PCGH - RX 480 fails at 1400MHz

[Tup3x]

First Gpus ever rolling out on finfet, everyone at Hardocp are already finfet experts and know how an unreleased gpu will react to voltage increases.
Interesting.

I said if. In other words speculation based on GPU-Z shots that show TDP.

 

[tybert7]

SemiAccurate Forums - View Single Post - Polaris 10 size / performance estimation


one report the limitation is the cooler more than the chip, but either way, this won't save nvidia below 300 dollars, even if it does not overclock to the moon.

 

[onedestinazn2]

Is this the 1600+ model?

Who parked the batmobile on top of a graphics card?

 

[primetime]

SemiAccurate Forums - View Single Post - Polaris 10 size / performance estimation
one report the limitation is the cooler more than the chip, but either way, this won't save nvidia below 300 dollars, even if it does not overclock to the moon.

I would love to see one with a xfx Triple Dissipation type cooler wether it OC better or not.

 

[Sheppard]

It's the law of physics and how FinFETs work. Intel has improved on how the 3D transistor is built up so it isn't so much a pyramid it once was with Ivy Bridge, but that really didn't help them that much.

It's the law of physics lol

 

[NKD]

This shit seems to be all over the place lol. Yesterday just saw the stress test running at similar speeds and it saying less than 120w. Oh well. I am just waiting for the reviews now. Lol

 

[NKD]

Just saw anther picture from that Chinese thread. One of the pictures has at 203 watt at 1200mhz rofl. So ether these guys are not using proper drivers or gpuz has some learning to do. Something ain't right. Reviews can't come soon enough.

 

[Sheppard]

All depends on the version of Gpuz I suppose

 

[NKD]

All depends on the version of Gpuz I suppose

Also they are using 16.5. The stage demo they put on video cardz I guess had hella crashes and errors. All these leaks coming from China are on the older drivers it seem cuz new drivers only are given to reviewers 16.6.2. They will released day one to public.

 

[Revdarian]

Just saw anther picture from that Chinese thread. One of the pictures has at 203 watt at 1200mhz rofl. So ether these guys are not using proper drivers or gpuz has some learning to do. Something ain't right. Reviews can't come soon enough.

That doesn't seem that it would be a true picture then, because even if you could go over the 150 w of the PCIe+6pin configuration, it shouldn't be 53 watts above it, at least not without frying the mobo and or the psu.

 

[Silverbolth]

Can't wait to replace my 7870 iceq ghz edition with it 2x 6 pins by this card!
Smaller foot print and more ram

 

[tviceman]

If the reference card doesn't overclock especially well, people on here saying to wait for 2x6 pin or 1x8 pin AIB cards for better OCing are going to be disappointed. There has never been an AIB air-based cooling cards that has demonstrated substantially better overclocking than reference models. Even the vaunted MSI lightning or ASUS Matrix cards are only marginally better, in terms of full OC potential.

MSI GeForce GTX 980 Ti Lightning 6GB Review
ASUS GTX 980 Matrix 4 GB Review
MSI R9 270X Gaming 2 GB Review
PowerColor R9 290X PCS+ 4 GB Review

Best case scenario, you're looking at maybe 5%. Often times the difference is absolutely zero.

 

[NKD]

If the reference card doesn't overclock especially well, people on here saying to wait for 2x6 pin or 1x8 pin AIB cards for better OCing are going to be disappointed. There has never been an AIB air-based cooling cards that has demonstrated substantially better overclocking than reference models. Even the vaunted MSI lightning or ASUS Matrix cards are only marginally better, in terms of full OC potential.

MSI GeForce GTX 980 Ti Lightning 6GB Review
ASUS GTX 980 Matrix 4 GB Review
MSI R9 270X Gaming 2 GB Review
PowerColor R9 290X PCS+ 4 GB Review

Best case scenario, you're looking at maybe 5%. Often times the difference is absolutely zero.

These chips work different. If with Finfet you got binned chips but not shit load of them that can tolerate higher clocks you can bet your ass they will go in to AIB cards to make more profit.

Plus alot of people such as you miss out on something. If the card is throttling and bios lock to keep it under certain tdp you will never get super high stable overclock out of it. That simple.

 

[Ieldra]

These chips work different. If with Finfet you got binned chips but not shit load of them that can tolerate higher clocks you can bet your ass they will go in to AIB cards to make more profit.

