Phil 'My Ass' Schiller Read Please: Our mMP whish list

[Mago]

[Updated]
- I don't know if you or actually somebody at cuppertino read this forum, how ever if the case this post some how arives to you, i personally promise never name you Phil "my ass" Schiler Again in case the mMP becomes at least an 80% satisfactory for us.

This is my personal Whish list:

GPUs: 1 or 2, I dont care if you use ISA PCIe GPUs or reuse the tMP proprietary design, as long it has a generous integrated cooling device its OK, MOST OF US WANT AN nVIDIA GPU OPTION, EVEN PASCAL GP100.

CPU: Intel Skylake Xeon with options upto Xeon Phi.

Storage: DUAL std m.2 form factor both PCIe NVME, RST (optane), and M.2. Sata Supported: consider this releases you the mess to honor warranty on such degradable devices, also saves you some money procuring the storage, a sensitive point, even more than GPUs.

RAM: 8 RDIMM ECC please.

Ports: DONT MISS HDMI 2.0b please, also a bunch of Thunderbol 3.

PSU: 1000W at least

Now on Mac OS:

API Support: CUDA8 OpenCL 2.0 and VULKAN (despite Vulkan announcedwill support compile to Metal), most developers cant justify/afford to re-write tons of CUDA/OpenCL code into Metal.

XCODE Update Python to 3.6, also LLVM and every common Toolchain to its latest STABLE versions, also support coding Python into Xcode, its pity some API are 3+ year old.

UPGRADES UPGRADES, Whatever the MODULARITY Concept you USE (propertary or Industry Standard), PLEASE SELL EVERY YEAR THAT MODULES TO RETROFIT older mMP.

For That Interim/Alternative iMac "PRO": support Xeon E3v6 64Gb ram and include some useful 8GB GPU from AMD RX570 to Nvidia GTX1070, build a 400W TDP capable iMac, so most Pro User could run marathon Render/Compute sessions, A DISCRETE (replaceable) SSD also its an Must.

update:
Enable Thunderbolt 3 Support for external GPU Accelerators, this will save the day along with an Xeon based iMac.

*******************************************************************

Forum: Guys Help to keep this post visible.

 

[Maxx Power]

AMD Zen based option would be very appealing for workstation usage, especially the 8 core variants. Also, I'm not sure about a nVIDIA Pascal based product in 2018. I would like to see an ability to use consumer/prosumer grade GPUs, instead of just Fire Pros/Radeon Pros/Quadros/etc. I would think a lot of us using the workstation would use the compute aspect of the GPUs more than gaming, but not everyone needs full performance double-precision floating point computations.

EDIT: Although SATA is going to be obsolete sooner or later, some form of cheaper/massive storage option would be nice to store a ton of media/data/etc that does not require fast access.

 

[Mago]

AMD Zen based option would be very appealing for workstation usage,

AMD is reading Zen-Naples Architecture aimed at Servers (workstations) with 16 to 32 cores, but Zen-Summit/Raven Ridge (aka Rizen) also its another good option since support ECC across the range from 4 to 16 cores.

Also time ago we speculated about another Zen which would arrive later this year or in 2018, the HPC oriented APU integrating a Naples or Summit Ridge 16 core CPU with an Vega GPU and 8GB HBM2 ram all on the same MCM module, upto 2 modules per system, it could be ideal for a modular MP very similar to the Trash Can, with 2 Zen APU each with independent cooling should shut up many mouths, and offer some flexibility and much lower cost than Intel Skylake Xeon, but never will reach the possibilities Xeon Skylake Offer (utpo 72 cpu cores, utpo 3 PCIe x16 GPUs).

 

[usna92]

I believe the key to professional use is standardization not innovation, so I would be pleased if Phil could have Apple use their head this time when designing this machine, rather than the aforementioned other parts.



"GPUs: 1 or 2, I dont care if you use ISA PCIe GPUs or reuse the tMP proprietary design, as long it has a generous integrated cooling device its OK, MOST OF US WANT AN nVIDIA GPU OPTION, EVEN PASCAL GP100."

