosystem Linux works in is capable of producing a stable filesystem.

                If your assessment is accurate then Apple is untoppable. And it has NOTHING to do with processors or architecture or process nodes at all.

                EDIT: If your assessment is accurate then it means Apple is on top ONLY because MS and Linux aren't capable of producing competitive software.
                Last edited by duby229; 10 December 2024, 11:09 AM.

                Comment

                • mSparks
                  Senior Member
                  • Oct 2007
                  • 2110

                  Originally posted by duby229 View Post

                  I think I mostly agree with you about Windows. But there is no alternative really.
                  There wasn't, there is now, and it is an order of magnitude better than you can get from windows.

                  Enjoy the videos and music you love, upload original content, and share it all with friends, family, and the world on YouTube.


                  windows is basically word, excel and powerpoint users at this point (#1 gaming device right now is the -Linux based- Meta quest 3S - get one, they are AMAZING). Right now there is no alternative to apple for that. at least double the battery life for a more responsive experience.

                  Originally posted by duby229 View Post
                  If your assessment is accurate then Apple is untoppable. And it has NOTHING to do with processors or architecture or process nodes at all.

                  EDIT: If your assessment is accurate then it means Apple is on top ONLY because MS and Linux aren't capable of producing competitive software.​
                  It's not about the "capability" - its about the cost of porting to a new architecture. Apple and Linux are based on industry agreed standards (POSIX), that result in the cost of porting to any architecture/OS basically zero. Microsoft invented all its own standards and the OS and windows specific applications cannot escape X86 performance/power without completely rewriting everything from scratch - windows is now facing the same challenges on desktop that it faced when mobile devices became decent computers, now non X86 chips are almost universally outperforming X86 chips at the same price point I just don't see an out for them.

                  Comment

                  • fotomar
                    Phoronix Member
                    • Jun 2024
                    • 103

                    Originally posted by duby229 View Post
                    NTFS might not eat your data, but BTRFS most definitely will. If you are using it right now, then you have silent corruption right now.
                    [citation needed]

                    Comment

                    • duby229
                      Senior Member
                      • Nov 2007
                      • 7783

                      Originally posted by mSparks View Post
                      There wasn't, there is now, and it is an order of magnitude better than you can get from windows.

                      Enjoy the videos and music you love, upload original content, and share it all with friends, family, and the world on YouTube.


                      windows is basically word, excel and powerpoint users at this point (#1 gaming device right now is the -Linux based- Meta quest 3S - get one, they are AMAZING). Right now there is no alternative to apple for that. at least double the battery life for a more responsive experience.



                      It's not about the "capability" - its about the cost of porting to a new architecture. Apple and Linux are based on industry agreed standards (POSIX), that result in the cost of porting to any architecture/OS basically zero. Microsoft invented all its own standards and the OS and windows specific applications cannot escape X86 performance/power without completely rewriting everything from scratch - windows is now facing the same challenges on desktop that it faced when mobile devices became decent computers, now non X86 chips are almost universally outperforming X86 chips at the same price point I just don't see an out for them.
                      I disagree with you premise that x86 is being universally outperformed. There may be some metrics you can point out that look better, but there is no way you can say x86 is universally outperformed. That's not true at all.

                      The bottom line is that Apples SOCs have tons of fixed function hardware that only work on their APIs. Linux on their hardware -will not- acheive the same performance. Like I said Apple can claim that they have optimized fixed function hardware that lowers power consumption and increases performance but they cannot say their compute performance is faster or lower power. It isn't true at all.

                      This is why macrobenchmarks do not actually represent computer performance.
                      Last edited by duby229; 10 December 2024, 12:27 PM.

                      Comment

                      Working...
                      X