If what you're saying is true then all that means is that there were indeed production issues, the expected norm would have been the high clocking chips and issues with the process led to this situation here.

Generally speaking though, there's nothing about finfet that leads to a larger variation in clocking

 

[NKD]

If what you're saying is true then all that means is that there were indeed production issues, the expected norm works have been the high clocking chips and issues with the process led to this

could be. May be the reason AMD picked the best clocks and hence the 6 pin connector. May be at the current process most chips are binned at 1330-1350. Looks like they picked the clocks that show best performance/watt compared to last gen and price it cheap. Nothing wrong with that, may be there will be some chips hit 1500 or so as the rumors were but may not be many.

 

[Tup3x]

It's not realistic at all to expect custom cards to suddenly OC 150-200 MHz better. I can't remember anything like that in recent history. Somehow I think AMD would have released a faster variant with 8-pin connector if the chip would reach those clock speeds with ease.

If power is the limiting factor then something slightly over 1,4 GHz is realistic in my opinion.

 

[MotionBlur]

It's not realistic at all to expect custom cards to suddenly OC 150-200 MHz better. I can't remember anything like that in recent history. Somehow I think AMD would have released a faster variant with 8-pin connector if the chip would reach those clock speeds with ease.

Agreed - they would have released a 485/490 variant if it was possible with an 8pin and sold it for $299 to help close the gap with the 1070s all the while keeping their 460,470,480 lineup the way it is.

 

[NKD]

It's not realistic at all to expect custom cards to suddenly OC 150-200 MHz better. I can't remember anything like that in recent history. Somehow I think AMD would have released a faster variant with 8-pin connector if the chip would reach those clock speeds with ease.

If power is the limiting factor then something slightly over 1,4 GHz is realistic in my opinion.

New process may be most chips are hitting 1350 stable but not all. If they do have binned chips that can do more you can bet they will be sold at a little premium. Its not unrealistic to think that they decided to go with what they could mass produce and have higher clock chips being sold at premium with custom cards. Not at all.

 

[NKD]

Agreed - they would have released a 485/490 variant if it was possible with an 8pin and sold it for $299 to help close the gap with the 1070s all the while keeping their 460,470,480 lineup the way it is.

what if they have those chips but not enough to produce in large quantities? Have you thought of that? That is a total possibility, may be they binned those chips for AIBs and let the play with it. That is totally possible.

If you are going to have a higher tier card meaning 485 or so you have to constantly produce it in volumes. May be all they can produce in large volumes is the reference and don't have enough chips to run at those clocks to sell to shit load of to people. They went with the safe bet that they could make the most of and sell as many as they can.

I am not saying they will clock higher and its still a rumor, but rumors about cards with dual bios switch could very well be true if they are setting aside binned chips for higher clocks.

We will know for sure mid july.

 

[Algrim]

I'm just waiting for the 29th for all the real reviews to come out and not have to rely on leakers.

 

[Pieter3dnow]

what if they have those chips but not enough to produce in large quantities? Have you thought of that? That is a total possibility, may be they binned those chips for AIBs and let the play with it. That is totally possible.
If you are going to have a higher tier card meaning 485 or so you have to constantly produce it in volumes. May be all they can produce in large volumes is the reference and don't have enough chips to run at those clocks to sell to shit load of to people. They went with the safe bet that they could make the most of and sell as many as they can.
I am not saying they will clock higher and its still a rumor, but rumors about cards with dual bios switch could very well be true if they are setting aside binned chips for higher clocks.
We will know for sure mid july.

The rumour about the binning that AMD let the AIB bin them so it could cut cost.


There was someone with an interesting train of thought on another forum why Polaris and why the pricing SemiAccurate Forums - View Single Post - Serious problem hits Nvidia's consumer Pascal GPUs

It makes a lot of business sense to rush out 1070 and 1080 when you're in Nvidia's situation. But I don't think it's going to be very effective in the face of RX 480 being such a good value. And anyone who follows this stuff will see that Vega will come in around $400 or so because AMD wouldn't leave that big price point empty if they priced Vega higher. In fact, AMD's moves probably signal the end of $500+ single GPU cards for quite some time. And Nvidia is probably rather worried about that.

The only thing is that Vega with HBM2 can not be "cheap". Fury was not "cheap" either.

