2016 nMP

[Zarniwoop]

16 TFLOPs Radeon Pro Duo for USD 1499. It is intended for VR content creation.. as is next MP most likely. So maybe it is Apple's will that developers could plug it to nMP v2? AMD's LiquidVR allows developers to use one GPU per eye.. to reduce latency. Power is delivered with three 8-pin PCI Express connectors.

 

[AidenShaw]

Power is delivered with three 8-pin PCI Express connectors.

So you'll need three MP7,1 cylinders to run one?

I think that the Achilles Heel (or "glass jaw") of the MP6,1 is its absurdly small power supply. Even if it didn't have proprietary form factor GPUs, the fact that those proprietary GPUs had to be under-clocked to keep the power consumption low enough for the "pretty form factor" will relegate the MP6,1 to one of the Edsels of computing - since automobile analogies are so popular.

​

 

[Bubba Satori]

I'm assuming that's a good news.

It's very good news for nMP owners who use OctaneRender or want to take advantage of the huge speed advantage of the Mercury Engine in Adobe CC.
http://www.nvidia.com/object/adobe-premiere-pro-cc.html
[doublepost=1458009820][/doublepost]

So you'll need three MP7,1 cylinders to run one?

I think that the Achilles Heel (or "glass jaw") of the MP6,1 is its absurdly small power supply. Even if it didn't have proprietary form factor GPUs, the fact that those proprietary GPUs had to be under-clocked to keep the power consumption low enough for the "pretty form factor" will relegate the MP6,1 to one of the Edsels of computing - since automobile analogies are so popular.

View attachment 621181​

That 450 watt PS had to cost Apple at least $10.
That's normally what you find in a $600 i5 pc.
What was the MP tower, around 900 watts?

 

[goMac]

16 TFLOPs Radeon Pro Duo for USD 1499. It is intended for VR content creation.. as is next MP most likely. So maybe it is Apple's will that developers could plug it to nMP v2? AMD's LiquidVR allows developers to use one GPU per eye.. to reduce latency. Power is delivered with three 8-pin PCI Express connectors.

Would be more interesting if Apple supported Crossfire or multi GPU render. If they announce multi GPU for Metal this year, it might not be a bad way to go. But high end Polaris would hit later in the year and immediately outdate 2x Furys by a considerable amount.

VR should be able to use the Mac Pros two GPUs now. Not sure why they aren't.

 

[pat500000]

It's very good news for nMP owners who use OctaneRender or want to take advantage of the huge speed advantage of the Mercury Engine in Adobe CC.
http://www.nvidia.com/object/adobe-premiere-pro-cc.html
[doublepost=1458009820][/doublepost]

That 450 watt PS had to cost Apple at least $10.
That's normally what you find in a $600 i5 pc.
What was the MP tower, around 900 watts?

Is that only for adobe cc? or adobe in general?

 

[tuxon86]

is that legal? (i mean, i assume it is or otoy probably wouldn't be boasting about their accomplishment)

..but, what gives? why are they able to do that?

There is nothing illegal in using a cross compiler or translation program. What would be illegal is if they were to port the whole Cuda engine code on another plateform without licensing it.

 

[goMac]

There is nothing illegal in using a cross compiler or translation program. What would be illegal is if they were to port the whole Cuda engine code on another plateform without licensing it.

It's like WINE basically. Nothing illegal there.

 

[koyoot]

Would be more interesting if Apple supported Crossfire or multi GPU render. If they announce multi GPU for Metal this year, it might not be a bad way to go. But high end Polaris would hit later in the year and immediately outdate 2x Furys by a considerable amount.

VR should be able to use the Mac Pros two GPUs now. Not sure why they aren't.

Apple will not allow to use dual GPU setup in Crossfire. However, what they will allow is improved idea of Split Frame Rendering. It will make dual GPU setup to be seen by application as one, big GPU. Both for graphics and compute.

 

[tuxon86]

It's like WINE basically. Nothing illegal there.

It's even better than Wine actually. Wine had to reverse engineer the windows API without pirating code from it. In my understanding what they did in regard to CUDA is to build a converter/translator for the resulting Cuda compilation binary result. This would be more like converting a document from one application to use on another instead of converting the application itself.

