that would be a good sendoff for eGPU users like myself and Mac Pro. everyone wins.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
MP All Models 40 core ASI with PCI-E slot is around the corner!!
- Thread starter Amethyst
- Start date
- Sort by reaction score
3 months later, and this still be currently latest in-development mac pro.
7,1 case, 6 pci-e slot, no ram slot.
Base ASi Mac Pro
- M2 Ultra SoC (3nm)
- 24-core CPU (16P/8E)
- 60-core GPU
- 32-core Neural Engine
- 96GB LPDDR5 SDRAM
- 800GB/s UMA bandwidth
- 1TB NVMe SSD (2 @ 512GB NAND blades)
- (6) PCIe slots
- US$5999
Fully-Loaded ASi Mac Pro
- M2 Ultra SoC (3nm)
- 24-core CPU (16P/8E)
- 76-core GPU
- 32-core Neural Engine
- 192GB LPDDR5 SDRAM
- 800GB/s UMA bandwidth
- 8TB NVMe SSD (2 @ 4TB NAND blades)
- (6) PCIe slots
- US$9999
as my pal have told me that their team have to launch MONSTER of CHIP for next mac pro for sake of making large step ahead of the others in Workstation market, so i think only 24 core CPU seem not enough (7000 amd threadripper and intel sapphire rapid is on the way)
I fully believed that we will see at lease 4*M2 max chip in 8,1 mac pro.
Base M2 Extreme Mac Pro:
- M2 Extreme SoC (N3B)
- 48-core CPU (32P/16E)
- 120-core GPU
- 64-core Neural Engine
- 192GB LPDDR5 SDRAM
- 800GB/s UMA bandwidth
- 1TB NVMe SSD (2 @ 512GB NAND blades)
- (6) PCIe slots
- US$12,499
- M2 Extreme SoC (N3B)
- 48-core CPU (32P/16E)
- 152-core GPU
- 64-core Neural Engine
- 384GB LPDDR5 SDRAM
- 800GB/s UMA bandwidth
- 8TB NVMe SSD (2 @ 4TB NAND blades)
- (6) PCIe slots
- US$17,499
The problem with multichip is its never 4 X the speed, like dual chip GPU's and SLI GPU's its always slower. Single 3.46ghz geekbench 2 score 15.000 dual 3.46ghz 23.000 not 30.000 it never scales right. but it will be faster than a single for sure.
hi, personally i don't care if the case is the previous one as much as it allows me to put in some external gpus, storages, and of course ram. I really do not know how apple is going to scale the ram limitation of the soc against the huge ram capacity of the previous machine and any other workstations' capacity in the market. For the pros, especially those creating 3D and simulation, there are so many cases we really really need lots of ram if we don't want to deal with huge cache file sizes in particle, fluid, rigid/soft body and etc simulations. Another thing is that even if Apple, lets say, comes up with a huge capacity, non-upgradability is still a problem for the pros. We generally tend to update the parts of our machines once a year, or once in 15 months or so when we have enough money so that we can work fast and comfortably. I think Apple will possibly addressing this in a way, however it is not going to make many people like us happy. Let's see.
After a lot of thinking I believe that the reasonable final Mac Pro 2023 will be based on the m2 max and ultra chips like the rumours have it, but that there will be a CXL like mechanism to add RAM in a PCIe slot for those that absolutely need it. This will be transparent to the user but I wonder what the perf. will be for large sims that need to be in ram all the time?
Then we will of course have max and ultra based PCIe gpus. single slot 1000$ for a max based one, dual/quad slot 2000$ for an ultra And we can stack 2 of these at least, ending up in up with up to 76*3 = 228 GPU cores.
Maybe there are even ultra duos for 6000$ a piece, so we end up with 380 cores with 192GB VRAM and 150 clean TF.
This solution would actually be quite bad ass and get bragging rights. Sure, the CPU would be a little bit weak compared to what we would have expected but for GPU it would be solved. This perf is around the same as a dual 4090 (or rather RTX 6000 ada pro cards) system but better in some ways and worse in others
If I was Apple this is what would be reasonable in order to not have to go hide in shame. Anything less would actually be a mocking of the MacPro name.
In the end, that would cover the whole desktop pro spectra, from those who just needs a m2 max and ram for instruments and slots for special audio cards to the heavy duty 3d rendering/vfx crowd.