 

[razor1]

no although it saves money on AMD's side, somewhat, it still increases AIB costs when they have to do it, you think AIB's will be happy about that? Also losing that kind of control on ensuring RMA levels stay at a certain amount isn't something that AMD would do just to drop initial costs, as it could come back and bite them in the ass if the AIB isn't careful. Doesn't matter who does the binning it always comes back into the cost of the cards.

For AMD if they created reference boards that were binned for higher clocks or lower voltage, something they have been doing for years, actually improves their margins as they charge the AIB's more when they bin GPUs for higher level GPU's or lower level voltage GPU's for specialty cards. So the cost always comes to the AIB anyways.

It all comes down to who is paying for what, and it always comes down to who ever is doing the binning it goes to the AIB partners paying for either model. But in this model where AIB's do the binning, AMD gets less money as they don't get the option of charging more for the specially binned chips.

 

[5150Joker]

As long as they can deliver the 4 GB cards at $200 in large quantities and get right at or below gtx 980 speeds at stock clocks, it should be fine even without overclocking headroom. Considering the GTX 980 is still selling for $350+, there's still a $100+ buffer there in price. I think the problem is a lot of people didn't temper their expectations because garbage sites like WCCFTech were spewing rumors right and left and unfortunately the AMD fanboys were gobbling it up.

 

[Pieter3dnow]

no although it saves money on AMD's side, somewhat, it still increases AIB costs when they have to do it, you think AIB's will be happy about that? Also losing that kind of control on ensuring RMA levels stay at a certain amount isn't something that AMD would do just to drop initial costs, as it could come back and bite them in the ass if the AIB isn't careful. Doesn't matter who does the binning it always comes back into the cost of the cards.

For AMD if they created reference boards that were binned for higher clocks or lower voltage, something they have been doing for years, actually improves their margins as they charge the AIB's more when they bin GPUs for higher level GPU's. So the cost always comes to the AIB anyways.

Why did you first state cost increase for AIB then say that it will in any case be added to the cost of the video card anyway , you made my point ?

 

[razor1]

Why did you first state cost increase for AIB then say that it will in any case be added to the video card anyway , you made my point ?

its the opposite of what that guy was saying, no matter who does the binning the cost is there in the end card, but by AIB's doing the binning, AMD doesn't get the extra margins it would have if they did it themselves and charged for it to the AIB's.

The explanation he gave doesn't fit with AMD giving the AIB's the binning, its not a cost savings thing from AMD's point of view because at the end they save on cost but lose the extra margins, at the end, its a loss for AMD not a gain.

 

[Pieter3dnow]

its the opposite of what that guy was saying, no matter who does the binning the cost is there in the end card, but by AIB's doing the binning, AMD doesn't get the extra margins it would have if they did it themselves and charged for it to the AIB's.

This might be true but look at it from another side if AMD knows that there not many chips that can be clocked higher why worry about the margins ?

 

[razor1]

This might be true but look at it from another side if AMD knows that there not many chips that can be clocked higher why worry about the margins ?

Well that is the thing, if it was that, then it would fit, cause the extra margins for the few chips that make it to the higher bins won't cover the costs of the binning.

 

[5150Joker]

Well that is the thing, if it was that, then it would fit, cause the extra margins for the few chips that make it to the higher bins won't cover the costs of the binning.

If these rumors are to be believed (and I tend not to believe them) and AIB are truly binning GPUs that can clock to 1400 MHz+, we'll probably see prices in excess of $350+ for chips that run fairly hot and frankly probably don't provide much more perf/$ vs the stock cards. I highly doubt we'll see RX480 AIB cards hitting 1600+ MHz even with the best binned cards unless it's on LN2 and done by some OC'ing team and even then I can barely picture it. I'd say the more realistic expectation is we may see AIB cards w/clocks around 1350 MHz with more phases and better cooling to justify the price increase and that's really about it.

Besides, if these chips fail, isn't it AMD that must warranty them even for AIB? So there has to be a level of constraint there else the AIBs would go nuts and AMD would be stuck w/an expensive warranty bill. I bet if AMD could've achieved higher clocks at acceptable temp/voltage, they'd have done it. This is probably close to as good as it gets with RX480 and the AIBs will just barely squeeze out some added performance and the real value will just be enhanced cooling.

