Sidebar Sidebar
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

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

Wouldn't it be really nice:

http://wccftech.com/apple-magic-keyboard-gets-a-vibrant-oled-touch-pad/
[doublepost=1466335648][/doublepost]I'm still thinking RX490 will not be Vega based but a possible 2560 core version of Polaris 10 with GDDR5X clocked higher.
It's said that Polaris 10 is only 2304 cores, full fat, but I have my doubts.
RX480 seems to come only with GDDR5 apparently, so the GDDR5X model will be RX490 for sure.
Vega will have other naming scheme, like Fury, for enthusiasts. Maybe Rage again?! :)
 
  • Like
Reactions: Mago
I wonder if the nnMP will be more like those Z series AIO's. They could go 5k with everything without the drama and a 27" - 30" could hold everything the tube does and it could lower production costs
 
lowendlinux said:
I wonder if the nnMP will be more like those Z series AIO's. They could go 5k with everything without the drama and a 27" - 30" could hold everything the tube does and it could lower production costs
Click to expand...

The people who want the 5K monitors built-in are already buying iMacs. There's no real benefit to be gained on Apple's side with new customers or higher margins.
 
You maybe want to know what is Project Greenland from AMD?

It is APU GPU chip, only. There is also quite huge chance that Greenland, or custom derivative of it with low clocks will be GPU of next Xbox.

Vega is completely different thing. People thought that it will be based on Greenland. Maybe parts of technology will be used in Vega. But the chips are not the same.

Remember what stands from rumors: 90 GFLOPs/watt. This is the number which is interesting in context of small Vega GPU.
 
Mago said:
FP32? ~ 10 TFlop ~ 3.3-5 TFlop FP64 ...:eek:
Click to expand...
We need some realistic CUDA/OpenCL benchmarks, rather than tossing out theoretical absolute peak TFLOP numbers and implying that an actual application will hit that bar.

(Mago, this is not directed at you....)

In real life, it's not some unobtainable theoretical numbers, but what real programs with real data can achieve.

And, it will never be the theoretical peak - at least on applications that you actually want to run.
 
Last edited:
Mago said:
FP32? ~ 10 TFlop ~ 3.3-5 TFlop FP64 ...:eek:
Click to expand...
Efficiency in GFLOPs/watt always is about FP32.

AidenShaw said:
We need some realistic CUDA/OpenCL benchmarks, rather than tossing out theoretical absolute peak TFLOP numbers and implying that an actual application will hit that bar.

(Mago, this is not directed at you....)

In real life, it's not some unobtainable theoretical numbers, but what real programs with real data can achieve.

And, it will never be the theoretical peak - at least on applications that you actually want to run.
Click to expand...
That is because... you are bottlenecked by software. Not hardware.
 
AidenShaw said:
it will never be the theoretical peak
Click to expand...
I'm aware how Paramount is to achieve more than 80% of such theoretical max performance, I just did an OpenCL accelerated app, optimizing the algorithms to efficiently use as many cores is not easy If possible, maybe on some trivial algorithms like brute force integrals calc, or scene renderings maybe..
 
This is interesting, from netkas.org:

MacOS 10.12 brings Fiji acceleration support finally

One can get working acceleration (and finally a full working driver) on amd Fiji cards (R9 Fury/X/Nano)

Just add device id (0x73001002) into Baffins section (AMDBaffinGraphicsAccelerator) in /System/Library/Extensions/AMDRadeonX4000.kext/Contents/Info.plist

Caveat – so far it only works on nmp as eGPU, some code that is checking for nmp is in place.
 
  • Like
Reactions: zephonic, Dean Yu, pat500000 and 2 others
Zarniwoop said:
This is interesting, from netkas.org:

MacOS 10.12 brings Fiji acceleration support finally

One can get working acceleration (and finally a full working driver) on amd Fiji cards (R9 Fury/X/Nano)

Just add device id (0x73001002) into Baffins section (AMDBaffinGraphicsAccelerator) in /System/Library/Extensions/AMDRadeonX4000.kext/Contents/Info.plist

Caveat – so far it only works on nmp as eGPU, some code that is checking for nmp is in place.
Click to expand...
Hmm I actually like this information :D.
Pretty huge amount of compute power. Edit: I think the Nano can work now because the deviceID of Fury is specifically designed/tailored for nMP framebuffer.

We were talking about Radeon graphics but I don't think anyone brought here this Info: http://videocardz.com/61213/nvidia-announces-tesla-p100-with-pci-express-interface

Finally.
 
Last edited:
koyoot said:
Hmm I actually like this information.
Pretty huge amount of compute power. Edit: I think the Nano can work now because the deviceID of Fury is specifically designed/tailored for nMP framebuffer.

