M2 Pro Mac mini vs Mac Pro - compared

Comments

• Reply 1 of 11

  sdw2001 Posts: 18,016member

  January 25, 2023 1:31PM

  So the top Mac mini can beat the base Mac Pro in many ways.  I have to say that is amazing.  It shows what an absolute monster an M2 (or M3?) Mac Pro will be. 

  
• Reply 2 of 11

  keithw Posts: 141member

  January 25, 2023 2:27PM

  The 2023(?) ASi "Mac Pro" must either be able to reach the 166,946 GB5 GPU results either with on-chip GPU cores or by a discrete graphics card like the existing Intel Mac Pro, otherwise, why bother to even release it?
  

  
• Reply 3 of 11

  mjtomlin Posts: 2,673member

  January 25, 2023 3:07PM

  keithw said:

  The 2023(?) ASi "Mac Pro" must either be able to reach the 166,946 GB5 GPU results either with on-chip GPU cores or by a discrete graphics card like the existing Intel Mac Pro, otherwise, why bother to even release it?
  

  
  I think this is why we haven't seen the new Mac Pro yet. The new GPU design for the A16 was supposed to see a huge performance increase (>50%, mainly due to implementing hardware based ray tracing), but it had to be pulled because apparently it wasn't meeting efficiency standards*. So If I had to guess, the M3 is going to skip the A16 and be based on the A17 generation of cores, so we should see a fairly substantial performance increase in the M3 and finally get the ASi based Mac Pro, which will be the first system with M3 generation SoCs with M3 Ultra and M3 Extreme.  (Both the A17 and M3 will also use TSMC's N3 process bringing further performance and efficiency enhancements.)
  
  *This is the issue that's going to have be addressed at some point in the future... trying to develop a single core for both mobile and desktop applications. More than likely, that new GPU would've been fine in a desktop system where thermal ceilings can be lifted with active cooling systems. I think Apple will eventually start "optimizing" actual CPU and GPU (more so) cores for their intended systems.
  

  edited January 2023
• Reply 4 of 11

  DAalseth Posts: 2,783member

  January 25, 2023 3:12PM

  A couple of decades ago I had a string of Mac Pros. Over time as they got more and more powerful, I started moving down the line. Today I’m doing graphics and such that are head and shoulders beyond what I was doing then. But right now I’m doing it on a nine year old iMac and am looking at a Mac Mini to replace it.
  
  What I do hasn’t gotten any less demanding, if anything the level is much higher. But the computers are so much more capable today that even the intro, bottom end Mac can easily take it and run. It’s really quite amazing.

  
• Reply 5 of 11

  rob53 Posts: 3,251member

  January 25, 2023 3:17PM

  keithw said:

  The 2023(?) ASi "Mac Pro" must either be able to reach the 166,946 GB5 GPU results either with on-chip GPU cores or by a discrete graphics card like the existing Intel Mac Pro, otherwise, why bother to even release it?
  

  Agree. The question is how Apple will be able to do this. I added some more comparisons to try and help answer your question but they're not complete. I din't see the Mac mini ever being configured to match a Mac Pro but we've already seen it match the Mac Studio's base configuration cost. Once the Studio gets a M2/M3 Max and Ultra, I can see the GPU results getting a lot closer to the existing Mac Pro. To replace the Mac Pro with Apple Silicon, these results (except single CPU core) need to be twice what the current Mac Pro is capable of production. The NVIDIA GeForce RTX 3090 Ti is running OpenCL at 229738, that's where the Mac Pro has to get to. The 3090 Ti uses 24GB of dedicated memory with 10752 cores. The M1 Ultra only uses 64c and gets an OpenCL score of 76040. That's 1188/GPU core while the 3090 only gets 21/GPU core. This might mean Apple would only need to produce a GPU with maybe 200 cores. Could the M2/M3 Ultra be enlarged to handle this many GPU cores? Would the Mac Pro be configured with multiple Ultras connected using a second generation unified memory architecture?
  