This is at least what I think of it.

 

[Bubba Satori]

Is that only for adobe cc? or adobe in general?

In the CC apps. One or two apps had it in CS6 I think.

 

[goMac]

Apple will not allow to use dual GPU setup in Crossfire. However, what they will allow is improved idea of Split Frame Rendering. It will make dual GPU setup to be seen by application as one, big GPU. Both for graphics and compute.

Sure, I think Apple would do Split Frame Rendering and not Crossfire (which is the way Windows is going too.) But they haven't shipped split frame rendering yet, so we're just kind of sitting here waiting for them to actually tap dual GPUs for stuff like games in OS X.
[doublepost=1458066283][/doublepost]

It's even better than Wine actually. Wine had to reverse engineer the windows API without pirating code from it. In my understanding what they did in regard to CUDA is to build a converter/translator for the resulting Cuda compilation binary result. This would be more like converting a document from one application to use on another instead of converting the application itself.

This is at least what I think of it.

Is the translator for CUDA byte code? That's pretty good. There's going to be some overhead involved, but probably nothing significant and likely only a one time hit per application run.

 

[beaker7]

Final nail in the OpenCL coffin.

 

[Mago]

Final nail in the OpenCL coffin.

The final nails are Vulkan and Metal, and python/numba for those on HPC biz.

Ahh, also those are the final nails on CUDAs coffin.

 

[goMac]

Final nail in the OpenCL coffin.

I'm pretty sure that it's just going to run against the existing backends like OpenCL or Metal. It's not bypassing them. OpenCL isn't mentioned as a supported backend, but Metal definitely is.

So not really a nail in the coffin of any non-CUDA platform. This translation layer needs those platforms to exist to do it's work.

(And as Mago mentioned, Vulkan and Metal are killing OpenCL anyway, and Vulkan in particular is a much larger threat to CUDA.)

 

[koyoot]

Sure, I think Apple would do Split Frame Rendering and not Crossfire (which is the way Windows is going too.) But they haven't shipped split frame rendering yet, so we're just kind of sitting here waiting for them to actually tap dual GPUs for stuff like games in OS X.

Ahem, Split Frame Rendering comes from Mantle . Similar feature is in Vulkan and LiquidVR, and if Metal is based on Mantle, also will sport this feature .

 

[t0mat0]

Apple will not allow to use dual GPU setup in Crossfire. However, what they will allow is improved idea of Split Frame Rendering. It will make dual GPU setup to be seen by application as one, big GPU. Both for graphics and compute.

AMD is moving away from Crossfire by the sounds of last night. MultiGPU support it is?

 

[goMac]

if Metal is based on Mantle, also will sport this feature .

It's not, and it doesn't support it. Apple could add it, but it's certainly not there now.

Metal is a totally separate API from Mantle. Just because OpenGL and DirectX share features, that doesn't mean they're based on each other. Metal and Mantle have the same relationship as OpenGL and DirectX.

 

[ManuelGomes]

It seems no HBM2 this year, Vega 10 and big Pascal.
It's becoming clearer that nMP will not see the light of day anytime soon, I guess.
Will Apple wait for the new stuff?

Kaby Lake seems to be on track for H2.
Still no news from Broadwell-EP, strange. Intel should bring Skylake-EP already and skip it.
With a PCH like the Kaby Lake's 200 series would really make it awesome.
[doublepost=1458073022][/doublepost]Beta 7 brings anything new to the table?

 

[JesperA]

Still no news from Broadwell-EP, strange. Intel should bring Skylake-EP already and skip it.

