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

[ManuelGomes]

Stacc, that lineup feels good, maybe a later option with Vega?
A lot of people is tying DP1.3 to Skylake which is nonsense really. TB3 doesn't support DP1.3 anyway, maybe it will in the future. The problem with the new TB3 display as far as I see it, is that Apple will want (and should) to have a 10b HDR display, which makes sense right now really, but that requires more bandwidth. DP1.4 introduces visually "lossless" compression but is this really a solution?
I'm betting on an all USB-C solution this time for the nMP. Like with all port type changes, first you get mad about the adapters, but then as it becomes common it becomes a blessing.

Aiden, I stand corrected, thanx. I knew what controller was but never looked up the specs. Seems constrained to me.

 

[Mago]

The above diagram does not agree with what Anandtech found by mapping the PCIe complexes.

View attachment 633855​

They put the USB 3.0 controller on the PCH.

In any event, there are four unused PCIe 2.0 lanes in the MP6,1. Either four from the PEX, or three from the PEX and one from the PCH. (Notice the unused PCH lane on your diagram.)

FL1100LX using PCIe x1.

I concede you the point, this leaves 4 PCIe2 ports unused 3 from the PEX one from PCH C602 (weird) this must be due the peculiar board arrangemnts, the I/O backplane having the TB2 and the USB3 its ilogicall they to wire the FL1100LX to the PCH having one very close.
[doublepost=1464793976][/doublepost]

I could also see a GPU lineup looking like:

D310: Polaris 11 XT, 4 GB VRAM, 7 TFLOPS total
D510: Polaris 10 cut, 8 GB VRAM, 11 TFLOPS total
D710: Polaris 10 XT, 16 GB VRAM, 12-13 TFLOPS total

I figured this arrangement earlier, hope we where wrong, Or Apple decide to launc Earlier the nMP only with D310 (polaris pro 10 4GB x2) and D510 (polaris xt 8 GB x2). and introduce later a D710 based on Vega.

...There is a rumor on the front page that says the thunderbolt display is in short supply in Apple stores. Hopefully it means we are getting a 5k thunderbolt 3 display at WWDC! Something strange to me is several websites have reported on this but when they speculate on a 5k display they think its impossible because skylake doesn't support displayport 1.3. It seems pretty clear to me that a 5k display would be driven by 2 displayport 1.2 streams via a single thunderbolt 3 cable. ....

DP1.3 isnt mandatory for 60Fps 5K, look at https://blog.startech.com/post/thunderbolt-3-the-basics/ there are plenty info on TB3 powering 5K displays (MST)

Macbook pro leaks show 4 USB-C connections and no standard USB-A connectors. If the mac pro follows suit than it will need all the extra PCIe bandwidth they can get if they do something like 10 USB-C connectors.

Alpine Ridge only supports 2 TB3 ports, the Skylake has 20 PCIe3 lines, either they are using 2 USB-C + 2 TB3, or they are ditching dGPU for 4 TB3 (and 12 PCIE3 lines unused)

 

[deconstruct60]

...

Alpine Ridge only supports 2 TB3 ports, the Skylake has 20 PCIe3 lines, either they are using 2 USB-C + 2 TB3, or they are ditching dGPU for 4 TB3 (and 12 PCIE3 lines unused)

Skylake ( the processor) has x16. The associated chipsets ( 100 series ) can have up to 20. 20+16 (36) is more than enough to provision x16 GPU , 2 x4 TBv3 and , x4 SSD and x1 Wifi/Bluetooth ( 16+8+4+1 = 29 )

http://arstechnica.com/gadgets/2015...-pcie-3-0-ssds-and-other-skylake-supplements/


If Apple picked at H170 , Z170 , or Q170 there is a budget. If they wanted to spread the load and take a mild GPU hit they use the z170/q170 and split the x16 into 2 x8 with the GPU and SSD over there.


