M2 and beyond: What to expect from the M2 Pro, M2 Max, and M2 Ultra

2»

Comments

  • Reply 21 of 29
    danoxdanox Posts: 3,065member
    Good starting-point and good comments—thanks everyone. There is already a precedent for basic disparities between M1 and the M1Max family — the memory-type difference mentioned above. So the idea that M2Max could use a different process node from M2 is not wildly far-fetched. 

    Daniel Eran Dilger’s main point when Apple Silicon was announced was that he thought it was unlikely Apple would adhere to a rigid release schedule for Macintosh silicon. That Apple would be in a position to shape the silicon to the needs of their hardware releases, instead of the other way around. I think that’s what we’re seeing here. So the Mac Pro awaits its silicon, and when M2Max is ready the new machine will be released. 
    Whatever Apple does they need to release the entire line of Mac’s in a more timely manner, dribbling it out like a old man trying to use the bathroom isn’t good enough. Getting rid of Intel appears to not have helped, is Apple allowing marketing placement decisions get in the way? I hope the problem isn’t technical and I don’t mean supply line problems. 

    The last chip transition took 13 months beginning to end. Right now Apple is ahead, but the competition won’t stay in the rear view mirror forever. They one more shot in the fall.
  • Reply 22 of 29
    I'm sorry, but this article is a serious failure due to ignorance of some of the basic underlying technologies.

    For example, the guesses about memory are completely off base. There is literally no chance at all that they're even close, based on the article's assumptions.

    The M1 has a bandwidth of ~68GB/s because it has a 128-bit memory bus and uses LPDDR4 memory at 4.266GT/s. The M1 Pro has higher bandwidth of ~200GB/s because it uses LPDDR5 memory at 6.4GT/s, and ALSO because it uses a double-wide bus (256 bits).

    The M2 has the same memory bus size (128 bits) as the M1, but it's already using LPDDR5 at 6.4GT/s. If there's an M2 Pro based on the same doubling as the M1 Pro was, it won't get any further benefit from the LPDDR5 (since the M2 already has that). It will have the same ~200GB/s bandwidth as the M1 Pro.

    Of course this all depends on timing - if the M2 Pro came out a year from now, higher-performance LPDDR5 might be common/cheap enough for the M2 Pro to use it, in which case you'd see additional benefits from that. But it DEFINITELY wouldn't get you to 300GB/s. LPDDR5 will never be that fast (that would require 9.6GT/s, which is not happening in the DDR5 timeframe - unless DDR6 is horribly delayed, years from now).

    You're also assuming Apple won't go with HBM, which is not at all a safe assumption. If they do they might well do better than 300GB/s for the "M2 Pro", if such a thing were built.

    Your entire article could have been written something like this:
    M1 Ultra = 2x M1 Max = 4x M1 Pro ~= 6x-8x M1, so expect the same with the M2 series.

    It's a really bad bet though.

    There are much more interesting things to speculate about! What are they doing for an interconnect between CPU cores, GPU cores, Neural Engine, etc? Improvements there are *critical* to better performance - the Pro, Max, and Ultra are great at some things but extremely disappointing at others, and that's mostly down to the interconnect- though software may also play some part in it (especially with the GPU).

    Similarly, the chip-to-chip interconnect for the Ultra is a *huge* advance in the state of the art, unmatched by any other vendor right now... and yet it's not delivering the expected performance in some (many) cases. What have they learned from this, what can they do better, and when will they do it?

    (Edit to add) Most of all, will desktop versions of the M2 run at significantly higher clocks? I speculated about this here when the A15 came out - that core looked a lot like something built to run at higher clocks than earlier Ax cores. I'd like to think that I was right, and that that's been their game all along. But... Apple's performance chart (from the keynote) for the M2, if accurate, suggests that I was wrong and that they don't scale clocks any better than the M1 did. That might still be down to the interconnect, though it seems unlikely. It's also possible that they're holding back on purpose, underestimating performance at the highest clocks, though that too seems unlikely (why would they?).

    For this reason, I suspect that the M2 is a short-lived interim architecture, as someone else already guessed. Though in terms of branding, they may retain the "M2" name even if they improve the cores further for the "M2 Pro" or whatever. That would go against all past behavior, but they don't seem terribly bound by tradition.
    This comment contains many of my thoughts as well!
    if you were to “extrapolate” that’s pretty boring actually. Yeah a linear-scaled guess is barely an article. You could do this for the M4, M5, M10 etc. it’s kind of useless and didn’t even go that far.

    at the same time the article ignores the most exciting questions like mentioned above. What can Apple and TSMC do to improve bandwidth on ArmV9? Fabric stacking? HBM over LPDDR5? 

