2016 nMP

[netkas]

One of the reasons why people need Dual CPU workstations is that using CUDA hardware - there is no hardware scheduling on hardware level.

In essence. CUDA hardware needs software to work properly, because it lacks hardware scheduling. Last GPU from Nvidia which had Hardware Scheduler was Fermi, and was hot inefficient and burnt to death. Thats why Nvidia got rid of it, because they were not able to control it properly. Kepler and Maxwell have GigaThread Engine which need drivers to work properly. It cannot adapt itself to application. If it would have Hardware Scheduling - it would. All of Scheduling is on CPU.

Oh please please, stop talking ******** and pretend you know things. Just please, stop. Seriously, stop, it's all ********.

Since kepler's GK110 nvidia has new GMU unit, it's in hardware.

Fermi was very hot mostly due to hot-clocks, when shaders run at almost twice the clock speed of rest of core.

This GMU is even able to provide Asynchronous Compute, but only with cuda. Nvidia ****ed up and made their GMU not compatible with DX12 requirements for bariers for efficient Async Compute queues.

 

[koyoot]

http://forums.anandtech.com/showpost.php?p=38057033&postcount=844
http://forums.anandtech.com/showpost.php?p=38057086&postcount=847
http://forums.anandtech.com/showpost.php?p=38057121&postcount=848

Is it enough for you? There is NO proper hardware scheduling on Nvidia hardware. It MUST USE Driver, because otherwise - it will not know what to do with application. It goes both for CUDA applications and graphics applications.

Edit. Here is the continuation: http://forums.anandtech.com/showpost.php?p=38057373&postcount=854
http://forums.anandtech.com/showpost.php?p=38057390&postcount=856

It still needs CPU and software interference into the execution.

P.S. The fact that Asynchronous Compute is able to be done through CUDA is exact emanation of what I have written. CPU and Drivers. This is main driver of scheduling on Nvidia hardware.

 

[SeaCaptainBiscuit]

I understand what you are saying, this actually has been a somewhat significant problem. Check out this picture of the AMD 390X. Look how many memory modules are around that card. For comparison, here is the D700. Basically with GDDR5 if you want to up the memory bandwidth you have to add memory modules. Its not impossible though, as Nvidia gets away with a reasonable amount of memory modules to go on the Titan X with 12 GB of VRAM.

Right, at this point the way things have played-out is that (for various reasons), the preferred design is to surround the GPU with a 1-deep "ring" of memory chips, finding a reasonable compromise between having them be closer-in (better for the electronics/wiring) and further-out (makes thermal management easier).

But that's "in hindsight". If you had to guess the designs of today from, let's say, 5 years ago, that looks like the probable outcome, but it's not entirely guaranteed; back then, you could've at least made an argument that the "right" tradeoff would be to have more chips -- maybe a "2-deep ring" -- essentially on the theory the right trade-off would have been to aim for "slower but wider" (and with more on-board memory).

Thus even though today it's obvious most of the board is basically "dead space", it'd be hard to be confident that's how it'll play out as long as video memory remains off-package.

Once you *believe* it's all going on-package in the near future, you can have a much higher confidence level that you won't really *need* most of that board area anymore, which would then make an nMP-esque design look like a lot less of a compromise.

However, I think the biggest constraint on GPUs in the mac pro is power/thermals and not PCB area. Reduced sized GPUs have existed for a long time in MXM format so the only thing different here is that Apple used their own design. Of course, it did restrict them from using obscene amounts of VRAM but Apple was never going to do that anyways. For instance, Apple did not use AMD's Hawaii GPU not because it wouldn't fit in the mac pro, but because it was too hot to get much benefit over Tahiti. (However, now looking at that picture it may because they couldn't get 8 GB of memory in there too).

