Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

Zarniwoop

macrumors 65816
Aug 12, 2009
1,038
760
West coast, Finland
Actually. Apple used in MP 500W PSU but advertised it as 450, for obvious reasons. So you may be right about that. It could be advertised as 520W PSU, not 580. So not 130W more for 2 GPUs. Only 70W more. 35W more headroom for GPUs putting into 165W range.

Fiji from S9300X2 and Radeon Pro Duo has 150W of TBW for each GPU and 850 MHz core clock...
I think that GPU's capable of DP1.3 or better are the only viable for nMP v2... so no Fiji.

And if 150W is the max TDP, then
Polaris 10 Pro and XT
Under clocked GTX 1080
 
Last edited:

Mago

macrumors 68030
Original poster
Aug 16, 2011
2,789
912
Beyond the Thunderdome
About the question on plugging a TB3 device on a USB-C only port, most TB3 cable I've seen comes with a blue light that brights when plugged into TB3 ports.

What's means 4 TB3 ports only on the nMP.

Each TB3 can be splitted into 2 TB2, so you can attach 12 TB2 peripheral to a single TB3 port.

Also implies NO PEX-Switch required if apple using PCIe2 NVMe's (and Darknet guy said it to have two of them).

Also implies a single GB Ethernet and USB3 Based Wifi/BT.

DG also insist all ports capable to deliver at least 4K video, I think some video channels ports should be shared either with TB3 ports or side-by side.
 

jonisign

macrumors regular
Jul 7, 2007
153
16
If the 1630 v4 base is 3.8 and Turbos up to 4.0 then there has been some progress. Otherwise, Intel is screwing the 4 core options pretty hard.

Yeah this is a good point. At least it isn't a downgrade... the IPC increase will make these CPUs faster at least than Ivy Bridge. Why do these 4 core CPUs have a 140W TDP if they are only clocked 3.7-3.8 GHz? You can purchase 4 core desktop CPUs with higher clocks and lower TDP.

I've heard that thees E5-16xx chips are actually unlocked and can be overclocked when supported by the motherboard. Has anyone poked around in the previous Mac Pro EFI to try anything like this out?
[doublepost=1464870793][/doublepost]
I think that GPU's capable of DP1.3 or better are the only viable for nMP v2... so no Fiji.

Lets seriously hope that Apple does not use DP1.2 GPUs–– the Mac Pro would be stuck at 4K/60Hz for the next 3 years!
 

Mago

macrumors 68030
Original poster
Aug 16, 2011
2,789
912
Beyond the Thunderdome
About PSU Power and CPU/GPU

I Think Apple Will adopt Dynamic Throtling installing High powered GPU (at 220W Max) but throtling as required to keep the set below 520 W.

When working on GPU-only compute both could be at full power while the CPU throthled to keep.

Also when using only a single GPU fully throtling the unused allows full CPU/GPU1.

Mixed load would mean general system throtling.
 

deconstruct60

macrumors G5
Mar 10, 2009
12,493
4,053
The guy from darkenet insist, the next mac pro to be all usb-c 4 TB3 only, also two NVMe, all ports capable to delivery 4K video, 5K only thru TB3.

' all usb-c 4 TBv3 only' for any flavor of USB or somewhat looney (has 1 port MacBook on the OCD brain ) 4 ports only on the Mac Pro. The latter would sure sign the lunatics are running the asylum.


Next Macbooks also to have mixed TB3 / USB3 only USB-C .

TBv3 means there is USB3 present. The USB3 controller aspect is built into the TB controller hardware.




only D710 GPU to be a real compute/pro gpu.

"real compute" .... What? the others only do imaginary number compute. "Real" as in has significant FP64 improvements? ( if the others scale up on FP32 that is still quite 'real' for a substantive number of workloads). Or 'real' in the only one worthy of bragging rights ( "can't innovate my ... " fodder ).



nMP 7,1 available later Q3, next Macbook pro Available earlier in August.

So neither one really needs WWDC stage time.

Only legacy non-retina iMac will be available with spinners hdd, every new Mac to be SSD only, including next mac mini, no more fusion drives also,

Either Apple is radically changing their SSD pricing markups or ...... good luck with that at the low end of the market.
 

Hellhammer

Moderator emeritus
Dec 10, 2008
22,164
582
Finland
Either Apple is radically changing their SSD pricing markups or ...... good luck with that at the low end of the market.

