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

AMD Confirms RX 480 At $199 USD, Other APU & Polaris Announcements

That looks a bit crazy to me.

If the card need to draw more then the designed limit. I think it make more sense to draw from the 6pin then the slot.

In a normal PC. That 6pin is directly connected to the PSU. And extra 20% power draw won't burn the cable. And most likely nothing will be broke.

In fact, OC a power hunger reference card that only has 6pin avail can easily achieve this situation (pull much more than 75W from the 6pin).

However, if that extra 20% is from the slot. It may kill the motherboard, that doesn't make any sense to me.
 

Good watch, especially when all of them get drunk.

It turns out that RX 480 GPU board power is really over 150W. The GPU itself uses around 110W. The problem is with VRAM, that draws around 40W alone. 8 GB of GDDR5X use 20W of power. The consumption now makes sense.

To give you a some form of perspective: HBM uses 8W of power under load.
 
Last edited:
SoyCapitanSoyCapitan said:
It's crazy how Linux driver support has left macOS in the dirt, but then it's not owned by closed system corporation
Click to expand...

What really nice is there is day 1 support in Linux for a new AMD card, I can't remember a time where that has happened. AMD has been trying to get back into HPC and clusters for a while and I think this day one support signals that they're ready to compete again. Hopefully they turn into another Intel where the Linux stuff comes quick and with quality.

As soon as there is support for this in OS X I'll pick one up from the states because my local store is selling the 8GB version for 299 Euro http://www.arlt.com/Hardware/PC-Komponenten/Grafikkarten/AMD-arlt/Radeon-RX-Serie/

--Edit--

I might not wait on OS X support I can throw a baby quadro in if I need to boot OS X
 
koyoot said:

Good watch, especially when all of them get drunk.

It turns out that RX 480 GPU board power is really over 150W. The GPU itself uses around 110W. The problem is with VRAM, that draws around 40W alone. 8 GB of GDDR5X use 20W of power. The consumption now makes sense.

To give you a some form of perspective: HBM uses 8W of power under load.
Click to expand...

HBM is why Nano is so efficient
 
To add to that misery of power consumption of GDDR5, AIB OC'ed versions of RX 480 will draw the same amount as Nano and still they will be slightly slower.

Well it may very well be cost efficient GPU. Power consumption - not entirely. Also it looks like the GPU heavily relies on memory bandwidth, which adds to the misery of this release in terms of efficiency.

Overall, if we compare performance per watt, RX 480 achieved 1.9 increase over previous generation of AMD GPUs of similar class of performance.

I do really hope that this amount of increase, at least, despite the fact that AMD touts 3 times higher performance per watt than 2015 GPUs, will be with Vega 10.
 
http://forums.anandtech.com/showpost.php?p=38327364&postcount=584

https://www.computerbase.de/2016-06...#diagramm-undervoltage-benchmark-in-2560-1440
This is proof that AMD does not learn on their mistakes.
Computerbase.de played a bit with WattMan part. Standard voltages for max thermals: 1.150V, they lowered it to 1.075V. What happens? The GPU looses 4% of performance, but power draw goes down 30-60W.

I have no idea whatsoever why they did this. It should have been from the start at that low voltage. The perception of the card would have been much, much better.
google Translate said:
In the normal power target the lower voltage makes no changes in the power consumption since the Radeon RX 480 still runs into the Power Target - just at higher clock rates. In the volume there are no differences, while the higher clock rates require a little tribute at the temperature. So the Polaris-10 GPU is then two degrees warmer.

Maximising the other hand, the power target, the voltage changes reduce the power consumption of the test system to 33 watts to 233 watts. The GPU is two degrees cooler and the volume is reduced by 3.5 decibels. Nevertheless, the graphics card is then still very loud.
Click to expand...
 
Last edited:
koyoot said:
I have no idea whatsoever why they did this. It should have been from the start at that low voltage. The perception of the card would have been much, much better.
Click to expand...

Most likely its one of the compromises needed to make the RX 480 a $200 card. Bump up those voltages and you don't have to throw away as many chips if yields are bad. If the power consumption is that much lower perhaps Apple can get some of the better binned chips and not have to underclock them to fit into the iMac and Mac Pro.
 
Yields have nothing to do with voltage and current thermal design for any silicon.

Also it turns out that OC the GPU... well, read it for yourself:
Videocardz said:
The benchmarks are clear: the increase in memory bandwidth by 15% affects in a performance gain 7-11%. This is quite remarkable, especially when you make yourself fully aware that the card would move ahead of some competitors.
Click to expand...

We are talking ONLY about OCing the memory, not core. OCing the core adds more performance as well.

Computerbase also mentioned that Overclockers site have learned through their sources that there might be 40 CU version of Polaris 10. And the speculation still exists...

AMD-Radeon-RX-400-series.jpg

RX 485, 490 and 495 still may come in future. One model is for sure that will come this year: RX 490.

P.S. It is however interesting that after number 9 it says: >256 Bit/4K.

So R9 490 might have wider memory bus?
 
Last edited:
koyoot said:
Computerbase.de played a bit with WattMan part. Standard voltages for max thermals: 1.150V, they lowered it to 1.075V. What happens? The GPU looses 4% of performance, but power draw goes down 30-60W.

I have no idea whatsoever why they did this. It should have been from the start at that low voltage. The perception of the card would have been much, much better.
Click to expand...

Lowering the voltage didn't affect the power draw at all when using the stock power target. Instead the performance was slightly increased because the GPU could maintain higher clocks before hitting the power target.

