RUMOR: Radeon 480 to be priced to replace 380, similar output as 390.

They wont rebrand a card with worse architecture and just 4GB unless they release a gpu that perform worse than Fiji XT

And news sites saidk hitman was running max settings

Polaris 11 is more like a 360x

Why a die shrink and an improved architecture cant reduce power consumption by 100+w more?



Tonga already used a smaller bus width given the color compression which isnt good on tonga a 280x can do a better job without any compression algorymth.

And this isnt a full flat die...
speculation based on a speculation article


How can you compare if they are using 2 different node sizes and different architectures?
Its called being a fanboy my friend. They already drew their conclusions still puzzled why they still posting.
 
They wont rebrand a card with worse architecture and just 4GB unless they release a gpu that perform worse than Fiji XT

And news sites saidk hitman was running max settings

Polaris 11 is more like a 360x

Why a die shrink and an improved architecture cant reduce power consumption by 100+w more?



Tonga already used a smaller bus width given the color compression which isnt good on tonga a 280x can do a better job without any compression algorymth.

And this isnt a full flat die...
speculation based on a speculation article


How can you compare if they are using 2 different node sizes and different architectures?



With no architectural changes, if you take a chip and get it on to 14nm, there is a savings of 60% in power or the inverse of that if you want to keep the same power envolope there is a max increase of 30% in performance. This was stated by Both AMD, Samsung+GF.

What does that tell you in best case scenarios this is what happens now best case scenario for AMD Polaris 10, is 2.5 the perf/watt over Hawaii. So lets look at a chip that uses 300 watts (using a nice round figure for ease of math). 60% reduction of 300 is how much?

A 180 watt reduction.


That would put a 300 watt card at 120 watts on 14nm, now what is Polaris 10 getting right now? Simply put there is enough information to out there to gather what Polaris 10 can do I'm expecting it to be right around r390x performance maybe a bit higher due to architectural changes and boost in clock speeds.

2.5 times the performance per watt is the same as 60% reduction in power consumption at the same performance level btw!
 
Last edited:
With no architectural changes, if you take a chip and get it on to 14nm, there is a savings of 60% in power or the inverse of that if you want to keep the same power envolope there is a max increase of 30% in performance. This was stated by Both AMD, Samsung+GF.

What does that tell you in best case scenarios this is what happens now best case scenario for AMD Polaris 10, is 2.5 the perf/watt over Hawaii. So lets look at a chip that uses 300 watts (using a nice round figure for ease of math). 60% reduction of 300 is how much?

A 180 watt reduction.


That would put a 300 watt card at 120 watts on 14nm, now what is Polaris 10 getting right now? Simply put there is enough information to out there to gather what Polaris 10 can do I'm expecting it to be right around r390x performance maybe a bit higher due to architectural changes and boost in clock speeds.

2.5 times the performance per watt is the same as 60% reduction in power consumption at the same performance level btw!
You are comparing die size not power consumption/perf

And yet these leaks are showing a 380 gpu replacement
 
I am not comparing die size, I am comparing perf/watt and what the new node provides (which has been stated by AMD and its fab partners) and what has been stated about Polaris by AMD themselves over Hawaii.

Yes if the performance is just about a 390x it will be just a replacement for the current x380

AMD stating 2.5 performance per watt

AMDGPU.jpg


Advanced Micro Device's New Polaris FinFET-Based Architecture Could Open New Doors For The Company

AMD talking about power savings with GF 14nm process

AMD is stating that FinFET offers a 20-35% improvement in terms of performance over 28A planar, which is a huge performance boost and much less variation over planar, meaning a much more reliable yields and performance. This performance improvement is also in conjunction with a power reduction of up to 50-60% compared to 28A planar, which is also massive when you consider what that can mean for GPUs in mobile form factors and servers.

Samsung talking about the 14nm process

Process Technology - 14 nm | Samsung Semiconductor Global Website

14LPE (Early edition) targets the early technology leaders and time-to-market customers such as mobile application SoCs to meet the latest mobile gadgets’ aggressive schedule and improved performance/power requirements. 14LPE is the first foundry process technology manufactured in the foundry industry with the successful volume ramp-up. 14LPE offers 40% faster performance; 60% less power consumption; and, 50% smaller chip area scaling as compared to its 28LPP process.
 
Last edited:
"AMD is stating that FinFET offers a 20-35% improvement in terms of performance over 28A planar, which is a huge performance boost and much less variation over planar, meaning a much more reliable yields and performance. This performance improvement is also in conjunction with a power reduction of up to 50-60% compared to 28A planar, which is also massive when you consider what that can mean for GPUs in mobile form factors and servers."