I suspect Apple is fully switching to their in-house PCIe NVMe SSD controller that is already being used in the 12" MacBook (and a derivative of it in iPhone/iPad). That would cut their costs quite a bit given that NVMe drives are still in short supply, which keeps the pricing higher compared to just buying plain NAND and combining that with an in-house controller.
 

Mago

macrumors 68030
Original poster
Aug 16, 2011
2,789
912
Beyond the Thunderdome
Yeah this is a good point. At least it isn't a downgrade... the IPC increase will make these CPUs faster at least than Ivy Bridge. Why do these 4 core CPUs have a 140W TDP if they are only clocked 3.7-3.8 GHz? You can purchase 4 core desktop CPUs with higher clocks and lower TDP.

I've heard that thees E5-16xx chips are actually unlocked and can be overclocked when supported by the motherboard. Has anyone poked around in the previous Mac Pro EFI to try anything like this out?
[doublepost=1464870793][/doublepost]

Lets seriously hope that Apple does not use DP1.2 GPUs–– the Mac Pro would be stuck at 4K/60Hz for the next 3 years!

Even current Mac Pro can use a 5K/60p display, using dual DP1.2 on MST, on TB3 you dont need two cables to do dual dp1.2 MST,

E5-16XX use to be highly overcolcked by HF Stock Traders, even HPE sells a proliant ready for single E5 cpu with special cooling, of course you' ll never see this on a legit mac.
[doublepost=1464872394][/doublepost]
' all usb-c 4 TBv3 only' for any flavor of USB or somewhat looney (has 1 port MacBook on the OCD brain ) 4 ports only on the Mac Pro. The latter would sure sign the lunatics are running the asylum.

Hi Dec, the 4xTB3 on the MP sure will rise debate, but given the whole system provides actually 8 TB2 ports, Apple has an argument.

All TB3 cables I've seen yet have an blue ring ligth when plugged to TB3 as witnes is properly plugged, I think this situation was foreseen by Intel long ago. (I'm not sure this to be a STD)

TBv3 means there is USB3 present. The USB3 controller aspect is built into the TB controller hardware.

?? the Macbooks Pro will have 2 TB3 capable to drive a 5K display as well 2 more USB-C capable to drive a 4K Displar non-thunderbolt, I see logic on provide Non-Thunderbolt USB-C ports, as a safe in case no-tb3 hardware is available, also saves 4 PCIe lines.

Ahh, also keep USB only ports its an saavy way to load USB recovery, and installers, turning arround any possible TB3 interference with recovery or system install.

"real compute" .... What? the others only do imaginary number compute. "Real" as in has significant FP64 improvements? ( if the others scale up on FP32 that is still quite 'real' for a substantive number of workloads). Or 'real' in the only one worthy of bragging rights ( "can't innovate my ... " fodder ).

4TFlops FP64 will put some Mac Pros on Research Desks again.

So neither one really needs WWDC stage time.
Agree.
Either Apple is radically changing their SSD pricing markups or ...... good luck with that at the low end of the market.

SSD prices are falling ludicrous fast, even seems 32GB will be the base for iPhone 7, at some time Apple has to give up with their 100$/16GB pricing.
[doublepost=1464873703][/doublepost]
I suspect Apple is fully switching to their in-house PCIe NVMe SSD controller that is already being used in the 12" MacBook (and a derivative of it in iPhone/iPad). That would cut their costs quite a bit given that NVMe drives are still in short supply, which keeps the pricing higher compared to just buying plain NAND and combining that with an in-house controller.

Neither Samsung nor Sandisk have "own" SSD controllers, most high-end SSD use ssd controller from Marvell, Silicon Motion, Micron, It's an higly specialized part that involves tricky patents etc, I douibt Apple to take the mess and high risck to buid it's own SSD Controller, what they could do (as surely does) is to license the a design just to integrate into their products as the ARM processors.
 
Last edited:

Hellhammer

Moderator emeritus
Dec 10, 2008
22,164
582
Finland
Neither Samsung nor Sandisk have "own" SSD controllers, most high-end SSD use ssd controller from Marvell, Silicon Motion, Micron, It's an higly specialized part that involves tricky patents etc, I douibt Apple to take the mess and high risck to buid it's own SSD Controller, what they could do (as surely does) is to license the a design just to integrate into their products as the ARM processors.