 

[Pieter3dnow]

SemiAccurate Forums - View Single Post - Polaris 10 size / performance estimation
one report the limitation is the cooler more than the chip, but either way, this won't save nvidia below 300 dollars, even if it does not overclock to the moon.

I didn't catch this one but how much cooler are those AIB ones, in general they are more quiet bit bigger but are they that much better with cooling (without raising price).

 

[razor1]

If these rumors are to be believed (and I tend not to believe them) and AIB are truly binning GPUs that can clock to 1400 MHz+, we'll probably see prices in excess of $350+ for chips that run fairly hot and frankly probably don't provide much more perf/$ vs the stock cards. I highly doubt we'll see RX480 AIB cards hitting 1600+ MHz even with the best binned cards unless it's on LN2 and done by some OC'ing team and even then I can barely picture it. I'd say the more realistic expectation is we may see AIB cards w/base clocks around 1350 MHz with more phases and better cooling to justify the price increase and that's really about it.

Besides, if these chips fail, isn't it AMD that must warranty them even for AIB? So there has to be a level of constraint there else the AIBs would go nuts and AMD would be stuck w/an expensive warranty bill. I bet if AMD could've achieved higher clocks at acceptable temp/voltage, they'd have done it. This is probably close to as good as it gets with RX480 and the AIBs will just barely squeeze out some added performance and the real value will just be enhanced cooling.

Yeah the warranties are covered by both AMD and the AIB, but most of the cost will be on AMD since the GPU is the most expensive part of the card.

 

[Pieter3dnow]

Yeah the warranties are covered by both AMD and the AIB, but most of the cost will be on AMD since the GPU is the most expensive part of the card.

If AMD sets limits (in the contract with AIB) to what you can do with Polaris then it is no problem ....

 

[NKD]

If these rumors are to be believed (and I tend not to believe them) and AIB are truly binning GPUs that can clock to 1400 MHz+, we'll probably see prices in excess of $350+ for chips that run fairly hot and frankly probably don't provide much more perf/$ vs the stock cards. I highly doubt we'll see RX480 AIB cards hitting 1600+ MHz even with the best binned cards unless it's on LN2 and done by some OC'ing team and even then I can barely picture it. I'd say the more realistic expectation is we may see AIB cards w/base clocks around 1350 MHz with more phases and better cooling to justify the price increase and that's really about it.

Besides, if these chips fail, isn't it AMD that must warranty them even for AIB? So there has to be a level of constraint there else the AIBs would go nuts and AMD would be stuck w/an expensive warranty bill.

350? thats stupid. lets just wait and see. At that point unless its meeting the 1070 its pointless or its a dual 480. lol. I doubt AIBs are that ignorant that they will price a 1400mhz card at 350

The stock cooler sucks, you slap some heat pipes on there and dual or triple fans it will probably drop 25c. Just look at it, its enough to get the job done at stock clocks.

 

[5150Joker]

350? thats stupid. lets just wait and see. At that point unless its meeting the 1070 its pointless or its a dual 480. lol. I doubt AIBs are that ignorant that they will price a 1400mhz card at 350

The stock cooler sucks, you slap some heat pipes on there and dual or triple fans it will probably drop 25c. Just look at it, its enough to get the job done at stock clocks.

I dunno, I think AIBs could slap on a twin or triple fan cooler w/vapor chamber and fancy LEDs, clock it at 1350-1400 MHz and price it at $320-$350 and many would bite. Yeah two stock cards would be faster but who wants to deal with Crossfire? You gotta remember, the stock 8 GB ones are $230 (or higher depending on price gouging) so $100+ for a custom AIB w/higher clocks wouldn't be too shocking. There's some custom 1070s that are $480 already which is $100 above the MSRP of a base 1070 (not the FE)

 

[razor1]

If AMD sets limits (in the contract with AIB) to what you can do with Polaris then it is no problem ....

True

 

[NKD]

I dunno, I think AIBs could slap on a twin or triple fan cooler w/vapor chamber and fancy LEDs, clock it at 1350-1400 MHz and price it at $320-$350 and many would bite. Yeah two stock cards would be faster but who wants to deal with Crossfire? You gotta remember, the stock 8 GB ones are $230 (or higher depending on price gouging) so $100+ for a custom AIB w/higher clocks wouldn't be too shocking.

