2016 nMP

[Zarniwoop]

Sure.. but we're talking about Apple here. OpenCL 2.1 sounds very logical for macOS.. but don't count on it.

Just for reference: Ask the developers, how they see OS X as a developer platform, how much they trust Apple. Where, when ever a new .1 update arrives, it might break your app. Or when a new OS version arrives, it certainly will break your app. So, once a year you are under a distress. With El Captain (non-beta), even MS Office stopped working. One of my clients still cannot update from Yosemite to El Captain, because their Music notation app (Sibelius) is not working on El Captain.

And ask from anybody who put all their apples to Cuda basket, how they feel. Ah, we've seen these guys crying out loud here since the start of the very first "Is there going to be a new Mac Pro in 2015".

Developers need a clear commitment from Apple in order to start to support something as so utterly difficult as openCL. When Apple delivered a buggy openCL, and without proper tools, it was a sign that they're not investing in to it too much.

Same time we got Swift and Metal.. all those mini OS's (tvOS etc) and now macOS... Apple is like a huge container ship on the ocean trying to turn around in full speed and losing some it's cargo in a way and the captain is counting beans and sailors drawing butterflies on its sides same time.

Eventually they will deliver, but what.. and where?

 

[antonis]

One of my clients still cannot update from Yosemite to El Captain, because their Music notation app (Sibelius) is still not working on El Captain.

Heh, don't get me started with the external usb audio interface adventures every time a new 10.x is launched.

 

[koyoot]

Zarniwoop:
I was not talking about Apple platform, but OpenCL generally .

About Apple platform I completely agree with you. It is quite funny but I was thinking about the same thing yesterday. Their hardware and what is more important software is completely outdated, and software will be outdated for at least another 4-6 months, depending when next OS version will come. At least, because we have absolutely no guarantee it will be updated with important features: HLSL, D3D12, new, better version of Metal, OpenCL 2.1, and that is only top of iceberg of problems with Apple platform.

It looks like Apple stopped completely updating their hardware and software in wait for something better to come alive. Maybe it is the case. In current state whole platform is pretty much useless for anything that is not iOS development.

Edit: Forni in his best form, as usual: http://www.bitsandchips.it/52-engli...he-company-with-the-best-marketing-department

 

[Zarniwoop]

Koyoot, that's why I've had all my imagineers lessons past weeks/months here.. Apple is on to something big. Something that need both, new software and new hardware. Are they re-inventing the whole computer concept?

 

[koyoot]

Maybe not computer concept, but computing, as a platform. That is my opinion of course. VR, mobile, smartphones, tablets, expandable computers, efficiency. All of which revolves around HSA 2.0 and... next generation HSA.

But this is only my humble opinion. I do not "know" what Apple wants to do.

 

[Zarniwoop]

https://forums.macrumors.com/threads/waiting-for-skylake-mbp-thread.1661246/page-446#post-22879652

That thread is way longer than nMP related.. turning to 4chan at the end. But at least the link goes directly to a leak about nMP v2.
[doublepost=1462879465][/doublepost]I knew that current rMBP is about to go.. Apple has pushed them to resellers with a discount along with 21.5" iMac's and the one 5k model with M380.

So, something is happening.

But those guys might disappoint, when it is the Macbook air replacement, the 14" Macbook with 15W Intel-U that comes out...

 

[Mago]

Apple distracted too many resources on Metal, metal in theory is much better for compute and rendering but lacks interoperability beyond Apple ecosystem, Apple should have adopted full Vulkan if they want to deprecate OpenCL (which depends more in driver support) also Apple should support opencl development OpenCL 2 has good toolchain in linux as Windows where debugging is not a pain in the ass as with opencl 1.x.

Apple has serious issues on how they manage the OSs.

1st no zero-day exploit bounty program, this its irresponsible.

2nd do not really enforce jailbreaking which independently on how popular is, its a cancer for security .

3rd do not allow a "developer mode" to enable users or developers do anything under the safe umbrella provided by Apple instead requiring some jailbreak sourced from dark china sites which compromises both the iDevices and everything it connects to.
OS/X has similar features limiting sandboxing on user/developer consent.