Again, I don't know how to say it any more simply than this: to adopt the nMP design you have to be able to make a colorable case it's not going to force too much compromise (on the metrics of interest), and so PCB size isn't a direct issue; it's only an issue if there's functionality that fits on a full-size board but not on the scaled-down custom boards (thus MXM isn't relevant either way: it's not "can you buy a small board" it's "do we need a big board to fit all the things we'll want?"). For HBM-based systems, no worries; for GDDR, you might have capacity constraints (especially if the slow-and-wide option was what the market had gone with...).

But, yes: the biggest miscalculation is the power envelope on the nMP. That higher-end--but still non-niche--GPUs have stayed at such high power levels was very unexpected and has permanently cramped the nMP's potential.

The expectation was you could fit a top-end GPU into it--HBM eliminates any risk of an nMP case being "too small"--and in a worst-case scenario down-clock it *a little bit* to, say, shave of 20-40% of the power draw in exchange for a 10-20% performance drop.

With higher-end GPUs stabilizing at current power draws, this trade-off just isn't there, with obvious consequences.

Eh, HBM1's biggest failing is that it only appeared on a graphics card that appeals to a very small niche. Fury is only a high end gaming card, and even 4 GB of VRAM is not that much for a card that sells for >=$500. I think they designed it as a competitor to the Nvidia GTX 980 and then Fury got delayed and trumped by the 980 Ti. HBM1 is not "dead-end tech", its just the first generation of new tech. It will be interesting to see if Fiji and HBM1 live on in AMD's next graphics lineup because of how big and expensive it is.

Again, as a technology HBM1 is fine; it's only "dead end tech" if you look at it as the product of "ecosystem", but in that broader context it's a dead man walking for sure.

 

[hollyhillbilly]

One of the reasons why people need Dual CPU workstations is that using CUDA hardware - there is no hardware scheduling on hardware level.

In essence. CUDA hardware needs software to work properly, because it lacks hardware scheduling. Last GPU from Nvidia which had Hardware Scheduler was Fermi, and was hot inefficient and burnt to death. Thats why Nvidia got rid of it, because they were not able to control it properly. Kepler and Maxwell have GigaThread Engine which need drivers to work properly. It cannot adapt itself to application. If it would have Hardware Scheduling - it would. All of Scheduling is on CPU.
Thats why the more CPU cores you have the better for Nvidia hardware.

On AMD Side: you don't need it. There are Hardware Schedulers. That makes that hardware capable of adapting itself to application. This is exactly why I have been writing for very long time that in future GPU farms will do the heavy lifting jobs, and CPUs will do much lighter jobs. You will not need that high power on CPU side. All you will need is: Thunderbolt 3 or even faster external connection, GPUs that are connected with coherent fabric like NVlink, OmniPath or what AMD currently works on, and CPU that can handle the work like draw calls. This is of course AMD vision of GPGPU and GPU farms. Nvidia can have different.

P.S. http://wccftech.com/amd-teases-standardized-external-gpu-solution-for-notebooks/

The second part of the video, will the razor core work with a future mbpro that has tb3?

 

[koyoot]

The second part of the video, will the razor core work with a future mbpro that has tb3?

http://wccftech.com/amd-teases-standardized-external-gpu-solution-for-notebooks/

 

[dpny]

Can't believe you guys are still having these kinds of threads; I admire your tenacity!

The fact is the nMP is based on two big bets. The first is that all but niche uses will transition to HBM-based video cards; this bet still seems right, but the transition is many, many years behind the expected schedule, and in the interim the nMP is making some rather large compromises.

The other bet is that "mid-sized" jobs aren't economically-significant enough to Apple for Apple to worry about when designing the nMP.

Good post.

I've seen this transition even on mid-size jobs. I worked with a guy who does 3D/Animation work. He got a freelance gig which required a fair amount of processing for a one-man shop, and definitely more than his cMP/MBP could handle. So, he built himself a dual 8-core, dual GPU monster rig, used it for the three to four months of the job as, basically, his personal render farm, then sold it when he was done and recovered most of the cost of building the thing. He used to joke that he only ran it at night because it would make the lights in his house dim.

