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

macOS Sierra Supports AMD R9 Fury

After reading two reviews that show voltage and test of voltage scaling on Fiji I conclude that it will be impossible to put this GPU into 125W thermal envelope, no matter what Apple will do. 150W - no problem at all, 1.169V on core and even 925 MHz of core clock at max.

What is more, at 850 MHz and 1.2V the GPU is running at 150W for each GPU in Radeon Pro Duo, and S9300X2.

Here are reviews:
http://www.anandtech.com/show/9421/the-amd-radeon-r9-fury-review-feat-sapphire-asus/17
http://www.anandtech.com/show/9621/the-amd-radeon-r9-nano-review/16
https://www.techpowerup.com/reviews/AMD/R9_Fury_X_Overvoltage/2.html
scaling.gif

Note the under voltage: -48mV and compare it to Asus Strix Fury voltage and power consumption. 216W TDP, at 1000 MHz of core clock and 1.169V, exactly 43mV less than stock voltage on Fury.
At 1.169V the power consumption of Fury goes 40W down, at 1070 MHz(20 more than stock clocks).
power.gif

Now lets look at Nano. 1.2V is the max voltage you get at load. At that voltage you get average core clock of 890 MHz, and 185W power consumption(because of thermal envelope).

Would it be possible of undervolting the GPU to 1.169V, and maintaining it at at least 850 MHz in 150W Thermal envelope?

Apple would need to up the PSU to 500W, at least, and stick water cooling in MP7.1 of cooling and highest possible stability of the GPUs.

One thing to note. Sapphire Fury that BareFeats used in their tests has 1.188V on core, and uses 29W of power more than Asus Strix Fury.

Edit: After this post I looked at this:
VFCurve.png

So to run the GPU at 925 MHz you need under 1.1V on core.

That might be very good news.
 
Last edited:
Still the same argument, who can predict that Apple can fit 2x 7970 6GB in a nMP. It's really hard to tell what and how they do it if they really want to fit 2x Fury in the 7,1.

If just follow those reviews, it's also impossible to make the 7970 a 125W graphic card.
 
It is possible ;)

HD 7970 at 925 MHz had 163W power consumption ;).
power_average.gif

What happened when they increased the stock clocks by 125 MHz to 1050 range? ;)
power_average.gif

So it was possible to put HD7970 into 129W thermal envelope ;).

But Apple not only down clocked the GPU, but also undervolted it.
 
  • Like
Reactions: ssgbryan
I personally don't think Apple will put a 28 nm GPU in the next mac pro. I think they might be updating the mac pro with each node shrink. Fiji is a non-starter for a lot of professional applications because its limited to only 4 GB of VRAM. They would also have to underclock it even further than the fury nano to make it fit in the thermal envelope. Maybe it could be one of the lower end options.

My preference would be for Apple to drop the mac pro to a single GPU. This means they could fit in any GPU whether its a 150 W AMD RX 480 or a 250 W Nvidia Pascal Titan X. Then they aren't stuck waiting for AMD or Nvidia to release GPUs in the small to moderate range that fit in the mac pro and instead can use the big power hungry GPUs like AMD's R9 390X, Fury or Nvidia's Titans. In a perfect world it would be nice to see them offer options for either AMD or Nvidia cards but since they must make custom PCBs for each so I doubt this happens.

If you want added GPU power then get an external thunderbolt enclosure which seems to have support in Sierra. It would have the added benefit of opening up some PCIE lanes so that they can increase the speed of the SSD(s) and have lots of thunderbolt 3 controllers.
 
koyoot said:
It is possible ;)

HD 7970 at 925 MHz had 163W power consumption ;).
power_average.gif

What happened when they increased the stock clocks by 125 MHz to 1050 range? ;)
power_average.gif

So it was possible to put HD7970 into 129W thermal envelope ;).

But Apple not only down clocked the GPU, but also undervolted it.
Click to expand...

I still couldn't get it.

If it can work on 7970, why impossible on Fury? Why can't Apple down clock and down volt the Fury at the same time to make it 125 TDP? I just couldn't get it.

You make up some numbers then say it's impossible, but why Apple need to use your number but not lower?
 
h9826790 said:
I still couldn't get it.

If it can work on 7970, why impossible on Fury? Why can't Apple down clock and down volt the Fury at the same time to make it 125 TDP? I just couldn't get it.