I have rarely seen a custom card with 100+ premium unless is like evga classified series. This card is not enthusiast card, so I don't see a 100 dollar premium tag on there. May be 50.

like I said if you are going to price a card at 350 then it better be a dual polaris card. Cuz the closer you get to msrp of 1070 the worse you look. Which I don't see happening unless it matches 1070 and I also know thats not happening unless its clocked at like 1650+, which is probably not happening either.

 

[5150Joker]

I have rarely seen a custom card with 100+ premium unless is like evga classified series. This card is not enthusiast card, so I don't see a 100 dollar premium tag on there. May be 50.

like I said if you are going to price a card at 350 then it better be a dual polaris card. Cuz the closer you get to msrp of 1070 the worse you look. Which I don't see happening unless it matches 1070 and I also know thats not happening unless its clocked at like 1650+, which is probably not happening either.

I agree, I'm just saying I wouldn't be surprised to see some upper tier ones reach the $100 mark like Zotac has done w/the 1070. It definitely loses it's value proposition as the price increases because I don't see the clocks scaling linearly with price and the focus will shift more towards cooling and looks. So an ACX 3.0 type of card might be $280-$300 with 1300-1350 MHz clocks and that would still sell well enough. Keep in mind the theoretical price of a base 1070 is $380 but none exists on the market and probably won't for a long time.

 

[jamesgalb]

I dont think even the very best coolers will cost over $299, maybe $309 absolute tops for some of the insane stuff... The reason we see 1070s jacked up $100 over $380 for a good cooler is the FE price and the 1080 price gap ($599+), as well as limited availability of both the 1070 and 1080.

Card manufactures will know that a $300+ 480 loses its appeal to a 1070, and wont sell well.

 

[CSI_PC]

SemiAccurate Forums - View Single Post - Polaris 10 size / performance estimation


one report the limitation is the cooler more than the chip, but either way, this won't save nvidia below 300 dollars, even if it does not overclock to the moon.

Only problem is that he is assuming voltage-thermal dissipation-performance window will not be an issue when pushing beyond the optimal performance window, some of this may already be contributing to the base cooler and power limit of the card (which will be shocking if it is going from an average 110W to around 150W with a 6-10% OC and requiring more voltage/cooling) - will not know more until seeing reviews but he is definitely making assumptions.
I am curious what he calls very good OC with the reference card, because 1350MHz (seems average so far by rumours) is only roughly 6% more than the 1266.
So is he hitting much more than this or rather a bit of a positive perspective from him with 1350MHz.

Just to clarify, I do think the card is solid for its performance and price point and is targetting a sweet spot, but still dubious with some claims or spin such as that post seems to be (IMO of course).
Cheers

 

[NKD]

Only problem is that he is assuming voltage-thermal dissipation-performance window will not be an issue when pushing beyond the optimal performance window, some of this may already be contributing to the base cooler and power limit of the card (which will be shocking if it is going from an average 110W to around 150W with a 6-10% OC and requiring more voltage/cooling) - will not know more until seeing reviews but he is definitely making assumptions.
I am curious what he calls very good OC with the reference card, because 1350MHz (seems average so far by rumours) is only roughly 6% more than the 1266.
So is he hitting much more than this or rather a bit of a positive perspective from him with 1350MHz.

Just to clarify, I do think the card is solid for its performance and price point and is targetting a sweet spot, but still dubious with some claims or spin such as that post seems to be IMO of course.
Cheers

There is also screenshots we have seen running at 1328 that are less than 120w and 1379 or what not, its a shit storm. Who knows. 3 days until reviews I will wait!

 

[CSI_PC]

There is also screenshots we have seen running at 1328 that are less than 120w and 1379 or what not, its a shit storm. Who knows.

Yep, Polaris 10 has had an insane number of 'leaks-rumours' put out there, why I have ignored nearly all of it until last few days before launch as that is the most likely to have some semblance of accuracy, and even then it is still bipolar with extremes.

Still really do not like the idea of this performance tier of cards having 4GB or 3GB models - applicable to both AMD and Nvidia.
I think longer term this is a potential pitfall for owner who would possibly struggle to justify replacing the card with the 8GB//6GB (for Nvidia) model due to the further cost incurred even if they sell the 4GB for an ok price.
Cheers