  Here's Mac Studio results that should have been added to the comparison even though the Mac mini isn't available with the Max or Ultra configuration. Includes Mac Pro. When I look at the Ultra's numbers, they're better than the Mac Pro CPU but while the GPU still lags behind the Radeon, it's almost twice as powerful as the Mac Pro mini numbers. This article should have included the M2 Max numbers from the MBP but Geekbench isn't listing the M2 in the Metal chart (yet). 

  
  

  Apple M1 Ultra @ 3.2 GHz (20 cores) single core CPU = 1754, Mac Pro 28c = 1152

  Apple M1 Ultra @ 3.2 GHz (20 cores) multi-core CPU = 23329, Mac Pro 28c = 20035

  Apple M1 Ultra GPU Metal = 94583, Mac Pro Radeon Pro W6900X = 166946

  edited January 2023
• Reply 6 of 11

  blastdoor Posts: 3,296member

  January 25, 2023 3:53PM

  Good comments discussion of GPU performance. 
  
  In order for the Mac Pro to compete with 'pro' level Windows/Linux systems using high-end discrete GPUs, I wonder if Apple needs to either (1) continue to include high-end discrete GPUs in the Mac Pro (which kind of runs contrary to Apple's strongly expressed preference for sharing memory between CPU and GPU cores, but perhaps so be it) or (2) reconfigure Apple Silicon so that CPU and GPU cores sit on different pieces of silicon and are linked together via 'UltraFusion', thereby perhaps improving chip yields since CPU and GPU cores could be on separate cores. 
  
  I'm inclined to think that option 2 is more appealing technically, but I'm not sure about the business/economics side. If Apple puts CPU and GPU on different dies linked by UltraFusion (or whatever they want to call their 'glue'), they would likely need to do that across more product lines than just the Mac Pro. Maybe only integrate CPU and GPU on a single piece of silicon for the generic M#, but for Pro, Max, Ultra, etc, put CPU and GPU on different dies linked. That would allow independent scaling of CPU and GPU power to better target the needs of users who either need more CPU or more GPU (or both). 

  
• Reply 7 of 11

  mjtomlin Posts: 2,673member

  January 25, 2023 5:31PM

  blastdoor said:

  Good comments discussion of GPU performance. 
  
  In order for the Mac Pro to compete with 'pro' level Windows/Linux systems using high-end discrete GPUs, I wonder if Apple needs to either (1) continue to include high-end discrete GPUs in the Mac Pro (which kind of runs contrary to Apple's strongly expressed preference for sharing memory between CPU and GPU cores, but perhaps so be it) or (2) reconfigure Apple Silicon so that CPU and GPU cores sit on different pieces of silicon and are linked together via 'UltraFusion', thereby perhaps improving chip yields since CPU and GPU cores could be on separate cores. 
  
  I'm inclined to think that option 2 is more appealing technically, but I'm not sure about the business/economics side. If Apple puts CPU and GPU on different dies linked by UltraFusion (or whatever they want to call their 'glue'), they would likely need to do that across more product lines than just the Mac Pro. Maybe only integrate CPU and GPU on a single piece of silicon for the generic M#, but for Pro, Max, Ultra, etc, put CPU and GPU on different dies linked. That would allow independent scaling of CPU and GPU power to better target the needs of users who either need more CPU or more GPU (or both). 

  
  This is what I thought they might do, but there's just not a market large enough to create a one-off design. The cost would be extremely high and price them out of market. I think their strategy of developing a single MxX SoC and creating 3 tiers (Max, Ultra, Extreme) of performance by stitching them together is an extremely cost effective design and one that is sustainable. An M1 Ultra tops out at about 220W that includes both the GPU and CPU. High-end desktop CPUs push 300W and high-end GPUs are over 400W now. And rarely does a workflow completely stress both GPU and CPU. So there's no reason an Mx "Extreme" can't top out at 400W inside a Mac Pro like case with a massive heat sink and fan.
  