You make up some numbers then say it's impossible, but why Apple need to use your number but not lower?
Click to expand...
It is simple. Fury X with core clock of 1050 is consuming 280W under load. It is Much more than HD7970. R9 Nano is locked to 175W of GPU TDP and average board power draw is 185W, at average 890 MHz core clock. 125W is impossible on this GPU unless you want extremely under clocked GPU. In 750 MHz range for example. It would make the GPU to give 5.7 TFLOPs of compute power. 0.2 TFLOPs less than RX 480. On the other hand another 25W of TDP makes gigantic difference. Currently Radeon Pro Duo and S9300X2 both are based on dual Fiji XT Setups and both are 300W GPU with power draw slightly under that value, because of locking the GPU clock, not TDP, to 850 MHz. Radeon Pro Duo draws under load maximum 293W. 13W more than Fiji XT in Fury X. You see? Dual FijiXT is able to draw slightly more power, for much more performance. FuryX has 8.6 TFLOPs of compute power, Radeon Pro Duo has 16 TFLOPs of compute power "rated" and S9300X2 has 13.9 TFLOPs of compute power.

Conclusion. If you are power gating the GPU you do not get very impressive numbers even if they can look like that. If you are clock gating the GPU, efficiency ratings go up much more, at least for Fiji ASIC.

As For Fiji in MP7.1 possibility. Lock it to 925 MHz and 1.05V and you get stability and 15 TFLOPs of compute power.
But I still think better setup we would get from dual GP104 GPUs, or Polaris 10/Vega 10 stack.
Stacc said:
My preference would be for Apple to drop the mac pro to a single GPU. This means they could fit in any GPU whether its a 150 W AMD RX 480 or a 250 W Nvidia Pascal Titan X. Then they aren't stuck waiting for AMD or Nvidia to release GPUs in the small to moderate range that fit in the mac pro and instead can use the big power hungry GPUs like AMD's R9 390X, Fury or Nvidia's Titans. In a perfect world it would be nice to see them offer options for either AMD or Nvidia cards but since they must make custom PCBs for each so I doubt this happens.
Click to expand...
As I have pointed out, you will benefit much more from dual GPU setup with lower clocks, and better efficiency, than single GPU.
 
koyoot said:
It is simple. Fury X with core clock of 1050 is consuming 280W under load. It is Much more than HD7970. R9 Nano is locked to 175W of GPU TDP and average board power draw is 185W, at average 890 MHz core clock. 125W is impossible on this GPU unless you want extremely under clocked GPU. In 750 MHz range for example. It would make the GPU to give 5.7 TFLOPs of compute power. 0.2 TFLOPs less than RX 480. On the other hand another 25W of TDP makes gigantic difference. Currently Radeon Pro Duo and S9300X2 both are based on dual Fiji XT Setups and both are 300W GPU with power draw slightly under that value, because of locking the GPU clock, not TDP, to 850 MHz. Radeon Pro Duo draws under load maximum 293W. 13W more than Fiji XT in Fury X. You see? Dual FijiXT is able to draw slightly more power, for much more performance. FuryX has 8.6 TFLOPs of compute power, Radeon Pro Duo has 16 TFLOPs of compute power "rated" and S9300X2 has 13.9 TFLOPs of compute power.

Conclusion. If you are power gating the GPU you do not get very impressive numbers even if they can look like that. If you are clock gating the GPU, efficiency ratings go up much more, at least for Fiji ASIC.

As For Fiji in MP7.1 possibility. Lock it to 925 MHz and 1.05V and you get stability and 15 TFLOPs of compute power.
But I still think better setup we would get from dual GP104 GPUs, or Polaris 10/Vega 10 stack.
As I have pointed out, you will benefit much more from dual GPU setup with lower clocks, and better efficiency, than single GPU.
Click to expand...

I think the locked clock speed doing nothing on this topic. Because that's just locked for normal PC card, not necessary the same on the Apple demanded card. Or is there any limitation that makes the nano already running at the absolute minimum clock speed, any lower speed will only increase power consumption?

Lower the clock looks bad? Yes, same on the 7970, but they did it.

Since we were talking about "possible" to make it 125W. It's actually possible. "Is it looks good?" or "Is it high performance?" doesn't really matter, right?

You can say that another 25W to make it a 150W card is a much better option, but that's your opinion, not Apple's. And it won't change the fact that it's actually possible to down clock and down volt the nano to make it 125W TDP.

I couldn't get your point, because you said "
I conclude that it will be impossible to put this GPU into 125W thermal envelope, no matter what Apple will do.". And obviously it's not the fact. If Apple folow the same track of 7970, down clock and down volt the card. They should able to do it.
 
What I was actually saying is that 925 MHz will be possible with Fiji XT(Full Fiji) ASIC at 150W.

