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

[Mago]

Again...

Mostly we know about 2016(hopefully) nMP 7,1 are personal assumptions and some "anonymous leaks".

Most rumors point to Introductions at WWDC with availability not earlier than Q4'16

• Should be Based on Intel C612 Server or X99 Workstation/ProSumer Chipset.
  
  
• GPUs x2 most leaks points to AMD GPU (Polaris and Vega) very few forum speculations still believe possible an nVidia based nMP.
  
  
• Thunderbolt 3 upgrade it's a logical evolution, as USB-C and HDMI2.
  
  
• CPUs Xeon E5v4 family most likely few rumors account on AMD Zen.
  
  
• Storage, it's also assumed as logical evolution either 1x NVMe on PCIe3 or 2xNVMe on PCIe2 2.5 GBps total throughput...
  
  
• RAM upto 256 GB possible as supported by C612 on 4 RDIMM DDR4 ECC Slot.
  
  
• Performance as Xeon E5v4 should be utpo 630GFlops compute FP64 (CPU only E5-2699v4).
  
  
• Performance on Upcoming GPUs speculatively should be (on each GPU) among 5.5Tflop FP32 (Polaris) to 9 TFlop FP32 (Vega) FP64 performance should be from 500GFlop to 4-5 TFlop.

The purpose of this thread is to discuss on possible configurations also share news and discussion on the probable nMP 7,1 components.

Everybody is welcome.

 

[AdamSeen]

Thanks for restarting the thread, Mago.

Any expectations on the GPU memory - HBM or GDDR? It wouldn't be a surprise if they held off for HBM2, if they are aiming for a 3 year refresh cycle.

 

[Stacc]

Good roundup Mago. Its probably wishful thinking but hopefully we can keep this thread focused on discussing future hardware for the mac pro.

Broadwell-E reviews are out today. This is important to the mac pro as these chips share similar configurations and performance to the Xeon E5-16XX line (which is Broadwell-EP).

Performance looks pretty good. Incrementally faster with lower power consumption. Another big win is that it looks like some of the higher core count chips still have fairly high turbo boost such that single threaded tasks don't suffer much performance loss compared to the lower core count chips. This is a common complaint with configurations like the 12 core mac pro where single threaded performance suffers significantly.

Thanks for restarting the thread, Mago.

Any expectations on the GPU memory - HBM or GDDR? It wouldn't be a surprise if they held off for HBM2, if they are aiming for a 3 year refresh cycle.

AMD is expected to release Polaris within the next month or so and its widely believed to be based on GDDR5 and GDDR5X. These are supposed to be "mainstream" parts so HBM is likely too expensive for this price bracket. AMD's vega chip expected to land late this year or early next year will have higher performance and be based on HBM.

AMD is holding a press event tonight so we should find out more about Polaris and maybe Zen. Its probably more of a teaser event since there is an NDA in place until the end of june so we likely won't see reviews on the actual cards until then.

 

[ManuelGomes]

OK, another thread blocked thanks to some people that keep firing up discussions.
Let's hope this time they stay clear.

That pretty much sums it up. Another month for Polaris based cards and Vega possibly in October.
All is geared towards a Late 2016 nMP.
We could see Polaris 10 (capped) with GDDR5 as D310, with GDDR5X as D510 and Vega 10 as D710, hopefully.
It's weird Intel hasn't announced 1600 v4 yet, 2600 is out for a while, -E just released but 1600 is MIA. They should be mostly based on LCC die, but maybe could also be "defective" MCC as well.
These should be the ones going in the nMP.

 

[Mago]

OK, another thread blocked thanks to some people that keep firing up discussions.
Let's hope this time they stay clear.

That pretty much sums it up. Another month for Polaris based cards and Vega possibly in October.
All is geared towards a Late 2016 nMP.
We could see Polaris 10 (capped) with GDDR5 as D310, with GDDR5X as D510 and Vega 10 as D710, hopefully.
It's weird Intel hasn't announced 1600 v4 yet, 2600 is out for a while, -E just released but 1600 is MIA. They should be mostly based on LCC die, but maybe could also be "defective" MCC as well.
These should be the ones going in the nMP.

