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

[antihelten]

It's a reference card with a single 6-pin power connector, so I'm not at all surprised that it can't overclock very much.

If custom cards can't overclock that would be a different story entirely.

There's not overclocking very much and then there's only overclocking by 5-7%. Even for a reference card that's pretty crappy.

 

[zentan]

4850 ($200) was 70-75% of a 280.
4870 ($300) was 85-90% of a 280.

If the stock 4GB version of 480 is equivalent to the 390X and the $300 versions come overclocked to 1500 MHz as rumoured, then performance will be roughly as follows:

Stock RX 480 ($200) is 70-75% of a 1070.
After-market RX 480 ($300) is 85-90% of a 1070.

So it would appear to me that this situation is actually perfectly comparable.

Of course one could argue that the proper equivalent of the 280 is the 1080 and not the 1070, but the pricing of the 280 at $500 (after the 4850/4870 forced Nvidia to drop the price) is arguably closer to the current prices of the 1070 ($410-480 on newegg) than those of the 1080 ($620-720 on newegg). Of course if Nvidia does a similar price drop on the 1080 as they did on the 280, then the 1080 would be the proper comparison point (the 1080 being about 20-25% faster than the 1070).

4870 beat the original gtx 260 and was priced 300$ at launch,150 lower to 450$ gtx260,also it came close to(85-90%) gtx 280 which was 650$ at 4870's launch. Yeah,it forced subsequent price cuts and later new cards like gtx260 core 216 and 4870 1GB also came into the play making it all more interesting.
Rx 480 8GB is coming at 230$ but whether it will have similar impact on Nvidia's pricing would depend of how much performance it can offer relatively.

 

[Erenhardt]

Detailed look at Crimson's new OC built in tool. Custom voltage for a range of power states along with custom fan curves.

Perfect. I always wanted to fine tune turbo clocks. I hope they implement such feature for CPUs aswell.

Voltage offset tool is very limiting. I can set -100mV under load and it runs fine, but as soon as I drop to idle clocks and volts, it becomes unstable. This new method should help with that!

 

[krumme]

There's not overclocking very much and then there's only overclocking by 5-7%. Even for a reference card that's pretty crappy.

I agree thats on the small side. But i was 100% sure it wouldnt hit 1500 on 150w. Wouldnt you agree? I would be happy about 1400 on 150w, so its more or less as expected for me.

The reason is its a low cost desktop card. They will raise voltage to have higher yields. A result of going for the low cost via the raised voltage is there is higher tdp. Thats how i expect it.
You are damn limited by the single 6 pin unless a bios hack can bypass it and the vrm are actually build for more than 150w.
No matter what the 1500 capable cards is surely AIB 6+8 and always have been. No free lunch.

 

[antihelten]

4870 beat the original gtx 260 and was priced 300$ at launch,150 lower to 450$ gtx260,also it came close to(85-90%) gtx 280 which was 650$ at 4870's launch. Yeah,it forced subsequent price cuts and later new cards like gtx260 core 216 and 4870 1GB also came into the play making it all more interesting.
Rx 480 8GB is coming at 230$ but whether it will have similar impact on Nvidia's pricing would depend of how much performance it can offer relatively.

Yes I know that the 260 and 280 launched at $450 and $650, but they had price drops almost immediately after the launch of the 4850 and 4870 (to $300 and $500 respectively) and several retailers and AIB partners even offered refunds for the difference. So the original launch prices for the 260 and the 280 aren't terribly relevant IMHO.

Now as I said, if the 1070 and the 1080 undergoes similar price drops after the launch of RX 480, then the proper comparison would of course be RX 480 to the 1080 and not the 1070. But I kinda doubt that this is going to happen.

I agree thats on the small side. But i was 100% sure it wouldnt hit 1500 on 150w. Wouldnt you agree? I would be happy about 1400 on 150w, so its more or less as expected for me.

The reason is its a low cost desktop card. They will raise voltage to have higher yields. A result of going for the low cost via the raised voltage is there is higher tdp. Thats how i expect it.
You are damn limited by the single 6 pin unless a bios hack can bypass it and the vrm are actually build for more than 150w.
No matter what the 1500 capable cards is surely AIB 6+8 and always have been. No free lunch.

Given AMD's historical OC ability, I would say that somewhere in the 15-20% range would have been what one might expect (the reference 7870 overclocked by about 20% and the reference 290X by about 15%). That would put the RX 480 at 1450-1500 MHz or about 150 MHz higher than what is being reported.

The 6 pin connector may very well be what is holding back the card, with the GPU itself capable of more. I guess we'll find out soon enough.

It is worth noting that the original 1500 MHz rumor said this:

I am told that all 6+8 pin Rx 480's are said to get this 1500mhz mark with software

So the rumor was never talking about the reference model.

