2016 nMP

[Zarniwoop]

It gets hotter and consumes more power and you might as well just use Intel chips? I don't really understand how anyone thinks ARM is some sort of magic bullet that will get around the same limitations Intel is hitting.

My point is that overclocking ARM / using more cores might make it useful chip for Macbook Air too. It uses 18W processor. If ARM can do same figures with 18W, then it is suitable to fill that market segment.

You see, if Apple can create 100-500 million extra profit by sacking Intel from low power computers, they'll have no pity for x86. Money is the god they trust. Follow the money... Apple is no different.

On the other hand, if they don't see it'll make profit, it wont happen. But I'm sure they'd love to keep the money they pour to Intel atm.

 

[goMac]

You see, if Apple can create 100-500 million extra profit by sacking Intel from low power computers, they'll have no pity for x86. Money is the god they trust. Follow the money... Apple is no different.

The cost for materials for ARM is much lower, but the spinup cost is much higher. Spinning up factories, making production lines to produce the chips... That's going to be very expensive, especially if Apple is trying to move production in house. It's not like they have a warehouse full of spare chips here.

It's probably a mistake to compare the materials cost of an ARM processor to an Intel processor. Once you've scaled up the design and set up manufacturing, the cost is probably more expensive. Not to mention, Intel has a lot of expertise in producing laptop and desktop scale chips that Apple doesn't. Apple doesn't even have that expertise in house yet for their iOS devices.

And if Apple wades into the ARM computer arena... ARM computers are cheap. Yes, they could generate extra profit in theory... But competing head to head with ARM devices would mean price drops.

I also just flat out don't see how an ARM chip would work well in the Mac Pro right now. There just isn't anything competitive with a 12 or 20 core Xeon in the ARM camp right now. It's kind of a silly notion.
[doublepost=1461451623][/doublepost]

This means, the code you compile it's linked with said library, very different to say that objective-c code runs on top of such runtime library, objective-c does the same thing as MS VC++ which link you code with the windows/vc++ MFC an spiritual doppelganger of the Objective-C runtime

"The document is intended for readers who might be interested in learning about the Objective-C runtime."

I don't really need to argue the point past that.

That means you don't need to care on the Objective-C runtime if you switch to Linux (a.e.) because the GNU Compiler will provide a custom implementation of the Objective-C runtime but KEEPING the API, that means you don't need to change a single line of code to interact with the library (for the purposes the library is provided only, don't include other os<->obj-c)

So you helped me to demonstrate my point, source code portability is not a problem of cpu platform.

You're mixing together operating systems and CPUs like the same thing. Talking about portability to Linux is not that same thing as talking about portability between CPUs. But regardless, the document itself even says portability to Linux is not guaranteed, so even then...

First I remember you the thread is related to migrate OSX from /x86 to something different as ARM, code portability among different OS It's another issue.

...Right?

Doing SSE AVX and other cpu optimizations it's matter of compiler switches and in few cases some "in-line" code which often it's c/c++/asm, and such in-line or pragmas well coded should be invisible as you link to other cpu platform.

https://developer.apple.com/library...nCL_MacProgGuide/Performance/Performance.html

Please take your time reading this, worth for you to educate on programming issues beyond "fast" reading or just Google on concepts you are no familiar with.

Why are you giving me an OpenCL document in response to SSE or AVX?

Even if you have in line switches, someone has to write the code for ARM that goes inside the ARM switch. There's no compiler flag that translates SSE or AVX to NEON, just as there was no magic compiler flag that translated Altivec into SSE. Like I said, this all happened before in 2005, and nothing has really changed since.

 

[Mago]

...

I don't have time to follow your arguments around to whats is fact to any proficient programmer, as your ridiculous asumptions on production cost setups, its obvious you dont have idea on how FABs "bake" ASICs, FYI the biggest cost is not in development, or "fab setup", its on IP LICENSING, not only from ARM or Imagination, a CPU use to have licensed parts from many patents, each and every one should be honored befor goint to the market.

then its scale economy, did you know R&D for A9 isnt more expensive than for its Chinese equivalent Helio X30 -despite cheaper chinese labor, cpu designers dont bill as locals-, the difference its Apple orders about 40 million A9 each to the fabs, while Mediatek orders (the same fabs)2 or 3million cpus (ok actually you order waffers, then you have to manage defective asics etc, often 1/3 to 1/10 of the products go on waste, as bigger the asic lower the yields.

I see what you do here, just move the thread where is no sense. trolling.