In Mac Pro context though there is about zero need for shorter USB slots because the case is getting "even thinner". Likewise the push to cover up a TB/USB socket with power-to-the-wall duties probably doesn't carry over from the MBP. On a stationary desktop system, that is beyond silly. A dedicated plug works just fine and provides a substantially wider power range.


5 TB controllers ( for 10 sockets ) is just plain ridiculous. Just plain out of control OCD that has exceedingly little to do with providing utility.

 

[ManuelGomes]

How many displays can I run at a time over a single Thunderbolt 3 connection?
A single connection can either support one 4K display (4096 x 2160) 30-bit @ 120 Hz, one 5K display (5120 x 2880) 30-bit @ 60 Hz with up to a bandwidth of 18Gbps of data downstream and 22Gbps upstream, or two 4K displays each (4096 x 2160) 30-bit @ 60 Hz with up to a bandwidth of 8Gbps of data downstream and 22Gbps upstream.

 

[Mago]

Skylake ( the processor) has x16. The associated chipsets ( 100 series ) can have up to 20. 20+16 (36) is more than enough to provision x16 GPU , 2 x4 TBv3 and , x4 SSD and x1 Wifi/Bluetooth ( 16+8+4+1 = 29 )

http://arstechnica.com/gadgets/2015...-pcie-3-0-ssds-and-other-skylake-supplements/


If Apple picked at H170 , Z170 , or Q170 there is a budget. If they wanted to spread the load and take a mild GPU hit they use the z170/q170 and split the x16 into 2 x8 with the GPU and SSD over there.


In Mac Pro context though there is about zero need for shorter USB slots because the case is getting "even thinner". Likewise the push to cover up a TB/USB socket with power-to-the-wall duties probably doesn't carry over from the MBP. On a stationary desktop system, that is beyond silly. A dedicated plug works just fine and provides a substantially wider power range.


5 TB controllers ( for 10 sockets ) is just plain ridiculous. Just plain out of control OCD that has exceedingly little to do with providing utility.

Thanks Dec. bad math, I see 20 PCIe3 lines, 8 for cpu, 8 remaining but I miss account each TB3 header needs only 4 lines, 2 Alpine Ridge actually fits the 8 available PCIe3 lines...., plus remaining 4 on the NVMe [bad math corrected]

Thanks Dec

 

[ManuelGomes]

The rumored 10 ports are not all TB. It was rumored to be 6 TB3 ports and 4 USB (also type -C) 3.0/3.1.
I personally would leave the GPUs at x16, use 2 TB3 controllers on the CPU and keep the SSD on the PCH.
Ideally a NVMe PCIe3 x4 would be used, a lot faster, but you can't have it all.
Could be better to use switch(es) for the remaining 8 CPU lanes and hook up all 3 TB3 controllers and the SSD. PCH PCIe2 lanes would be used for dual GbE, WiFi and BT. USB3 from the PCH as well.
[doublepost=1464802115][/doublepost]Those 100 series PCH, only Z170 actually, have 20 PCIe 3 lanes and the processor has extra 16, these are not subtracted from the total. But of those 20 lanes on the PCH some are hardwired, not available to assign functions as per FlexIO. Still, enough to get 2 TB3 controllers and SSD.

 

[Mago]

The rumored 10 ports are not all TB. It was rumored to be 6 TB3 ports and 4 USB (also type -C) 3.0/3.1.
I personally would leave the GPUs at x16, use 2 TB3 controllers on the CPU and keep the SSD on the PCH.
Ideally a NVMe PCIe3 x4 would be used, a lot faster, but you can't have it all.
Could be better to use switch(es) for the remaining 8 CPU lanes and hook up all 3 TB3 controllers and the SSD. PCH PCIe2 lanes would be used for dual GbE, WiFi and BT. USB3 from the PCH as well.
[doublepost=1464802115][/doublepost]Those 100 series PCH, only Z170 actually, have 20 PCIe 3 lanes and the processor has extra 16, these are not subtracted from the total. But of those 20 lanes on the PCH some are hardwired, not available to assign functions as per FlexIO. Still, enough to get 2 TB3 controllers and SSD.

some lines from the PCH can be re-assigned as you dont need native USB3 suuport, neither Sata storage, Z170 allows these deals, somewhere at Intel there is an diagram on that.