They are talking about transistor performance, not gpu performance. Compared to 28nm planar, the transistors consume 50-60% less power, given a power envelope the transistors will perform 20-35% higher

The transistors.

GPU performance gains come from packing more transistors on the die primarily
 
yes so if you have the same architecture and same GPU and crunch it down you have the same amount of transistors, its either or of course that is why you can't directly correlate with a new architecture, but AMD is also saying 2.5x performance/ watt for its new architecture Polaris that gives us numbers on what is best case in either side, end reality it will be somewhere in the middle.
 
"AMD is stating that FinFET offers a 20-35% improvement in terms of performance over 28A planar, which is a huge performance boost and much less variation over planar, meaning a much more reliable yields and performance. This performance improvement is also in conjunction with a power reduction of up to 50-60% compared to 28A planar, which is also massive when you consider what that can mean for GPUs in mobile form factors and servers."

They are talking about transistor performance, not gpu performance. Compared to 28nm planar, the transistors consume 50-60% less power, given a power envelope the transistors will perform 20-35% higher

The transistors.

GPU performance gains come from packing more transistors on the die primarily
14nmLeakage.png
power.jpg
 
According to some of the posts in this thread Polaris will defy all laws of physics and common sense. Truly exciting times.

Some believe Neo exists as a engineer at AMD's Radeon Technology Group. Quite frankly, I'm not so sure (although I sure wish this was true, so that prices can come down on video cards from both sides).
 
My Magic 8 ball says that both companies video cards will be released in the near future!
 
Yeah, if you have perf/watt numbers over the current generation of cards, then you look at die size on the same node of what its competing against then look at what we now know what gm104 is replacing, those rumors are pretty much accurate what they are replacing.... Polaris doesn't fit into the same bracket, it just can't. Polaris has less ALU's than Hawaii, smaller die size than gp104, only way it can reach gp104 is if its clocked higher a lot higher, if that is the case it can't keep to the 100 to 130 watt power envelope and if it won't fit into that power envelope then it won't fit with the 2.5 perf/watt increase (and that 2.5 is best case, if its lower than that without frame rate locks, it will slip even more)

So far what we know of polaris, its focused on VR at low power consumption, something that a high end notebook would be able to push. All marketing points to this. If they decide to up the clocks to compete with gp104's brackets they might be able to get to the lowest one the card that is replacing gtx 970, this replacement card should just around the 980ti in performance. nV has increased their clock rates for pascal, but they also kept their thermal envolope close to what they were before, they did this by increasing efficiency and increasing ALU amounts.

AMD has not done one of these they dropped ALU amounts at least if its a replacement for the 390x, but if it was to replace Tonga, its the same amount of ALU's or close to it. So for Polaris, they have the advantage of dropping power useage but increasing performance slightly. nV took another route, get as much performance as possible within the same power envelope. That's why Polaris won't be in the same bracket as GP104
 
Last edited:
"AMD is stating that FinFET offers a 20-35% improvement in terms of performance over 28A planar, which is a huge performance boost and much less variation over planar, meaning a much more reliable yields and performance. This performance improvement is also in conjunction with a power reduction of up to 50-60% compared to 28A planar, which is also massive when you consider what that can mean for GPUs in mobile form factors and servers."

They are talking about transistor performance, not gpu performance. Compared to 28nm planar, the transistors consume 50-60% less power, given a power envelope the transistors will perform 20-35% higher

The transistors.

GPU performance gains come from packing more transistors on the die primarily
Exactly, 20-35% performance gain WITHOUT even tweaking the architecture. Get a boost clock working like Nvidia already did and as patents suggest, that could be another 50%. That would be a 95% increase in performance and may be a bit conservative.

yes so if you have the same architecture and same GPU and crunch it down you have the same amount of transistors, its either or of course that is why you can't directly correlate with a new architecture, but AMD is also saying 2.5x performance/ watt for its new architecture Polaris that gives us numbers on what is best case in either side, end reality it will be somewhere in the middle.
How is it either or? It's 30% faster WITH 50-60% less power consumption. How much faster is it with the same power consumption? That 2.5x perf/watt is a 60% power reduction IF you got perfectly linear scaling from the transition to FINFET.

if that is the case it can't keep to the 100 to 130 watt power envelope and if it won't fit into that power envelope then it won't fit with the 2.5 perf/watt increase (and that 2.5 is best case, if its lower than that without frame rate locks, it will slip even more)
Why is it keeping the 100-130 watt power envelope? Why not push it a bit to 200W for a higher tier part? All they said was one of the parts was yielding 2.5x perf/watt.
 