E5-16XX is on roadmap, should be announced before WWDC.

I bet on Koyoot's theory on D310 being a Polaris XT on GDDR5 at 4GB, and D510/D710 both based on Vega 10 and HBM2 4/8GB each.
[doublepost=1464708242][/doublepost]

AMD's vega chip expected to land late this year or early next year will have higher performance and be based on HBM.

AMD is holding a press event tonight so we should find out more about Polaris and maybe Zen. Its probably more of a teaser event since there is an NDA in place until the end of june so we likely won't see reviews on the actual cards until then.

Few week ago was leaked AMD to hurry Vega 10 launch to Q3, from this we speculate the nMP 7,1 logically should wait for Vega10/HBM2 and launch on Q4.
http://videocardz.com/59808/amd-vega-gpu-allegedly-pushed-forward-to-october

 

[koyoot]

If you want to know what is important for MP7.1 read this article from Anandtech: http://www.anandtech.com/show/10387/hsa-11-specification-launched-multi-vendor-support

 

[Draeconis]

I still think that it's more likely to see dual Fiji than Polaris / Vega, given how Mac Pro releases have historically been 2 years behind GPU wise.

With this timing though the MacPro7,1 could well use an SM961, since they're available apparently in the 'second half of 2016'. 3,200MB/s read, yes please.

CPU is a little more hard to determine. Historically Apple have released some products with Intel chips that are brand new, been invented specifically for the form-factor or haven't even been announced yet. Whatever it is, hopefully we get enough PCIe Lanes.

 

[Mago]

My own speculation on nMP7,1 buses configuration:

1st no PEX-Switch:

Having 40 PCIe3 lines available and 8 PCIe2 from CPU and PCH respectively, the NMP could be wired as this:

GPUs 1 on x16, 1 on x8 (both having an independent interconnect fabric)
SSD: 1 NVMe on PCIE3 x4 (or 2 on PCIe2 x4)
Thunderbol 3 Alpline Ridge, 3x PCIe3 x4
4 PCH's USB3.0 on USB-C connetors
HDM2 wired to GPU1 either on a dedicated line or sharing TB3-1 DP signal.
2 10GbT taking the final x4 PCIe3
alternatively, with only 2 TB3 its possible to Wire both GPUs to x16 PCIe2, this way:


GPUs 2 on x16 (no independent interconnect fabric)
SSD: 2 NVMe on PCH's PCIe2 x4)
Thunderbol 3 Alpline Ridge, 2x PCIe3 x4
6 PCH's USB3.0 on USB-C connetors
only a singe GB Nic
10GbT network taken fro TB3's 10GbT integrated MAC.

This is my favorite non PEX.

With an PEX-91xx, either these configurations are possible:

Only 8 PCIE3 lines switched from PEX ( 8 pcie3 to 5 PCIe3 x4)
GPUs 2 on x16 cpu's PCIe3
SSD: 1 NVMe on PCIE3 x4 from PEX
Thunderbol 3 Alpline Ridge, 3x PCIe3 x4 from PEX
4 PCH's USB3.0 on USB-C connetors
2 10GbT taking the final x4 PEX PCIe3

Alternatively, all 40's CPU's Lines Wired to 1 or 2 PEX switches (as done by Asrock on they x99 WS Mobo), Bandwidth management done by PEX's logic.
this one is my Fav PEX- based nMP 7,1.

 

[Stacc]

Few week ago was leaked AMD to hurry Vega 10 launch to Q3, from this we speculate the nMP 7,1 logically should wait for Vega10/HBM2 and launch on Q4.
http://videocardz.com/59808/amd-vega-gpu-allegedly-pushed-forward-to-october

This is one of those rumors that I put in the "I will believe it when I see it" category. GPUs tend to get pushed back not forwards. Not saying its impossible, just that I would be surprised.

I still think that it's more likely to see dual Fiji than Polaris / Vega, given how Mac Pro releases have historically been 2 years behind GPU wise.

