AMD Polaris architecture – GCN 4.0

[MyBrainz]

You got a nice deal for the 7950,but anyway we have both had very good value for what we paid.

I just wish amd would give us details of there new cards performance and price.

I agree.. and i just want the info's now.. gimme specs to drool over, AMD!

 

[Seanspeed]

You're doing it again, reading something and then applying your own slant on it, by this measure your prediction is even more out.

Given that Fiji is a 250 Watt card; if Polaris is a 250 Watt card it will end up 2x the performance of a Fury-X

Half a Fury-X (Fury-X performance at Polaris power consumption) is 125 Watts

So lets do the Maths.

Polaris at 125 Watts = Fury-X performance < 2x the performance per Watt
Polaris at 190 Watts = 1.5x Fury-X performance < 2x the performance per Watt
Polaris at 250 Watts = 2x Fury=X performance < 2x the performance per Watt

whats the power-consumption of P10? is it 50 watts?, or 150 watts? maybe 200 Watts? no body knows.....

Do people even bother to think about they are saying anymore?

We do know this, the 125mm^2 P11 is a 75 Watt GPU, so the 235mm^2 must be about 135 Watts? so that must mean P10 is 1.15x Fury-X, right? No.... we don't know enough about anything to draw conclusions like that.
Oh but wait a minute, the Fur-Nano is only a 150 Watt card and is only about 15% slower than a Fury-X, now i can argue we will have 3x Fury-X performance with Polaris 10! yay for AMD fans.....

What a load of old.....

The main point here is that GP104 is going to be some 40% bigger than Polaris 10. Both on *mostly* the same new process. Both new architectures. I'm not saying the performance difference is going to be directly and exactly proportional, but I dont think it's unreasonable to assume the top end GP104 is going to be the higher performer, probably by a somewhat reasonable amount at the least.

I dont know why you think this is such a ridiculous conclusion to make. :/

 

[CAT-THE-FIFTH]

The main point here is that GP104 is going to be some 40% bigger than Polaris 10. Both on *mostly* the same new process. Both new architectures. I'm not saying the performance difference is going to be directly and exactly proportional, but I dont think it's unreasonable to assume the top end GP104 is going to be the higher performer, probably by a somewhat reasonable amount at the least.

I dont know why you think this is such a ridiculous conclusion to make. :/

Nearly 40% comparing die areas directly but probably closer to 20% to 25% when taking into account the process node density differences.

Something that has been bugging me is that due to the new node,Polaris 10 probably has more transistors than Hawaii,and unlike the latter is probably going to not really be strong at DP compute and has a smaller 256 bit memory controller,so it should be more optimised towards gaming.

So taking that into account,it should really be able to outperform an R9 390X easily at 1080p and 2560p,so either GF/Samsung 14NM cannot clock reliably above 1GHZ,yields are rubbish so they are only selling die salvaged parts,or AMD is trolling us.

 

[TheRealDeal]

Have to be careful when trying to compare different architectures. The big one i remember was the 4870 and gtx280. The 4870 was 256 mm² and 956 million transistors. The gtx280 was 576 mm² and 1,400 million transistors. This never materialized into a huge lead as you can see here at tpu. That's a chip twice the size but no where near twice the transistors. Even with 50% more transistors the lead the gtx280 had was nothing special.

http://www.techpowerup.com/reviews/Diamond/HD_4870/23.html

The architectures back then were very far apart compared to these days but it just shows size is not everything.

 

[humbug]

The main point here is that GP104 is going to be some 40% bigger than Polaris 10. Both on *mostly* the same new process. Both new architectures. I'm not saying the performance difference is going to be directly and exactly proportional, but I dont think it's unreasonable to assume the top end GP104 is going to be the higher performer, probably by a somewhat reasonable amount at the least.

I dont know why you think this is such a ridiculous conclusion to make. :/

For a start GP104 is 16nm while Polaris is 14nm, but they probably arent on the same performance level anyway.

Its the whole premise that Polaris 10 is equivalent in performance to a 390X that is ridiculous. not just because we don't know enough about it but also because your equations don't add up to the little we do know.

 

[CAT-THE-FIFTH]

https://translate.google.co.uk/tran...m/p/4523923961?qq-pf-to=pcqq.group&edit-text=

Ellesmere XT still belongs to A0 chip testing phase frequency? Stream processing units ? Sorry, now measured in simple terms is that these are now filled with now is a transistor chip, as it is to edit the number of stream processors with much frequency much power run ah, now just do this thing, global workforce in RTG under the collaboration, we expect the third week of may to provide A1 official version of the chip for testing at that time, performance, power, basic specifications can be determined, and the rest is driven adjustment and optimization of the product before the A1 chip out, any run points are guessing right, RTG has now issued a PCB design reference suggests, the various AIB has entered card pre-production stage, relatively speaking, Baffin PCB's proposal is very short and simple, 4-layer PCB + single fan on can, but also shorter than the Nano
Ellesmere's PCB recommendations are slightly longer than the Nano, but as long as the 6-layer PCB + single landlord asked AIB fan who can not engage in gas stove, AIB say can consider engage cooker ...
AIB landlord asked who can not engage in a three-fan, AIB say we going to put two fans when the display it ...
Finally Ellesmere Pro now with Baffin A0 are not out, do not rush it, RTG efforts in the
PS. Which help me to take a message to the USG Ishimura, to see if he says the 1070 have been suspended or beaten 980Ti, Polaris represents less people first

So,Polaris 11 reference cards are shorter than an R9 Nano and the Polaris 10 reference cards are slightly longer than an R9 Nano.