Honestly this would also go some way to explain AMDs claims of covering the entire segment from $100 to $300. Personally I always though that having a reference version at $200 and the AIB versions covering the rest of the gap up to $300 was very unusual (I can't think of any card that ever had aftermarket versions priced 50% higher than the reference version, with the exception of water cooled versions), which is why I thought the (now debunked) theory of Polaris 10 actually having 40 CUs was plausible as that would leave enough headroom to reach the $300 price point. However if the reference 480 is power limited by it's 6-pin connector, thus leaving AIB partners plenty of headroom for significantly overclocked aftermarket versions (significant as in 20% or so, i.e. 1500 MHz), then that would make sense as this would also leave enough performance headroom on the table to reach the $300 price point.

 

[zentan]

Yes I know that the 260 and 280 launched at $450 and $650, but they had price drops almost immediately after the launch of the 4850 and 4870 (to $300 and $500 respectively) and several retailers and AIB partners even offered refunds for the difference. So the original launch prices for the 260 and the 280 aren't terribly relevant IMHO.

Now as I said, if the 1070 and the 1080 undergoes similar price drops after the launch of RX 480, then the proper comparison would of course be RX 480 to the 1080 and not the 1070. But I kinda doubt that this is going to happen.

Yeah,we are probably looking at the same thing with different perspectives.
I would consider launch prices of 280/260 relevant because without RV770 being so good and timely, Nvidia very likely would have retained those prices for longer.

 

[antihelten]

Yeah,we are probably looking at the same thing with different perspectives.
I would consider launch prices of 280/260 relevant because without RV770 being so good and timely, Nvidia very likely would have retained those prices for longer.

Sure if RV770 hadn't happened, those prices would have undoubtedly stuck around. My point is simply that RV770 did in fact happen and we know that Polaris 10 will also happen.

We don't know the performance of Polaris 10, nor do we know if Nvidia will drop the prices of the 1080 and 1070 after the launch of RX 480. But if we assume that the RX 480 is roughly equal to the 390X in performance and furthermore assume that there will not be any price drops, then the comparison of aftermarket ($300) RX 480 to the 1070 (at it's current prices), would fall closer to the 4870/280 comparison than the RX 480 to 1080 comparison would.

If the performance of RX 480 differs significantly from that of 390X and/or Nvidia has price drops on the 1070/1080, then we are of course in a completely different situation.

 

[PeckingOrder]

Detailed look at Crimson's new OC built in tool. Custom voltage for a range of power states along with custom fan curves.

Fan is very low RPM.

A pair of RX 480 in CF:

damn, those temps... doesn't look good to me

 

[antihelten]

damn, those temps... doesn't look good to me

Remember, that is a crossfire setup, you always tend to get higher temperatures there.

 

[trane]

RV770 is the greatest graphics chip, at least this decade. That and R300. Polaris 10 looks pretty fantastic too, but it doesn't devastate Nvidia like RV770 did. To do that, 480 would have to beat 1070. We know it's going to fall short. Not to say that it isn't an awesome chip, but it isn't the miracle RV770 was.

 

[PeckingOrder]

Remember, that is a crossfire setup, you always tend to get higher temperatures there.

fully aware of that, but considering those cards were stressed for a couple of seconds, it still looks terribad to me

 

[William Gaatjes]

damn, those temps... doesn't look good to me

It is possible that if those temperatures are anything as third party tools displaying APU temperatures in the past, there is going to be need some work done in GPU z for the correct temperature.
I remember my APU being anywhere from -20C to +100C in third party tools in the past. But never the value it really was.


EDIT:
I do have to mention that while temperature inside the case is around 27C and the average core temperature is according to AMD overdrive 36C, my (APU) GPU temperature is 14C according to AMD overdrive... :hmm:
Even with the most recent AMD overdrive.
It still does not make any sense.

I guess there is a Peltier effect going on in the APU...

 

[zinfamous]

well, terrible news for UK and European buyers--current collapse of the lb and expected contraction of the euro will probably mean some ugly costs for this next couple of months of GPUs

 

[lukart]

Yeah. But if it's 1379mhz, it's either without +vcore or Polaris is also a crap overclocker like Pascal.

Well, you will be lucky if you get 1379Mhz with default voltage :sneaky:
The card strugles to go anything beyond 100mhz oc.

 

[Erenhardt]

damn, those temps... doesn't look good to me

tell that to all these R9 290s mining cryptos at 94'C 24/7

 

[krumme]

The 6 pin connector may very well be what is holding back the card, with the GPU itself capable of more. I guess we'll find out soon enough.

It is worth noting that the original 1500 MHz rumor said this:



So the rumor was never talking about the reference model.

