AMD Radeon RX 480 Supplies at Launch

Status
Not open for further replies.
if you think 4% in 3 quarters is a good then man I don't know what to say. 4% in 3 quarters is what, 1.3% per quarter, yeah that isn't because of good products. That is because they have products in channel vs not having products in channel, but those products in channel are still not very attractive.

I made no judgement on it, simply stated the fact. And 4% higher market share without new products and inferior performance at the high-end is certainly a better result than one would expect.

Its not amount of sellable products end users can buy at the moment, its in the entirity of the life time of the product we are talking bout and when certain things are being launched.

Your question was very specific about profit within a single quarter, my comments are based on that initial assumption.

In regards to the rest, let's view it from another perspective, with actual numbers taken from the current Steam hardware survey. After adjusting the survey results for discrete graphics cards only, the performance segment ( GTX970 / R9 200 and higher) consists of less than 15%. 85% of the market is mainstream and entry-level.

Now let's imagine you know that you will release a new product roughly at the same time as your competitor and you know that they will release into the 15% first, with no product for the 85%. Do you try to sell into the same 15% or take a hit to margins and sell into the 85% where you have no competition for a while?

Was that their original plan? No idea. But given the current market situation, they stand to gain more from their strategy than from going after nVidia's new cards.
 
I thought they all become irrelevant cost wise as soon as FPGA/ASIC cards dedicated to mining came out.
Any way, 20MH/s is pretty crappy considering you can pull 700MH/s out of a 7970. (Although not efficiently)

In one of the other threads some folks were talking shit about how RX 480 would be bought up by miners so no one could buy it. I have had my doubts even one of them suggested that AMD would get bogged down by miners RMA and they were all serious about it as well :( .
 
I made no judgement on it, simply stated the fact. And 4% higher market share without new products and inferior performance at the high-end is certainly a better result than one would expect.

They did have new products, the 300 series are all new microcodes but also they were inferior with power usage, they made changes in the silicon to create those changes,and you are right inferior performance higher end cards.

That is why they didn't weren't compelling products. Just because they are in channel and people that don't know too much about graphics cards are picking them up or they are in prebuilt systems.

In regards to the rest, let's view it from another perspective, with actual numbers taken from the current Steam hardware survey. After adjusting the survey results for discrete graphics cards only, the performance segment ( GTX970 / R9 200 and higher) consists of less than 15%. 85% of the market is mainstream and entry-level.

You can't use steam survey because those numbers don't have a range of dates of who is buying what at what dates, you have cards on there that are more than one gen old, are those cards even being sold at in last year? Probably not, then you have take out the integrated market numbers too. You can't get a good picture of what is being sold now or in recent Q's

Now let's imagine you know that you will release a new product roughly at the same time as your competitor and you know that they will release into the 15% first, with no product for the 85%. Do you try to sell into the same 15% or take a hit to margins and sell into the 85% where you have no competition for a while?

First off we are looking at 50 50, 50% for performance, 50% from midrange and value together.

Using those numbers, because we know those are per Q numbers of last quarter, I would go for the higher margin products first, I want my investment back as quickly as possible and put my competitor at a disadvantage by hurting them for the next 6 months to a year in that segment and also keep the halo going for my lower end products when they come out.


Was that their original plan? No idea. But given the current market situation, they stand to gain more from their strategy than from going after nVidia's new cards.

Their original plan, lets wait and see what we can do with Polaris if we can get it into the performance segment 300 bucks and higher great that is our best option, if not we have another plan that won't be as good but still we can deliver a good product and at good price range with no competition to worry about in the short term *we don't know when the competitor will be launching their cards to compete in this segment but we have at least 1 month.

Even now if they were able to get Polaris with reasonable wattage compared to the competition at Nano level of performance they can charge 300 bucks. They just need to bin of P10, one that can clock higher at less voltage lol. But that isn't happening even though people are saying they can get clocks up to 1600 mhz.

Come on the 1070 the cheapest it is, is 379 bucks which we haven't seen yet, at least I haven't on Newegg, wouldn't a $300 alternative with performance around 15% of a 1070 look good?

That way they can have the low end performance and the midrange, and the low end too all at the same time.

pretty much they couldn't do it. They couldn't make a product that would be compelling enough at 300 bucks as a reference card which those reason we will see why once reviews come out.
 