 

[tuxon86]

I don't have time to follow your arguments around to whats is fact to any proficient programmer, as your ridiculous asumptions on production cost setups, its obvious you dont have idea on how FABs "bake" ASICs, FYI the biggest cost is not in development, or "fab setup", its on IP LICENSING, not only from ARM or Imagination, a CPU use to have licensed parts from many patents, each and every one should be honored befor goint to the market.

then its scale economy, did you know R&D for A9 isnt more expensive than for its Chinese equivalent Helio X30 -despite cheaper chinese labor, cpu designers dont bill as locals-, the difference its Apple orders about 40 million A9 each to the fabs, while Mediatek orders (the same fabs)2 or 3million cpus (ok actually you order waffers, then you have to manage defective asics etc, often 1/3 to 1/10 of the products go on waste, as bigger the asic lower the yields.

I see what you do here, just move the thread where is no sense. trolling.

You're not even trying to make any sense now...
Your lack of knowledge in regard to cross compiling and cpu architecture is just sad.

 

[Mago]

You're not even trying to make any sense now...
Your lack of knowledge in regard to cross compiling and cpu architecture is just sad.

LOL

 

[wallysb01]

My point is that overclocking ARM / using more cores might make it useful chip for Macbook Air too. It uses 18W processor. If ARM can do same figures with 18W, then it is suitable to fill that market segment.

You see, if Apple can create 100-500 million extra profit by sacking Intel from low power computers, they'll have no pity for x86. Money is the god they trust. Follow the money... Apple is no different.

On the other hand, if they don't see it'll make profit, it wont happen. But I'm sure they'd love to keep the money they pour to Intel atm.

The cost for materials for ARM is much lower, but the spinup cost is much higher. Spinning up factories, making production lines to produce the chips... That's going to be very expensive, especially if Apple is trying to move production in house. It's not like they have a warehouse full of spare chips here.

It's probably a mistake to compare the materials cost of an ARM processor to an Intel processor. Once you've scaled up the design and set up manufacturing, the cost is probably more expensive. Not to mention, Intel has a lot of expertise in producing laptop and desktop scale chips that Apple doesn't. Apple doesn't even have that expertise in house yet for their iOS devices.

And if Apple wades into the ARM computer arena... ARM computers are cheap. Yes, they could generate extra profit in theory... But competing head to head with ARM devices would mean price drops.

What goMac said. Plus, they won't be as good as intel chips at the same power. Not in CPU power, not GPU, not feature set... no way, no how. Just for example, how are you going to get Thunderbolt 3 on ARM?

 

[Mago]

Just for example, how are you going to get Thunderbolt 3 on ARM?

Thunderbolt it's just an Pcie bridge, ARM already supports pcie, so supporting Thunderbolt (besides Intel licensing) only requires drivers for the Thunderbolt controller, actually Thunderbolt it's a way to transparently interconnect pcie peripherals so an Pcie peripheral previously supported should be supported by Thunderbolt too (alimony the vendor implemented the full pcie std which includes hot-plug required to work on TB. AMD switched few patents to gain access to Thunderbolt, maybe the only way.

 

[wallysb01]

Thunderbolt it's just an Pcie bridge, ARM already supports pcie, so supporting Thunderbolt (besides Intel licensing) only requires drivers for the Thunderbolt controller, actually Thunderbolt it's a way to transparently interconnect pcie peripherals so an Pcie peripheral previously supported should be supported by Thunderbolt too (alimony the vendor implemented the full pcie std which includes hot-plug required to work on TB. AMD switched few patents to gain access to Thunderbolt, maybe the only way.

Thanks. That was an actual question and I was hoping for an answer. I was aware that ARM has PCIe, but didn't know much about how that would get extended to TB3. And that does seem like a pretty large complication for ARM and seems like a significant price increase. After some digging, do any ARM SoCs have enough PCIe bandwidth to support TB3? I'm only seeing relatively small numbers of PCIe 2.0 lanes... is that right? Seems reasonable though, since they are really only there for the storage, correct?

 

[mattspace]

(besides Intel licensing)

Satan will be iceskating to work before intel approves TB3 for anything ARM based.

 

[goMac]

Thunderbolt it's just an Pcie bridge, ARM already supports pcie, so supporting Thunderbolt (besides Intel licensing) only requires drivers for the Thunderbolt controller, actually Thunderbolt it's a way to transparently interconnect pcie peripherals so an Pcie peripheral previously supported should be supported by Thunderbolt too (alimony the vendor implemented the full pcie std which includes hot-plug required to work on TB. AMD switched few patents to gain access to Thunderbolt, maybe the only way.