Commodity components have gotten so cheap that the equipment has become almost disposable.

 

[ManuelGomes]

VESA announced DP1.4, too bad DP1.3 hasn't picked up yet though.
Another year and a half to start coming up in real products?
Allows 8K@60Hz HDR, probably due to the "lossless" (visually that is) compression.


http://www.techpowerup.com/220478/vesa-publishes-displayport-1-4-standard.html

 

[AidenShaw]

Commodity components have gotten so cheap that the equipment has become almost disposable.

So that's why a hex core MP6,1 with only 16 GiB of RAM and a 256 GB SSD starts at $4000. Components have gotten "so cheap".



I think of "disposable income" as $40 for a lunch for two. Not $4000 for a "lower mid-size" non-upgradeable computer.

 

[koyoot]

I disagree, While those custom APU usually ends on game console and may end at some VR related devices, this will not be the case at Apple, Apple TV will remain with Ax chips same for VR, since Apple develop those chip in-house no need for amd help, not the case for Mac hardware since they need amd64 compatibility they need either Intel or AMD, Intel don't sell customized cpu, so AMD Is the only available vendor.

http://forums.anandtech.com/showpost.php?p=38065865&postcount=37

You wanted, here you have a bit . This guy is well informed. It also looks like that Zen will have Haswell-E levels of performance. It is from the thread in which this post came along.

Edit. One more thing. http://www.kitguru.net/components/g...etails-on-fury-x2-and-polaris-gpus-this-week/

 

[dpny]

So that's why a hex core MP6,1 with only 16 GiB of RAM and a 256 GB SSD starts at $4000. Components have gotten "so cheap".



I think of "disposable income" as $40 for a lunch for two. Not $4000 for a "lower mid-size" non-upgradeable computer.

I wasn't addressing nMP price, but the idea of Apple designing for either small- or large-scale jobs, while abandoning the middle segment. The argument makes sense to me. For small-scale work the nMP is enough. For large-scale work, with the majority of the hard work done on server farms, it has a place. But, even for mid-range work, the guy I worked with found a solution, and was able to recoup most of the build costs just be selling his machine after he was done with it.

 

[AidenShaw]

http://forums.anandtech.com/showpost.php?p=38065865&postcount=37

You wanted, here you have a bit .

That post is nonsense. What do isolated guesses about intra-chip bus peak bandwidths have to do with system performance?

Edit. One more thing. http://www.kitguru.net/components/g...etails-on-fury-x2-and-polaris-gpus-this-week/

Not nonsense - just an announcement of a PR event. I'm sure that we'll be able to read the details (albeit filtered) here.

 

[Mago]

http://forums.anandtech.com/showpost.php?p=38065865&postcount=37

You wanted, here you have a bit . This guy is well informed. It also looks like that Zen will have Haswell-E levels of performance. It is from the thread in which this post came along.

Edit. One more thing. http://www.kitguru.net/components/g...etails-on-fury-x2-and-polaris-gpus-this-week/

Amazing, my only concern is that all that much promise actually being delivered by AMD on schedule.

That 8c/12t + Polaris Zen APU either is targeted to an console (PS5?) or the nMP 8.1 it's really impressive, long time I wasn't surprised by an cpu development since Intel Core 2 quad maybe.

I'll check on the Polaris Gpu details on the AMA, it's what we could see sooner on a Mac since no Mac will get Zen until November or early next year.

Also Zen will arrive with PCIE 4 From the start, let's see what does Intel now since the Skylake successor seems will be not enough against Zen.

 

[koyoot]

That post is nonsense. What do isolated guesses about intra-chip bus peak bandwidths have to do with system performance?
Not nonsense - just an announcement of a PR event. I'm sure that we'll be able to read the details (albeit filtered) here.