Last edited:
How good did the R9 290 look when it matched or beat the GTX 780 for $250 less?
Even post-price cut ($500), the 780 was 25% more expensive for the same performance.
 
290 didn't get great reviews though initially I think that is what hurt it most.
 
290 didn't get great reviews though initially I think that is what hurt it most.
The 780 also OC'd better, used less power, and AMD still had issues with their DX11 drivers (plus no DX12 advantage).
Normalizing performance-per-dollar: The 1070 is 42% faster than the 390X, using that same performance value for the RX 480 as well as a $230 initial price you end up at $326, about $100 cheaper than the 1070 currently... The same price difference between the 290 and 780 at launch.

The 480 is going to be positioned a lot better this time around, though. I wouldn't expect Nvidia to be concerned with it unless it can reach 1070 levels sub-$300.
 
It's not unreasonable for them to have designed Polaris to be around Fury X levels when mildly overclocked, Nano levels stock.
I disagree. The card is just too small, and I just don't agree that Polaris 10 would have been the first die ever to fill the 3 slots of 490,480,470. There is no room for such a card in the lineup given Vega 10 and Vega 11. P11 would be entry-level (and compared to the 950), but suddenly P10 would be Fury level?

A RX490 with Nano performance stock, >= Fury X performance when overclocked, at around 130W. That, to me, sounds about right for a node jump + FinFET + architectural changes.
It would've been amazing and unbelievable given the perf/watt of Grenada. AMD are already claiming close to 3x perf/watt from their previous arch.
According to your performance numbers where would P11 end up? An entry level 50Watt card at 390 performance?

If Polaris 10 had Nano performance stock, Fury X overclocked, it would place full Polaris 10 squarely in the enthusiast tier. And with a cut down die for the 480 for the mainstream segment.
And it would mean that P10 would clock ~1.8ghz/1.9ghz. I don't see it. They would also have to use HBM2 or GDDR5x. It doesn't make sense.

I have nothing more to add regarding this or anything new to add. I don't really get the "it is not unreasonable" debate. There is no evidence that P10 should've been enthusiast or anywhere close to Fury X, except some misunderstood (indeed, maybe even intentionally misleading and vague) slides and quotes (and maybe also some wishful thinking). No direct AMD quotes, no nothing. Perhaps AMD were hoping for a faster Polaris 10 card, but not as much as people are insinuating. They explicitly state (multiple times) how they're not going for performance but for perf/watt. They give a talk about Polaris in January about how Polaris is all about perf/$ (i.e. no HBM, no costly memory), explicitly marking the 970 if I recall correctly. Also, again, the card is just too small.

We shall see how P10 compares to GP106 in the future.

Now if AMD assumed 1080 to be = 980Ti, and 1070 to be = 980, then full Polaris 10 at Nano/Fury X level stock/OCed and cut Polaris 10 at Fury/Nano levels would've made pretty much every single gamer in the world drool, especially if they priced it at, say, $500 and $400.
I've already stated it a few times. Only someone uninformed would assume that 1080=980ti. The 1080 had to be at least 25% better than the 980ti (stock), as AIBs were already releasing 20% faster 980tis.
 
  • Like
Reactions: NKD
like this
They did have new products, the 300 series are all new microcodes but also they were inferior with power usage, they made changes in the silicon to create those changes,and you are right inferior performance higher end cards.

That is why they didn't weren't compelling products. Just because they are in channel and people that don't know too much about graphics cards are picking them up or they are in prebuilt systems.



You can't use steam survey because those numbers don't have a range of dates of who is buying what at what dates, you have cards on there that are more than one gen old, are those cards even being sold at in last year? Probably not, then you have take out the integrated market numbers too. You can't get a good picture of what is being sold now or in recent Q's



First off we are looking at 50 50, 50% for performance, 50% from midrange and value together.

Using those numbers, because we know those are per Q numbers of last quarter, I would go for the higher margin products first, I want my investment back as quickly as possible and put my competitor at a disadvantage by hurting them for the next 6 months to a year in that segment and also keep the halo going for my lower end products when they come out.




Their original plan, lets wait and see what we can do with Polaris if we can get it into the performance segment 300 bucks and higher great that is our best option, if not we have another plan that won't be as good but still we can deliver a good product and at good price range with no competition to worry about in the short term *we don't know when the competitor will be launching their cards to compete in this segment but we have at least 1 month.