Its a fair point but Fiji isn't a good fit for the mac pro. Its expensive, has a relatively high TDP (even the nano is 175 W), is very VRAM limited, and has limited double precision compute capabilities. Apple chose Tahiti not because it was a generation old, but because it offered the best balance of performance per watt at their target TDP.

Polaris (and maybe Vega) fit the bill pretty well here in that they should be efficient chips, compute targeted chips. Personally I hope they just go with Polaris as I don't want to wait until the end of the year.

CPU is a little more hard to determine. Historically Apple have released some products with Intel chips that are brand new, been invented specifically for the form-factor or haven't even been announced yet. Whatever it is, hopefully we get enough PCIe Lanes.

Something to keep in mind here is that the one time Apple got a special chip is that the chip itself was the same as what was announced or soon to be announced it was just packaged on a slightly smaller PCB. Intel doesn't design custom CPUs for Apple. The best we can hope for is that Apple gets first dibs at a new chip. Something along the lines of the quad core skylake chip with iris graphics or maybe the broadwell-ep workstation chips.

My own speculation on nMP7,1 buses configuration:

1st no PEX-Switch:

Having 40 PCIe3 lines available and 8 PCIe2 from CPU and PCH respectively, the NMP could be wired as this:

GPUs 1 on x16, 1 on x8 (both having an independent interconnect fabric)
SSD: 1 NVMe on PCIE3 x4 (or 2 on PCIe2 x4)
Thunderbol 3 Alpline Ridge, 3x PCIe3 x4
4 PCH's USB3.0 on USB-C connetors
HDM2 wired to GPU1 either on a dedicated line or sharing TB3-1 DP signal.
2 10GbT taking the final x4 PCIe3
alternatively, with only 2 TB3 its possible to Wire both GPUs to x16 PCIe2, this way:


GPUs 2 on x16 (no independent interconnect fabric)
SSD: 2 NVMe on PCH's PCIe2 x4)
Thunderbol 3 Alpline Ridge, 2x PCIe3 x4
6 PCH's USB3.0 on USB-C connetors
only a singe GB Nic
10GbT network taken fro TB3's 10GbT integrated MAC.

This is my favorite non PEX.

With an PEX-91xx, either these configurations are possible:

Only 8 PCIE3 lines switched from PEX ( 8 pcie3 to 5 PCIe3 x4)
GPUs 2 on x16 cpu's PCIe3
SSD: 1 NVMe on PCIE3 x4 from PEX
Thunderbol 3 Alpline Ridge, 3x PCIe3 x4 from PEX
4 PCH's USB3.0 on USB-C connetors
2 10GbT taking the final x4 PEX PCIe3

Alternatively, all 40's CPU's Lines Wired to 1 or 2 PEX switches (as done by Asrock on they x99 WS Mobo), Bandwidth management done by PEX's logic.
this one is my Fav PEX- based nMP 7,1.

These all seem reasonable with the exception of I don't think we will see 10 GbT this generation. Its still not a standard feature of any X99 motherboard so until Intel natively supports it this will be a thunderbolt accessory.

There is a new rumor with leaked specs of Polaris. Basically 5.5 TFLOPS with 2304 stream processors. This is fairly underwhelming even for a chip that is supposed to be mainstream. For the general retail market the only way this is a big winner is if they price it something like $250-$300 since I think even Nvidia's GTX 1070 would probably beat it definitively.

I am still clinging on to hope that Apple will get exclusive access to the full 2560 stream processor Polaris chip which would push its compute performance > 6 TFLOPS. It wouldn't be unheard of since Apple basically got exclusive use of full Tonga for roughly a year. If this were the case a new Polaris based mac pro would have 12-13 TFLOPS GPU performance compared to the current 7 TFLOPS which would be pretty good.

 

[beaker7]

I don't see them integrating 10GbE and dedicating their limited PCI-E lanes to them. On a machine that tries as hard as it can to push connectivity external, that is the ideal thing to push out.