While they are doing right in Swift building a solid and modern coding tool, the entire toolchain requires attention, key areas are OpenCL despite Vulkan/metal superiority Apple should Fund its development as they fund Mozilla so they receive incentives for better support OSX also I think Metal should never ever been, instead Apple should adopt Vulkan instead just fund and watch, Metal has sense for Apple only apps, but most developer wants portability specially complex component as accelerators, I don't see Adobe coding Vulkan for premiere coding again the same component on Metal, Apple should kill metal and commit it into vulkan and adopt vulkan as well support OpenCL 2 to sponsor its much needed toolchains in OSX.

These are just few things I consider Apple should do.

 

[Mago]

Again I consider Apple computer being managed by vacuum salesman minded CEO and not real IT people with open vision on the future . That's very concerning

 

[koyoot]

If Metal is based on Mantle, and it is, then there is no point for Vulkan to be adopted by Apple. Because Vulkan is Mantle.

People say that Metal was developed for Mobile platform first. Well, Mantle was developed for every platform that supports HSA foundation. Vulkan is already on Android. And it came from Mantle. You see the big picture?

 

[Mago]

If Metal is based on Mantle, and it is, then there is no point for Vulkan to be adopted by Apple. Because Vulkan is Mantle.

People say that Metal was developed for Mobile platform first. Well, Mantle was developed for every platform that supports HSA foundation. Vulkan is already on Android. And it came from Mantle. You see the big picture?

I know both metal and Vulkan come from mantle, that's why I wrote "Apple should commit metal with vulkan", the problem with Apple us they believe Developers will only code for Apple, developer main interest is to reach the biggest audience on the lower cost, to lower cost a main priority is portable code.

Even if Apple fully opensources Metal its already dead , economics kill metal, only big budget projects can justify coding twice some components, Metal would be a niche in big name titles , small publishers has to decide develop twice o develop vulkan and have presence on Android Steam (Linux) Windows also with same code.

 

[Zarniwoop]

I think that is why macOS is predicted to get UiKit. For those small projects... and it should be easy to port between iOS and macOS.

 

[Mago]

I think that is why macOS is predicted to get UiKit. For those small projects... and it should be easy to port between iOS and macOS.

Developers actually wants to port between iOS OSX Linux Wincr@p ...


Ok, about GPUs, I just chated some newsman I meet which cover tech for a magazine , he told me they expect today some official pre-release news on Polaris or at least big "leaks" about, AMD is urged to cool down nVidia's moment, and catch audience focus before gtx1080 reviews fill media rooms instead Polaris news (both should happen at same time, that's not good anyway, so either AMD hurry and launches this or early next week its Polaris gpu or at least enoug info will leak just ti catch attention before gtx1080 reviews reaches media rooms)

 

[koyoot]

If there will be any "leak" there will be something small to put people in expectation and bring attention to them, but still remain completely silent about upcoming GPUs. Yes, thats how bad their marketing team is .

 

[Mago]

If there will be any "leak" there will be something small to put people in expectation and bring attention to them, but still remain completely silent about upcoming GPUs. Yes, thats how bad their marketing team is .

Whatever should be enough to water the extended nVidia party.

 

[goMac]

Apple distracted too many resources on Metal, metal in theory is much better for compute and rendering but lacks interoperability beyond Apple ecosystem, Apple should have adopted full Vulkan if they want to deprecate OpenCL (which depends more in driver support) also Apple should support opencl development OpenCL 2 has good toolchain in linux as Windows where debugging is not a pain in the ass as with opencl 1.x.

The problem is Apple is only willing to support open source standards they control. WebKit, LLVM, and Swift are great for Apple because Apple controls them. Apple did have a fair amount of control over the OpenCL spec, but if you look at OpenCL 2, there were a few changes that probably resulted in loss of control for Apple. My guess is that's why OpenCL 2 support is never coming, and why higher ups pushed Metal instead of OpenCL. Apple handed OpenCL over to Khronos, and then OpenCL was tweaked in ways they didn't like.

