2016 nMP

[ManuelGomes]

Again, really?
http://wccftech.com/amd-radeon-m400-mobility-lineup-leak/
[doublepost=1461071500][/doublepost]1600 CPUs are usually 140W parts. I expect no different this time round.
[doublepost=1461071677][/doublepost]Remember 2600 series have lots of options, workstation, freq. optimized, etc

 

[koyoot]

M480 and higher looks to be Polaris .

Yeah, it was predictable that they would rebrand the lower end GPUs, again.

 

[Zarniwoop]

Intel Broadwell-EP Lineup:
SKU Name Cores/Threads Base Clock L3 Cache (LLC) TDP Price
Intel Xeon E5-2699 V4 22/44 2.2 GHz 55 MB 145W $4115 US
Intel Xeon E5-2698 V4 20/40 2.2 GHz 50 MB 135W $3228 US
Intel Xeon E5-2697A V4 16/32 2.6 GHz 40 MB 145W $2891 US
Intel Xeon E5-2697 V4 18/36 2.3 GHz 45 MB 145W $2702 US
Intel Xeon E5-2695 V4 18/36 2.1 GHz 45 MB 120W $2424 US
Intel Xeon E5-2696 V4 22/44 2.1 GHz 55 MB 150W $2321 US
Intel Xeon E5-2690 V4 14/28 2.6 GHz 35 MB 135W $2090 US
Intel Xeon E5-2687W V4 12/24 3.0 GHz 30 MB 160W $2141 US
Intel Xeon E5-2683 V4 16/32 2.1 GHz 40 MB 120W $1846 US
Intel Xeon E5-2680 V4 14/28 2.4 GHz 35 MB 120W $1745 US
Intel Xeon E5-2667 V4 8/16 3.2 GHz 35 MB 135W $2057 US
Intel Xeon E5-2660 V4 14/28 2.0 GHz 35 MB 105W $1445 US
Intel Xeon E5-2650L V4 14/28 1.7 GHz 35 MB 65W $1329 US
Intel Xeon E5-2650 V4 12/24 2.2 GHz 30 MB 105W $1166 US
Intel Xeon E5-2643 V4 6/12 3.4 GHz 10 MB 135W $1552 US
Intel Xeon E5-2640 v4 10/20 2.4 GHz 25 MB 90W $939 US
Intel Xeon E5-2637 v4 4/8 3.5 GHz 5 MB 135W $996 US
Intel Xeon E5-2630 v4 10/20 2.2 GHz 25 MB 85W $667 US
Intel Xeon E5-2630L v4 10/20 1.8 GHz 25 MB 55W $612 US
Intel Xeon E5-2623 V4 4/8 2.6 GHz 5 MB 85W $444 US
Intel Xeon E5-2620 v4 8/16 2.1 GHz 20 MB 85W $417 US
Intel Xeon E5-2609 V4 8/8 1.7 GHz 20 MB 85W $306 US
Intel Xeon E5-2603 v4 6/6 1.7 GHz 10 MB 85W $213 US

http://ark.intel.com/products/codename/38530/Broadwell#@Server

The 135W line is a good example what I mean, when one core consumption goes over 17W. For Eight core, it is quite exactly that 17W. On six core it is 22,5W. That's like 30% increase.. and what do you get for that extra 30% electricity bill? 6% more computing power. Even better (worse) with 4 core.. 34W per core. 100% increase on TDP per core. To get 0.3GHz. You see my point? For a computer like nMP, you look for the best perf/watt ratio as possible. To reach those final megahertz, you have to pay from the thermal reserve.

That is a second reason, I don't belive Apple is after more CPU cores.. it is just too expensive thermal sense. (First one is, that too few software has a need for more than eight cores..)

Most likely those 135W processors are 130W chips on 1600 series. And true, in real world the candidates for nMP.

BUT, I was playing with a different scenario, that if Apple shifts its focus on GPU's.. and keeps the old Tube. THEN you'd need to calculate very carefully the CPU's TDP per core scenario.

This is the reason why Apple needs two or three categories for nMP, if they are stubborn with the thermal and PSU needs. With four core, you could use more powerful GPU, and vice verse with eight core less powerful GPU. Those are the numbers. Apple took the middle way with current nMP so that you can mix any CPU with any given GPU combination. Better would be, that you could choose more powerful CPU, but then the most powerful GPU is not available any more. And so on.. but would it be too difficult to Apples core customers? Who don't really care anything else but that it works.