850 MHz might not be possible at 125 or 129W.
 
h9826790 said:
I think the locked clock speed doing nothing on this topic. Because that's just locked for normal PC card, not necessary the same on the Apple demanded card. Or is there any limitation that makes the nano already running at the absolute minimum clock speed, any lower speed will only increase power consumption?

Lower the clock looks bad? Yes, same on the 7970, but they did it.

Since we were talking about "possible" to make it 125W. It's actually possible. "Is it looks good?" or "Is it high performance?" doesn't really matter, right?

You can say that another 25W to make it a 150W card is a much better option, but that's your opinion, not Apple's. And it won't change the fact that it's actually possible to down clock and down volt the nano to make it 125W TDP.

I couldn't get your point, because you said "
I conclude that it will be impossible to put this GPU into 125W thermal envelope, no matter what Apple will do.". And obviously it's not the fact. If Apple folow the same track of 7970, down clock and down volt the card. They should able to do it.
Click to expand...

Its possible, but you get diminishing returns on taking such a big die and down clocking it. Typically cards of similar architectures (i.e. Tahiti, Hawaii, Tonga, Fiji) on the same process (28 nm) have similar performance per watt metrics. Thats why Apple went with Tahiti (the HD 7970) even though Hawaii (R9 290X) was released before the mac pro shipped. It fit their target power envelope much better. You can take those big dies like Fiji (Fury) and down clock them, but the small increase in performance compared to a smaller die is not worth the added cost.
 
koyoot said:
You are asking in the context of MP 5.1 or 7.1(Can)?
Click to expand...

I was referring to the 5,1 models, Ive never owned a 7.1 trash can and I thought they were just like the imacs, minimal upgrades and i had no idea you could upgrade the gpus on them. I thought the gpu's were built into it and not expandable... I guess i was wrong if your asking this question.
 
Mac Pro 7.1 is upcoming, potentially, model of can Mac Pro. Current Mac Pro 2013 is MP 6.1. And no they are not upgradeable. I was talking about upcoming Mac Pro.
 
koyoot said:
Mac Pro 7.1 is upcoming, potentially, model of can Mac Pro. Current Mac Pro 2013 is MP 6.1. And no they are not upgradeable. I was talking about upcoming Mac Pro.
Click to expand...
Seriously, are you that sure that Apple won't just announce the end of the Mac Pro soon?

The MP6,1 is such a compromised design that it's almost impossible to upgrade, and clearly impossible to upgrade to state-of-the-art GPUs - whether red or green. (And the red team has clearly fumbled their latest chips - but will Apple's pride allow it to embrace the green team?)

The single socket design eliminates half of the potential PCIe lanes, and there's almost no headroom to increase the power supply wattage without overloading the thermal envelope.

I think that two things are likely:
  • No MP7,1 - Apple stops building trucks
  • MP7,1 is dual-socket mid-tower, with room and power inside for dual double-wide GPUs and a half-dozen or so NVMe SSDs, and T-Bolt for the big storage. (Phil's ass will go on-and-on about "re-imagining" the tower - as he shows a Z-620 clone.)
I doubt that Apple's ego will allow them to bring back a tower, so I think EOL is more likely.
 
AidenShaw said:
Seriously, are you that sure that Apple won't just announce the end of the Mac Pro soon?

The MP6,1 is such a compromised design that it's almost impossible to upgrade, and clearly impossible to upgrade to state-of-the-art GPUs - whether red or green. (And the red team has clearly fumbled their latest chips - but will Apple's pride allow it to embrace the green team?)

The single socket design eliminates half of the potential PCIe lanes, and there's almost no headroom to increase the power supply wattage without overloading the thermal envelope.

I think that two things are likely:
  • No MP7,1 - Apple stops building trucks
  • MP7,1 is dual-socket mid-tower, with room and power inside for dual double-wide GPUs and a half-dozen or so NVMe SSDs, and T-Bolt for the big storage. (Phil's ass will go on-and-on about "re-imagining" the tower - as he shows a Z-620 clone.)
I doubt that Apple's ego will allow them to bring back a tower, so I think EOL is more likely.
Click to expand...
Why would they do this? What are the reasons they would kill the line that is meant for Final Cut Pro, selling point of potential Thunderbolt Display 5K or even 8K, and Metal API software that is benefiting from Mac Pro design?

Only for people with needs similar to yours it is compromised design. To be fair, even tower is compromised design, in term of efficiency.

You rumble around about how it is compromised from your perspective. Have you ever thought of looking at this from APPLE perspective, and what they want to do with it? What they want to achieve with this design? What they target particularly?

