[WCC] AMD Rolling Out New Polaris GPU Revisions With 50% Better Perf/Watt

[kraatus77]

What? You're mixing completely different things with eatch other. What does IPC have to do with power consumption?

LOL, because they are related ? ( performance per watt)

There are 2 ways to increase p/w.

1. increase performance at same wattage ( aka more ipc ).
or
2. reduce power consumption at same performance.

1st one is not happening here since the chip and architecture is same, and 2nd is already pretty close to what it should be, so i don't really understand where is this 50% less power consumption is coming from.

people still don't understand, polaris have good enough power consumption for the amount of teraflops it provides. but those teraflops doesn't provide the amount of performance they should. so the result is lower performance/ watts.

what wccftech saying is, that new cards will have 50% lower power consumption resulting 50% more performance/watt. which is not possible. if they said new cards will have 50% more performance resulting 50% more performance/watt i could believe that if only there's going to be an architecture change. which again is not happening.

i can't explain it any better. just remember 1.6ghz talk what happened to it. you people are the reason amd fails because you believe in hype and expect too much.

 

[rainy]

I wish to understand why people are reading that trashy site.

10 percent or a bit more higher clocks together with 10 percent lower power consumption could be achievable as combination of a new revision and improved process - 50 percent is completely unrealistic.

Pascal can be undervolted for dramatic perf/watt gains, too.

I think, that gains from undervolting Polaris are bigger than in the case of Pascal.
Btw, dramatic sounds a bit hilarious to me.

 

[Glo.]

Polaris chips run better with undervolting. AMD have basically pumped more voltage to qualify as many dies as possible. With changes to process and a new chip revision combined with better binning I think we could see a 30-40% perf/watt increase. If AMD bring G5x on Rx 485 , a 50% perf/watt increase is possible.

It isn't possible. Revision of the silicon will bring 15-20% lower power consumption, at best. GDDR5X will reduce power consumption of 8 memory chips by 8W. That is all what you will get in terms of efficiency.
In real world, perfect world, good process, and GDDR5X would end up in RX485 being 1.4 GHz chip with 150W power consumption.

I think, that gains from undervolting Polaris are bigger than in the case of Pascal.
Btw, dramatic sounds a bit hilarious to me.

No. Pascal has much higher clocks, therefore it can be pushed much further back, hence GP104-400 with 5.5 TFLOPs can be within 75W TDP, whereas Ellesmere XT with 5.8 TFLOPs can be within 95W TDP.

 

[Gikaseixas]

Improvements can be achieved this soon but no one can clearly quantify them. Im impressed with the RX 480 GTR and most likely that amateur site is using that as basis and add more to generate clicks. Quit bickering guys

Sent from my HUAWEI MT7-L09 using Tapatalk

 

[Riek]

Strangely (or not?) the rumours are in line with what we expected from polaris from the beginning.

There we assumed from leaks and others that a 480 would be closer to 90-100W (related to 480 CF being more efficient than 1080 with slightly better performance in a specific game, while in reality this combination drew 100W to much to be called more efficient in that game).

So given this leak, wouldn't this just bring 460/480 to the level which we hoped/anticipated before they released those chips?

While I find it unlikely, as we were fooled by the above during the 480 release. It isn't unimaginable because we expected those ballpark figures based on statements and initial leaks.
So if we found it possible then, the same should be true now. (Doesn't mean it will happen or is real...)

 

[Piroko]

LOL, because they are related ? ( performance per watt)

There are 2 ways to increase p/w.
1. increase performance at same wattage ( aka more ipc ).

Again, IPC is short for "instructions per clock" and does not magically improve power consumption.

2. reduce power consumption at same performance.

That's what this thread is about.

1st one is not happening here since the chip and architecture is same, and 2nd is already pretty close to what it should be

No, it's close to what you think it should be.

 

[Glo.]

On the other hand, that would make more logical the refresh of Nvidia GPUs, that is also rumored, and that rumor came actually first, before the rumors of revised AMD GPUs.

http://videocardz.com/63764/editorial-nvidia-geforce-20-series

The revised RX 485 about which I posted higher, in DX12 games/Vulkan Games would be 20% faster than RX 480. And that would make it effectively close to GTX 1070. Nvidia will not let that happen thats why they might want to refresh the GPUs with higher clocks, specs, memory, and keep the mindshare image of being "better" than AMD.

And we have to keep in mind that Vega - 2 next generation chips from AMD are coming out.

But all of this is just speculation, and finding logic in some moves in the industry.

 

[kraatus77]

Again, IPC is short for "instructions per clock" and does not magically improve power consumption.

