MP 7,1 Waiting for Mac Pro 7,1 A1991 (no more)

[skippermonkey]

Another suggestion might be that the prices announced at WWDC have created big backlogs for HP Z-series as the VIP customers jump ship. Apple might not be under any pressure from the VIPs to ship....

Surely the ship-jumpers have all swam away by now. I think there are only about three people left who want one of these, me being one of them.
[automerge]1572741318[/automerge]

I don’t think Intel’s new CPUs that they are using have become generally available yet. The CPUs they are using weren’t supposed to even ship until Fall.

The expanded PCIe lane count for the new CPUs seems pretty critical to the 8 slot design of the Mac Pro.

Looking at Intel’s site, this CPU which I think is the base CPU “launched” in Q2. But Intel “launch” dates are generally a quarter or two behind general availability.

Intel® Xeon® W-3223 Processor (16.5M Cache, 3.50 GHz) - Product Specifications | Intel

Intel® Xeon® W-3223 Processor (16.5M Cache, 3.50 GHz) quick reference with specifications, features, and technologies.

ark.intel.com

I don’t think HP has even launched their updated workstations with these CPUs yet.

Well that seems pretty reasonable – a lack of a CPU would indeed slow down manufacture somewhat.

 

[Stephen.R]

since Apple is an American company, that it would be talking about 'Fall' as in how Americans define it.

Right I forgot Apple has no customers outside America. How silly of me.

 

[thisisnotmyname]

Right I forgot Apple has no customers outside America. How silly of me.

Well if a Thailand based company that happens to sell its products internationally made an announcement in Thailand in the Thai language I guess I wouldn't be upset if they didn't use American norms in their presentation.

In any case, it's coming "soon," there's no need for us to parse the meaning of "fall" in multiple regions and cultures.

 

[Pro7913]

I don’t think Intel’s new CPUs that they are using have become generally available yet. The CPUs they are using weren’t supposed to even ship until Fall.

The expanded PCIe lane count for the new CPUs seems pretty critical to the 8 slot design of the Mac Pro.

Looking at Intel’s site, this CPU which I think is the base CPU “launched” in Q2. But Intel “launch” dates are generally a quarter or two behind general availability.

Intel® Xeon® W-3223 Processor (16.5M Cache, 3.50 GHz) - Product Specifications | Intel

Intel® Xeon® W-3223 Processor (16.5M Cache, 3.50 GHz) quick reference with specifications, features, and technologies.

ark.intel.com

I don’t think HP has even launched their updated workstations with these CPUs yet.

Intel Xeon W-3223 Octa-core (8 Core) 3.50 GHz Processor - OEM Pack CD8069504248402: Microprocessors - COLAMCO.com

Buy Intel Xeon W-3223 Octa-core (8 Core) 3.50 GHz Processor - OEM Pack CD8069504248402 at COLAMCO.com: INTEL® XEON® W PROCESSORS Intel® Xeon® W processors deliver optimized performance for the needs of mainstream workstation professionals. Hardware-enhanced workload performance, se...

www.colamco.com

What about here?

 

[fendersrule]

I can't wait to destroy this with my Dune Pro AMD build.

 

[Stephen.R]

In any case, it's coming "soon," there's no need for us to parse the meaning of "fall" in multiple regions and cultures.

The last several pages of this thread indicate otherwise, given that even you lot can't agree when it is.

Unless you're selling a ****ing snow blower or something else that's inherently related to the weather of a given season, why on earth would you reference a season in your announcement.

You might as well say "Sometime after the first turkey day and before the second turkey day".

 

[Korican100]

Right I forgot Apple has no customers outside America. How silly of me.

How exactly would you want them to release a product then? Do you want there to be a separate keynote spoken in every language? And on top of that, they are making it in Texas lol. If they want to say Fall, then its up to the other countries to translate it to what makes sense to them. Just stay on your thailand website (which does state "later this year"), and let this American company talk how it wants.

 

[Stephen.R]

How exactly would you want them to release a product then?

I don't know. Using terminology that doesn't have people questioning whether it's snowing yet?
[automerge]1572752602[/automerge]

And on top of that, they are making it in Texas lol.

Ok, so I guess you'll have no problem when the next iPhone is released with all Chinese information then?

 

[Korican100]

I don't know. Using terminology that doesn't have people questioning whether it's snowing yet?
[automerge]1572752602[/automerge]

Ok, so I guess you'll have no problem when the next iPhone is released with all Chinese information then?

