4 ms·

Should be fine to compare to an EPYC 72F3 or 73F3 (Zen3, 8/16 cores, 8 channel DDR4-3200 RDIMM (204.8 GB/s theoretical ceiling), $2500/$3500). If memory latenc

by namibj 5y ago

Should be fine to compare to an EPYC 72F3 or 73F3 (Zen3, 8/16 cores, 8 channel DDR4-3200 RDIMM (204.8 GB/s theoretical ceiling), $2500/$3500). If memory latency is that important, one would AFAIK have to compare to Zen3 Threadripper, because EPYC (and afaik Threadripper Pro, the other 8-channel variant of Zen3) is rather locked down in regards to memory "overclock"s. (Note that this is more of a latency/efficiency tradeoff, an any moderate OC is trivial to cool.)

dragontamer 5y ago
> 8 channel DDR4-3200 RDIMM (204.8 GB/s theoretical ceiling) But the M1 Max is 8-channel LPDDR5 400GBps, or literally twice the bandwidth of that EPYC
- rsynnott 5y ago
  Per the article, it can't use more than about 250GB/s for the CPU, though, so it probably isn't an unreasonable comparison.
- gigatexal 5y ago
  Is anyone else finding it hilarious people are comparing these mid-tier consumer chips to server grade EPYCs? :rofl:
  - dragontamer 5y ago
    EPYC almost certainly has more compute power though. Honestly, the memory bandwidth is imbalanced. Its mostly there to support the GPU, but the CPU also gains benefits from it. Its hard enough to push an EPYC to use all 200GBps in practice. EDIT: For workstation tasks however, 64GB is huge for a GPU, while 400GBps is huge for a CPU. Seems like win/win for the CPU and GPU. Its a very intriguing combination. GPU devs usually have to work with much less VRAM, while CPU devs usually have to work with much less bandwidth. 64GB is small for CPU workstation tasks however. Its certainly a strange tradeoff.
    - hajile 5y ago
      The article in question shows that a mere 8 big cores and 2 little ones can use 243gb/s. I'm guessing Apple will go with HBM3/4 before too long due to the lower power consumption and great performance.
      - gigatexal 5y ago
        I thought HBM was a power sucking tech?
        dragontamer 5y ago
        HBM is very low-clock speed and super efficient. HBM's downside is that it requires many, many, many pins. Each channel is 1024-pins of communications (and more pins for power). In practice, the only thing that can make HBM work are substrates. (Typical chips have 4x to 6x HBM stacks, for well over 4096 pins to communicate, plus more pins for power / other purposes) But HBM is among the lowest power technologies available. Turns out that clocking every pin at like 500MHz (while LPDDR5 is probably a 3200 MHz clock) saves a lot on power. Because DRAM has such high latency, the channel speed is more for parallelism more so than anything else. (DDR4 parallelizes RAM into 4-bank groups, each with 4-banks. All 16 can be accessed in parallel across the channel). HBM just does this parallel access thing at a lower clock rate, to save on power. But spends way more pins to do so.
        namibj 5y ago
        Only if you measure by the GB of capacity. Flips around if you measure by the GB/s of bandwidth.
      - namibj 5y ago
        Which is why I suggested comparing to the EPYC 73F3, which is a 5950X clocked at 3.5 - 4 GHz, with 4x the L3$, 4x the memory bandwidth (if you don't overclock it), and 5~6x the IO bandwidth. We know a 5950X is roughly on-par with an M1 Max (at least ignoring the latter's 2 efficiency cores). If the occasional wins of the M1 Max are due to memory bandwidth, this should more-or-less turn the tables.
  - ChuckNorris89 5y ago
    Not if you actually understand the technical constraints vs the business applications of each product in their specific market segments. A Ford F-150 can tow more weight than a Lamborghini despite having less power and being significantly cheaper but each is geared towards a different use case so direct comparisons are just splitting hairs.
    - dragontamer 5y ago
      That's also a good point. EPYC has 2TB or 4TB of DDR4 RAM support. That being said: its amusing to me to see the x86 market move into the "server-like" arguments of the 90s. x86 back then was the "little guy" and all the big server folks talked about how much bigger the DEC Alpha was and how that changes assumptions. It seems like "standard" servers / systems have grown to outstanding sizes. The big question in my mind is if 64GB RAM is large enough? Moana scene (https://www.disneyanimation.com/resources/moana-island-scene/ https://www.disneyanimation.com/resources/moana-island-scene...) is 131 GBs of RAM for example. It literally wouldn't fit on the M1 Max. And that's a 2016-era movie, more modern movies will use even more space. The amount of RAM modern 3d artists need is ridiculous!! Instinctively, I feel like 64GB is enough for power-users, but not in fact, the digital artists who have primarily been the "Pro" level customers of Apple.
      - gigatexal 5y ago
        I am waiting for the photography pros of YouTube to weigh in on that last bit but the Disney bit about 131GB of ram usage is intense. Surely a speedy disk can page but likely not enough to make the 64GB of ram a bottleneck. Maybe things like optane or ssds will get so much quicker that we'll see a further fusion of io down to disks and one day we'll really have a chip that thinks all it's storage is ram/storage and doesn't really distinguish between it. Sure it's unlikely SSDs will get to 400GB/s in speed but if they could get to 10GB/s or more sustained that latency could be handled by smart software probably. I think for that cohort any future iMac Pro or Mac Pro with the future revisions of these chips will surely increase the ram to 128GB maybe even 256GB or more. I am super curious how Apple will tackle those folks who want to put 1TB of ram or more into their systems if they'll do an SOC with some ram plus extra slotted ram as another layer?