Those 33W power reduction were achieved in comparison to stock voltage @ maximum power target.

They should have lowered the power target while undervolting to match the stock performance. This would have shown us what a better reference design might have looked like.
 
Fl0r!an said:
Lowering the voltage didn't affect the power draw at all when using the stock power target. Instead the performance was slightly increased because the GPU could maintain higher clocks before hitting the power target.

Those 33W power reduction were achieved in comparison to stock voltage @ maximum power target.

They should have lowered the power target while undervolting to match the stock performance. This would have shown us what a better reference design might have looked like.
Click to expand...
Yes, and what is the Max Power target for the GPU? 150W TDP, and 163 W of power consumption. This is exactly what we are seeing in every review that test total power consumption at the wall.

Computerbase.de test computer was taking from the wall 266W of power. What are results for other sites?
1467185872F5hoIVuh4I_12_1.gif

RX-480-ABC-91.jpg

Both are perfectly in line with what has computerbase shown. See the point? ;)
SoyCapitanSoyCapitan said:
They should have just did a Nano with 14nm process and higher clockspeed. Even at 1Ghz it beats a 1700mhz GTX1080 at Luxmark.
Click to expand...
Well if AMD will get even to 60 GFLOPs/watt the Vega 10 GPU with HBM2, we will see 11 TFLOPs GPU with 185W of power consumption. That is GP102 territory.

HBM2 will help to drive the power down quite a lot, however I do not believe the clocks will go up heavily, compared to Polaris. I expect something like 1350 MHz on core clock for V10.

https://techaltar.com/amd-rx-480-gpu-review/4/

110W consumed.

And one more thing. There will be GDDR5X version of the RX 480.
 
Last edited:
koyoot said:
Computerbase also mentioned that Overclockers site have learned through their sources that there might be 40 CU version of Polaris 10. And the speculation still exists...
Click to expand...

This is wishful thinking. Rumors have persisted forever that there is a Polaris 10 with > 36 CUs but we have seen zero evidence of it and Anandtech reports that the RX 480 is a fully enabled Polaris 10.

koyoot said:
And one more thing. There will be GDDR5X version of the RX 480.
Click to expand...

Again, why? If Polaris 10 has some sort of potential that isn't being unlocked why is AMD holding it back? What we have seen in reviews is most likely what we are getting from AMD until Vega comes next year. Hopefully when AMD goes back to TSMC to produce Vega that will help with some efficiency improvements that seem to be lacking in Polaris.

SoyCapitanSoyCapitan said:
They should have just did a Nano with 14nm process and higher clockspeed. Even at 1Ghz it beats a 1700mhz GTX1080 at Luxmark.
Click to expand...

This is basically Vega. It is rumored to have the same number of CUs as Fiji/Fury. The nano shrunk to 14nm would be roughly a 300 mm2 chip while Polaris comes in at 232 mm2.
 
koyoot said:
Yes, and what is the Max Power target for the GPU? 150W TDP, and 163 W of power consumption. This is exactly what we are seeing in every review that test total power consumption at the wall.

Computerbase.de test computer was taking from the wall 266W of power. What are results for other sites?
Click to expand...

No, Computerbase clearly stated that "Max Power Target" is 50% more than the stock PT (page 5).
Their complete system power consumption was 209W with a stock RX 480, and it went up to 266W on stock clocks when maxing out the power target. This was then reduced to "just" 233W by undervolting the GPU.

This might have been lost in translation for you, as a German native speaker I'm sure this is what they've written.
 
Stacc said:
Again, why? If Polaris 10 has some sort of potential that isn't being unlocked why is AMD holding it back? What we have seen in reviews is most likely what we are getting from AMD until Vega comes next year. Hopefully when AMD goes back to TSMC to produce Vega that will help with some efficiency improvements that seem to be lacking in Polaris.



This is basically Vega. It is rumored to have the same number of CUs as Fiji/Fury. The nano shrunk to 14nm would be roughly a 300 mm2 chip while Polaris comes in at 232 mm2.
Click to expand...
Because there is a source over the internet that says so? He even said how it will be named. Im sure it will land on sites really soon, especially on italian ones.

As for die size of Vega - 350mm2.

Fl0r!an, thanks for explanation.
 
koyoot said:
Because there is a source over the internet that says so? He even said how it will be named. Im sure it will land on sites really soon, especially on italian ones.
Click to expand...

Given that the rumors for the RX 480 turned out to be false (roughly 100 W power consumption for R9 390X-Fury performance) I am not about to start believing more random forum posts.
 
lowendlinux said:
Jayztwocents like it he's sorta my default youtuber in regards to GC and the like
Click to expand...
Gave up on him for being a general dufus.

I just had a 1070 and a pair of 480s in my hands and had to make me mind up. I know the Crossfire 480 will tear the 1070 in OpenCL work but I don't like the idea of the power draw from the slot potentially hurting my system.

I have bought the PNY Founder Edition 1070 for my PC and will wait for a better solution for the cMP.
 
SoyCapitanSoyCapitan said:
Gave up on him for being a general dufus.

I just had a 1070 and a pair of 480s in my hands and had to make me mind up. I know the Crossfire 480 will tear the 1070 in OpenCL work but I don't like the idea of the power draw from the slot potentially hurting my system.

I have bought the PNY Founder Edition 1070 for my PC and will wait for a better solution for the cMP.
Click to expand...

That dufus act is why I like him..:D
 
  • Like
Reactions: SoyCapitanSoyCapitan
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