Ok if its both 30% performance boost and 60% drop in power consumption (which isn't actually, its transistor performance, not end GPU performance of the same GPU, something you don't seem to want to delineate) Anarchist, I will like to see your math, of 2.5 perf/watt increase over hawaii, can you make those number fit? NO you can't!

That is the problem, why wouldn't AMD show more that 2.5 perf/watt then?

Why would they stay in that 100-130 watt range, notebooks, what have they been targeting marketing wise right now........

Are they going to have different chips for notebooks vs. desktop, can you tell me the financial implications for that, bottom line, top line and margins? If they had different chips for both market segments, it wouldn't fit at all with the stratagies that AMD have laid out, If you want look over everything Raji has stated so far.

Funny thing is you ate up async performance (which was total ephemeral marketing) from their marketing department but you aren't eating this up (which are solid numbers provided by them)?

And you still haven't even gotten to the fact Polaris 10 has less ALU's than Hawaii.
 
Last edited:
Why would they stay in that 100-130 watt range, notebooks, what have they been targeting marketing wise right now........
Perhaps for the ITX/SFF market (based on that Hitman demo)... and a revisit to the good old days of 6" card.
 
Another thing to consider is the possibility of the 490/x being dual gpu cards
That would piss a lot of people off since it defies the naming scheme. Even now I still see people who don't realize cards like the GTX 690 or 295 X2 are dual GPU.
Plus that seems like a really cheap way to fill out a performance tier.
 
well with SLi and CX not working at all in most new games that won't be a smart move
 
I think they are probably making something on them at their current prices.

You'll have to show me where in there I said it will just be a replacement for fiji. Fiji performance for the smaller Vega, but technically replacing the 390x(Polaris 10 being 390 and lower). I'm guessing Polaris 10 and little Vega are about the same size. Big Vega with HBM2 should be the new high end to actually replace Furies. Complete with high price and water cooling. Small/mobile vega will probably be the x2 parts as well 390X with it's extra memory.

by the way, when i say the new 390x, i mean it's equivalent in the new lineup , not that it performs like it

so the new 480/x could be 390/x performance at 100-150w and $200-$250.

Sounds like a good start...


and the new 970/980 will be around 980ti/titanx performance ±10% presumably, it seems like they'll both be uncontested in their respective price ranges
 
Last edited:
AMD has been sitting around the 290/290X performance level with a ~$300+ cost for multiple years now. I really don't see how any of this is considered a "rumor", it's just common sense. If AMD and Nvidia launched new GPUs with the same performance and prices as the current generation of cards, they wouldn't sell well. Each new generation of GPU releases moves the bar, it's how technology works.
 
AMD has been sitting around the 290/290X performance level with a ~$300+ cost for multiple years now. I really don't see how any of this is considered a "rumor", it's just common sense. If AMD and Nvidia launched new GPUs with the same performance and prices as the current generation of cards, they wouldn't sell well. Each new generation of GPU releases moves the bar, it's how technology works.

None of this is hard to believe if you look at the current state of VR. The few demos that were free on Steam look like the most basic bare bone experiences graphically. The GPU bar has to be raised to support better visuals before people like me consider a headset.
 
AMD has been sitting around the 290/290X performance level with a ~$300+ cost for multiple years now. I really don't see how any of this is considered a "rumor", it's just common sense. If AMD and Nvidia launched new GPUs with the same performance and prices as the current generation of cards, they wouldn't sell well. Each new generation of GPU releases moves the bar, it's how technology works.

If they lower the price ~$100 and the TDP to 100-150, the 480 at 390 performance would be major upgrade... Im sure the 1070 will be 980 performance and some power savings at ~$100 less, same with the 1080 to 980ti ~$100 less with power savings... Whats the gripe?
 
If they lower the price ~$100 and the TDP to 100-150, the 480 at 390 performance would be major upgrade... Im sure the 1070 will be 980 performance and some power savings at ~$100 less, same with the 1080 to 980ti ~$100 less with power savings... Whats the gripe?

There is no gripe, just pointing out that you don't need a rumor to figure out both AMD will launch faster cards at each price point. They need to have 390 type performance at a lower price point so they can launch faster cards at the 390's price point. It doesn't matter what the second number in the product name is, improving performance for the customer's dollar is what generates sales.
 