    Also, at 3nm the next Pro and Max chips are not the linear improvement the article suggests. Why was it written like that? 
    programmer
  • Reply 23 of 29
    lwiolwio Posts: 110member
    This may be awkward in a publicity sense as far as a Mac Pro is concerned.
    If they release it with m1 ultra chips then it looks like it’s using older technology than Apples base laptop although it of course would be screaming past it in processing power. So will they jump to an m2 ultra straight away without the delayed release as with the m1.
  • Reply 24 of 29
    tmaytmay Posts: 6,416member
    danox said:
    Good starting-point and good comments—thanks everyone. There is already a precedent for basic disparities between M1 and the M1Max family — the memory-type difference mentioned above. So the idea that M2Max could use a different process node from M2 is not wildly far-fetched. 

    Daniel Eran Dilger’s main point when Apple Silicon was announced was that he thought it was unlikely Apple would adhere to a rigid release schedule for Macintosh silicon. That Apple would be in a position to shape the silicon to the needs of their hardware releases, instead of the other way around. I think that’s what we’re seeing here. So the Mac Pro awaits its silicon, and when M2Max is ready the new machine will be released. 
    Whatever Apple does they need to release the entire line of Mac’s in a more timely manner, dribbling it out like a old man trying to use the bathroom isn’t good enough. Getting rid of Intel appears to not have helped, is Apple allowing marketing placement decisions get in the way? I hope the problem isn’t technical and I don’t mean supply line problems. 

    The last chip transition took 13 months beginning to end. Right now Apple is ahead, but the competition won’t stay in the rear view mirror forever. They one more shot in the fall.
    Funny, but Apple stated precisely, that they would require two years to deliver the full line, and it appears that Apple will meet its commitment.

    As for the competition, yeah, still in the rear view mirror, except when it comes to kilowatt consuming gaming machines, but Apple's growing market isn't gaming machines. 
    mattinoz
  • Reply 25 of 29
    lwio said:
    This may be awkward in a publicity sense as far as a Mac Pro is concerned.
    If they release it with m1 ultra chips then it looks like it’s using older technology than Apples base laptop although it of course would be screaming past it in processing power. So will they jump to an m2 ultra straight away without the delayed release as with the m1.
    I’ll guess it’s mostly about putting the Mac Pro truly into a class of its own for certain workloads, and the M1 family just didn’t get it done. It was good, even great, but not good enough. Too many compromises, and too close to the Mac Studio’s performance.

    So early on, the M3 Max/Ultra/Heavy (I tend to agree with the person above who argues it will be called M3) was designed for the Mac Pro — it will also appear in the Studio and the flagship MacBook Pro, but there will be no compromises in the M3 Mac Pro.
  • Reply 26 of 29
    mobird said:
    How does the M2 compare to the latest desktop chips from Intel?
    It destroys everything in performance per watt, competes well in raw performance against laptop chips, but gets destroyed by the top end power hungry desktop chips. M2 Ultra will likely compete well against current top end desktop chips, but AMD/NVIDIA/Intel aren’t sitting still either, so by this time next year when M2 Ultra is released, who knows? MacPro may go beyond M2 Ultra, so if they can release a chip based on M2 that’s 4X an M2 Max or double an Ultra, that will likely take the performance crown no matter what (some unoptimized gaming comparisons notwithstanding).
  • Reply 27 of 29
    I'm sorry, but this article is a serious failure due to ignorance of some of the basic underlying technologies.

    For example, the guesses about memory are completely off base. There is literally no chance at all that they're even close, based on the article's assumptions.

    The M1 has a bandwidth of ~68GB/s because it has a 128-bit memory bus and uses LPDDR4 memory at 4.266GT/s. The M1 Pro has higher bandwidth of ~200GB/s because it uses LPDDR5 memory at 6.4GT/s, and ALSO because it uses a double-wide bus (256 bits).

    The M2 has the same memory bus size (128 bits) as the M1, but it's already using LPDDR5 at 6.4GT/s. If there's an M2 Pro based on the same doubling as the M1 Pro was, it won't get any further benefit from the LPDDR5 (since the M2 already has that). It will have the same ~200GB/s bandwidth as the M1 Pro.

    Of course this all depends on timing - if the M2 Pro came out a year from now, higher-performance LPDDR5 might be common/cheap enough for the M2 Pro to use it, in which case you'd see additional benefits from that. But it DEFINITELY wouldn't get you to 300GB/s. LPDDR5 will never be that fast (that would require 9.6GT/s, which is not happening in the DDR5 timeframe - unless DDR6 is horribly delayed, years from now).

    You're also assuming Apple won't go with HBM, which is not at all a safe assumption. If they do they might well do better than 300GB/s for the "M2 Pro", if such a thing were built.

    Your entire article could have been written something like this:
    M1 Ultra = 2x M1 Max = 4x M1 Pro ~= 6x-8x M1, so expect the same with the M2 series.