Thunderbolt requires support in the chipset. It's not impossible that they could add Thunderbolt to an ARM chipset, but as it's currently deployed, ARM doesn't work with Thunderbolt. You can't just stick Thunderbolt on any old PCI-E bus. That's why there are no cMP upgrade cards.

The A9X has PCIe, but it's also important to remember that Intel has done a lot of work on optimizing the i7's ability to push data around at high speed. I'd be surprised if the A9X has that level of optimization that is primarily only important on desktops, or if Intel hadn't patented their work pretty deeply. The Core 2 Duo's inability to push data around well compared to the i7 is one reason the 3,1 Mac Pro is finally starting to age out of use.

Again, these things maybe could be acceptable in a Macbook if you got over the ARM software compatibility issues (and Microsoft couldn't even get Adobe to port Photoshop to ARM.) If we're talking about a Mac Pro desktop it's really not a pretty picture.

 

[Mago]

And that does seem like a pretty large complication for ARM and seems like a significant price increase. After some digging, do any ARM SoCs have enough PCIe bandwidth to support TB3? I'm only seeing relatively small numbers of PCIe 2.0 lanes... is that right? Seems reasonable though, since they are really only there for the storage, correct?

TB3 asic cost about 30-40$ besides the usb-c and USB3 & display port video signals.

All pcie3 enabled ARM SOC has enough bandwidth for TB3, and much more, in fact Qualcomm is developing a Xeon Phi rival named Hydra based on 48 cores Cortex Kyro (same as the dual core snapdragon 820) delivering about 1 TFlop on 15% of the power, it's bandwidth it's so high (about 100GT/s vs 40GTs on Xeon E5v4, nVidia Pascal still lead at 400GT and actually Xeon Phi KL it's about 200GT) that requires an Fabric sourced from Mellanox to interconnect on clusters.

Calvium also is shipping it's ThunderX with 48 cores cortex a53.

The actual problem with those powerful high core count arm cpu isn't having enough gflops for compute, it's the human interaction, while its true they can deliver comparable flops to the most powerful Xeon, the truth is that a typical desktop Application would be miserably slow on those monsters, because most desktop apps relies on a single thread for its human interaction and most (if not all) it's internal operations, this is the real barrier blocking ARM from desktop computers this is something you'll never note on an iPad because the os and gui are highly optimized for slow thread and most work it's done by the gpu.

I consider arm need to twice it's Cortex A72 single thread speed before it could be considered a desktop part, this is not a problem on servers which actually are highly optimized for parallelism or multiple thread of relatively slow process.

This it's also the same issue present by High Frequency Traders, they use to build custom servers on a single cpu but highly overclocked just to reach the fastest single thread speed possible, at the point some use bare i7 instead Xeon, and some are using FPGA for this, this it's another extreme market banned for ARM.

 

[t0mat0]

Sounds like there's a nice FCP X update coming (10.3?), maybe prior to/alongside a new Mac Pro.
https://forums.creativecow.net/thread/335/88480#88598 Apple at FCPExchange at NAB this Monday as far as I can see.

 

[ManuelGomes]

Polaris targets:
http://wccftech.com/amd-polaris-10-desktop-polaris-11-notebook-gpu/

 

[Zarniwoop]

iMac discounts been around, similar like Macbook's one month ago.. is Apple emptying its stock in order to release an iMac update before WWDC.. Maybe 21" will get Skylake treatment. But I think 27" should get its update when Polaris 10 is out... is it time for iMac Pro at WWDC?

Also MBA price was dropped, without Skylake update. It is the last call for that series... it's Apples best selling Mac, so what is going to replace it.. has to be something great.

Haswell rMBP 15", Broadwell 13" MBA & rMBP... I think these will be get new design that's going to be introduced at WWDC.. hopefully with nMP.

 

[Hank Carter]

So, what do we think? New MP at WWDC or do we all move to WIN10 in July?

 

[Zarniwoop]

Hmm... what if iMac Pro would be a 135W Xeon with Polaris 11 based FirePro D310.. and then you could buy a Dual Radeon Pro extension trashcan next to it if you need... connected with TB3/USB type-c. Could Apple kill MP for that?

PS. Polaris 10 is just too perfect for nMP and iMac, that I'd believe it could happen. Now.
[doublepost=1461527043][/doublepost]

So, what do we think? New MP at WWDC or do we all move to WIN10 in July?

At WWDC we should at least hear news about nMP. If not, we can draw our conclusions.