Towers maybe universal, when your needs are... "undecided", and you need to make your mind what you will use it for. If you want highest possible performance in lowest possible thermal envelope, you edit videos, create music, edit photos, do graphics work, program for iOS and Mac, what can you get that is better for ALL of this tasks in Apple lineup?

How they can improve thermal envelope? Liquid Cooling. It would make the computer even quieter, more reliable because of much lower temperatures on all of the hardware(CPU and GPUs).

All people do on this forum is rumble about how awful is design of Mac Pro without even bloody considering for what it was designed for.

Final Cut Pro X upcoming update(XI?) appears to be most interesting in this context.

P.S. Tell me one thing Aiden. What you will do when Intel, AMD, Nvidia will not allow you to change parts inside your computer? What you will do when all you will be allowed to buy are "devices", without internal expansion?
 
AidenShaw said:
Seriously, are you that sure that Apple won't just announce the end of the Mac Pro soon?

The MP6,1 is such a compromised design that it's almost impossible to upgrade, and clearly impossible to upgrade to state-of-the-art GPUs - whether red or green. (And the red team has clearly fumbled their latest chips - but will Apple's pride allow it to embrace the green team?)

The single socket design eliminates half of the potential PCIe lanes, and there's almost no headroom to increase the power supply wattage without overloading the thermal envelope.

I think that two things are likely:
  • No MP7,1 - Apple stops building trucks
  • MP7,1 is dual-socket mid-tower, with room and power inside for dual double-wide GPUs and a half-dozen or so NVMe SSDs, and T-Bolt for the big storage. (Phil's ass will go on-and-on about "re-imagining" the tower - as he shows a Z-620 clone.)
I doubt that Apple's ego will allow them to bring back a tower, so I think EOL is more likely.
Click to expand...

Neither of those are likely. Apple will release another mac pro, likely in a similar form factor to the existing. They could tweak it to accept slightly beefier GPUs or maybe drop it to one GPU and use the rest of the PCIe bandwidth for thunderbolt 3 controllers and dual SSDs.

I don't think a dual socket mac pro is ever coming back. Not in the world where Intel has 22 core processors. I could see some flavor of a mac pro with a PCIe slot but I think its unlikely.
[doublepost=1470602053][/doublepost]
koyoot said:
How they can improve thermal envelope? Liquid Cooling. It would make the computer even quieter, more reliable because of much lower temperatures on all of the hardware(CPU and GPUs).
Click to expand...

Liquid cooling doesn't make any sense. Apple tried this in the power mac and it was disastrous. It only makes sense in the PC world because its PC manufacturers are forced to use commodity components that limit cooling options. Liquid cooling just transports the heat from one place to a radiator pointing directly outside the case. The mac pro is already designed to directly remove the heat from the case with its thermal core design.
 
  • Like
Reactions: ManuelGomes
Stacc said:
Liquid cooling just transports the heat from one place to a radiator pointing directly outside the case. The mac pro is already designed to directly remove the heat from the case with its thermal core design.
Click to expand...
That is exact problem I have with the design. The heat from components is transferred to case, then it is transferred outside the case. If Apple would ditch the thermal core, and go for Liquid Cooling with pipes directly connected to the xPU's, liquid would go to radiator, and then dissipated by the fan.
 
koyoot said:
That is exact problem I have with the design. The heat from components is transferred to case, then it is transferred outside the case. If Apple would ditch the thermal core, and go for Liquid Cooling with pipes directly connected to the xPU's, liquid would go to radiator, and then dissipated by the fan.
Click to expand...

Again, this doesn't make sense. Currently you have:

CPU heat -> large metal heat dissipation object -> large cooling fan -> heat ejected from case

With liquid cooling you are just adding extra transport steps

CPU heat -> water block -> liquid transportation -> large metal heat dissipation object -> large cooling fan -> heat ejected from case

You aren't solving any problems here, just creating them by adding failure points. Look up PowerMac G5 cooling leaks. You are also adding noise from a pump mechanism and liquid being pushed around.
 
You have just presented this like all of these steps are separated from each one ;). And unfortunately I do not agree with this point of view.

About G5 and its problems, you are looking for evidence in the wrong place. Modern liquid setups are much, much better than they were ever before. They are quiet, efficient and effective.
 
koyoot said:
You have just presented this like all of these steps are separated from each one ;). And unfortunately I do not agree with this point of view.
Click to expand...

This is not a point of view, its the mechanics of the system.

koyoot said:
About G5 and its problems, you are looking for evidence in the wrong place. Modern liquid setups are much, much better than they were ever before. They are quiet, efficient and effective.
Click to expand...

I agree that modern all in one liquid coolers are better than in the past, but they are still solving a problem that the mac pro has been designed to not need.
 
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