We were talking about Radeon graphics but I don't think anyone brought here this Info: http://videocardz.com/61213/nvidia-announces-tesla-p100-with-pci-express-interface

Finally.
Click to expand...
It's a Tesla, so it has no graphics - compute-only ....
NVIDIA-Tesla-P100-PCIE-2-900x621[1].jpg

4.7 TFLOPS FP64
9.3 TFLOPS FP32
18.7 TFLOPS FP16
​
​
 
Last edited:
  • Like
Reactions: Mago
So...there is actual hope? MacOS with some hidden Fury code
Looks like only 6,1 actually worked. So no Polaris on next Mac.
 
Last edited:
koyoot said:
You still can buy MP 7.1 and throw into it P100 in external enclosure ;).
Click to expand...

More likely What I'll do, a MP with dual Polaris and the best CPU/SSD combo available, but I wrote this idea to shown a concept, I'm sure more than one will consider, even more than have only nVidia GPU (a p100 is overkill for software R&D - my case), even a Intel Xeon Phi or some Altera FPGA cards....

Even the nnMP could host at least 2 TB3 external cages (4x tb3 ports / 2 headers), the tcMP will enable a lot of flexibility as long AMD, Intel and nVidia provides the necessary support for Cuda, OpenCL 2.x which evidently Apple wont provide.

At least Intel PE for macOS seems will enable Xeon Phi KL toolchains on macOS (reading Intel MIC related forums)
 
Last edited:
Depends on compute performance. If it would be stock GPU - I would definitely rather Fiji because it will have more compute power.

If it will be OC'ed. 1.4 GHz would be around 6.5 TFLOPs, and that would not be bad. Fiji has also that problem that it lacks DP1.3 and HDMI 2.0.
 
koyoot said:
Edit: I think the Nano can work now because the deviceID of Fury is specifically designed/tailored for nMP framebuffer.
Click to expand...

Remember the leak that talked about the nMP having a Fury GPU, may have basis on this info previously known under NDA by many people, Netkas shouldn't be the only capable to mine this info from system's files.

My hope, this "Fury" actually being a Polaris or a Vega GPU.
[doublepost=1466466554][/doublepost]
ManuelGomes said:
Yep, I wouldn't go Fiji now even with the better compute performance.
Click to expand...
I doubt it actually being a true Fury, it should be a camo for another GPU unreleased (an uprated Polaris or a Vega, Vega is coming on Q3/Q4 that matches Apple schedules)
 
  • Like
Reactions: pat500000
Is Vega a process-shrunk Fiji? Could be a better fit for a MacPro7,1 if it had slightly lower power requirements like Polaris.

Currently you'd have to down-clock Fiji to fit nicely inside the power envelope generated by a SFF PSU. Could have Polaris as entry-level options :) Think I'd prefer 2x 1070s though ;)
 
Draeconis said:
Is Vega a process-shrunk Fiji? Could be a better fit for a MacPro7,1 if it had slightly lower power requirements like Polaris.

Currently you'd have to down-clock Fiji to fit nicely inside the power envelope generated by a SFF PSU. Could have Polaris as entry-level options :) Think I'd prefer 2x 1070s though ;)
Click to expand...
Nobody knows for sure at this moment. It was speculated that Vega is the same thing as greenland, however that chip died and went as APU chip only. Some parts of technology from Greenland can land in Vega GPUs, however.
Mago said:
I doubt it actually being a true Fury, it should be a camo for another GPU unreleased (an uprated Polaris or a Vega, Vega is coming on Q3/Q4 that matches Apple schedules)
Click to expand...
We don't know that, we can only speculate about this.
Zarniwoop said:
It could be, that Apple & AMD have delivered eGPU boxes with Fury chips to some key developers to test new macOS features, like Metal v2.
Click to expand...
Apple may have ordered engineering samples from AMD. AMD only could deliver to Apple them, nowhere else.
 
Last edited:
Fury is discontinued. If you want to buy Fury series GPU, do it now, because only available stock will be for buy. When it is sold - thats it. There will be no more Fury series GPUs available.

Which is very interesting in the context of Mac Pro. Would Apple buy GPUs that went out of production yesterday?

About Vega. Fiji with 4096 GCN cores directly ported to 14 nm FinFet would consume under load around 140W while having full 1050 MHz on core, constantly. It is without new architecture changes, without shaders, without power gating technology to adapt to load of the GPU.

61.44 GFLOPs/watt. So that might give you some perspective what we can expect. I do think Vega 10 might have specs like this: 150W power consumption, with 8 pin power connector, 4096 GCN cores, 8 GB of HBM2, 96 ROPS, 1350 MHz of core clock, with 11 TFLOPs of compute power, at around 350 mm2 die size, and 1/4 FP64 performance.

This GPU would not be competitor to GP104, but to GP102.
 
Last edited:
You must log in or register to reply here.
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.
Top Bottom
Back