Samsung does everything in-house (NAND, DRAM, controller, firmware...). SanDisk uses Marvell and Silicon Motion. Micron uses Marvell and Silicon Motion, although they may have an in-house controller after their acquisition of Tidal last year. Intel uses in-house in enterprise and third party in client.

Apple already has their own SSD controller, which as I said is being used in the 12" MacBook. Hence it would make sense for Apple to expand the usage of that controller to other Macs as well.

http://www.anandtech.com/show/9136/the-2015-macbook-review/8
 

zephonic

macrumors 65816
Feb 7, 2011
1,314
709
greater L.A. area
The 6850K cost $600+ . The E5 1620-1630 v3 cost $292-372. Apple isn't going to "eat" the $200+ ( plus 30% mark up) between those two. Unless Intel bumps the base 1600 series up to 6 cores and keeps the price the same .... Apple is probably going to pick the new 1620-1630 v4 that is the same price as the 1620 v2 they have in the base systems now.
Intel is increasing the prices anyway.Doubtful Apple is going to "double down" on that start off $200+ higher still in the component line up.

Yeah, makes sense. A bit of wishful thinking on my part, I guess...
 

ManuelGomes

macrumors 68000
Dec 4, 2014
1,617
354
Aveiro, Portugal
Skylake-W (yes, no more -EP for 1S systems) will use Kaby Lake PCH, like I predicted long time ago. Basin Falls platform, different from Purley for 2S+ systems.
Great specs, that would be awesome for nMP, but still too far off!! :-(
Finally PCIe 3 on the PCH.
Same socket though, 2011-3.
Intel-Skylake-W-Platform-Details-635x583.png


Also, -E will be -X for Skylake and Kaby Lake will also be -X for the desktop.

Intel's roadmap is getting so messed up...

28 cores, 2667 DDR4, 48 PCIe3 lanes - insane!!
Compatibility with the still to come CannonLake-W
[doublepost=1464902842][/doublepost]Benches of P100 out, not gaming benches of course :)
Also, mobile 10x0 parts will be the same as desktop, maybe lower clocks for 30W lower TDP.
 
Last edited:

spaz8

macrumors 6502
Mar 3, 2007
492
91
... For what we're doing (machine learning and AI), fast FP16 is much more interesting than FP64.


Sorta off topic, but I'm curious, are you training models on the mac? and if so are you using OpenCL? .. all the API's I know of are CUDA based, currently with OpenCL maybe on the roadmap. Which is too bad cuz I could use the 14x speed up over the CPU.
 

ManuelGomes

macrumors 68000
Dec 4, 2014
1,617
354
Aveiro, Portugal
Mago, I sure hope those rumors are true about the nMP.
4 TB3 ports on the remaining 8 lanes would be sweet, the other 6 ports would be the PCH's USB3 using USB-C of course :)
But 2 SSDs on the 8 PCH PCIe 2 lanes is not possible, you always need 1 lane per GbE port - we had this discussion before and dec pointed it out correctly. So, 2 for dual GbE - I doubt they'll go only 1 port now. Maybe the 2nd SSD will be slower on 2 lanes only?!
Or Apple will wait another year and have it all on board with Skylake-W :)
All accounted there will be 68 PCIe lanes in the platform, 48 from the CPU and another 20 on the PCH.
That would be awesome, dual GPU x16, dual SSD x4, dual 10GbE, 6 TB3 ports and some to spare.
Another year of waiting is crazy though...
 

Mago

macrumors 68030
Original poster
Aug 16, 2011
2,789
912
Beyond the Thunderdome
Looks like we might be waiting a little longer for the MacPro as 9to5 Mac reckon WWDC is mainly a software event.
http://9to5mac.com/2016/06/02/apple-wwdc-no-external-display-software/
I mean the MacPro isn't a huge fanfare stage hogging intro so it could still happen but this started to make me a little concerned. 9to5 have quite a good track history which doesn't help matters!
Fingers crossed they are wrong.
Read te fine print, they clean hands ONLY on Thunderbolt Display with eGPU rumor, the same time they contition: "mostly...."