Well i bet the people of Puerto Rico who wish to purchase a Mac Pro are real confused, even though it's US land and never snow's or experiences autumn. Are you just arguing because you are restless? If so, me too, but c'mon dude. Let them say Fall.

 

[Ludacrisvp]

I think we're crap out of luck on that. This passed my mind too. Apple stopped the cheaper BTO options with the 3,1 IIRC. That was the last time you could get a Mac Pro for like, $2,399. They haven't had that option back since.

you could get a single processor 4,1/5,1 for almost as cheap. $2499...

Mac Pro "Quad Core" 3.2 (2012/Nehalem) Specs (Mid-2012, MD770LL/A, MacPro5,1, A1289, 2629): EveryMac.com

Technical specifications for the Mac Pro "Quad Core" 3.2 (2012/Nehalem). Dates sold, processor type, memory info, hard drive details, price and more.

everymac.com

Apple -might- do other BTO options again with cheaper setups.

 

[Coyote2006]

you could get a single processor 4,1/5,1 for almost as cheap. $2499...

Mac Pro "Quad Core" 3.2 (2012/Nehalem) Specs (Mid-2012, MD770LL/A, MacPro5,1, A1289, 2629): EveryMac.com

Technical specifications for the Mac Pro "Quad Core" 3.2 (2012/Nehalem). Dates sold, processor type, memory info, hard drive details, price and more.

everymac.com

Apple -might- do other BTO options again with cheaper setups.

How would that cheaper BTO option look like? Even smaller SSD? Less RAM? No GPU?

The only option would be to be able to buy a raw system, even without Apple SSD and add your own CPU, GPU, RAM and SSD. But that Vega II is optimized for this machine, also the SSD. Therefore we don't even have that much flexibility. We have to buy a modular system with main parts from Apple. If not, you're not using the system's optimized features.

 

[goMac]

Intel Xeon W-3223 Octa-core (8 Core) 3.50 GHz Processor - OEM Pack CD8069504248402: Microprocessors - COLAMCO.com

Buy Intel Xeon W-3223 Octa-core (8 Core) 3.50 GHz Processor - OEM Pack CD8069504248402 at COLAMCO.com: INTEL® XEON® W PROCESSORS Intel® Xeon® W processors deliver optimized performance for the needs of mainstream workstation professionals. Hardware-enhanced workload performance, se...

www.colamco.com

What about here?

Again, launched does not necessarily equal mass availability with Intel.

Sure, you can find one individually. Last I checked you can't find them at HP at though. It's hard to say that Apple is behind in launching this when HP hasn't launched either. It's not like all the PC vendors have launched their new Xeons and Apple is just way behind.

 

[jinnyman]

Based on the way it goes, I think Apple is not yet seriously doing the stuff required for them to do if they were to be considered as a serious business computer company. They should open up its secrecy on the plan and expected schedule for their 'real" professional model.

 

[fendersrule]

And allow nVidia. I will bet never again for MacOS.

 

[ct2k7]

How exactly would you want them to release a product then? Do you want there to be a separate keynote spoken in every language? And on top of that, they are making it in Texas lol. If they want to say Fall, then its up to the other countries to translate it to what makes sense to them. Just stay on your thailand website (which does state "later this year"), and let this American company talk how it wants.

the website is translated or localised, and they all say the equivalent of fall or Autumn in my case

 

[Mago]

They should appear to the systems as two GPUs. It isn't like an eGPU in a substantive way at all. If had two modern PCI-e v3 ( or v4 ) GPU cards stuffed into a legacy MP 2010-2012) they'd appear as two GPUs at reduced (from native) bandwidth; there is nothing particularly "external" about that at all.

GPUs interconnected with AMD's infinity fabric appears as a single logic GPU, this is for instinct mi50/60 shouldn't be different for Vega II s. FYI.

 

[Mago]

And allow nVidia. I will bet never again for MacOS.

The biggest loser with Apple's move to takeaway nVidia have been precisely AMD, not nVidia, not Intel, nVidia has a very strong dominance on the hpc/stem/so market, meanwhile Intel having it's own strategy is moving along fpga and s new GPU architecture build around sycl/mkl as API, further nVidia is developing an compute "appliance or peripheral" running on risc-v and GPGPU cores (like s Jetson on steroids) both Intel and nVidia are playing hard to promote it's API, meanwhile AMD efforts in ROCm still don't catch traction, besides few botched supercomputers no one in research neither production is using it seriously, if the hip compiler where compatible with metal (or vice versa) maybe ROCm would have been a chance having macOS as r&d platform and Linux as production platform, even while AMD id doing good with cpu, with regard compute accelerators npu/tpu, they don't have a strong committed strategy, beyond rumours on the ngc-ii being optimized for compute (as nVidia did long ago) there is nothing more from AMD (even apple relies on arm IP for Ax's neural coprocessor)

 

[danwells]

Anybody for whom Windows is acceptable would have gone Z-series long ago, while the trashcan sat on the market. The ~$1000 or less difference between a base Mac Pro and an equivalent Z-series (ignoring ridiculous 4-core Zs with 8 GB of RAM and a hard drive as a boot device) is chicken feed to these companies.

