AMD Polaris Thread: Radeon RX 480, RX 470 & RX 460 launching June 29th

[flopper]

470 King of the 1080p
480 King of the 1440p
Coming king Vega for 4k

Its a Home run for amd this year a true slam dunk

 

[lukart]

Where are the RX 480 reviews you got this from ??

Let's just say I didnt sign NDA.

1266 boost clock.

 

[William Gaatjes]

the power is split into 6+1 phases, which is actually more than GTX 1080 (5+1).

http://videocardz.com/61091/amd-radeon-rx-480470-pcb-pictured-up-close
Does that make sense? what is the meaning of more phases?

I will try to explain.
Well, the gpu runs on a given voltage for example VDD = 1V.
But the input voltage is 12V. To get that 12V down to 1V, a synchronous buck converter is used. This is a power converter.
What is done, is that by making use of a very smart circuit, the 12V is converted down to 1V. Let say we need 1V at 24A. In real life, nothing is for free. But we assume that the efficiency of the buck converter is 100%. In reality, this is between 85 and 95 % for what i know. With a static load, higher efficiency numbers can be reached up to 99%, but the gpu chip is a dynamic load.
Meaning, that the amount of current drawn depends on how busy the gpu is kept with doing calculations.
But forget about that for now. We assume for calculation ease, 100%.
The input is 12V 2A. The buck converter converts that down to 1V@24A.
P = U * I. So it is just plugging in the numbers. Pout = Pin @100%.
In reality, Pout = Pin * efficiency.
so , 1 * 24 = 12 * 2 * 100 % efficiency.


In reality, it is not 100% efficiency. There is some energy lost when converting from 12V to 1V for any given current. This energy is lost in the components from the buck converter design that do the heavy lifting. Like the n-mosfet switches, the output capacitors, the current sense resistor and the inductor.
The energy lost, is usually converted into heat. This is called dissipation.
The buck converter is designed to have its highest efficiency at maximum load since the losses are the highest at maximum load (meaning the maximum output current has to be delivered).

Now, the amount of phases used, is dependent on the components choice.
You can generally say that the more phases, the more the amount of power that is delivered is spread over the amount of output phases. In reality, the energy lost during conversion is also spread over more phases. This means that the power (heavy lifting) components in each phase, have to take only a fraction of the total dissipation.

It all depends on the design. How much output power (output current x output voltage) you need, which components you choose. The higher grade components are generally speaking more expensive. The size of the components is also a limiting factor. Sometimes the designer is restrained to a limited printed circuit board area or component height. There exists for example smt inductors (From Wurth elektronik) that can do 42A easily. But this comes with a certain size. Now since a cooler must be mounted on top of the gpu, there is a limit here as well when it comes to component height.

The choice of which controller chip to use for the buck converter design. The amount of regulation needed. Now the regulating capability of the buck converter is also the most important thing. The output voltage has to stay as close to the designed output voltage as possible independent of the dynamically changing load.

All in all, the amount of phases say not that much in a linear fashion.
So, i would not worry about that.
What it could mean :
The dissipation is spread out over all phases. This can mean that the power components individually get less hot.
The chosen buck converter is a 6 phase model. Six phases must be used.
Lower cost components.
Smaller size components.
Availability of components.
Better regulation ? (This depends on the buck converter controller)

As you can see, there are a lot of variables.


Here is an in depth article about the benefits of using multi phase converters with respect to single phase converts.
Basically, it is better response to dynamic changing loads.

http://www.eetimes.com/document.asp?doc_id=1273224

The inductors are the grey blocks.
The output capacitors are the little drums with a blue /white shade.
The n-mosfet switches are fitted on the backside of the board for as far as i can see. (they have to be there ).

 

[lukart]

Says everything anyone needs to know about your contributions. Welcome to the ignored list



1.5+Ghz would be truly impressive. But have to wait and see if this turns out to be based in reality or not

Ok, Ive tried to understand why you said that, you BOLD the text that said AMD to me failed, which after I said that if at 150W they cannot get near the performance to the 150W nvidia card .
To me looks like something went wrong, specially if they are saying they focused on power efficiency.

I guess I must be really wrong about this line of thinking...

 

[Glo.]

Ok, Ive tried to understand why you said that, you BOLD the text that said AMD to me failed, which after I said that if at 150W they cannot get even close to 150W nvidia card.
To me looks like something went wrong, specially if they are saying they focused on power efficiency.

I guess I must be really wrong about this line of thinking...

TDP is not equal to power draw. According to WCCFTech the GPU draws around 100W. Also AMD slide from 31st may about crossfire RX480 says the same thing, that crossfire setup has better efficiency than 180W GPU.

 

[lukart]

TDP is not equal to power draw. According to WCCFTech the GPU draws around 100W. Also AMD slide from 31st may about crossfire RX480 says the same thing, that crossfire setup has better efficiency than 180W GPU.

Alright, my bad. Then we shall take conclusions when the articles come out.

 

[Glo.]

I will try to put some things into some form of perspective. I was trying to build a PC on Xeon E5v5, and Fury Nano. And then I realized that I may be way better if I will wait for RX 480 on price, efficiency, and performance.

Money Saved I can put into i7 6800K, and water cooling for it. Funny times are coming, I have to say.