Also says cards have entered pre-production at AIB partners.

 

[pete910]

Out of interest, How long from pre-production to available ?

I have a feeling this round of GPU's may be a bit underwhelming for us enthusiast/power hungry types

 

[humbug]

Out of interest, How long from pre-production to available ?

I have a feeling this round of GPU's may be a bit underwhelming for us enthusiast/power hungry types

The thinking is launch @ E3 (Jun 14 - 16) on the shelves soon after.

 

[LoadsaMoney]

That is great as long as that is not their top end Polaris. If it is and all they can do is match the performance of a card that has been out for what 3 or so years, then that would be a bit sad.

If its not the top end Polaris, then that would probably be 390X performance, just above, as everyone has been saying.

 

[GFX-Kid]

The architectures back then were very far apart compared to these days but it just shows size is not everything.

True mate, true. I bet that's what she tells you eh?

Jokes aside, can't wait until June

If its not the top end Polaris, then that would probably be 390X performance, just above, as everyone has been saying.

Top end Polaris needs to match or beat 980Ti for me.

 

[ICDP]

Top end Polaris needs to match or beat 980Ti for me.

Agreed but if it only matches Fury X/980TI it better be considerably cheaper.

 

[LoadsaMoney]

AMD Polaris 10 has 390/390X performance

For less than $299

AMD is about to launch two GPUs soon. One is codenmamed Polaris 10 and will replace Radeon 390 cards and the other, the Polaris 11 should replace the Radeon 370 and take on Geforce 950 performance.

Our well-informed sources are confident that Polaris 10 should match or outperform Radeon R9 390 cards and in some cases even give the Radeon R9 390X a good kicking.

AMD currently wants $400 for Radeon R9 390X, and and its Radeon R9 390 could go as low as $310. The Polaris 10 should end up at $299 price at launch, and it will focus on the low power and power per wat where the Nvidia card is vulnerable.

Both the Geforce Pascal cards and AMD Polaris 10 and 11 are expected in June. We would not be surprised if one or both launched at Computex, as this show is focused on component business.

AMD was focusing on performance per watt when it showcased Polaris 11, its Geforce GTX 950 competitor in January timeframe and a month later with Polaris 10.

AMD might be anticipating Nvidia to take a while in replacing Geforce GTX 960 and GTX 950 cards with its 16nm Pascal-based variants, but it will have a tough battle against any successor of the Geforce GTX 970 and GTX 980. Some people dare to call these cards the “Geforce GTX 1080” and “GTX 1070”, but we will see if that is the brand that Nvidia uses.

http://www.fudzilla.com/news/graphics/40585-amd-polaris-10-packs-390-performance

 

[psychas]

if the power usage so low they should as well make polaris 10 x2 card as soon as possible , would be good for consumers who needs more power at respectful price

 

[Dave2150]

if the power usage so low they should as well make polaris 10 x2 card as soon as possible , would be good for consumers who needs more power at respectful price

With the state of SLI/Crossfire, it would be completely pointless of AMD to do so.

 

[Finners]

Match a 390? That would be a huge disappointment

 

[AlamoX]

Match a 390? That would be a huge disappointment

you will have pro and XT version of the chip anyway, i think the pro might even be slightly slower than 390 but much cheaper than the XT version that should be around fury nano/980 perf.
i think this is as low as they can allow the 480/X perf to be, anything lslower would be a mistake, and they really shouldn't name it 490, if they do it would be an even bigger mistake to have your 90s card slower than nvidia's mid range, the perception of it wouldn't be good, so even if they dont have a 90 segment stay at 80 and wait for vega to introduce 490. or 590

 

[Toss3]

Match a 390? That would be a huge disappointment

for £200? Much lower power consumption as well so more overclocking headroom. I'm guessing these will actually overclock really well and be a bargain at £200.

 

[Mtom]

Match a 390? That would be a huge disappointment

If this will be the 480, then matching the 390(x) is the one step up usually happens in every new generation.
Also as mentioned already, these should clock higher, and have better OC headroom.

 

[Steampunk]

I think AMD have made a marketing mistake by making everyone think Polaris 10 will be hugely powerful, when it's really only going to be Fury(ish) levels of performance on a smaller, cheaper, cooler chip. We think Polaris 10 is high end and Polaris 11 is low end, but in fact, there's also Vega coming at the end of the year, and that will be the true high end product that takes the place of the current Fury with something much better.

People are going to be disappointed in Polaris because it isn't a jump from Fury. AMD will be showing the price and efficiency and expecting us to be impressed, and we'll all be asking where the next-gen jump in performance is. It won't be coming until the end of the year or early 2017 when Vega arrives.

I'm sure Polaris will be great for what it is and what market segment it's aimed at, but it's not the high end product we are waiting for.

 

[TheRealDeal]

I think AMD have made a marketing mistake by making everyone think Polaris 11 will be hugely powerful, when it's really only going to be Fury(ish) levels of performance on a smaller, cheaper, cooler chip. We think Polaris 10 is high end and Polaris 11 is low end, but in fact, there's also Vega coming at the end of the year, and that will be the true high end product that takes the place of the current Fury with something much better.

People are going to be disappointed in Polaris because it isn't a jump from Fury. AMD will be showing the price and efficiency and expecting us to be impressed, and we'll all be asking where the next-gen jump in performance is. It won't be coming until the end of the year or early 2017 when Vega arrives.

I'm sure Polaris will be great for what it is and what market segment it's aimed at, but it's not the high end product we are waiting for.

As far as i am aware AMD's marketing has always been Polaris 10 is designed to bring in VR performance at a lower price point.