What hurt the nMP the most was a lack of ability or willingness to continue to innovate as the GPU market evolved. I would not want to see Apple to perpetuate this mistake in a whole new line of machines. Commercially available GPUs (and the ability to fit more than one) is the key. Proprietary designs only work if you are going to stick with the market. Apple is not committed to that kind of change rate (just look at the lack of progress in keeping up with CPU design), so don't continue to go down that road. Allow professional users to determine what video cards they need, the easy access to, or change in EFI design that makes them more universal, and step back. Let the market decide which GPU technology works the best for for a given application. Some people need a build farm, some people need a FCPX machine, let people decide at multiple times what they need, rather than only at first purchase.


CPU: Intel Skylake Xeon with options upto Xeon Phi.

Agreed.

Storage: DUAL std m.2 form factor both PCIe NVME, RST (optane), and M.2. Sata Supported: consider this releases you the mess to honor warranty on such degradable devices, also saves you some money procuring the storage, a sensitive point, even more than GPUs.

Also agreed. here. Stick with STANDARD schemes. I do not need a proprietary interface to a standard sized SSD. No one does. Please stop doing that. Space for a full height 3.5" drive or drive system would be nice. More SATA ports than slots for drives would be great to allow for double stacking of SSDs etc.

RAM: 8 RDIMM ECC please.

Sure, or at least support for it. Whatever they pick please make it readily available on the market. Doesn't has to be cheapest, just available.

Ports: DONT MISS HDMI 2.0b please, also a bunch of Thunderbol 3.

Here are I torn. Do I want improved HDMI, sure (but won't that be on my GPU?). I am more concerned with Thunderbolt3. I know we are going to get TB3 ports, but can we get a few regular USB as well. You know, for old times sake? The new adapters from TB3 to TB2 and USB are quirky and not 100% reliable.

PSU: 1000W at least

Yes to support the GPUs above.

Now on Mac OS:

API Support: CUDA8 OpenCL 2.0 and VULKAN (despite Vulkan announcedwill support compile to Metal), most developers cant justify/afford to re-write tons of CUDA/OpenCL code into Metal.

Not my wheel house, so sure, why not.

XCODE Update Python to 3.6, also LLVM and every common Toolchain to its latest STABLE versions, also support coding Python into Xcode, its pity some API are 3+ year old.

Yes please. Additionally, can we ensure the newest compiler is optimized for a multicore machine? I am looking at you Swift compiler performance.

UPGRADES UPGRADES, Whatever the MODULARITY Concept you USE (propertary or Industry Standard), PLEASE SELL EVERY YEAR THAT MODULES TO RETROFIT older mMP.

Yes, Yes, YES!. Except for the part about proprietary interfaces. DO NOT DO THAT.

For That Interim/Alternative iMac "PRO": support Xeon E3v5 64Gb ram and include some useful 8GB GPU from AMD RX570 to Nvidia GTX1070, build a 400W TDP capable iMac, so most Pro User could run marathon Render/Compute sessions, A DISCRETE (replaceable) SSD also its an Must.

I am in the camp that I can wait for whatever comes, but whatever you do with this machine, do not say 6 months later, this is close enough to a pro, we don't need to do another machine. I completely understand that the Pro market is a small slice of your mac sales, let alone overall sales, but this is the group that makes the apps that make the rest of the ecosystem great. Give us the tools to allow developers and content creators to continue to do that. Pretty is nice in a pro machine, but not at the expense of usability. This is not an area where it needs to be thinner. It just needs to work like a dog, 24x7, and then be upgradable with things I can actually purchase when I need more horsepower.

 

[Mago]

What hurt the nMP the most was a lack of ability or willingness to continue to innovate as the GPU market evolved.

the tcMP GPU card form factor wasnt an issue, it actually solves an recurrent issue with PCIe slots: shortcuts due slot miss alignment, a morelesss common issue in PCs.

Also the tcMP GPU card, drives all the signals back to the motherboard (PCIe, and Video Output).

Also consider TB3 need somehow 2 DP1.2 signals for each GPU headers, PCIe dont provide, and I dont believe Apple will include a dp1.2 pass trough cable/mb connector to feed the TB3 connectors, so the best solution its either an revision on the tcMP GPU card design or adopt MXM (which its much more constrained about power than tcMP's GPU cards).

Of course recycle tcMP gpuCard doesnt means to re-edit the real problem: the Thermal Core, even with an Generous TDP, it isnt modular (unless Apple is happy with users messing with thermal paste - unlikely-).

Of course its an full year until Apple releases the mMP 7,1, lets see whats happens.

 

[mattspace]

the tcMP GPU card form factor wasnt an issue, it actually solves an recurrent issue with PCIe slots: shortcuts due slot miss alignment, a morelesss common issue in PCs.