Honestly this would also go some way to explain AMDs claims of covering the entire segment from $100 to $300. Personally I always though that having a reference version at $200 and the AIB versions covering the rest of the gap up to $300 was very unusual (I can't think of any card that ever had aftermarket versions priced 50% higher than the reference version, with the exception of water cooled versions), which is why I thought the (now debunked) theory of Polaris 10 actually having 40 CUs was plausible as that would leave enough headroom to reach the $300 price point. However if the reference 480 is power limited by it's 6-pin connector, thus leaving AIB partners plenty of headroom for significantly overclocked aftermarket versions (significant as in 20% or so, i.e. 1500 MHz), then that would make sense as this would also leave enough performance headroom on the table to reach the $300 price point.

What i dont understand is. Except for binning how is those 300 cards gona sell in numbers when everyone and his brother can make cheap 6 plus 8 230 usd cards add some minor voltage and get eg 1450?

I dont beliewe the plus 280 usd will sell at all. The power is plenty low already, noise even under heavy oc is easily controlled with 2 cheap fans, and who cares if a card is eg. 10% less efficient than some fancy msi edition.

One thing is an intention to sell 300 usd cards another is reality and in numbers that matters.

 

[flopper]

Y

The 6 pin connector may very well be what is holding back the card, with the GPU itself capable of more. I guess we'll find out soon enough.

It is worth noting that the original 1500 MHz rumor said this:



So the rumor was never talking about the reference model.

People mix different info.
reference model will be a great default card but unleashing the beast requires a different solution and card custom made.

you see Pascal throttle with FE and OC.
finfet smaller area so harder to really cool when pushed.

 

[JDG1980]

Detailed look at Crimson's new OC built in tool. Custom voltage for a range of power states along with custom fan curves.

Very nice! Should give everything enthusiasts need to fine-tune cards. Goodbye MSI Afterburner!

Regarding the GPU-Z screenshots, I'm not worried about it running at 94C (that's within the design parameters), but the fan speeds are a bit higher than I'd like. 1500-1800 RPM seems pretty loud. Hopefully that's only with a 100% GPGPU/FurMark torture test and not actual gaming. Of course, if it proves to be an issue, AIB cards will fix that. Even a fanless Arctic Accelero S1 Plus or S3 cooler should do the trick.

Hopefully they fixed excessive power consumption in multi-monitor idle and UVD decoding.

 

[Shivansps]

6 pin has nothing to do with it, stop using that excuse.

 

[antihelten]

What i dont understand is. Except for binning how is those 300 cards gona sell in numbers when everyone and his brother can make cheap 6 plus 8 230 usd cards add some minor voltage and get eg 1450?

I dont beliewe the plus 280 usd will sell at all. The power is plenty low already, noise even under heavy oc is easily controlled with 2 cheap fans, and who cares if a card is eg. 10% less efficient than some fancy msi edition.

One thing is an intention to sell 300 usd cards another is reality and in numbers that matters.

Whilst the AIB could certainly throw together a 6+8 (or 6+6) card and sell it at $230, they could probably also get away with selling it for $300 assuming that it is indeed 20% faster than the reference version (1500 MHz would be about 20% higher than reference).

No AIB is going to sell their cards for less than they have to, since they actually care about that little thing called margins. And if the only way to get a 6+8 pin card is to pay $280 or more, then I don't see why it wouldn't sell decently.

6 pin has nothing to do with it, stop using that excuse.

And you know this to be the case, how exactly?

 

[maddie]

Well, you will be lucky if you get 1379Mhz with default voltage :sneaky:
The card strugles to go anything beyond 100mhz oc.

That's the key isn't it?

I do believe that AMD is running with little voltage margin unlike in the past and relying on an extensive use of their adaptive clocking technology to alleviate temporary voltage drops. This is part of the AMD technologies up to 2.8X perf/W increase

If true, this will allow little to no spare voltage at stock and we will need to increase voltage for every increase in frequency.

 

[SSBrain]

According to the previous screenshot voltage is 1150mV @ 1266 MHz, isn't this a bit high compared to the GTX 1070/1080?

 

[Yakk]

Powercolor working on a new Dual RX480 GPU Devil Card?

http://www.powercolor.com/event/A_New_Devil_is_Born/index.asp

Beastly!

 

[oussama-tn]

Powercolor working on a new Dual RX480 GPU Devil Card?

http://www.powercolor.com/event/A_New_Devil_is_Born/index.asp

Beastly!

I hope it's not another r9 390 devil 13

 

[Sweepr]

http://videocardz.com/61396/new-amd...exclusive-first-look-at-new-overclocking-tool

That's in line with what I saw in chinese forums. If 6-pin is the limiting factor, then this card is not using only 100W in typical gaming loads.

Bonus:

- Radeon RX 480 OC @ 1360 MHz / 2050 MHz (3DMark Fire Strike Extreme GPU Score): 6104
- Radeon R9 390X OC @ 1120 MHz / 1700 MHz (3DMark Fire Strike Extreme GPU Score): 6157