We don't know what Apple's going to charge for non-base processors, but they have a huge advantage over the Z8 (and Z6?) in what they can charge. By being single-processor only, they're using Xeon-W. The Z8 is stuck with Xeon-SP, which carries a 300% price premium for the same processor to enable multiprocessing. Even single processor Z8 configurations use the SP line.

That 28-core where we're afraid Apple will only offer the $7000 high memory capacity version (when there's a $4000 version capped at a gigabyte)... HP's stuck with a $20,000 high-memory version because they're using SP. They do also offer the $10,000 1 GB version - still a notable price premium over the $7000 high-memory Xeon-W, and a huge premium over the $4000 base Xeon-W.

 

[wallysb01]

Anybody for whom Windows is acceptable would have gone Z-series long ago, while the trashcan sat on the market. The ~$1000 or less difference between a base Mac Pro and an equivalent Z-series (ignoring ridiculous 4-core Zs with 8 GB of RAM and a hard drive as a boot device) is chicken feed to these companies.

We don't know what Apple's going to charge for non-base processors, but they have a huge advantage over the Z8 (and Z6?) in what they can charge. By being single-processor only, they're using Xeon-W. The Z8 is stuck with Xeon-SP, which carries a 300% price premium for the same processor to enable multiprocessing. Even single processor Z8 configurations use the SP line.

That 28-core where we're afraid Apple will only offer the $7000 high memory capacity version (when there's a $4000 version capped at a gigabyte)... HP's stuck with a $20,000 high-memory version because they're using SP. They do also offer the $10,000 1 GB version - still a notable price premium over the $7000 high-memory Xeon-W, and a huge premium over the $4000 base Xeon-W.

? but with those DP machines you get two processors... two 6230s gets you 40 cores for about $4K (yes it’s a bit less than 2K each). Top end turbos are 4GHz. For a bit more, $4.5K, on Xeon W, the 3275 gets you 28 cores and a top end turbo of 4.4GHz.

Being stuck to Xeon W is not an advantage. Particularly for high RAM tasks. Now I need crazy high density DIMMs to keep up with duel socket boards that have 24 DIMMs vs the 12 in the new Mac Pro.

 

[defjam]

We don't know what Apple's going to charge for non-base processors, but they have a huge advantage over the Z8 (and Z6?) in what they can charge. By being single-processor only, they're using Xeon-W. The Z8 is stuck with Xeon-SP, which carries a 300% price premium for the same processor to enable multiprocessing. Even single processor Z8 configurations use the SP line.

At Z8 price points and configuration levels I doubt any cost savings of using a W series processor over that of a SP series is of concern. The Z8 was intended for the highest levels of workstation computing and therefore the lack of a single socket CPU option is a non-issue. If price is a concern then there's always the Z6 which officially support the W series Xeons.

 

[ssgbryan]

Anybody for whom Windows is acceptable would have gone Z-series long ago, while the trashcan sat on the market. The ~$1000 or less difference between a base Mac Pro and an equivalent Z-series (ignoring ridiculous 4-core Zs with 8 GB of RAM and a hard drive as a boot device) is chicken feed to these companies.

Not necessarily. I didn't get serious about jumping to windows until Apple actually showed the 7,1.

 

[wallysb01]

Not necessarily. I didn't get serious about jumping to windows until Apple actually showed the 7,1.

I jumped to Linux in early 2013 when the cMP was behind then.

 

[goMac]

GPUs interconnected with AMD's infinity fabric appears as a single logic GPU, this is for instinct mi50/60 shouldn't be different for Vega II s. FYI.

It does not, at least in macOS. Apple has already documented this.

 

[deconstruct60]

GPUs interconnected with AMD's infinity fabric appears as a single logic GPU, this is for instinct mi50/60 shouldn't be different for Vega II s. FYI.

Using which drivers? If the drivers are different then the expectation that they have to be 100% has problems. Other platforms supported Crossfire/SLI for several generations. macOS hasn't been a part of that. Apple is probably going to "walk before they run" with evolutionary changes to macOS driver stack.