He is also on another forum, and provided a lot of information about Processes from GloFo, and from TSMC in the past. That were correct to add. If he posts anything AMD related it is 99% correct.

About performance, in the upper right corner there is link to the thread. Its good lecture, if we ignore typical derailing. http://forums.anandtech.com/showthread.php?t=2465645

 

[AidenShaw]

He is also on another forum, and provided a lot of information about Processes from GloFo, and from TSMC in the past. That were correct to add. If he posts anything AMD related it is 99% correct.

About performance, in the upper right corner there is link to the thread. Its good lecture, if we ignore typical derailing. http://forums.anandtech.com/showthread.php?t=2465645

Whether accurate or not, the post is still nonsense. Micro-technical discussions like that have little bearing on system performance.

Would you pick car "A" over car "B" because "A" has a 57mm diameter exhaust pipe and "B" has a 51mm exhaust pipe?

Or would you pick "B" because it has a turbo and 40 more horsepower and gets 10% better fuel economy?

Micro-specs are nonsense. But they're what one cites when one doesn't have system specs to cite.

 

[pat500000]

Here's a serious question: "To buy or not to buy....that's the real question."

 

[Zarniwoop]

I suppose Apple is not behind with Pro Mac (laptops and nMP) updates just because there's no good HW available, but also because OS X is not mature yet.. it is still under transition to Metal. Metal v2 is needed in order that Apple will move all its Pro software to work on top of it. Also, openGL will most likely run on top of Metal v2 sooner or later (if not, an open source project could do it). Then game programmers etc. don't need to worry how to program multi GPU support to legacy and new apps. Current openGL implementation is also so slow, that it should only improve on top of Metal v2. Also, Apple will save money when Intel, AMD and Nvidia don't need to build separate drivers for openGL, but just one for Metal.

But what happens to openCL? Will Metal v2 introduce something that will replace it, or will openCL 2.1 be part of the package? This is also showing that OS X is not mature yet.. transition is going on. Apple has been patching openCL in every OS X update recently (10.11.1, 10.11.2 and even 10.11.3) so they have not given up with it.

If Apple doesn't deliver Metal v2 as El Captain update, we have to wait all the way to next fall for OS X 10.12 to have something new in the Pro sector.

There weren't any GPU driver updates in 10.11.3.. so if the same happens with .4, Apple is creating something new behind the scenes.

 

[MH01]

Whether accurate or not, the post is still nonsense. Micro-technical discussions like that have little bearing on system performance.

Would you pick car "A" over car "B" because "A" has a 57mm diameter exhaust pipe and "B" has a 51mm exhaust pipe?

Or would you pick "B" because it has a turbo and 40 more horsepower and gets 10% better fuel economy?

Micro-specs are nonsense. But they're what one cites when one doesn't have system specs to cite.

exhaust pipe diameters affect the engine performance, Especially on turbo cars. They are constrained out of the factory due to omission laws.

Increasing the diameter and removing the cat converter will get you performance gains .

Maybe you wanted to say the exhaust pipe tip.

One thing everyone turns to while talking tech is car anologies , they really don't work most of the time
[doublepost=1456920868][/doublepost]

Here's a serious question: "To buy or not to buy....that's the real question."

If it meets you current needs and you need one, buy. Good refurb deals to be had.

A new Mac Pro may not arrive till 2017, potentially never. Since 2013 so many pros and video studios have jumped to PCs , that the pro might the way of the 17 MBP , iPod classic, removed.

Personally I think apple will bail now or after the next update. While in 2013 there was so many looking forward to the new model, in 2016 is there enough of the niche users that are left to continue the product ? Dunno.

 

[tuxon86]

exhaust pipe diameters affect the engine performance, Especially on turbo cars. They are constrained out of the factory due to omission laws.

Increasing the diameter and removing the cat converter will get you performance gains .

Maybe you wanted to say the exhaust pipe tip.