by namibj 5y ago

dragontamer 5y ago

> 8 channel DDR4-3200 RDIMM (204.8 GB/s theoretical ceiling) But the M1 Max is 8-channel LPDDR5 400GBps, or literally twice the bandwidth of that EPYC

rsynnott 5y ago
Per the article, it can't use more than about 250GB/s for the CPU, though, so it probably isn't an unreasonable comparison.
gigatexal 5y ago
Is anyone else finding it hilarious people are comparing these mid-tier consumer chips to server grade EPYCs? :rofl:
- dragontamer 5y ago
  EPYC almost certainly has more compute power though. Honestly, the memory bandwidth is imbalanced. Its mostly there to support the GPU, but the CPU also gains benefits from it. Its hard enough to push an EPYC to use all 200GBps in practice. EDIT: For workstation tasks however, 64GB is huge for a GPU, while 400GBps is huge for a CPU. Seems like win/win for the CPU and GPU. Its a very intriguing combination. GPU devs usually have to work with much less VRAM, while CPU devs usually have to work with much less bandwidth. 64GB is small for CPU workstation tasks however. Its certainly a strange tradeoff.
  - hajile 5y ago
    The article in question shows that a mere 8 big cores and 2 little ones can use 243gb/s. I'm guessing Apple will go with HBM3/4 before too long due to the lower power consumption and great performance.
    - gigatexal 5y ago
      I thought HBM was a power sucking tech?
      - dragontamer 5y ago
        HBM is very low-clock speed and super efficient. HBM's downside is that it requires many, many, many pins. Each channel is 1024-pins of communications (and more pins for power). In practice, the only thing that can make HBM work are substrates. (Typical chips have 4x to 6x HBM stacks, for well over 4096 pins to communicate, plus more pins for power / other purposes) But HBM is among the lowest power technologies available. Turns out that clocking every pin at like 500MHz (while LPDDR5 is probably a 3200 MHz clock) saves a lot on power. Because DRAM has such high latency, the channel speed is more for parallelism more so than anything else. (DDR4 parallelizes RAM into 4-bank groups, each with 4-banks. All 16 can be accessed in parallel across the channel). HBM just does this parallel access thing at a lower clock rate, to save on power. But spends way more pins to do so.
      - namibj 5y ago
        Only if you measure by the GB of capacity. Flips around if you measure by the GB/s of bandwidth.
    - namibj 5y ago
      Which is why I suggested comparing to the EPYC 73F3, which is a 5950X clocked at 3.5 - 4 GHz, with 4x the L3$, 4x the memory bandwidth (if you don't overclock it), and 5~6x the IO bandwidth. We know a 5950X is roughly on-par with an M1 Max (at least ignoring the latter's 2 efficiency cores). If the occasional wins of the M1 Max are due to memory bandwidth, this should more-or-less turn the tables.
- ChuckNorris89 5y ago
  Not if you actually understand the technical constraints vs the business applications of each product in their specific market segments. A Ford F-150 can tow more weight than a Lamborghini despite having less power and being significantly cheaper but each is geared towards a different use case so direct comparisons are just splitting hairs.
  - dragontamer 5y ago
    That's also a good point. EPYC has 2TB or 4TB of DDR4 RAM support. That being said: its amusing to me to see the x86 market move into the "server-like" arguments of the 90s. x86 back then was the "little guy" and all the big server folks talked about how much bigger the DEC Alpha was and how that changes assumptions. It seems like "standard" servers / systems have grown to outstanding sizes. The big question in my mind is if 64GB RAM is large enough? Moana scene (https://www.disneyanimation.com/resources/moana-island-scene/ https://www.disneyanimation.com/resources/moana-island-scene...) is 131 GBs of RAM for example. It literally wouldn't fit on the M1 Max. And that's a 2016-era movie, more modern movies will use even more space. The amount of RAM modern 3d artists need is ridiculous!! Instinctively, I feel like 64GB is enough for power-users, but not in fact, the digital artists who have primarily been the "Pro" level customers of Apple.
    - gigatexal 5y ago
      I am waiting for the photography pros of YouTube to weigh in on that last bit but the Disney bit about 131GB of ram usage is intense. Surely a speedy disk can page but likely not enough to make the 64GB of ram a bottleneck. Maybe things like optane or ssds will get so much quicker that we'll see a further fusion of io down to disks and one day we'll really have a chip that thinks all it's storage is ram/storage and doesn't really distinguish between it. Sure it's unlikely SSDs will get to 400GB/s in speed but if they could get to 10GB/s or more sustained that latency could be handled by smart software probably. I think for that cohort any future iMac Pro or Mac Pro with the future revisions of these chips will surely increase the ram to 128GB maybe even 256GB or more. I am super curious how Apple will tackle those folks who want to put 1TB of ram or more into their systems if they'll do an SOC with some ram plus extra slotted ram as another layer?