  I do think Apple will need to support 3rd party GPUs (accelerators) in the Pro as some work flows require them. I also believe that with doubling the Ultra, the Extreme variant will have 16 memory channels. 8 will be used for "on die" RAM, and the other 8 will be used for expandable RAM.
  

  
• Reply 8 of 11

  keithw Posts: 141member

  January 25, 2023 6:00PM

  blastdoor said:

  Good comments discussion of GPU performance. 
  
  In order for the Mac Pro to compete with 'pro' level Windows/Linux systems using high-end discrete GPUs, I wonder if Apple needs to either (1) continue to include high-end discrete GPUs in the Mac Pro (which kind of runs contrary to Apple's strongly expressed preference for sharing memory between CPU and GPU cores, but perhaps so be it) or (2) reconfigure Apple Silicon so that CPU and GPU cores sit on different pieces of silicon and are linked together via 'UltraFusion', thereby perhaps improving chip yields since CPU and GPU cores could be on separate cores. 
  
  I'm inclined to think that option 2 is more appealing technically, but I'm not sure about the business/economics side. If Apple puts CPU and GPU on different dies linked by UltraFusion (or whatever they want to call their 'glue'), they would likely need to do that across more product lines than just the Mac Pro. Maybe only integrate CPU and GPU on a single piece of silicon for the generic M#, but for Pro, Max, Ultra, etc, put CPU and GPU on different dies linked. That would allow independent scaling of CPU and GPU power to better target the needs of users who either need more CPU or more GPU (or both). 

  
  I can see no technical reason Apple can't have both a powerful on-chip set of GPUs as well as supporting discrete PCIe GPU boards for people who need it.  I've been running an AMD RX 6900 XT graphics card in a Thunderbolt 3-based eGPU enclosure made by Sonnettech, and getting similar GB 5 metal results to the current Mac Pro.  This is with a 5-year-old iMac Pro.
  

  
• Reply 9 of 11

  tenthousandthings Posts: 1,037member

  January 25, 2023 8:52PM

  keithw said:

  blastdoor said:

  Good comments discussion of GPU performance. 
  
  In order for the Mac Pro to compete with 'pro' level Windows/Linux systems using high-end discrete GPUs, I wonder if Apple needs to either (1) continue to include high-end discrete GPUs in the Mac Pro (which kind of runs contrary to Apple's strongly expressed preference for sharing memory between CPU and GPU cores, but perhaps so be it) or (2) reconfigure Apple Silicon so that CPU and GPU cores sit on different pieces of silicon and are linked together via 'UltraFusion', thereby perhaps improving chip yields since CPU and GPU cores could be on separate cores. 
  
  I'm inclined to think that option 2 is more appealing technically, but I'm not sure about the business/economics side. If Apple puts CPU and GPU on different dies linked by UltraFusion (or whatever they want to call their 'glue'), they would likely need to do that across more product lines than just the Mac Pro. Maybe only integrate CPU and GPU on a single piece of silicon for the generic M#, but for Pro, Max, Ultra, etc, put CPU and GPU on different dies linked. That would allow independent scaling of CPU and GPU power to better target the needs of users who either need more CPU or more GPU (or both). 

  
  I can see no technical reason Apple can't have both a powerful on-chip set of GPUs as well as supporting discrete PCIe GPU boards for people who need it.  I've been running an AMD RX 6900 XT graphics card in a Thunderbolt 3-based eGPU enclosure made by Sonnettech, and getting similar GB 5 metal results to the current Mac Pro.  This is with a 5-year-old iMac Pro.
  

  I think if we step back, it’s evident the current Mac Pro must have been designed to house Apple Silicon. They introduced it only a year before the announcement (shipping only months before it) — the planning and preparation for such a consequential move must have been well underway in 2017 and 2018 when the current Mac Pro was created. If so, then the plan all along has been to support MPX modules with Apple Silicon, including GPUs. 
  

