blastdoor

About

Username
blastdoor
Joined
Visits
337
Last Active
Roles
member
Points
6,913
Badges
1
Posts
3,878

Reactions

5.2KLike168Dislike414Informative

  • Apple's C1 modem is a quiet game-changer that's mostly flying under the radar

    blastdoorblastdoor
    Apple's silicon design team appears to be among the strongest teams in the company, and among the best silicon design teams in the world (if not the best). 

    I wonder if Apple will buy or build their own fab so that they are both designing and manufacturing their chips. I know what the knee jerk reaction to that suggestion will be, but TSMC margins are steadily going up, which means that's profit Apple is missing out on. Time and time again, we have seen Apple identify suppliers with fat profit margins and then take over that business themselves. 

    Maybe a way to start could be a joint venture with TSMC or Intel to build an Apple fab in the US. Apple could finance and own the fab and pay their partner some patent licensing and management/operation fees. Eventually Apple could then take over the management and operation. 
    watto_cobradanox

  • Mac Studio review roundup: Still the fastest on the block

    blastdoorblastdoor
    Oh, and yeah, m4 cores are faster. 
    watto_cobra

  • Mac Studio review roundup: Still the fastest on the block

    blastdoorblastdoor
    Penzi said:

    @Blastdoor I, also, am intrigued by your use case where, I assume, you wire two Mac minis together (Ethernet? Thunderbolt?) and also as to why, for the exact same price, having the exact same number of CPU cores make the mini more interesting than the Studio to you. Straight up that they are the faster M4 cores? I know that in AI neither Ethernet nor Thunderbolt has sufficient bandwidth to make two minis as efficient/competent as one Studio but I’d love to learn more!
    Kids today might call my work “data science,” I’d call it statistics or econometrics. 

    My use for many cores is bootstrapping or Monte Carlo simulations. These things are embarrassingly parallel. They can be distributed across multiple machines with little loss in performance. 

    Apple’s cores are good for this kind of work and surprisingly cost competitive. Plus I just like Macs. 

    It’s a bit more convenient to have a single computer, but multiple can also work. The small size of the mini is a bonus.
    watto_cobra

  • Mac Studio review roundup: Still the fastest on the block

    blastdoorblastdoor
    I was poking around Apple's website and an interesting comparison is the M4 pro Mac mini to the base M3 Ultra Mac Studio. 

    For the same price as the base M3 Ultra Studio (ie, $4000), you can get two $2000 M2 Pro Mac minis with the 14 core CPU, 20 core GPU, 48 GB of RAM, and 512GB SSD (so, a total of 28 cores CPU, 40 cores GPU, 96 GB RAM, and 1 TB SSD). 

    The Studio has the advantage of an extra 20 GPU cores but the disadvantage of being M3 rather than M4. 

    My workloads want CPU more than GPU, so dual minis might actually make more sense for me. 
    watto_cobra

  • Mac Studio review roundup: Still the fastest on the block

    blastdoorblastdoor
    MacPro said:
    Had the latest Studio been an M4 Ultra, I'd have changed up my M2 Ultra,  I don't see the 30% gain is worth the cost (trying to convince myself here).  I really should wait for what comes next in the Ultra range.  M5 or M6?
    When you say 30% gain, are you referring to the GB6 multicore? 

    You might want to pay more attention to GB5 multicore -- it's nearly 50% faster. 

    https://browser.geekbench.com/v5/cpu/compare/23395021?baseline=23393292

    It depends on the nature of your workloads, of course, but the GB6 multicore is more limited in the extent to which it can take advantage or more cores -- ie, Amdahl's Law hits it pretty hard. GB5 is more indicative of how 'embarrassingly parallel' workloads benefit from more cores. 
    watto_cobra