 

[ManuelGomes]

What? No announcement? Is this a sign?
http://wccftech.com/macbook-lineup-gets-refreshed-with-more-colors-and-more-battery-life/

8GB RAM (DDR3 really?), new SKL procs, new colors... same USB-C Gen 1.

 

[Zarniwoop]

What? No announcement? Is this a sign?
http://wccftech.com/macbook-lineup-gets-refreshed-with-more-colors-and-more-battery-life/

8GB RAM (DDR3 really?), new SKL procs, new colors... same USB-C Gen 1.

Apple marketing team must have demanded to save DDR4 for the Pro machines...

Yes, what a let down... DDR4 = power savings, more bandwidth.

 

[ManuelGomes]

Does this mean that there will be no changes to the rMBP MBA lineup as already speculated?

 

[netkas]

Yet they advertise DP1.2 support but their laptop can do 4k at 30 hertz max, like an old macmini 2014

so, at wwdc we will see new mac pro, ivy bridge xeon cpu, tahiti based gpus, now in rose gold

 

[lowendlinux]

I still like that little thing

 

[Zarniwoop]

Actually Skylake m-processors don't support DDR4, so... wasn't Apple's marketing team this time. ;-)
http://ark.intel.com/products/88197

BUT, Skylake m-processors DO support 4k / 60Hz. If Apple not.. now we have to look at the marketing dept.

UPDATE: Weird, Apple states they're using Core-m3 1.1GHz.. but Intel says, they have only 0,9GHz core-m3. Mistake on Apples webshop? Core-m5 is 1.1 GHz

 

[Mago]

Actually Skylake m-processors don't support DDR4, so... wasn't Apple's marketing team this time. ;-)
http://ark.intel.com/products/88197

BUT, Skylake m-processors DO support 4k / 60Hz. If Apple not.. now we have to look at the marketing dept.

UPDATE: Weird, Apple states they're using Core-m3 1.1GHz.. but Intel says, they have only 0,9GHz core-m3. Mistake on Apples webshop? Core-m5 is 1.1 GHz

those are ultra-low power 2 core (5W) you'll see only on MBr12, other new macbooks -pro (including the new mac mini) to be based on mobile "i-" non "m-" skylake (28W dual core, 45W quad core) which supports DDR4
http://ark.intel.com/products/codename/37572/Skylake#@Mobile

 

[Zarniwoop]

those are ultra-low power 2 core (5W) you'll see only on MBr12, other new macbooks -pro (including the new mac mini) to be based on mobile "i-" non "m-" skylake (28W dual core, 45W quad core) which supports DDR4
http://ark.intel.com/products/codename/37572/Skylake#@Mobile

Yes, and Apple has overclocked its core-m3 chips. Intel sells them only at 0.9GHZ. Apple sells 1.1GHz.

 

[Mago]

Yes, and Apple has overclocked its core-m3 chips. Intel sells them only at 0.9GHZ. Apple sells 1.1GHz.

At the moment I wrote this I didn't aware the updated Macbook retina just come out.

P.D. the new Macbook still sucks

 

[Stacc]

The 135W line is a good example what I mean, when one core consumption goes over 17W. For Eight core, it is quite exactly that 17W. On six core it is 22,5W. That's like 30% increase.. and what do you get for that extra 30% electricity bill? 6% more computing power. Even better (worse) with 4 core.. 34W per core. 100% increase on TDP per core. To get 0.3GHz. You see my point? For a computer like nMP, you look for the best perf/watt ratio as possible. To reach those final megahertz, you have to pay from the thermal reserve.

That is a second reason, I don't belive Apple is after more CPU cores.. it is just too expensive thermal sense. (First one is, that too few software has a need for more than eight cores..)

Most likely those 135W processors are 130W chips on 1600 series. And true, in real world the candidates for nMP.

BUT, I was playing with a different scenario, that if Apple shifts its focus on GPU's.. and keeps the old Tube. THEN you'd need to calculate very carefully the CPU's TDP per core scenario.

This is the reason why Apple needs two or three categories for nMP, if they are stubborn with the thermal and PSU needs. With four core, you could use more powerful GPU, and vice verse with eight core less powerful GPU. Those are the numbers. Apple took the middle way with current nMP so that you can mix any CPU with any given GPU combination. Better would be, that you could choose more powerful CPU, but then the most powerful GPU is not available any more. And so on.. but would it be too difficult to Apples core customers? Who don't really care anything else but that it works.

I think Apple looked at it a different way. They saw that 90% of workstations sold have a ~130 W Xeon Processor and ~250 W of GPU power + ~50 W of other (motherboard + SSD).