At WWDC Apple covered the new Metal Features for "Pro Apps" ( including the Inifintiy Fabric peer group transfer API extensions added. ).

Clipped from the transcript there.
"...
Then I'll review a few Metal peer groups transfer strategies, and some of these are already in use today by Pro App developers.

And finally I'll walk you through an example use case to show how Infinity Fabric Link can help unlock challenging workflows. So let's start by looking at the transfer rates for our key connection points. ..."


If there was just one logical (opaque ) GPU then there really wouldn't be much of a need for multiple strategies of putting together peers. You'd stick in your hardware and it would automagically work as a single GPU.

Again from the transcript for the same presentation.

"...
Now let's take a quick look at how to detect multi GPU configurations.

Metal exposes device properties to identify the location, location number, and max transfer rate for each device. This information could be used to say determine the fastest possible hosted device transfer device. It can also be used to determine if a device is integrated, discreet, or external or even low power or headless. "...

Why would there be a need for an App to detect the multiple GPUs if the API was hiding them as just one logical device ?

Metal's general approach is to hand more work to the app developers. This allows more customization , but it also more work ( unless using someone else's API to mask some of that off . another Metal path which is support more 3rd party libraries that play that role ( game engines , cross platform frameworks, etc. ) . )


P.S. MI50/60 only have Linux OS support. Limiting them to just one allows AMD to put time and effort into the "one virtual" GPU component to their driver stack. It seems likely though there are at least a mechanism to provide hints of how to balance the data or it would be curious how well relatively blindly automagically doing balance would work on a wide variety of workloads.

 

[Mago]

It does not, at least in macOS. Apple has already documented this.

You mean this?

Metal for Pro Apps - WWDC19 - Videos - Apple Developer

Metal is the platform-optimized graphics and compute framework at the heart of GPU acceleration on Apple platforms. Learn key aspects of...

developer.apple.com

Actually you can either use each GPU independently even transfer data among GPU by infinity fabric with specific commands, or link multiple GPUs on s single workload and let both infinity fabric and GPU work as a single device , it's named "metal peer group API".

What I see metal 2 is boring complicated compared to CUDA, let's see how it evolves with metal 3. (Metal is moreless s sycl knockoff, while sycl it's an blatant CUDA knockoff .
[automerge]1572802382[/automerge]

Using which drivers? If the drivers are different then the expectation that they have to be 100% has problems. Other platforms supported Crossfire/SLI for several generations. macOS hasn't been a part of that. Apple is probably going to "walk before they run" with evolutionary changes to macOS driver stack.


At WWDC Apple covered the new Metal Features for "Pro Apps" ( including the Inifintiy Fabric peer group transfer API extensions added. ).

Clipped from the transcript there.
"...
Then I'll review a few Metal peer groups transfer strategies, and some of these are already in use today by Pro App developers.

And finally I'll walk you through an example use case to show how Infinity Fabric Link can help unlock challenging workflows. So let's start by looking at the transfer rates for our key connection points. ..."


If there was just one logical (opaque ) GPU then there really wouldn't be much of a need for multiple strategies of putting together peers. You'd stick in your hardware and it would automagically work as a single GPU.

Again from the transcript for the same presentation.

"...
Now let's take a quick look at how to detect multi GPU configurations.

Metal exposes device properties to identify the location, location number, and max transfer rate for each device. This information could be used to say determine the fastest possible hosted device transfer device. It can also be used to determine if a device is integrated, discreet, or external or even low power or headless. "...

Why would there be a need for an App to detect the multiple GPUs if the API was hiding them as just one logical device ?

Metal's general approach is to hand more work to the app developers. This allows more customization , but it also more work ( unless using someone else's API to mask some of that off . another Metal path which is support more 3rd party libraries that play that role ( game engines , cross platform frameworks, etc. ) . )


P.S. MI50/60 only have Linux OS support. Limiting them to just one allows AMD to put time and effort into the "one virtual" GPU component to their driver stack. It seems likely though there are at least a mechanism to provide hints of how to balance the data or it would be curious how well relatively blindly automagically doing balance would work on a wide variety of workloads.

We watched the same, yes you have to first detect multiple GPU and then you have multiple strategy to leverage the workload, among them are "metal peer groups", I bet as metal evolves this strategy will summarize on automated peer groups or something alike as CUDA did (again: long ago).

The problem with apple and GPGPU is they are just following CUDA, don't lead neither introduce nothing New, just another redundant proprietary API.