 

[Bubba Satori]

AnandTech tests Broadwell-E.
http://www.anandtech.com/show/10337...x-6900k-6850k-and-6800k-tested-up-to-10-cores

Stuff one of those and a 1070 in a Shuttle size box and call it the Mini Pro.
Bazinga.

 

[rockyromero]

Mostly we know about 2016(hopefully) nMP 7,1 are personal assumptions and some "anonymous leaks".

Most rumors point to Introductions at WWDC with availability not earlier than Q4'16

The mood for an announcement has shifted from hopeful to more likely.

What is unknown is the pricing.

More than likely, would the pricing be in line with what we have already?

 

[deconstruct60]

Broadwell-E reviews are out today. This is important to the mac pro as these chips share similar configurations and performance to the Xeon E5-16XX line (which is Broadwell-EP). ...

One non technical point is that the price points seems to be climbing higher.

" ... The i7-6850K and i7-6800K are priced at $617 and $434 respectively, marking an increase in price in order to get onto the HEDT ladder from previous generations. ... "
http://www.anandtech.com/show/10337...x-6900k-6850k-and-6800k-tested-up-to-10-cores

Since Apple isn't likely to "eat the costs" that probably means a trickle down increase in price for the Mac Pro's too. Also since the entry model is knee-capped with 28 PCI-e v3 lanes, seems likely the E5 1620-1630 will be capped at 4 cores also ( hopefully with higher turbo range than 1630 v3. )

I suspect this is partially because Intel fully expects the 4 core max, "lots of max single thread drag race" folks just to buy the Core i7 x700K series anyway ( i7 6700K ) . The "enthusiasts" who have a high need for the 'latest, greatest' tech will want to be following the micro arch updates on faster pace anyway.

AMD Zen

Intel cranking the prices higher does give AMD more breathing room. AMD could just do Intel's now 'old' price points and still make decent money.

Seems a bit doubtful that Zen is really going to compete with E5 1600 v4.

https://semiaccurate.com/2016/05/22/38688/
One of the 5/22 updates put the L3 cache size at 8MB per quad cluster (2MB/core). That's relative to 2.5MB /core of the Xeon E5's.

I suspect it is going to be closer to the Core i7 x700 series; only will scale to higher core counts.

AMD is expected to release Polaris within the next month or so and its widely believed to be based on GDDR5 and GDDR5X. ....
AMD is holding a press event tonight so we should find out more about Polaris and maybe Zen. ...

Sigh the last thread locked up due to GPU fanboy flame wars.. If there are a high number of GPU component posts again here then extremely likely to get the exact same results.

There isn't good movement, leak, or otherwise as to which way Apple is going. So just give this particular component a rest. It does exceedingly little to invite another round of the fanboy flame wars.
[doublepost=1464715991][/doublepost]

My own speculation on nMP7,1 buses configuration:

....
GPUs 2 on x16 (no independent interconnect fabric)
SSD: 2 NVMe on PCH's PCIe2 x4)
Thunderbol 3 Alpline Ridge, 2x PCIe3 x4
6 PCH's USB3.0 on USB-C connetors
only a singe GB Nic
10GbT network taken fro TB3's 10GbT integrated MAC.

...

TB3 integrated 10GbT MAC ? Err, where is this documented?
Previous descriptions suggested an emulated 10GbE connection. Not specifically that is was base T specific physical implementation.

It would be cheaper for Apple just to deliver 10GbE over the same Type-C connector using for the the TB ports.
[ Probably going to take another CPU/Chipset iteration to get to 10GbE as default on Mac Pro's. More PCI-e v3 lanes and it will be easier to do. Meanwhile same old 3rd party add-on vendors will work. Maybe Apple will do a TBv3 dongle that is affordable. ]


Also somewhat doubtful that Apple will do Type-C connectors for plain USB 3.0. Higher risk of confusion between TB and USB 3.0 only ports if look exactly the same. [ Apple has never bought into the different color inserts for different rated USB ports. Either all went to same standard or stayed the same. ] Yeah Ive & Company OCD drama might go with the make-them-all-the-same route but also going to stick the users with more potential for confusion. Also pissing even more folks off since folks with TB v2 will need dongles anyway. Even more USB Type A-to-C dongles isn't gong to make anyone happy..... especially just to get plain old USB 3.0 anyway. Same old speed, just gratuitously different shape. Even Apple is going to have trouble selling that kool-aid. Different shape for more functionality is one thing. Different shape and primarily useful for just Type-A devices; does what besides piss people off?