They looked at Intel's current and upcoming xeon lineup and saw 12+ core processors. This reduced the need of dual processors when you can get 18 cores on a single chip and that number continues to go up.

Instead of 1 big, hot GPU they divided their GPU envelop in 2 and got 2 GPUs with more computational power than one single big one. Then they talked to AMD and was told that after 28 nm GPUs they weren't going to be able to make big die GPUs. This further justifies the dual GPU approach.

Sure, this isn't the old mac pro which is all things to all people but like most Apple products, it is a well focused design that is more than suitable if you are doing work that requires a lot of CPU/GPU power. The downside is that you will probably pay more than you want because you want that 12 processor or you need thunderbolt peripherals. However, Apple has never been in the business of making a strong value proposition (at least compared to commodity hardware) so they were obviously willing to make these tradeoffs.

The CPU + GPU tradeoffs you discuss are already present in the current mac pro. Both the CPU and GPU have the capability to "boost" in that they run over their base frequency if the thermal headroom allows. Thus if you are doing a CPU limited task with idle GPUs you can get the maximum boost frequency out of the CPU. The watt per core argument doesn't quite hold up because there are shared resources on the CPU that generate a lot of heat (think L2 cache).

 

[Hank Carter]

I still have one of these in my closet, and it still works. Though the rubber cylinder wheels one the "MegaPixel" (4 color gray scale) display have merged with the glue that welded them onto the hubs, and turned into a semi-viscous tar like substance. Messy and sticky..

I also have a Cube and the monitor stashed away. It was a brilliant machine, but way overpriced and initially only monochrome, where color had become the norm.

 

[Mago]

I think Apple looked at it a different way. They saw that 90% of workstations sold have a ~130 W Xeon Processor and ~250 W of GPU power + ~50 W of other (motherboard + SSD).

They looked at Intel's current and upcoming xeon lineup and saw 12+ core processors. This reduced the need of dual processors when you can get 18 cores on a single chip and that number continues to go up.

Instead of 1 big, hot GPU they divided their GPU envelop in 2 and got 2 GPUs with more computational power than one single big one. Then they talked to AMD and was told that after 28 nm GPUs they weren't going to be able to make big die GPUs. This further justifies the dual GPU approach.

Sure, this isn't the old mac pro which is all things to all people but like most Apple products, it is a well focused design that is more than suitable if you are doing work that requires a lot of CPU/GPU power. The downside is that you will probably pay more than you want because you want that 12 processor or you need thunderbolt peripherals. However, Apple has never been in the business of making a strong value proposition (at least compared to commodity hardware) so they were obviously willing to make these tradeoffs.

The CPU + GPU tradeoffs you discuss are already present in the current mac pro. Both the CPU and GPU have the capability to "boost" in that they run over their base frequency if the thermal headroom allows. Thus if you are doing a CPU limited task with idle GPUs you can get the maximum boost frequency out of the CPU. The watt per core argument doesn't quite hold up because there are shared resources on the CPU that generate a lot of heat (think L2 cache).

I think Apple will go up to 145w cpu on the Mac Pro, and upto 110w each gpu plus an uprated psu to 500-550w and faster fan as AidenShaw suggest.

And both gpu most likely will be based on Polaris 10, and I don't discard an nVidia GPU, the problem with nVidia it's their latest mobile quadro still on Maxwell/28nm and 1/32 fp64 not the best choice for an workstation neither competitive against Polaris, an mobile or low power pascal I don't see it coming soon enough unless nVidia's skunk works surprises everybody.

 

[Stacc]

I think Apple will go up to 145w cpu on the Mac Pro, and upto 110w each gpu plus an uprated psu to 500-550w and faster fan as AidenShaw suggest.

And both gpu most likely will be based on Polaris 10, and I don't discard an nVidia GPU, the problem with nVidia it's their latest mobile quadro still on Maxwell/28nm and 1/32 fp64 not the best choice for an workstation neither competitive against Polaris, an mobile or low power pascal I don't see it coming soon enough unless nVidia's skunk works surprises everybody.

It would be nice to get a little bump in the PSU. It would add a little headroom to the CPU and GPUs. Might have a little added noise but its not the end of the world.

Given that Apple has exclusively used AMD GPUs for the last couple years and the recent rumors that AMD has gotten design wins from Apple I would say there is something like a 90% they go with AMD. But given that both AMD and Nvidia's new GPUs are supposed to land in the beginning of June it doesn't sound like either are going to come out significantly earlier than the other.