Base Max based machine starting at 4999$, m2 ultra base machine at 6999$ , Full ultra with 192 RAM: 8999$
A system like this would have a lot of flexibility and I can only see that the most CPU bound tasks get left out of the equation with the "extreme" chip not being there anymore.
Wild cards: hardware Raytracing support, dual slot/daughter cards for CPUs
Come on Apple, let this be the system so I can get a m2ultra with an extra ultra GPU
and maybe add an m3 based GPU when the time comes. Solves needs right now while being relevant in 5 years time since upgrades are available.
Then we will of course have max and ultra based PCIe gpus. single slot 1000$ for a max based one, dual/quad slot 2000$ for an ultra And we can stack 2 of these at least, ending up in up with up to 76*3 = 228 GPU cores.
Maybe there are even ultra duos for 6000$ a piece, so we end up with 380 cores with 192GB VRAM and 150 clean TF.
This solution would actually be quite bad ass and get bragging rights. Sure, the CPU would be a little bit weak compared to what we would have expected but for GPU it would be solved. This perf is around the same as a dual 4090 (or rather RTX 6000 ada pro cards) system but better in some ways and worse in others
If I was Apple this is what would be reasonable in order to not have to go hide in shame. Anything less would actually be a mocking of the MacPro name.
In the end, that would cover the whole desktop pro spectra, from those who just needs a m2 max and ram for instruments and slots for special audio cards to the heavy duty 3d rendering/vfx crowd.
Base Max based machine starting at 4999$, m2 ultra base machine at 6999$ , Full ultra with 192 RAM: 8999$
A system like this would have a lot of flexibility and I can only see that the most CPU bound tasks get left out of the equation with the "extreme" chip not being there anymore.
Wild cards: hardware Raytracing support, dual slot/daughter cards for CPUs
Come on Apple, let this be the system so I can get a m2ultra with an extra ultra GPU
and maybe add an m3 based GPU when the time comes. Solves needs right now while being relevant in 5 years time since upgrades are available.
I expect hardware ray-tracing to come with the M3 family of SoCs...
Regarding dual slot/daughter cards, Johny Srouji explained why that (multi-socket system) was Not Good in the M1 Ultra introduction...
Still holding out for the "differing SoC" solution; one "regular" SoC and one "GPU-specific" SoC per Mn Ultra...
Two of these mixed SoC Mn Ultras to make a Mn Extreme SoC...
Multiples of the "GPU-specific" SoC on a PCIe card to make an ASi GPGPU...
These would be like the MPX format, mainly for the power delivery than the "TB fold-back", and for the massive heat sink...
This could explain the 7.1 chassis with six slots rather than eight, the MPX GPGPUs each take four slots worth of width, and the actual PCIe slots that were covered by MPX GPUs in the 2019 Intel Mac Pro would not be there, hence six slots rather than eight...
Hopefully, we will find out what the first generation of Apple silicon Mac Pro headless desktops has for us in a few months...! ;^p
Regarding dual slot/daughter cards, Johny Srouji explained why that (multi-socket system) was Not Good in the M1 Ultra introduction...
Still holding out for the "differing SoC" solution; one "regular" SoC and one "GPU-specific" SoC per Mn Ultra...
Two of these mixed SoC Mn Ultras to make a Mn Extreme SoC...
Multiples of the "GPU-specific" SoC on a PCIe card to make an ASi GPGPU...
These would be like the MPX format, mainly for the power delivery than the "TB fold-back", and for the massive heat sink...
This could explain the 7.1 chassis with six slots rather than eight, the MPX GPGPUs each take four slots worth of width, and the actual PCIe slots that were covered by MPX GPUs in the 2019 Intel Mac Pro would not be there, hence six slots rather than eight...
Hopefully, we will find out what the first generation of Apple silicon Mac Pro headless desktops has for us in a few months...! ;^p
You mention this a lot, can I ask - do you understand the value of Ray Tracing in the modern context isn't that it's hardware-based, but that it's realtime.
Like, hardware based ray tracing in a GPU that can't power the scale and complexity and framerate of a 3d scene isn't actually a useful thing. All the hardware based ray tracing in the world won't be of utility, if the GPU itself isn't of a class that the rest of the industry has adopted as the standard for the complexity of work being done.
There are rumors floating around about Blender having real-time ray-tracing for the Metal view port at some point, one would assume dedicated ray-tracing hardware would help with this...?