Yes, that's what i said "increase performance at same wattage". do you even read ? or just trolling ?

That's what this thread is about.

Thanks captain.

No, it's close to what you think it should be.

Both actually, a transistor is not going to magically decrease it's power consumption by 50% compared to similar other transistors unless we are comparing 28 vs 14/16nm. ( 10-15% is possible but 50% is just not going to happen, never at same node)

again just remember, where is 1.6ghz rx480 this website was talking about ? why do you believe in such fairytales ?

 

[R0H1T]

It isn't possible. Revision of the silicon will bring 15-20% lower power consumption, at best. GDDR5X will reduce power consumption of 8 memory chips by 8W. That is all what you will get in terms of efficiency.
In real world, perfect world, good process, and GDDR5X would end up in RX485 being 1.4 GHz chip with 150W power consumption.

No. Pascal has much higher clocks, therefore it can be pushed much further back, hence GP104-400 with 5.5 TFLOPs can be within 75W TDP, whereas Ellesmere XT with 5.8 TFLOPs can be within 95W TDP.

It's like you totally forgot about the Nano & Fury X

Granted we're also dealing with air vs water cooling but even then Fury Tri-X numbers should give you an indication as to what the best binned GCN cards can do. Then we have Nvidia reference cards that stick very close to the TDP so I believe their efficiency will not improve drastically when the clocks are lowered, at least not as much as AMD in the years since GCN cards debuted.

 

[Glo.]

Cut.

I have not forgot about Nano.

The problem is that you do not see the evidence we already have. GTX 1080 pushed to 1000 MHz core clock is able to be run only through PCI-ex power, because it has around 75W TDP. GTX 1060 with just 200 MHz under clocking(to 1500MHz) goes from 125W to 65W of power consumption.

That is because Pascal GPUs have extremely high core clocks. That is also why they have gigantic thermal headroom for undervolting and saving power.

 

[Det0x]

i'm saying this again, just ban that website and don't allow any posts referring to them as source. simple as that.

Thread is marked with [WCC] in title, and in the first line i say its rumors only..

You know what your getting when your browsing this thread

But i have to say, i find this wishful thinking/speculation from WCC more interesting, then indie benchmarks from hardocp in (vr) games with a wooping ~100 players total, made out to be a big deal

 

[Arachnotronic]

Thread is marked with [WCC] in title, and in the first line i say its rumors only..

You know what your getting when your browsing this thread

Bit i have to say, i would rather read this wishful thinking/speculation from WCC, then indie benchmarks from hardocp in (vr) games with a wooping ~100 players total, made out to be a big deal

Considering AMD and NVIDIA are both hyping VR to the moon, I would think that VR benchmarking is relevant.

 

[Piroko]

Yes, that's what i said "increase performance at same wattage".

The problem is that this explanation needs more magic than the other one. It needs serious changes in the design itself and it has to assume that there was something seriously wrong in the concept of Polaris to begin with. And even then you have to assume that this improvement doesn't make your cores "work harder" (feed them better = more work = more switching transistors = more power consumption) but "work smarter", so not just some easy bottleneck fixing. That "work harder" part is why I said improving, well, "practically achievable IPC?" itself has nothing to do with power consumption. I think a couple of Pentium 4 variants had improved IPC while being even less efficient than the "non-improved version".

Both actually, a transistor is not going to magically decrease it's power consumption by 50% compared to similar other transistors unless we are comparing 28 vs 14/16nm. ( 10-15% is possible but 50% is just not going to happen, never at same node)

Actually, a combination of lower voltage and lower temperature can go a long way in reducing power consumption at the same clocks, certainly further than 15% as demonstrated by many mobile SKUs as well as a couple of desktop bins. That is, assuming that current binning is too lax, which doesn't seem that far reached.

again just remember, where is 1.6ghz rx480 this website was talking about ? why do you believe in such fairytales ?

I don't. I would like to see it happen, as that would make this forum very lively for a couple of weeks. But I do see it as it is, a rumor from an otherwise unreliable source.

 

[kraatus77]

you are making a big fuss over nothing, i already said "more performance at same power" just like what happened with kepler - maxwell. maxwell delivers 35% more performance at same amount of teraflops without increased power consumption.
secondly, even with combination of both (t&v) 50% is unheard without arch/node changes. as i said 10-15% is possible. maybe 20% in some cases but not 50%. that's damn close to a node shrink.

speculative discussion is something not possible in this forum. specially when it comes to anything positive about amd. because usual suspects take it as "hyping" and when the said product launches you will get every thread bombarded with "hurr durr , so fail, much hype". this is why i'm against it. more so when the source is wccftech and what they are saying is pretty much unheard of.

i would be happy to see it too, but since i'm not the one making it, and it doesn't even help amd or me a bit, i'm not going to fuel more positive thoughts in it. if it's true everyone will benefit from it. but if it doesn't, well you know what's happening in last 5 years.