We have at least one reliable source on the nMP being announced next WWDC, also the new MBPs. Other sources as "DarkentGuy" never predicted TB Display for WWDC only as a possibility and had never handle info on monitor withs eGPU yet.
[doublepost=1464904711][/doublepost]
Mago, I sure hope those rumors are true about the nMP.
4 TB3 ports on the remaining 8 lanes would be sweet, the other 6 ports would be the PCH's USB3 using USB-C of course :)
But 2 SSDs on the 8 PCH PCIe 2 lanes is not possible, you always need 1 lane per GbE port - we had this discussion before and dec pointed it out correctly. So, 2 for dual GbE - I doubt they'll go only 1 port now. Maybe the 2nd SSD will be slower on 2 lanes only?!
Or Apple will wait another year and have it all on board with Skylake-W :)
All accounted there will be 68 PCIe lanes in the platform, 48 from the CPU and another 20 on the PCH.
That would be awesome, dual GPU x16, dual SSD x4, dual 10GbE, 6 TB3 ports and some to spare.
Another year of waiting is crazy though...

The PCIe lines match if you discard using a discrete Ethernet controller instead the one provided by the PCH, the drawback is the MP 7,1 will have only one Gb Ethernet, they will overcome the cricits based on the inmediate availability at no cost of the ultra-fast Thunderbolt 3 Ethernet, (and actually its an unicorn a nMP plugged on both eth using Link Aggregation )
 

AidenShaw

macrumors P6
Feb 8, 2003
18,667
4,677
The Peninsula
Sorta off topic, but I'm curious, are you training models on the mac? and if so are you using OpenCL? .. all the API's I know of are CUDA based, currently with OpenCL maybe on the roadmap. Which is too bad cuz I could use the 14x speed up over the CPU.
Our new systems have 72 cores (144 threads), 1 TiB of RAM and 4 Titan X cards.

They're not Apples. And yes, there's a CUDA library or package for whatever we need.
 

fuchsdh

macrumors 68020
Jun 19, 2014
2,028
1,831
Looks like we might be waiting a little longer for the MacPro as 9to5 Mac reckon WWDC is mainly a software event.
http://9to5mac.com/2016/06/02/apple-wwdc-no-external-display-software/
I mean the MacPro isn't a huge fanfare stage hogging intro so it could still happen but this started to make me a little concerned. 9to5 have quite a good track history which doesn't help matters!
Fingers crossed they are wrong.

WWDC is more often than not just a software event (and certainly that's its focus) but they've released some kind of hardware there almost as often.
 

flat five

macrumors 603
Feb 6, 2007
5,580
2,657
newyorkcity
WWDC is more often than not just a software event (and certainly that's its focus) but they've released some kind of hardware there almost as often.

9to5 isn't saying "WWDC is generally a software event so don't expect to see much hardware"
they're saying this particular event -> WWDC 2016
..is going to be highly focused on software and light on the hardware.

"sources say that WWDC this year will be a “software year”, indicating the show will be light on new hardware announcements in general. "

(of course, this doesn't mean there will be zero hardware shown... it's just that many rumors have seemed to imply we will see a whole lot of new hardware at wwdc2016.. mac pro, mbp, mini, tbolt display, etc.
when in reality, we might see only one or two of those things)
 

Mago

macrumors 68030
Original poster
Aug 16, 2011
2,789
912
Beyond the Thunderdome
9to5 isn't saying "WWDC is generally a software event so don't expect to see much hardware"
they're saying this particular event -> WWDC 2016
..is going to be highly focused on software and light on the hardware.

"sources say that WWDC this year will be a “software year”, indicating the show will be light on new hardware announcements in general. "

(of course, this doesn't mean there will be zero hardware shown... it's just that many rumors have seemed to imply we will see a whole lot of new hardware at wwdc2016.. mac pro, mbp, mini, tbolt display, etc.
when in reality, we might see only one or two of those things)

they are just pushing back the risk of being m0r0ns with predictions they cant rely, the TrashCan MacPro was a big surprise on 2013, no leak (sice it was US Made), 9to5mac , iMore has asian sources where its hard to enforce NDA, not the same as the on state side.
 

pat500000

Suspended
Jun 3, 2015
8,523
7,515
Looks like we might be waiting a little longer for the MacPro as 9to5 Mac reckon WWDC is mainly a software event.
http://9to5mac.com/2016/06/02/apple-wwdc-no-external-display-software/
I mean the MacPro isn't a huge fanfare stage hogging intro so it could still happen but this started to make me a little concerned. 9to5 have quite a good track history which doesn't help matters!
Fingers crossed they are wrong.
Who knows...maybe phill will talk about the nMP and will tell it will be available on October or something. I'm tired of Apple having self abused mentality.
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.