Personally, i actually think thunderbolt is part of the problem. Intel is never going to progress display standards in TB as fast as the gpu makers. I'd rather have my displays on a desktop plugged into my gpu, and keep tb for storage and laptop displays.

If Apple can't do custom gpus with the same price and performance of the retail cards from Nvidia, FCPX is more commoditisable than the highest performance in GPUs, expecially given Apple doesn't rely on that product for the company's survival, the way Avid, Adobe, and to an extent Davinci do.

 

[Maxx Power]

AMD is reading Zen-Naples Architecture aimed at Servers (workstations) with 16 to 32 cores, but Zen-Summit/Raven Ridge (aka Rizen) also its another good option since support ECC across the range from 4 to 16 cores.

Also time ago we speculated about another Zen which would arrive later this year or in 2018, the HPC oriented APU integrating a Naples or Summit Ridge 16 core CPU with an Vega GPU and 8GB HBM2 ram all on the same MCM module, upto 2 modules per system, it could be ideal for a modular MP very similar to the Trash Can, with 2 Zen APU each with independent cooling should shut up many mouths, and offer some flexibility and much lower cost than Intel Skylake Xeon, but never will reach the possibilities Xeon Skylake Offer (utpo 72 cpu cores, utpo 3 PCIe x16 GPUs).

I agree. If Apple can pass the cost savings to the consumer, it will be a real win for both parties. Additionally, I would welcome any of that Zen-based APUs into the Mac Minis and iMacs.

 

[ActionableMango]

I'm going to be humble and just hope for a couple of PCIe slots and internal storage with standard industry connectors.

 

[joebclash]

Keep jony ive away from the hardware specs. He can design the buttons and handles but keep him away from hardware specs.

 

[JesperA]

If the next Mac Pro will use Skylake-EP (purley) then we could put in Xeon Phi ourself even though Apple would not have a BTO option for it (yeah yeah, thermals).

But, do anyone know if macOS can handle 72 CPU cores with 288 threads? (im talking about main CPU, not Phi as a coprocessor)

 

[Mago]

IBut, do anyone know if macOS can handle 72 CPU cores with 288 threads? (im talking about main CPU, not Phi as a coprocessor)

Sure, Xcode/ObjectiveC complies thru LLVM, the problem are the Apps itself, not every logic design can benefit from such massive parallelism, actually very specific algorithms maybe splitted into as many threads (cryptography, machine learning, video processing, some (not all) 3-D related compute)

 

[OS6-OSX]

[Updated]

Forum: Guys Help to keep this post visible.

 

[PortableLover]

Hopefully two xeon cpus. In addition to m.2 a few 3.5 hdd slots

 

[AidenShaw]

Hopefully two xeon cpus. In addition to m.2 a few 3.5 hdd slots

Drop the proprietary EFI boot interface, and support UEFI like everyone else. Boot screens for any standard card without needing special proprietary firmware. EFI was deprecated in 2005....

 

[violst]

Drop the proprietary EFI boot interface, and support UEFI like everyone else. Boot screens for any standard card without needing special proprietary firmware. EFI was deprecated in 2005....

That is absolutely the best thing they can do! And to have support for up to 4 Nvidia/AMD GPU's. I myself would use 4 Nvidia 1080 Ti's if I could!

 

[Mago]

Drop the proprietary EFI boot interface, and support UEFI like everyone else. Boot screens for any standard card without needing special proprietary firmware. EFI was deprecated in 2005....

Too optimistic... but Ok its legit.

I believe the Modularity on the Next Mac Pro will go on two choices: a custom GPU<=>PCIe interface non ISA (likely the current GPU design with few revisions about Power/TDP and Modular (independent) cooling, or some sort of PCIe Slot Interposer to hold std PCIe GPUs and connect it with all its Display Port buses both tunneled to the Main board (remember the GPU has to Feed the TB3 chipset).

 

[mavericks7913]

1. More PCIE lines
2. Dual M.2 SSD
3. More RAM
4. Dual CPU
5. Better cooling
6. Bigger case
7. Software support and update especially Nvidia

Etc.

I wonder if apple is going to throw out the current trash mac pro. Why not re-use it for another Mac desktop positioned between iMac and Mac Pro?

 

[Mago]

I wonder if apple is going to throw out the current trash mac pro. Why not re-use it for another Mac desktop positioned between iMac and Mac Pro?