With an PEX-91xx, either these configurations are possible:
.

The PEX-91xx does not do much for a Mac Pro device. It is aimed at a different class of machines with substantially different problems trying to address.

 

[ManuelGomes]

D310 having only 4GB, although lame at this point (besides HBM 1st gen which cards come with 4GB? Excluding lower end of course) could happen, unfortunately. 8GB for Vega D510 and 16GB for Vega D710 are more reasonable.
We could get a full Tonga with 6GB (with the full 384b mem bus) as D310 but I hope not, no more 28nm tech please. I guess Apple will want to go full 14nm this time. D510 could be Polaris and D710 Vega.
I don't expect CPUs to change, Xeons it is. Prices will rise if to maintain speeds, although at same speed it should be faster. Let's see how 1600 v4 comes in terms of speed and price. 1620 v3 lowered from 3.7 to 3.5 keeping the same price range, going to 3.7 v3 costs a bit more, but performance is better. Will Apple go for price or performance? Silly question!!
I'm betting GPUs will stick to x16, the remaining x8 will be shared between 3 TB3 controllers, and possibly the NVMe SSD, although I believe the SSD will be on the PCH.
I would like to see an all USB-C ports nMP, although it would indeed be a nightmare for some when it comes to adapters. But sooner or later this will happen, so why not now? Apple ditches legacy in a snap, I must admit I like that.
10GbE might not be standard at this time, I believe they'll have a USB-C adapter for it, from a TB3 port. Having 6 ports gives you the flexibility to choose to have 1 or 2 10GbE ports, a few displays and still some spare port for anything else. Pick the used ports right and you'll hardly notice the bandwidth struggle if you're lucky. In some cases you might be constrained but I believe in most it will be alright.
[doublepost=1464718028][/doublepost]Polaris in that slide shows as DP1.4 compliant!!
The naming has changed, I assumed the X in RX would be eXtreme but guess not.
This could be a D310 but I hope it's the full Polaris 10, and Vega 10 the rest of the lineup.

 

[deconstruct60]

Most rumors point to Introductions at WWDC with availability not earlier than Q4'16.

Why intro the product 3-6 months before release in complete violation of normal Apple policy?

Last time it was primarily because of a major form factor change and while keeping the same name the product targeting was substantially changing.

If the same market with new parts .... why? All it primarily does is kill off sales now. "Buy it now because it won't be for sale anymore in 4-6 months" is different than ".... don't buy now because the current is extremely dated and once-in-a-blue-moon update is coming 4-6 months from now... " .

Given Apple's standard practices, any announcement is most likely means shipping within a month (maybe two). Weaving in components that aren't going to ship for more than Quarter is highly unlikely if tightly coupled to a WWDC announce date. Not impossible, just unlikely.

 

[Mago]

Why intro the product 3-6 months before release in complete violation of normal Apple policy?

Last time it was primarily because of a major form factor change and while keeping the same name the product targeting was substantially changing.

If the same market with new parts .... why? All it primarily does is kill off sales now. "Buy it now because it won't be for sale anymore in 4-6 months" is different than ".... don't buy now because the current is extremely dated and once-in-a-blue-moon update is coming 4-6 months from now... " .

Given Apple's standard practices, any announcement is most likely means shipping within a month (maybe two). Weaving in components that aren't going to ship for more than Quarter is highly unlikely if tightly coupled to a WWDC announce date. Not impossible, just unlikely.

Just like 3yr ago when the nMP 6,1 was introduced at WWDC aimed for Q3'13 availability (which actually ends on late q1'14 due the huge backlog), the WWDC it's a main focus event for developers and power users, those that even dont use an iPhone but a Macintosh and dont miss an WWDC while iPhone events dont worry miss..