    It's a really bad bet though.

    There are much more interesting things to speculate about! What are they doing for an interconnect between CPU cores, GPU cores, Neural Engine, etc? Improvements there are *critical* to better performance - the Pro, Max, and Ultra are great at some things but extremely disappointing at others, and that's mostly down to the interconnect- though software may also play some part in it (especially with the GPU).

    Similarly, the chip-to-chip interconnect for the Ultra is a *huge* advance in the state of the art, unmatched by any other vendor right now... and yet it's not delivering the expected performance in some (many) cases. What have they learned from this, what can they do better, and when will they do it?

    (Edit to add) Most of all, will desktop versions of the M2 run at significantly higher clocks? I speculated about this here when the A15 came out - that core looked a lot like something built to run at higher clocks than earlier Ax cores. I'd like to think that I was right, and that that's been their game all along. But... Apple's performance chart (from the keynote) for the M2, if accurate, suggests that I was wrong and that they don't scale clocks any better than the M1 did. That might still be down to the interconnect, though it seems unlikely. It's also possible that they're holding back on purpose, underestimating performance at the highest clocks, though that too seems unlikely (why would they?).

    For this reason, I suspect that the M2 is a short-lived interim architecture, as someone else already guessed. Though in terms of branding, they may retain the "M2" name even if they improve the cores further for the "M2 Pro" or whatever. That would go against all past behavior, but they don't seem terribly bound by tradition.
    This comment contains many of my thoughts as well!
    if you were to “extrapolate” that’s pretty boring actually. Yeah a linear-scaled guess is barely an article. You could do this for the M4, M5, M10 etc. it’s kind of useless and didn’t even go that far.

    at the same time the article ignores the most exciting questions like mentioned above. What can Apple and TSMC do to improve bandwidth on ArmV9? Fabric stacking? HBM over LPDDR5? 

    Also, at 3nm the next Pro and Max chips are not the linear improvement the article suggests. Why was it written like that? 
    I concur the memory bandwidth projection is a blunder, and the article puts a lot of words together to explain what could be explained more concisely, but the speculation that Apple may do things with their desktop class chips to further differentiate them from mobile chips is just that-pure speculation. The near  future is likely a lot more boring, with M2 Max and M2 Ultra chips being largely as described in the article. Apple probably won’t sink that level of silicon engineering investment for such a small market. Taking existing IP developed for iPhones, bumping the clock a little, and adding cores, chiplets, and a couple extra features is unfortunately the name of the game. The next Pro and Max chips will not be 3nm, there is no evidence for that, but there is a chance Apple incorporates 5nm++ and A16 cores in the M2 Pro and Max after the iPhone 14 comes out. Even if that’s the case, A16 cores are likely to bring the smallest performance jump ever, so the general scaling idea won’t be that far off. The only major thing really wide open for speculation is the next MacPro, but that’s going to start at $$$$ and very few people will need/buy it. M2 series will roll out over the next year at least, no M3 series chips until the end of 2023 and that will include 3nm.
  • Reply 28 of 29
    I'm sorry, but this article is a serious failure due to ignorance of some of the basic underlying technologies.

    For example, the guesses about memory are completely off base. There is literally no chance at all that they're even close, based on the article's assumptions.

    The M1 has a bandwidth of ~68GB/s because it has a 128-bit memory bus and uses LPDDR4 memory at 4.266GT/s. The M1 Pro has higher bandwidth of ~200GB/s because it uses LPDDR5 memory at 6.4GT/s, and ALSO because it uses a double-wide bus (256 bits).

    The M2 has the same memory bus size (128 bits) as the M1, but it's already using LPDDR5 at 6.4GT/s. If there's an M2 Pro based on the same doubling as the M1 Pro was, it won't get any further benefit from the LPDDR5 (since the M2 already has that). It will have the same ~200GB/s bandwidth as the M1 Pro.

    Of course this all depends on timing - if the M2 Pro came out a year from now, higher-performance LPDDR5 might be common/cheap enough for the M2 Pro to use it, in which case you'd see additional benefits from that. But it DEFINITELY wouldn't get you to 300GB/s. LPDDR5 will never be that fast (that would require 9.6GT/s, which is not happening in the DDR5 timeframe - unless DDR6 is horribly delayed, years from now).

    You're also assuming Apple won't go with HBM, which is not at all a safe assumption. If they do they might well do better than 300GB/s for the "M2 Pro", if such a thing were built.

    Your entire article could have been written something like this:
    M1 Ultra = 2x M1 Max = 4x M1 Pro ~= 6x-8x M1, so expect the same with the M2 series.

    It's a really bad bet though.

    There are much more interesting things to speculate about! What are they doing for an interconnect between CPU cores, GPU cores, Neural Engine, etc? Improvements there are *critical* to better performance - the Pro, Max, and Ultra are great at some things but extremely disappointing at others, and that's mostly down to the interconnect- though software may also play some part in it (especially with the GPU).