There was a slide at Open Compute Summit a couple of days ago featuring Intel E5-V4 CPU:s which i guess is "Broadwell-EP"? I will see if i can find the slides again. So it seems like it is on its way and Apple should, in that case, be able to update the Mac Pro soon


Edit:

At 9:48, the server to the right, but maybe it is a typo

 

[ManuelGomes]

v4 is indeed BDW.
It's supposed to be out in Q1, which is almost over.
I'm not sure Vega 10 was postponed because of Radeon Pro (to give it some time in the market before Vega based cards take the crown again), or if indeed it's related to yields and cost. Probably a bit of both.
Vega in 2017 makes it weird for Apple not to have it as a high end option on the nMP, or maybe they'll just release the lower end parts now and a higher end one next year, but unlikely.
If it comes out this year, GPUs will be all GDDR5X based most likely.
Still, there is no complete lineup, Polaris 10 for D310, Vega 10 should be for D510 and D710, following the current setup.
[doublepost=1458076246][/doublepost]Probably is, we just lack official announcement.

 

[Mago]

v4 is indeed BDW.
It's supposed to be out in Q1, which is almost over.
I'm not sure Vega 10 was postponed because of Radeon Pro (to give it some time in the market before Vega based cards take the crown again), or if indeed it's related to yields and cost. Probably a bit of both.
Vega in 2017 makes it weird for Apple not to have it as a high end option on the nMP, or maybe they'll just release the lower end parts now and a higher end one next year, but unlikely.
If it comes out this year, GPUs will be all GDDR5X based most likely.
Still, there is no complete lineup, Polaris 10 for D310, Vega 10 should be for D510 and D710, following the current setup.
[doublepost=1458076246][/doublepost]Probably is, we just lack official announcement.

There are two possibilities :

1- Polaris 11 comes available as for July (Polaris 10 is for notebook and seems almost confirmed for new rMBP) and then both the new rMBP and nnMP are announced at WWDC with availability for Q3.

2- Apple decided just to skip both Xeon and Polaris and go for a full AMD powered nMP at year end loaded with Zen and Vega on HBM2.


Whatever the new rMBP and nnMP should be announced at WWDC, next week we should see new retina MacBooks with just cpu/color updates.

 

[ManuelGomes]

Mago, Vega will only be out next year.
I'm guessing another year without a new nMP, which is absurd, or availability with current tech (Broadwell and Polaris only). A higher end GPU option would be lacking though.
The outlook doesn't really look good at the moment.
For one, waiting would be the preferred choice, but it would also add to the already long, long wait...

 

[Mago]

Mago, Vega will only be out next year.
I'm guessing another year without a new nMP, which is absurd, or availability with current tech (Broadwell and Polaris only). A higher end GPU option would be lacking though.
The outlook doesn't really look good at the moment.
For one, waiting would be the preferred choice, but it would also add to the already long, long wait...

That what I'm said when I said year's end, and given coincides Vega and Zen intro.

On the other hand, the video cards on the nMP aren't properly workstation gpu just under clocked enthusiasts gaming gpu, a nMP with Vega I don't see it to come soon without an PSU/Thermal redesign, so I'm confident the nnMP (7.1) will come with Xeon E5v4 and Fiji (dual fury nano) as base cpu and Polaris 11 at higher end (maybe only two gpu choices also very likely the intermediate gpu D500 replacement won't see the light).

 

[Derpage]

If they extend this much longer, a lot of those old tubes will go up for resale with absolutely zero warranty coverage. Meaning, a lot of very expensive repairs for the second hand market.

 

[Zarniwoop]

At next WWDC Apple has to reveal its cards about VR, AR and some of their base technologies, like Metal v2. Current Metal is buggy and incomplete, it doesn't support even Shader Model 5.0 let alone asynchronous graphics/compute with multi-GPU environment - current version is just a developers playground. Also, DSP audio support is essential for VR. AMD mentioned TrueAudio Next in their yesterdays slides.

So these key-tech features should come with OS X 10.12 "Disneyland" (just came up with that name, but hey, it's American landmark and original VR concept, and sounds better than El Captain). Same time Apple launches beta versions of Final Cut VR and Logic Pro VR. Works only with Metal v2 and TrueAudio next.

After yesterdays AMD slides and demos, I'd say that in WWDC we'll see MP v2 (if tube concept stays):
- Broadwell EP
- DDR 4
- Duo Polaris 11 4GB and Duo Polaris 10 8GB with TrueAudio next (no more third GPU option)
- USB type-c with TB3 and DP 1.3

Apple's consumer VR gadgets could come Q4/15. With Zen and Vega.