 

[askunk]

Well well well... it appears that TSMC and Samsung have confirmed that 10nm architectures will be available in H1 2017 and 7nm ready by H1 2018. Good times are coming ^__^

@

 

[goMac]

Boom.

https://www.macrumors.com/2016/04/19/amd-polaris-2016-macs/

 

[Synchro3]

Bad for native OS X driver support of future Nvidia cards…

As expected, unfortunately.

 

[Mago]

Boom.

https://www.macrumors.com/2016/04/19/amd-polaris-2016-macs/

The future of the Mac Pro is less certain, though there will certainly be suitable high-end chips from AMD manufactured on TSMC's 16nm process this year.

We still at the same point, this only reinforced the rumours on nVidia comeback to the nMP.

A major fact is mainstream software publisher as adobe readies a Cuda8 specific product targeted at OSX for VR edition, but where is that Mac with nVidia GPU capable for VR content producing? Don't look at the past, either based on pascal or maxwell I have the good feeling that the nMP L2016 will arrive on nVidia cards.

 

[fuchsdh]

We still at the same point, this only reinforced the rumours on nVidia comeback to the nMP.

A major fact is mainstream software publisher as adobe readies a Cuda8 specific product targeted at OSX for VR edition, but where is that Mac with nVidia GPU capable for VR content producing? Don't look at the past, either based on pascal or maxwell I have the good feeling that the nMP L2016 will arrive on nVidia cards.

From a business standpoint, I can certainly see Apple offering a SKU with native Nvidia support as a way of boosting adoption as well as encouraging upgrades. I guess the question is who has more leverage in the relationship and if Apple sees the extra resources and effort as worth it (and the same going for Nvidia).

As for a larger PSU in the next revision, I'd say it's possible, but who knows. The tube is already dead silent for 90% of the tasks I've ever run on it (haven't done super-long multi hour renders, though) so it's entirely possible a slightly louder fan wouldn't make any real-world difference to that very welcome element of it.

 

[koyoot]

You forget that CUDA might be possible to run on AMD GPUs, Mago. Also it might be only for... Mac Pro 5.1 with Pascal GPUs.

 

[Mago]

You forget that CUDA might be possible to run on AMD GPUs, Mago. Also it might be only for... Mac Pro 5.1 with Pascal GPUs.

Sadly this project still Vaporware IMHO.

I don't believe Adobe Build plans for a 4 year old hardware, despite how good was the cMP, those workhorse are seeing it's sundown.

 

[Stacc]

We still at the same point, this only reinforced the rumours on nVidia comeback to the nMP.

A major fact is mainstream software publisher as adobe readies a Cuda8 specific product targeted at OSX for VR edition, but where is that Mac with nVidia GPU capable for VR content producing? Don't look at the past, either based on pascal or maxwell I have the good feeling that the nMP L2016 will arrive on nVidia cards.

I still maintain that Apple does not want to tie itself to a single GPU vendor especially one who's code would only run on a small fraction of macs with discrete GPUs. Apple would rather developers leverage OpenCL and Metal and not CUDA.

Another factor is performance. AMD's cards seem to have better OpenCL and OpenGL performance while Nvidia does better in cases where they can tailor their drivers to perform well in directx11 games and CUDA. AMD GPUs are also very good at encoding video.

One of the downsides to the Mac Pro's form factor is its not easy to add in multiple vendors for the GPUs. Each one is a custom form factor which requires significant cost. With any luck this option could be added with external GPU support over thunderbolt 3.

 

[Mago]

I still maintain that Apple does not want to tie itself to a single GPU vendor especially one who's code would only run on a small fraction of macs with discrete GPUs. Apple would rather developers leverage OpenCL and Metal and not CUDA.

Another factor is performance. AMD's cards seem to have better OpenCL and OpenGL performance while Nvidia does better in cases where they can tailor their drivers to perform well in directx11 games and CUDA. AMD GPUs are also very good at encoding video.

One of the downsides to the Mac Pro's form factor is its not easy to add in multiple vendors for the GPUs. Each one is a custom form factor which requires significant cost. With any luck this option could be added with external GPU support over thunderbolt 3.

At 100w TDP Polaris 10 Elsmere Pro /XT should offer much better performance until nVidia has some pascal derived cars for lower TDP (yet announced/rumored pascal products are rated well over 150w).

The real problem here is metal, developers don't/can't wait until Metal delivers, opencl on the other hand don't offer the resources you find on cuda.

That's why the (named by koyot) Boltzmann initiative seeks an compiler that allow deploy cuda apps on OpenCL hardware (amd and maybe Xeon phi), Boltzmann compiler still Vaporware, and initially it's targeted at Linux clusters.