In service of what though? it's still going to be MASSIVELY limited by the GPU itself. Maybe it's useful for ray tracing a small object model in a blackbox environment... but honestly this just reeks of "We've got ray tracing too, folks! Folks?"
Or, it's going to be for something 3D people aren't expecting - like faster lighting simulation for the facetime camera, or real world lightsource measurement and simulation in the UI like Bob Borrough's demos a couple of years back. Either way, If you're looking for RTX-level ray tracing, I suspect you're going to be disappointed, or "yeah, but NEXT year's M(x).."
just a note: hw rt also accelerate regular renderers not just realtime ones. When Rt hw came to nvidias 2000 series 5 years ago it took a year a then redshift and octane got a great boost for most scenes. If you go to https://render.otoy.com/octanebench...le_by=linear&filter=&singleGPU=1&showRTXOff=1 you can see that in general rt hw gives 30% boost but in some scenes with simple shading a lot of geometry the boost is much larger. In blender there seems to be an insane benefit to use rt hw(optix) instead of using thr older cuda path.
All in all, on a mac, rt hw would mainly benefit these kind of workloads since the mac Gpus will not be powerful
Enough for realtime rendering of RT effects (only a few high end gpus are so there is no shame). Oh, and there are very few games anyway 😂
All in all, on a mac, rt hw would mainly benefit these kind of workloads since the mac Gpus will not be powerful
Enough for realtime rendering of RT effects (only a few high end gpus are so there is no shame). Oh, and there are very few games anyway 😂
Another reason why you want 3rd party GPUs:
www.techspot.com
DirectStorage benchmark shows massive transfer speed improvements
Tests from PC Games Hardware show that Microsoft's DirectStorage API can help NVMe SSDs load assets significantly faster than SATA SSDs. They also offer the enormous advantages...
www.techspot.com
And don´t forget PCI-E 5.0 which has twice the speed of PCI-E 4.0Another reason why you want 3rd party GPUs:
DirectStorage benchmark shows massive transfer speed improvements
Tests from PC Games Hardware show that Microsoft's DirectStorage API can help NVMe SSDs load assets significantly faster than SATA SSDs. They also offer the enormous advantages...
View attachment 2141628
I remember when I got my 2019s, they were touted as being eGPU-capable, but there were no eGPUs on the market that it’d work with and I forgot all about the idea. Sitting on a few 2019s yet, I should look into what the best options for eGPUs turned out to be now that the dust has settled. Perhaps that’ll tide me over til Apple delivers on all my 3nm hopes and dreams.
7.1 Mac Pro with eGPUs, maybe the 7.1 is cosplaying as a 6.1 Mac Pro...? ;^p
…oops, was actually talking about the 2019 iMac & MacBook Pro’s. The big selling point (to us) at the time was going to be the ability to be mobile but then plug them in @ the desktop and add a huge boost of eGPU power, but then none were compatible so it was a bit moot. I figure someone eventually must have sorted out the most powerful option possible though, before the M series derailed everything.
Hardware raytracing give you real-time performance on gaming engine and similar, using proper raytracer on anything complex you may get a reasonably fast feedback but it won’t run in realtime no matter what hardware you throw at it.
<<CAUTION! this is my opinions>>
From my thread in OCT22.
as the oct 22 macbook pro prototype have been use for production model.
and no new mac pro prototype since then, i quite believed that production mac pro unit is somewhat decisively.
From my thread in OCT22.
as the oct 22 macbook pro prototype have been use for production model.
and no new mac pro prototype since then, i quite believed that production mac pro unit is somewhat decisively.
Ouch, so we can probably expect the Mac Pro to be a real turd then.A single m2 ultra really is way way way to slow in the GPU department if we compare to even the old 2019 MP. Surely Apple cannot release this with any sense of pride.
Apple will not want the mac pro to have any user storage slots so they can change
$200 to go from 512 to 1TB SSD storage
I think it‘s impossible that the new Mac Pro will just come with a single M2 Ultra chip. If it‘s like that, the mox modules will be entire m2 ultra clusters that you can add in there to just double the performance or something. Otherwise it also makes ZERO sense to keep the current design.
It‘s gotta be like this, they can‘t just release an m2 studio in the old chassis with the ability to add 3 pci cards. No way.
It‘s gotta be like this, they can‘t just release an m2 studio in the old chassis with the ability to add 3 pci cards. No way.