Yes, I'll love another trashcan with non-xeon hardware and a couple of RX480 GPUs

 

[Spacedust]

Bring back Windows 7 support!

 

[Synchro3]

Bring back Windows 7 support!

Hahahaha. Microsoft prevents actively Kaby Lake support for Windows 7 and 8, all security updates will be blocked next month. Therefore I don't think Windows 7 is a topic for Apple.

However: The new ASRock Z270 boards for Kaby Lake have Windows 7 driver support: http://www.asrock.com/MB/Intel/Fatal1ty Z270 Gaming-ITXac/index.us.asp#osW764

Build a PC/Hackintosh with a Windows 7 partition.

 

[Mago]

Hahahaha. Microsoft prevents actively Kaby Lake support for Windows 7 and 8, all security updates will be blocked next month. I don't think Windows 7 is a topic for Apple.

However: The new ASRock Z270 boards have Windows 7 driver support: http://www.asrock.com/MB/Intel/Fatal1ty Z270 Gaming-ITXac/index.us.asp#osW764

Build a Hackintosh with a Windows 7 partition.

also can run W7 into a VM as I use to do, a lot safer and given i use Windows for few non demanding apps, the Virtualization overhead doesnt taxes the system performance.

 

[slughead]

I agree with others:

- Ryzen really is a great value for the mid-low end workstation, PLUS the x1800 uses ECC out of the box
- You do NOT need a 1000w power supply for a single GPU, or even most dual setups. In fact I don't even know if dual CPU + Dual GPU with modern procs could use that much (probably close). that just adds cost
- Apple's proven they will NOT churn out a custom PCIe GPU, best to use standard PCIe cards with dual slots
- eGPUs are stupid on a desktop. There, I said it. adds expense, points of failure, and bottlenecks. Watch a 1080Ti get strangled by TB3, it's sad

IMO they should just make a regular workstation and put an Apple stamp on it. 5 slots at least with spacing to accommodate 2 dual slot GPUs, plus room for 10GbE or whatever you want. Having AMD and Intel options for CPU is totally not going to happen but would be great to capture the low end Ryzen and high end xeon - preferably dual CPU as an option.

Although I'm very excited about the future of AMD dual CPU, it's going to be freaking amazing.

 

[now i see it]

My wish is that the the Title of this thread didn't have the word "wish" misspelled and that there weren't so many typos in the first post.

 

[jblagden]

I agree with others:

- Ryzen really is a great value for the mid-low end workstation, PLUS the x1800 uses ECC out of the box
- You do NOT need a 1000w power supply for a single GPU, or even most dual setups. In fact I don't even know if dual CPU + Dual GPU with modern procs could use that much (probably close). that just adds cost
- Apple's proven they will NOT churn out a custom PCIe GPU, best to use standard PCIe cards with dual slots
- eGPUs are stupid on a desktop. There, I said it. adds expense, points of failure, and bottlenecks. Watch a 1080Ti get strangled by TB3, it's sad

IMO they should just make a regular workstation and put an Apple stamp on it. 5 slots at least with spacing to accommodate 2 dual slot GPUs, plus room for 10GbE or whatever you want. Having AMD and Intel options for CPU is totally not going to happen but would be great to capture the low end Ryzen and high end xeon - preferably dual core.

Although I'm very excited about the future of AMD dual core, it's going to be freaking amazing.

If you want to see external graphics done right, check out Alienware. Somehow they managed to get a full PCIe slot's bandwidth going through that cable.

 

[slughead]

If you want to see external graphics done right, check out Alienware. Somehow they managed to get a full PCIe slot's bandwidth going through that cable.

I'll have to check it out, last time I checked the TB overhead was up to 15% in high FPS situations PLUS the GTX 1080 had another 10-20% hit from the 4gbps limitation (even through a direct PCIe slot). The TB overhead I could see being reduced maybe but at the end of the day, you're still trapped at 4gbps.

Not too shabby, for sure, however for high end GPUs (soon to be mid-range speeds), TB is a bottleneck.

I'm not being down on TB, and eGPUs for laptops and all-in-ones aren't a bad idea. However for an actual tower desktop it's absolutely stupid. Even if you don't hit a wall, you still have the issue of ANOTHER wall wart and ANOTHER case fan and MORE cables. All expensive and points of failure.

It's just dumb.