    Similarly, the chip-to-chip interconnect for the Ultra is a *huge* advance in the state of the art, unmatched by any other vendor right now... and yet it's not delivering the expected performance in some (many) cases. What have they learned from this, what can they do better, and when will they do it?

    (Edit to add) Most of all, will desktop versions of the M2 run at significantly higher clocks? I speculated about this here when the A15 came out - that core looked a lot like something built to run at higher clocks than earlier Ax cores. I'd like to think that I was right, and that that's been their game all along. But... Apple's performance chart (from the keynote) for the M2, if accurate, suggests that I was wrong and that they don't scale clocks any better than the M1 did. That might still be down to the interconnect, though it seems unlikely. It's also possible that they're holding back on purpose, underestimating performance at the highest clocks, though that too seems unlikely (why would they?).

    For this reason, I suspect that the M2 is a short-lived interim architecture, as someone else already guessed. Though in terms of branding, they may retain the "M2" name even if they improve the cores further for the "M2 Pro" or whatever. That would go against all past behavior, but they don't seem terribly bound by tradition.
    This comment contains many of my thoughts as well!
    if you were to “extrapolate” that’s pretty boring actually. Yeah a linear-scaled guess is barely an article. You could do this for the M4, M5, M10 etc. it’s kind of useless and didn’t even go that far.

    at the same time the article ignores the most exciting questions like mentioned above. What can Apple and TSMC do to improve bandwidth on ArmV9? Fabric stacking? HBM over LPDDR5? 

    Also, at 3nm the next Pro and Max chips are not the linear improvement the article suggests. Why was it written like that? 
    I concur the memory bandwidth projection is a blunder, and the article puts a lot of words together to explain what could be explained more concisely, but the speculation that Apple may do things with their desktop class chips to further differentiate them from mobile chips is just that-pure speculation. The near  future is likely a lot more boring, with M2 Max and M2 Ultra chips being largely as described in the article. Apple probably won’t sink that level of silicon engineering investment for such a small market. Taking existing IP developed for iPhones, bumping the clock a little, and adding cores, chiplets, and a couple extra features is unfortunately the name of the game. The next Pro and Max chips will not be 3nm, there is no evidence for that, but there is a chance Apple incorporates 5nm++ and A16 cores in the M2 Pro and Max after the iPhone 14 comes out. Even if that’s the case, A16 cores are likely to bring the smallest performance jump ever, so the general scaling idea won’t be that far off. The only major thing really wide open for speculation is the next MacPro, but that’s going to start at $$$$ and very few people will need/buy it. M2 series will roll out over the next year at least, no M3 series chips until the end of 2023 and that will include 3nm.
    You're right that discussion of desktop-class chips is speculation, but so is "likely a lot more boring". If you think anything at all about chip development in the next couple of years is boring, you're really not paying attention. The chiplet revolution is well underway, but there's a LOT more runway yet, as Apple has shown with the interconnect on the Ultra - which is both an amazing achievement and a (probably) mediocre first effort. It's amazing in terms of raw bandwidth, blowing past anything else ever done. On the other hand the end result is not very good for some types of workloads. So there's a LOT more to do there in terms of raw design, and if you think Apple won't put in the effort to get it right, you're wrong. That kind of performance will be *table stakes* for advanced VR. Everyone's going to need it, and it's going to be everywhere, not just in some high-end desktops like Studios and Pros.

    BTW, this explains Apple's insistence on efficiency - if the real end-game is VR everywhere, they're going to need good efficiency for mobile solutions, starting with the rumored headset but not ending there. Whatever we get for the Mac Pro will be a great workstation, but you can count on most of the tech in there being eventually useful in high-powered mobile devices.

    It may be that, as you say, next-gen cores will have very little performance improvement - though I wouldn't count on it! There's still a ton of room for improving the interconnects, which are a huge part of efficiency and overall performance. Apple's shown they can do brute force better than anyone. Now they have to add finesse, and there's *so much* that can be done here.
    tenthousandthingsdewme
  • Reply 29 of 29
    programmerprogrammer Posts: 3,458member
    I agree with the idea that we probably aren't just going to see the M1 chip lineup re-done with the M2.  ARM v9 has been defined for quite a while now (slightly over a year), and it offers the potential for quite a bit more compute scalability.  This would be something well suited to the MacPro.  The new A16 cores could be v9, with M-series variants that have much more powerful vector units.  The MacPro may well use a more advanced memory type -- such techs are expensive, but the high margins on the MacPro could justify its use.  The volume of production needed for the MacPro is also not very high (and the Mac in general, compared to iOS devices), so it should be worth Apple carving out some of the 3nm volume from their phones.
Sign In or Register to comment.