  
• Reply 10 of 11

  tenthousandthings Posts: 1,037member

  January 25, 2023 9:34PM

  mjtomlin said:

  keithw said:

  The 2023(?) ASi "Mac Pro" must either be able to reach the 166,946 GB5 GPU results either with on-chip GPU cores or by a discrete graphics card like the existing Intel Mac Pro, otherwise, why bother to even release it?
  

  
  I think this is why we haven't seen the new Mac Pro yet. The new GPU design for the A16 was supposed to see a huge performance increase (>50%, mainly due to implementing hardware based ray tracing), but it had to be pulled because apparently it wasn't meeting efficiency standards*. So If I had to guess, the M3 is going to skip the A16 and be based on the A17 generation of cores, so we should see a fairly substantial performance increase in the M3 and finally get the ASi based Mac Pro, which will be the first system with M3 generation SoCs with M3 Ultra and M3 Extreme.  (Both the A17 and M3 will also use TSMC's N3 process bringing further performance and efficiency enhancements.)
  
  *This is the issue that's going to have be addressed at some point in the future... trying to develop a single core for both mobile and desktop applications. More than likely, that new GPU would've been fine in a desktop system where thermal ceilings can be lifted with active cooling systems. I think Apple will eventually start "optimizing" actual CPU and GPU (more so) cores for their intended systems.
  

  That leak about the A16 graphics for iPhone is believable, but it may miss the forest for the trees. An M2 Extreme could use the A16 foundation on N4 with new graphics, and let’s not forget we don’t actually know anything about the M2 Ultra. Your hypothetical optimized cores may be closer than we know…
  

  edited January 2023
• Reply 11 of 11

  doggone Posts: 380member

  March 2, 2023 5:23PM

  blastdoor said:

  Good comments discussion of GPU performance. 
  
  In order for the Mac Pro to compete with 'pro' level Windows/Linux systems using high-end discrete GPUs, I wonder if Apple needs to either (1) continue to include high-end discrete GPUs in the Mac Pro (which kind of runs contrary to Apple's strongly expressed preference for sharing memory between CPU and GPU cores, but perhaps so be it) or (2) reconfigure Apple Silicon so that CPU and GPU cores sit on different pieces of silicon and are linked together via 'UltraFusion', thereby perhaps improving chip yields since CPU and GPU cores could be on separate cores. 
  
  I'm inclined to think that option 2 is more appealing technically, but I'm not sure about the business/economics side. If Apple puts CPU and GPU on different dies linked by UltraFusion (or whatever they want to call their 'glue'), they would likely need to do that across more product lines than just the Mac Pro. Maybe only integrate CPU and GPU on a single piece of silicon for the generic M#, but for Pro, Max, Ultra, etc, put CPU and GPU on different dies linked. That would allow independent scaling of CPU and GPU power to better target the needs of users who either need more CPU or more GPU (or both). 

   keithw said:

  I can see no technical reason Apple can't have both a powerful on-chip set of GPUs as well as supporting discrete PCIe GPU boards for people who need it.  I've been running an AMD RX 6900 XT graphics card in a Thunderbolt 3-based eGPU enclosure made by Sonnettech, and getting similar GB 5 metal results to the current Mac Pro.  This is with a 5-year-old iMac Pro. 
  

  
  

   The market for MacPros is very small.  Maybe 50K units per year.  It will be very hard to justify making a special chip.  Again if this special chip use is expanded to the higher versions of other Mac products it dilutes the economy of scale for the current chip format.  For truly high end needs, using the eGPU solution the gap in GPU performance can be closed at a fraction of the cost of the current MAcPro solution.
  
  What amazes me is that the current M2 chips are close to the performance of the current Mac Pro. Even my M1 MBP works better with Intel programs than the previous Intel machine did.  That's after all the rosetta conversions are done. Battery life is out of this world.  I can be unplugged for days and still never get below 50% with average internet and video use.  
  
  I think we are at a paradigm shift on how PCs can be efficiently run.  This is truly Apple to make us think differently.  Performance and battery life are unbeatable with a M# chip.  On chip GPU can address most of the standard needs for mid and high end machines.  Where exceptional performance is needed external solutions can fill the gap.