One thing everyone turns to while talking tech is car anologies , they really don't work most of the time
[doublepost=1456920868][/doublepost]

If it meets you current needs and you need one, buy. Good refurb deals to be had.

A new Mac Pro may not arrive till 2017, potentially never. Since 2013 so many pros and video studios have jumped to PCs , that the pro might the way of the 17 MBP , iPod classic, removed.

Personally I think apple will bail now or after the next update. While in 2013 there was so many looking forward to the new model, in 2016 is there enough of the niche users that are left to continue the product ? Dunno.

While it does have an impact it isn't really a factor that will make or break your decision when it comes to buying said car. The overall performance is. You missed the forest for the trees...

 

[ActionableMango]

Here's a serious question: "To buy or not to buy....that's the real question."

Easy answer. With some rare exceptions, Mac model pricing stays level over the months or years that they are available. So Macs are their best value when they are a brand new model and their worst value right before an update. Right now the MP is 2-3 years old. The price is really high for what you get. So wait, if you can. Probably there will be an update this year and for the same price you'll get a much better MP.

Of course if you can't wait, then it's not a question. You have to do what you have to do in order to get your business done.

 

[pat500000]

Easy answer. With some rare exceptions, Mac model pricing stays level over the months or years that they are available. So Macs are their best value when they are a brand new model and their worst value right before an update. Right now the MP is 2-3 years old. The price is really high for what you get. So wait, if you can. Probably there will be an update this year and for the same price you'll get a much better MP.

Of course if you can't wait, then it's not a question. You have to do what you have to do in order to get your business done.

Totally makes sense. The reason that I was using shakesperes's quote was because the last rumor about nMP stated that there was a clue that it would arrive within OS X El Capitan updates.and now we are in the month March and oh wait.....I just remembered ...new OS X upgrade wouldn't arrive till fall, right?

 

[ActionableMango]

Totally makes sense. The reason that I was using shakesperes's quote was because the last rumor about nMP stated that there was a clue that it would arrive within OS X El Capitan updates.and now we are in the month March and oh wait.....I just remembered ...new OS X upgrade wouldn't arrive till fall, right?

Oh, haha that obviously went right over my head.

My guess would be a new nMP announcement at WWDC, but I know nothing.

• The average time between Mac Pro updates is 449 days.
• The longest time between updates, prior to the current model, was 685 days.
• As of today we are now at 804 days without an update.

 

[pat500000]

Oh, haha that obviously went right over my head.

My guess would be a new nMP announcement at WWDC, but I know nothing.

• The average time between Mac Pro updates is 449 days.
• The longest time between updates, prior to the current model, was 685 days.
• As of today we are now at 804 days without an update.

Now 804 days is crazy. I hate when Apple don't announce things ahead of time. Lol

 

[wallysb01]

Oh, haha that obviously went right over my head.

My guess would be a new nMP announcement at WWDC, but I know nothing.

• The average time between Mac Pro updates is 449 days.
• The longest time between updates, prior to the current model, was 685 days.
• As of today we are now at 804 days without an update.

Wow, I hadn't realized it had added up to be that long. Though the 685 days is really kind of, 685 plus the next update time, which was mostly just a price drop and a small spec bump. So the longest time, could arguably be set to 1241. Though the current model could at least get a similar price drop.... there isn't much available to get a spec bump without moving to a new generation of chips from Intel though.

What a sad state this currently is.

 

[ManuelGomes]

The crazy thing is that we're almost at the end of Q1 and no sign of Broadwell-EP as per the roadmap.
Trouble?!

 

[ActionableMango]

Though the 685 days is really kind of, 685 plus the next update time, which was mostly just a price drop and a small spec bump. So the longest time, could arguably be set to 1241.

That's a good point, as that was indeed only a spec bump.

But at least Apple could have thrown a "spec bump" bone bone to anyone who had to buy right now or even last year. It's easy to do and would have provided a better value today.