 

[koyoot]

I assume those are respectively the 4gb(199$) and 8gb(229$) versions.

I figure the nMP7,1 to have a slightly underclocked version of the 4GB model rebranded ad D310, and the 8GB version Rebranded as D510, introducing Vega as D710 later as the Darkent source commented, I would preffer to have an lower end 8GB Vega as D510, lets see what Apple thinks.
[doublepost=1464785903][/doublepost]
You're wrong Aiden, take your time to read the article from Arstechnica on the MP6,1 it details onlu 1 PCIe 2 line was left unused.

All of this can be moot, because Apple can troll us all and use two options: Fiji Pro and Fiji XT in MP. Fiji Pro as D510 as base model option and Fiji XT as D710 option, for 600 bucks more.

 

[Mago]

I was at lunch when I fast read Dec, post then I gen into a mess.

I see, now:

http://www.intel.com/content/www/us/en/chipsets/mainstream-chipsets/hm170-qm170-chipset-brief.html

Mobile Skylake PCH HM170 and QM170 only Allows 16 PCIe lines,

What can you allocate on 16 PCIe lines?

at X4 2 Alpine Ridge + 1 NVMe + (1 Wifi/Bt + 1 Touch ID)

of course, some brands actually distribute laptops with PEX-switches and everithing loaded.

I bet there will be no dGPU on the next macs... only Intel's and the new Thunderbolt Display will be the one with dGPU on board.

Now back to nMP7,1

 

[Stacc]

Alpine Ridge only supports 2 TB3 ports, the Skylake has 20 PCIe3 lines, either they are using 2 USB-C + 2 TB3, or they are ditching dGPU for 4 TB3 (and 12 PCIE3 lines unused)

I believe its not uncommon to run mobile GPUs at 8x PCIe 3.0 partially to save on power requirements and the performance is such that it would not see a measurable penalty.

5 TB controllers ( for 10 sockets ) is just plain ridiculous. Just plain out of control OCD that has exceedingly little to do with providing utility.

Agreed. I don't think we would see 5 TB 3 controllers. If Apple wants all their macs to have USB-C I could see a USB-C only machine. I would rather see some combination of USB-A, USB-C/TB 3 and TB 2 though. I am sure there would be more than a few mac users who would not be thrilled to buy a $100 TB 2 to TB 3 adapter.

I bet there will be no dGPU on the next macs... only Intel's and the new Thunderbolt Display will be the one with dGPU on board.

Maybe, maybe not. Apple is pushing GPUs as a way to accelerate certain tasks on OS X. While certainly a minority of portable macs have a GPUs it at least makes sense to keep it as a way to encourage development for this class of apps. Plus, 14/16 nm GPUs like Polaris 11 should have great performance compared to even skylake iris graphics. Who knows, maybe within a year or two AMD can make a competitive APU that includes an efficient CPU, fast GPU and HBM all on the same package.

 

[koyoot]

Mago: http://9to5mac.com/2016/06/01/apple...goes-out-of-stock-may-feature-integrated-gpu/

Have you not discussed this, already?

 

[AdamSeen]

Mago: http://9to5mac.com/2016/06/01/apple...goes-out-of-stock-may-feature-integrated-gpu/

Have you not discussed this, already?

It makes sense for laptops that are at the same desk each day. But not sure how this would fit when using Mac Pro lineup.

However, the eGPU is an inevitability (either apple or third party). It would make more sense if you can slot into it, take it away with you and upgrade separately to the monitor, rather than being completely integrated in to it. You're not going to have the option to take a monitor with you, it's useless for mac pro setups and would be a waste to replace your monitor that has a lifespan of over 5 years, just to upgrade the GPU with a 2-3 year lifespan.