 

[koyoot]

People think there will be only two Polaris Chips, but more and more evidence is coming up that there will be 3rd, bigger GPU. With huge surprises. AMD already has silicon of it. And what comes from this is that Vega is not exactly what people at this point think it is. I cannot give any more specifics.

 

[Zarniwoop]

But this third GPU is not coming this summer, right? Radeon Pro is not out yet, and we're already expecting its funerals this summer... something is not right in this picture. Vega = Zen?

 

[koyoot]

No. Vega is GPU.

Previously Raja said there will be only 2 GPUs from Polaris architecture. Baffin and Ellesmere. There was 3 GPU - Greenland, and was rumored to use Vega architecture with HBM2.

It will not. Vega is brand new architecture. Greenland died, and it was replaced by third GPU from Polaris. I do not have any more specifics, that I can disclose, at this point.

 

[Zarniwoop]

iMac discounts been around, similar like Macbook's one month ago.. is Apple emptying its stock in order to release an iMac update before WWDC.. Maybe 21" will get Skylake treatment. But I think 27" should get its update when Polaris 10 is out... is it time for iMac Pro at WWDC?

There are discounts for all 21" iMacs AND 27" with M380. Could it mean, that Iris Pro 580 is going to replace the 27" started model too... Iris Pro 580 is going to be quite a beast then.. UPDATE: Current Iris Pro 6200 is 883 GFLOPs and 580 is going to be 1152 GFLOPs. UPDATE2: M380 is 1152 GLOPS too.. what a coincidence.

Not that this has much to do with nMP, unless dGPU iMacs will get refreshment later along with nMP spiced up with Polaris.

 

[koyoot]

I know the source credibility, take it with gigantic grain of salt.

The AMD Polaris 10 GPU has a maximum TDP of 175W but cards will actually consume much less than that. The GPU was initially built to support HBM memory but AMD chose to go the GDDR5/X route since it offers a better value currently. We will get to see HBM on AMD GPUs when Vega launches but until then, only Fury series will have HBM support. The Polaris 10 GPU is said to have 3DMark Firestrike Ultra performance around 4000 points which is about what a Radeon R9 Fury X and GeForce GTX 980 Ti score. By 4000 points, we don’t mean exactly 4000 but it’s actually quite a bit less but that’s the number we were told.

 

[Mago]

nVidia readies too GP104 on same time frame, a 8GB GPU .

I think three are serious possibility the iMac also joins to the update party.

 

[koyoot]

nVidia readies too GP104 on same time frame, a 8GB GPU .

I think three are serious possibility the iMac also joins to the update party.

AMD will be a bit faster with release. Qualification silicon of GP104 is dated for 14th week of 2016. Which means we have at least still 3 months to release of GP104, which falls perfectly in line with information provided by Bits&Chips about release schedule of Nvidia GPUs.

 

[Stacc]

People think there will be only two Polaris Chips, but more and more evidence is coming up that there will be 3rd, bigger GPU. With huge surprises. AMD already has silicon of it. And what comes from this is that Vega is not exactly what people at this point think it is. I cannot give any more specifics.

Wishful thinking. AMD has 2 chips coming out in 2016, Polaris 10 and 11. Polaris 10 is looking like a 125 W - 175 W chip with performance between the 390X and Fury X. If Polaris 10 can achieve Fury X performance at <= 150 W this should be a great chip and perfect for the mac pro especially if its compute oriented.

It sounds like AMD knew it would be tough to make a big 14/16 nm chip so instead they went with a smaller 230 mm2 mainstream part. Nvidia is coming out with a bigger 320 mm2 part that should take the performance crown but it sounds like it will have a high cost associate with it. Given that the best selling GPU lately has been the Nvidia GTX 970, targeting a $300 price point sounds like a pretty reasonable move for AMD.

 

[Mago]

Wishful thinking. AMD has 2 chips coming out in 2016, Polaris 10 and 11. Polaris 10 is looking like a 125 W - 175 W chip with performance between the 390X and Fury X. If Polaris 10 can achieve Fury X performance at <= 150 W this should be a great chip and perfect for the mac pro especially if its compute oriented.

AMD stands Polaris Twice on Perf/Watt the preceding gpu generation, assuming that as true, given R9-FuryX delivers 8601 TF Fp32 at 275W, Polaris Elsmere XT at 150W should deliver 9.4TF fp32 and 6.3TF at 100W (the MacPro TDP) with 1:4 FP64 it its near 1.5TFlop Fp64 or ˜3Teraflop (2x) Total on the MacPro Setup, this is an Order Of magnitude leap from current D700., and close the performance of a real Workstaton