If Metal is based on Mantle, and it is, then there is no point for Vulkan to be adopted by Apple. Because Vulkan is Mantle.

Metal is not based on Mantle. If it was, we wouldn't be having this conversation. They share a few concepts, but there isn't any source of one being based on the source of another.

I know both metal and Vulkan come from mantle, that's why I wrote "Apple should commit metal with vulkan", the problem with Apple us they believe Developers will only code for Apple, developer main interest is to reach the biggest audience on the lower cost, to lower cost a main priority is portable code.

I don't think we'll see Vulkan this year, but it would be nice to be surprised.

 

[ManuelGomes]

Let's see if we get some info on Polaris today then.
It would be nice to have at least some preliminary info.

 

[pat500000]

Let's see if we get some info on Polaris today then.
It would be nice to have at least some preliminary info.

Are they announcing Polaris info?

 

[ManuelGomes]

Don't know, Mago said someone told him that they might have some info out today.

 

[koyoot]

And Mago was right. Well, sort of.

http://videocardz.com/59808/amd-vega-gpu-allegedly-pushed-forward-to-october
http://www.3dcenter.org/news/amd-zieht-den-vega-launch-angeblich-auf-oktober-2016-vor
Vega in October?

They did not pushed themselves a lot, did they?

 

[ManuelGomes]

It seems 14nm is leading, will AMD come out first this time?
So, this is why NVidia paper launched big Pascal so soon? Since they'll be last to have products ready, they thought it would be a good idea to disclose them first. Smart move.
And possibly why they pushed 1080 so high, to be closer to Vega when it ships?
[doublepost=1462904130][/doublepost]So, we can still see Vega based D710?

 

[Zarniwoop]

I think Vega is too big for nMP with its current form. If there's theoretically 130W per GPU available, Polaris 10 will be the work horse in this category. But, what prevents from making a new design for Vega.

Anyway, I feel that AMD gave false impression that Vega is a Q1/17 product to mislead Nvidia a bit. The chart was made so, that if you're a postive person, it'd be Q4/16 and Q1/17 for the rest.

 

[ManuelGomes]

I'd say Vega can be made to fit the max power envelope for a GPU on the nMP.
They did it with the current D700, why not Vega? Being a 14nm part it should scale down well to fit 110W or so.
We wouldn't see the nMP before September anyway, so the timing seems perfect. This time around we could see 2 boards (D310 and D510) baed on Polaris 10 (Po and XT) and D710 based on Vega 10 Pro probably. They could cut it down to 3072 cores, maybe slow them down and add 16GB HBM2 maybe also clocked down to keep power draw lower.
And voila, great setup.

 

[MacVidCards]

So let's sum it up.

Nvidia currently ships worlds fastest GPUs.

Nvidia starts selling an even faster GPU in 2 weeks, AMD won't even have a glimpse of the taillights.

AMD is about to release a more efficient mid level GPU. (Only people who rave for mid level power efficiency are OEMs making cheaper machines/laptops, etc)

AMD is hoping to release something fast in October.

Must seem like an eternity away to those poor bxxxxxds.

As far as nMP 7,1 goes, only suitable AMD GPU in next couple months is replacement for D300. (See previous "eternity" comment)

Don't know about you guys, seems like Nvidia will be throwing a party not to be missed.

 

[sigmadog]

Not impressed with Vega. Gonna wait for Pinto.

 

[koyoot]

I'd say Vega can be made to fit the max power envelope for a GPU on the nMP.
They did it with the current D700, why not Vega? Being a 14nm part it should scale down well to fit 110W or so.
We wouldn't see the nMP before September anyway, so the timing seems perfect. This time around we could see 2 boards (D310 and D510) baed on Polaris 10 (Po and XT) and D710 based on Vega 10 Pro probably. They could cut it down to 3072 cores, maybe slow them down and add 16GB HBM2 maybe also clocked down to keep power draw lower.
And voila, great setup.

They would not need to cut down Vega GPU to 3072 GCN cores to fit in that thermal envelope.