There is no gripe, just pointing out that you don't need a rumor to figure out both AMD will launch faster cards at each price point.
The rumor narrows it down to the midrange though, I was expecting Polaris to also compete in the high end, with enthusiast class gpus in 2017
 
If they lower the price ~$100 and the TDP to 100-150, the 480 at 390 performance would be major upgrade... Im sure the 1070 will be 980 performance and some power savings at ~$100 less, same with the 1080 to 980ti ~$100 less with power savings... Whats the gripe?
They raised the price by $100 when they transitioned from the 200-series to the 300-series.
The 290 and 290X were sub-$300 shortly after the GTX 970's release a year and a half ago. These 400-series GPUs are effectively the same thing with about 1/2 power usage and with the additional VRAM.

To anyone excited about a $250 480X (equals 390X), where were you back then? Are you stuck on a weak PSU or are you suddenly desperate for 8 GB?

After comparing the specs with Pascal, it seems like Nvidia is either going to counter their DX12 shortcomings with brute force or AMD is about to get wrecked. Unless AMD has a bigger "Polaris 9" hidden up their sleeve. It's going to be especially ironic since they spent the last 3 years being criticized for their power consumption, only to finally fix it and be outshined by Nvidia's much beefier GPUs (which will probably also consume more power).

Maybe Nvidia decided to stop diving into their cash pools Scrooge McDuck-style and actually brought their A-game this time.
 
AMD is not going to get wrecked, (at least I will hold out on saying this till we actually see power consumption figures form Polaris midrange and low high end Pascal (1070 or what ever it will be called) Polaris is in a totally different bracket as Pascal at least at launch. Now if nV can counter the mid range and low end Polaris quickly enough, AMD will not gain much marketshare and this is AMD's goal to gain as much marketshare as possible because that will give mindshare too. We know AMD won't come out with any other cards till next year, which Vega will be their high end and enthusiast level GPU's, so they lose somethings here, margins and potential sales, but the advantage if nV can't counter AMD in the mid range and low end Polaris for more than a quarter the marketshare growth for AMD will be in double figures, more than 10% which will give their bottom line a healthy jump.
 
Whomever creates the fastest single GPU card will win this round. Consumers will ALWAYS compare the 980ti vs the Fury X and get the affordable lower tier card from that brand name. Just how it works. The performance of the lower tier card doesn't matter to them. It's what they can afford and from the company that makes the fastest GPU.
 
Whomever creates the fastest single GPU card will win this round. Consumers will ALWAYS compare the 980ti vs the Fury X and get the affordable lower tier card from that brand name. Just how it works. The performance of the lower tier card doesn't matter to them. It's what they can afford and from the company that makes the fastest GPU.

I think people put a bit more thought into it than that. Right now the 380/380X cards are a better value just above $200, I can't imagine Nvidia outselling that with the 960 cards by 4:1 or anything.

Though I will say it certainly helps Nvidia when card partners like EVGA absolutely blanket the market with a larger variety of cards than AMD board makers offer.
 
Last edited:
AMD has been sitting around the 290/290X performance level with a ~$300+ cost for multiple years now.
2013 - 290X ~300W TDP
2015 - 390 ~300W TSP (still at 290X performance)
2016 - 480 ~150W TDP (still at 290X performance?)
 
If they start packing the 390/x performance at $200-$250, and the Fury/X performance at $300-$500... All with TDP improvements... Isnt that a big upgrade? Lower price points and power TDP.

Then Vega will be their new tech, HBM2 and all. $600+ Titan competitor.
 
Anyone who already owns a 290 or better won't see a reason to upgrade to Polaris. Anyone interested in the $300+ tier (Fury+) will probably just wait for Vega.

Fury and Fury X don't seem to be very popular cards. Maybe the 390, 390X, and 980 Ti have stolen the spotlight and a price cut will make them more appealing. Or maybe it's the 4GB vram. They would also be the only last-gen rebrands in the line up, making them even less desirable.

Should probably refer to them as Fiji XT and Pro since they could be rebranded/refreshed as the 490X and 490.
 
Anyone who already owns a 290 or better won't see a reason to upgrade to Polaris. Anyone interested in the $300+ tier (Fury+) will probably just wait for Vega.

Fury and Fury X don't seem to be very popular cards. Maybe the 390, 390X, and 980 Ti have stolen the spotlight and a price cut will make them more appealing. Or maybe it's the 4GB vram. They would also be the only last-gen rebrands in the line up, making them even less desirable.

Should probably refer to them as Fiji XT and Pro since they could be rebranded/refreshed as the 490X and 490.

In reality, the 980Ti is just a much better card for the same price.

The Fury (nonX) is too close to the 390X performance wise to matter, and the 390X is a clear winner in it's price-tier.
 
Back
Top