 

[goMac]

However, the eGPU is an inevitability (either apple or third party). It would make more sense if you can slot into it, take it away with you and upgrade separately to the monitor, rather than being completely integrated in to it. You're not going to have the option to take a monitor with you, it's useless for mac pro setups and would be a waste to replace your monitor that has a lifespan of over 5 years, just to upgrade the GPU with a 2-3 year lifespan.

The most I could see is a low end GPU that just provided some sort of boost, not an actual gaming level GPU or anything. Mostly for MacBooks and MacBook Airs. Just enough to draw on the display in 2D. That removes the need to worry about GPU performance or upgrades.

But if they are considering it, it does possibly clear the way for external GPUs. It wouldn't surprise me if stuff like Asus's external Thunderbolt GPUs worked out of the box, assuming the drivers for the GPUs themselves are there. But it would create a cottage industry of external boxes that you could add your GPUs to. Places like LaCie and OWC would be all over that. I don't really expect Apple to wade into that themselves.

 

[pat500000]

I wonder how Apple is gonna talk about these rumors within 2 hrs of keynote Especially with Mac lineups (if they will).
Apple could probably troll us by talking about how much "green saving Heros" they are.

Obviously OS X 12 or macOS will be discussed and next iOS .

They would talk about how much products they sold so far which would take time from 2 hrs keynote.

How will Apple be able to talk about Mac Pro and other Mac products if they are limited in time to discuss all this?

It's probably just me but something just doesn't add up...unless they extend the time or give us the "one more thing" (I loved that moment).

For the moment, looks like amd is pretty good for what it is: performance/price...

 

[AidenShaw]

For the moment, looks like amd is pretty good for what it is: performance/price...

Remember when the PowerMac and MacPro were all about performance, and not performance per dollar or performance per watt ?

The ATI announcements turned out to be a big yawn. "It's quite a bit slower than Nvidia, but only $199."

 

[pat500000]

Oops I meant per dollar...
[doublepost=1464815189][/doublepost]

Oops I me

Remember when the PowerMac and MacPro were all about performance, and not performance per dollar or performance per watt ?

The ATI announcements turned out to be a big yawn. "It's quite a bit slower than Nvidia, but only $199."

i meant to say in dollars.

Ati was boring and barely anyone clapped.

 

[s-hatland]

i, too, was a bit let down by AMDs offering. was really hoping they had something up their sleeve. however, if 200-230 bucks can get my windows machine into the VR world, it makes buying a vive or rift a little easier to swallow. for a mac pro, on the other hand, would much prefer nvidias current offerings. it doesn't add up, nvidia wins on performance (and watts?). apple must have major beef with nvidia or a major vested interest in AMD if they don't at least offer any green GPUs this go around. but, whichever way they go (hopefully both), we should at least [finally] be VR capable on the Mac side of things, no?

 

[AidenShaw]

...but, whichever way they go (hopefully both), we should at least [finally] be VR capable on the Mac side of things, no?

First, you need hardware that's capable. Currently, only the MP6,1 with D700s seems to be at least close. (It's technically under the performance threshold, but there are reports that running Windows it can drive the headsets.)

Second, you need the software - unless you want to pay the Apple hardware tax only to run Windows for the software.

So the short-term answer will be "No, unless you run Windows". If Apple shows some serious hardware and a commitment to maintain it - then the software will come over to the Apple side.

Don't hold your breath, though. If Apple comes out with another nonexpandable system with some ATI cards that barely qualify - the VR software people won't assume that it will be less that 4 years before the next update.

 

[koyoot]

i, too, was a bit let down by AMDs offering. was really hoping they had something up their sleeve. however, if 200-230 bucks can get my windows machine into the VR world, it makes buying a vive or rift a little easier to swallow. for a mac pro, on the other hand, would much prefer nvidias current offerings. it doesn't add up, nvidia wins on performance (and watts?). apple must have major beef with nvidia or a major vested interest in AMD if they don't at least offer any green GPUs this go around. but, whichever way they go (hopefully both), we should at least [finally] be VR capable on the Mac side of things, no?