 

[pat500000]

Based on what we know...who (customers) would they be targeting for? FCPX users or film makers? VR users? I could see Polaris as gpu.

 

[ManuelGomes]

http://wccftech.com/intel-skylake-e-lga-3647-hexa-channel-memory/

Next-next nMP?
Not really... server stuff
[doublepost=1464719974][/doublepost]Not for nMP but great anyway:

http://wccftech.com/samsungs-512gb-pm971nvme-ssd-insanely-small-lightning-fast/

 

[deconstruct60]

It's weird Intel hasn't announced 1600 v4 yet, 2600 is out for a while, -E just released but 1600 is MIA. They should be mostly based on LCC die, but maybe could also be "defective" MCC as well.

Not strange at all. The original E5 1600 rolled out about a quarter staggered after the same baseline Core i7 x9xx .
They probably aren't trying to make both variants at the same time.

The other thing is the split at v5 ( Skylake) between the multiple socket chipsets ( E5 2600 v5 ) and single socket ( E5 1600 v5 ) where Purley ( multiple ) splits off from Basin Falls ( single socket workstation). If work on Basin Falls is secondary then Intel isn't in a hurry to get v4 out because v5 is coming later ( or substantially sooner depending of just shim the consumer chipset onto the 1600 v5 line with Basin Falls or not ).
[doublepost=1464722421][/doublepost]

http://wccftech.com/intel-skylake-e-lga-3647-hexa-channel-memory/

Next-next nMP?
Not really... server stuff

If the tweak town article is right. ( instead of the second hard wccftech)

http://www.tweaktown.com/news/52336....it&utm_medium=twitter&utm_campaign=tweaktown

that is a picture of a Skylake-E motherboard. If Intel is reusing the socket pin numbers for the single socket version then perhaps not so much just server stuff. For example, get the Purley memory upgrade but not the built in Omnipath , 10GbE , FPGA , etc stuff. Just a simpler link back to Kaby Lake chipset and a subset of the top end I/O functionality. Pretty decent chance some of those pins go no where ( like the 2nd set of QPI pins for the 1600 do now. )

Would be interesting if there is some eDRAM on that package (since it is now bigger to provision more pins ).

[doublepost=1464723727][/doublepost]

Just like 3yr ago when the nMP 6,1 was introduced at WWDC aimed for Q3'13 availability (which actually ends on late q1'14 due the huge backlog), the WWDC it's a main focus event for developers and power users,

Apple pre-announces cancellations. XServe got a 4-6 month notice. So did the "old style" Mac pro three years ago. That was not primarily a new product announcement. It is was primarily we cancelled the old product one.

Every Spring on this forum folks talk about how WWDC is about "Pro users" and Apple is going to announce some new super, elite high end pro device ...... even broken analog clocks are correct twice a day. 2013 was a broken clock mark for that annual Spring ... new Mac Pro coming crowd.


Like pros would not buy the Mac Pro if announced around NAB or in November? If the product is ready and the components ship around WWDC then it could get weaved into WWDC. Otherwise probably not. There is nothing in Apple's rather consistent behavior that indicates otherwise.

E5 1600 v4 is quite late at this point. If Apple was targeting a release somewhat synchronized to v4 availability it should be ready whatever it is at this point.


What is more flawed in the current Mac Pro is more so Apple's execution of the concept rather than the design. Release and then disappear into a rabbit hole for 3 years. They could iterate on the GPUs ( and CPUs if weren't coupled to the dead end the tick-tock cycle ). If putting on a current tech parade every June was extremely criticial that's what they would have been doing for the past 5 years... they have not.

For developers new opportunities to sell more software for more revenues (and/or more efficiently ) is what of primary interest at WWDC .... not newest tech porn.