 

[railven]

you are making a big fuss over nothing, i already said "more performance at same power" just like what happened with kepler - maxwell. maxwell delivers 35% more performance at same amount of teraflops without increased power consumption.
secondly, even with combination of both (t&v) 50% is unheard without arch/node changes. as i said 10-15% is possible. maybe 20% in some cases but not 50%. that's damn close to a node shrink.

speculative discussion is something not possible in this forum. specially when it comes to anything positive about amd. because usual suspects take it as "hyping" and when the said product launches you will get every thread bombarded with "hurr durr , so fail, much hype". this is why i'm against it. more so when the source is wccftech and what they are saying is pretty much unheard of.

i would be happy to see it too, but since i'm not the one making it, and it doesn't even help amd or me a bit, i'm not going to fuel more positive thoughts in it. if it's true everyone will benefit from it. but if it doesn't, well you know what's happening in last 5 years.

Kudos! Well said. The day I stopped listening/believing the hype-posters, the more I realized AMD wasn't as bad company. Expectations get shot to the moon echoed by the same usual posters (on both sides) but unfortunately for AMD, often their sources are so far off the mark they get the black eye while NV just gets the "LOL."

 

[Headfoot]

wccctech rumor = made up by wccftech based on one misrepresented, out of context fact somewhere multiplied by all the unmet expectations of who always want more

 

[Arachnotronic]

Kudos! Well said. The day I stopped listening/believing the hype-posters, the more I realized AMD wasn't as bad company. Expectations get shot to the moon echoed by the same usual posters (on both sides) but unfortunately for AMD, often their sources are so far off the mark they get the black eye while NV just gets the "LOL."

Mostly true, but AMD's claims of "2.8x perf/watt" and "we designed a completely new architecture for FinFET and called it Polaris" only to get what we saw with Polaris didn't help.

 

[rainy]

Mostly true, but AMD's claims of "2.8x perf/watt"

It's not correct - that was up to 2.8x and this changes situation quite a bit.

 

[gamervivek]

1.4Ghz for Polaris 11 should've been the lede but Khalid has to hype AMD to the moon. He should get paid by AMD if he isn't already.

 

[crisium]

Don't believe it for a second.

 

[lixlax]

The efficiency of Polaris was a letdown, also the clocks so there should be room for improvement. I can believe 20-30% over such a short period of time but ~50% sounds quite unbelievable and means they screwed up big time (or rushed the still immature chips out which can true since it will put the efficiency right where AMD promised it to be in the first place).

 

[Bacon1]

Now I think the 50% claim is ridiculous and WCCF tech in general, but stock 480s were overvolted which caused them to throttle.

People that bought them undervolted (lower power) while also gaining performance which increased their perf/watt. Add in GDDR5x for lower power usage and you have some good gains just from those two and no other architecture.

 

[Kapav]

I don't understand how the engineering team at AMD would, the first time around, miss a relatively quick design and manufacturing change to get 50% better efficiency. That's not rushing to market, that's sprinting without any analysis.

 

[Magee_MC]

I don't understand how the engineering team at AMD would, the first time around, miss a relatively quick design and manufacturing change to get 50% better efficiency. That's not rushing to market, that's sprinting without any analysis.

I also think that the 50% number isn't realistic, however, what if AMD didn't miss a quick design change, but instead GF was having problems moving their process to mass production? AMD might have been working with what were effectively golden samples of GF's implementation of Samsung's process, and had tuned the cards to those samples. When GF went to ramp up they ran into problems scaling the process to full production. That would explain why AMD had to run with higher voltages across the board in order to increase the binning.

If GF has straightened out the problem, it might be that we would finally be seeing the Polaris chips that AMD designed for the process that they were expecting. All of this could be just another example of the GF WSA pulling AMD down by the neck.

Fixed stupid typo.

 

[itsmydamnation]

You don't get 50%+ perf/watt improvements from process revisions, sorry. They'd be lucky to get 10% out of process enhancement alone, and that would probably require more substantial changes than...well, nothing.

while i dont buy WCCF crap, Nelaham D0 says hello..........
Better question is if there is a polaris rev comming is it just process, is it new metal layer or a Complete New steeping. Option One and Two are far mroe likely then three.