Bottom end for VR is 3.5 TFLOPs of compute power, and Asynchronous Compute capabilities on the hardware level. It is important because Context Switching without it creates latency and your performance tanks.

Second part of this is you need software with as low intervention from Drivers to the hardware and application as you can, because that creates latency, again. It is for the Application to drive the hardware with help of API. Not the drivers telling the application and the hardware what to do.

 

[AidenShaw]

Bottom end for VR is 3.5 TFLOPs of compute power, and Asynchronous Compute capabilities on the hardware level. It is important because Context Switching without it creates latency and your performance tanks.

So those people doing VR on Nvidia cards before Pascal weren't doing VR? And is that 3.5 TFLOPs in FP16, FP32 or FP64?

I never trust any statement that you need "3.500202 or more MegaMuggles per MicroFortnight" to do X. The real world isn't that precise.

Second part of this is you need software with as low intervention from Drivers to the hardware and application as you can, because that creates latency, again. It is for the Application to drive the hardware with help of API. Not the drivers telling the application and the hardware what to do.

Drivers always tell the hardware what to do. That's why they're called drivers.

Drivers never tell the application what to do, or they wouldn't be drivers.

Maybe you meant to say something meaningful, but you failed.

 

[spaz8]

So what are the odds that a NMP 7,1 will mean a trickle down GPU upgrade for the NMP 6,1? (i.e. 310,510,710).

 

[koyoot]

So those people doing VR on Nvidia cards before Pascal weren't doing VR? And is that 3.5 TFLOPs in FP16, FP32 or FP64?

I never trust any statement that you need "3.500202 or more MegaMuggles per microfortnight" to do X. The real world isn't that precise.


Drivers always tell the hardware what to do. That's why they're called drivers.

Drivers never tell the application what to do, or they wouldn't be drivers.

Maybe you meant to say something meaningful, but you failed.

All of what I have said is constantly told by game developers. Ask them if you do not believe. Go to game dev forum. Read what they say. I told technical reasons behind VR, that are brand agnostic.

Ask devs. why Low-Level APIs came to life. Ask them what they will use for VR games.

You described situation where you have Hardware Scheduling. Nvidia GPUs do not Hardware Scheduler, they rely on Static Scheduling over the CPU to schedule tasks. Therefore, the higher amount of cores on CPUs, the better for Nvidia GPUs, and it is already seen. Technically both of us are correct.

 

[AidenShaw]

So what are the odds that a NMP 7,1 will mean a trickle down GPU upgrade for the NMP 6,1? (i.e. 310,510,710).

Zero.

Apple's business model is to sell a new one, and send the older model to eWaste.

The MP6,1 hardware architecture also makes it difficult. The sole internal SSD plugs into one of the graphics cards. If the MP7,1 has PCIe 3.0 NVMe SSDs - will the connectors be compatible?

If the MP7,1 puts a second SSD in the gaping hole over the other GPU - will the connectors be compatible?

T-Bolt has the incredibly stupid design of putting display signals on the PCIe expansion bus. Will T-Bolt 3 be compatible with anything?

 

[Mago]

Mago: http://9to5mac.com/2016/06/01/apple...goes-out-of-stock-may-feature-integrated-gpu/

Have you not discussed this, already?

I'm aware Apple patented this idea time ago, but I dont believe they 'll build an monitor with integrated GPU, means too many compromises, and actually isn't neccesary for Mac Pro, iMac neither High end Macbook Pro, At least I dont believe Apple to launch only an Thunderbolt Display with integrated eGPU, at least they 'll launch two versions, a regular Thunderbolt display 5K, and such eGPU Display.
[doublepost=1464823473][/doublepost]

So what are the odds that a NMP 7,1 will mean a trickle down GPU upgrade for the NMP 6,1? (i.e. 310,510,710).

50:50

 

[Peace]

Who's to say Apple won't go with the new 6-core i7 as the base system ?