WWDC get hardware stuff typically as a tickle down from Computex where Intel/AMD/Nvidia and other major PC component vendors have released stuff to wide distribution. Apple's laptops are way past due and the non TB v3 sysetms is getting more than strange. More likely will see than Mac Pro. Mac Pro only if it has been sitting waiting on E5 v4 and current GPU to go.

 

[Mago]

Seems a bit doubtful that Zen is really going to compete with E5 1600 v4.

https://semiaccurate.com/2016/05/22/38688/
One of the 5/22 updates put the L3 cache size at 8MB per quad cluster (2MB/core). That's relative to 2.5MB /core of the Xeon E5's.

I suspect it is going to be closer to the Core i7 x700 series; only will scale to higher core counts.

Zen CPU Family its Aimed at both Enthusiast, Server and HPC Markets, according AMD coomunications and roadmap, First batch of ZEN CPUs are 8 core models aimed at Workstations/Servers/Enthusiast, having IPC comparable or better than Intel Skylake.

TB3 integrated 10GbT MAC ? Err, where is this documented?
Previous descriptions suggested an emulated 10GbE connection. Not specifically that is was base T specific physical implementation.

I see it on some PDF, while TB3 Emulated Network Mode Actually offers 20 GBPs coonection (20 up, 20dn).

PD. I'm in debt with this reference, ASAP I find again it I'll quote you again.

The PEX-91xx does not do much for a Mac Pro device. It is aimed at a different class of machines with substantially different problems trying to address.

Previously discussed, I come with an link form Anandtech, where explain the feature of such PEX switches, further the Mac pro 6,1 uses one to multiply 8 PCIE3 into 12 PCIE 2 for the older TB2 Falcon Ridge controller (3x).

The same configuration I propose is being selling by ASROCK on its X99 WS motherboard, also Gigabyte has one identical with a couple of PEX- enabilg 7 16line each PCIe3 connections concurrent (switched).

The New PEX improves its features with advanced DMA etc
[doublepost=1464725410][/doublepost]@koyoot

 

[koyoot]

There isn't good movement, leak, or otherwise as to which way Apple is going. So just give this particular component a rest. It does exceedingly little to invite another round of the fanboy flame wars.

If you would ask me, 3 possibilities:
Polaris and Vega with October releases.
GTX 1070 renamed to Quadro or Tesla.
Fiji in every possible flavor.

@Mago I was expecting more.

 

[AidenShaw]

If you would ask me, 3 possibilities:
Polaris and Vega with October releases.
GTX 1070 renamed to Quadro or Tesla.
Fiji in every possible flavor.

@Mago I was expecting more.

Tesla cards don't have graphics, so that would be OK for the compute card. Quadro cards do have graphics, so that would be OK for the graphics card (assuming that Apple keeps the compute/graphics segregation).

 

[ManuelGomes]

Yep, Basin Falls seems to even to being related to Purley - in that it's quite different that is.
And considering that there won't be Kaby Lake Xeon...

 

[Draeconis]

Its a fair point but Fiji isn't a good fit for the mac pro. Its expensive, has a relatively high TDP (even the nano is 175 W), is very VRAM limited, and has limited double precision compute capabilities. Apple chose Tahiti not because it was a generation old, but because it offered the best balance of performance per watt at their target TDP.

Polaris (and maybe Vega) fit the bill pretty well here in that they should be efficient chips, compute targeted chips. Personally I hope they just go with Polaris as I don't want to wait until the end of the year.

I may well be wrong, just seen the MR post about ACD supplies drying up. Since a possible replacement 5K ACD would almost certainly be designed around DP1.3, and Fiji only supports 1.2, I now feel Polaris/Vega is more likely.

Sadly, this could still mean OS X doesn't get Fiji drivers, but then if Polaris/Vega is available, I guess that isn't such an issue anyway.

 

[ManuelGomes]

http://9to5mac.com/2016/05/31/new-thunderbolt-display-possible/
[doublepost=1464731324][/doublepost]Maybe a new keyboard is indeed coming with the OLED function keys. MBP at least looks to go that way:
https://www.macrumors.com/2016/05/31/macbook-pro-leaked-photos-oled-touch-panel/