 

[lukart]

I will try to put some things into some form of perspective. I was trying to build a PC on Xeon E5v5, and Fury Nano. And then I realized that I may be way better if I will wait for RX 480 on price, efficiency, and performance.

Money Saved I can put into i7 6800K, and water cooling for it. Funny times are coming, I have to say.

For sure, but the true test is when nvidia releases their 1060 card, I'm very curious how tell play this card. I bet they didn't expect AMD to have such low prices from the start.

 

[Glo.]

For sure, but the true test is when nvidia releases their 1060 card, I'm very curious how tell play this card. I bet they didn't expect AMD to have such low prices from the start.

To compete with RX480 Nvidia would have to cut down from GP104. If they will release GP106 with half the amount of cores from GP104-400 it has no chances with RX 480.

GP106 will compete only with RX 470, then. And it still can loose on efficiency, and cost.

 

[lixlax]

Let's just say I didnt sign NDA.

1266 boost clock.

Then you have no problem showing us some hard evidence.

 

[trane]

To compete with RX480 Nvidia would have to cut down from GP104. If they will release GP106 with half the amount of cores from GP104-400 it has no chances with RX 480.

GP106 will compete only with RX 470, then. And it still can loose on efficiency, and cost.

I'm hoping GP106 ships with 1440 shaders or so. But seeing how heavily cut 1070, and how they treated GM206, it seems in all likelihood we're looking at 1280 shaders with 128-bit bus; although I hope it's 192-bit.

Yes, that's not going to compete with RX 480. Nvidia have no other choice but to hustle a further cut GP104, something like 1664 shaders retaining a 256-bit bus. That will also need 1070 and 1080 to massively drop in price.

 

[AtenRa]

Let's just say I didnt sign NDA.

1266 boost clock.

Im talking about this one specifically,

150W with half performance of 150W 1070 card,

Where did you see that ??

 

[Glo.]

http://wccftech.com/radeon-rx-460-crossfire-3dmark-11-benchmarks/

Can anyone explain this:

The graphics score of the RX 460 graphics card ranges between 7000 points to 8000 points on the 3DMark 11 benchmark. This is roughly half of what the RX 470 is capable off (~13500 points) – which makes perfect sense, considering the Polaris 11 GPU is exactly half of Polaris 10.

Polaris 10 - 36 CU's
Polaris 11 - 16 CU's

?
Is nowhere near "exactly" half of P10.

Unless... both of GPUs are cut down versions of full dies...

 

[coercitiv]

Can anyone explain this:

Ignore them, they just meant unit count of RX 470 vs RX 460, not Polaris 10 vs Polaris 11.

 

[dark zero]

Also... they are delaying GP 106 a lot... is like they can't scale down.

 

[monstercameron]

Also... they are delaying GP 106 a lot... is like they can't scale down.

Dude what?

 

[cytg111]

Dude what?

"Well that's just like your opinion man."
- dark zero, probably.

 

[iiiankiii]

Let's just say I didnt sign NDA.

1266 boost clock.

You can't make statements like this and provide no proof. Since you're not under NDA, why don't you post some pictures of the card with your screen name and run some benches for us. That would be really awesome.

Or just don't mention it in the first play if you're unwilling to do so.

 

[sze5003]

I wonder what the beast mode cards will go up against. Those are suppose to be the $300 range versions. Makes sense if their 460,470,480 are priced up to $199.

http://wccftech.com/amd-rx-480-1500mhz-overclocking-tool-voltage-control/

 

[f2bnp]

He has nothing, he's merely speculating.

That NDA comment he made is completely out of the blue.

 

[DownTheSky]

It would be funny to see after GTX 1060 arrives, AMD launching RX480 and RX470 1.4Ghz edition. Like they did with the 7970 series after gf 680.

AMD Well yeah, we could have clocked them from start @ 1.4ghz, all cards do that, but you see, there's like 5% better efficiency @ 1.266.

 

[sze5003]

It would be funny to see after GTX 1060 arrives, AMD launching RX480 and RX470 1.4Ghz edition. Like they did with the 7970 series after gf 680.

AMD Well yeah, we could have clocked them from start @ 1.4ghz, all cards do that, but you see, there's like 5% better efficiency @ 1.266.

Yeat it will be the 480 GHz versions which probably means more waiting after the initial release.

 

[Ares202]

Would the 8GB version of the R480 definitely be worthwhile? or is 4GB sufficient for now on a mid range card?

I think I will go back to AMD if the benchmarks are proved to be accurate, the R480 would be a steal at $199/239

 

[sze5003]

Would the 8GB version of the R480 definitely be worthwhile? or is 4GB sufficient for now on a mid range card?

I'm going to go for the 8gb version since I like keeping my cards a long time. Also probably better resale value, well not really since they aren't that expensive to begin with.

 

[Qwertilot]

Seems sense. Unless you've got a specific/shortish term upgrade plan in mind I'd reckon on getting an 8GB, decent AIB cooler version. Cost a little bit more but not major amounts and nicer for the longer term.


As for tricks with special versions - that sort of trick works much better at the top of the market!

Not a mid range thing. That bit of the market actually likes efficiency quite a bit and it's very likely that really pushing the clocks on these things will destroy that.

Someone will likely do that but the point will be to try and nip at the coat tails of stock 1070's to eat into a bit of that market.