Even now if they were able to get Polaris with reasonable wattage compared to the competition at Nano level of performance they can charge 300 bucks. They just need to bin of P10, one that can clock higher at less voltage lol. But that isn't happening even though people are saying they can get clocks up to 1600 mhz.

Come on the 1070 the cheapest it is, is 379 bucks which we haven't seen yet, at least I haven't on Newegg, wouldn't a $300 alternative with performance around 15% of a 1070 look good?

That way they can have the low end performance and the midrange, and the low end too all at the same time.

pretty much they couldn't do it. They couldn't make a product that would be compelling enough at 300 bucks as a reference card which those reason we will see why once reviews come out.

you know for a fact they don't have binned chips for 300?

How about AIB cards coming out at 1500 stock and that may clock up to 1600? At that point we are talking about with in striking distance of 1070.

Yes we will see if that happens, but you can't say they don't have binned chips any more than I can say they will because rumors point towards that, when we already have rumors or AIBs coming out with 1500mhz stock cards. If that happens they are using binned chips most likely or simply it needs more juice, either way I will wait to make a judgement on that until after I see those cards at 300. Those might be the ones coming creeping up to 1070.
 
you know for a fact they don't have binned chips for 300?

How about AIB cards coming out at 1500 stock and that may clock up to 1600? At that point we are talking about with in striking distance of 1070.

Yes we will see if that happens, but you can't say they don't have binned chips that will overclock well when we already have rumors or AIBs coming out with 1500mhz stock cards.
Anything's possible but there's no reason to assume that's true since there's no precedent for it.
 
Nvidia went from ~1250 MHz Maxwell stock boost to about 1800 MHz Pascal stock boost, an increase of 44%. Apply the same to AMD's stock boost of 1000 MHz and you get 1440 MHz.

So with the RX 480 actually being clocked at 1266 MHz, then we might assume AMD is aggressively binning chips or heavily under-clocking to maximize efficiency. 1440 MHz with an additional 10% factory OC (typical) makes 1584 MHz.

Pure napkin math, however.
 
I disagree. The card is just too small, and I just don't agree that Polaris 10 would have been the first die ever to fill the 3 slots of 490,480,470. There is no room for such a card in the lineup given Vega 10 and Vega 11. P11 would be entry-level (and compared to the 950), but suddenly P10 would be Fury level?


It would've been amazing and unbelievable given the perf/watt of Grenada. AMD are already claiming close to 3x perf/watt from their previous arch.
According to your performance numbers where would P11 end up? An entry level 50Watt card at 390 performance?


And it would mean that P10 would clock ~1.8ghz/1.9ghz. I don't see it. They would also have to use HBM2 or GDDR5x. It doesn't make sense.

I have nothing more to add regarding this or anything new to add. I don't really get the "it is not unreasonable" debate. There is no evidence that P10 should've been enthusiast or anywhere close to Fury X, except some misunderstood (indeed, maybe even intentionally misleading and vague) slides and quotes (and maybe also some wishful thinking). No direct AMD quotes, no nothing. Perhaps AMD were hoping for a faster Polaris 10 card, but not as much as people are insinuating. They explicitly state (multiple times) how they're not going for performance but for perf/watt. They give a talk about Polaris in January about how Polaris is all about perf/$ (i.e. no HBM, no costly memory), explicitly marking the 970 if I recall correctly. Also, again, the card is just too small.

We shall see how P10 compares to GP106 in the future.


I've already stated it a few times. Only someone uninformed would assume that 1080=980ti. The 1080 had to be at least 25% better than the 980ti (stock), as AIBs were already releasing 20% faster 980tis.

You can beat this around all you want. But you won't be able to make anyone understand. I posted an anandtech article that stated polaris will have a gddr5 part first which is exatly what happened and you are right. After Kyle's article there are few here defending his stance on that article. Nothing wrong with that. But its a cult, there is no evidence from anywhere, not a word from amd that polaris was suppose to enthusiast card. But other will push their agenda with no proof,
 
Nvidia went from ~1250 MHz Maxwell stock boost to about 1800 MHz Pascal stock boost, an increase of 44%. Apply the same to AMD's stock boost of 1000 MHz and you get 1440 MHz.

So with the RX 480 actually being clocked at 1266 MHz, then we might assume AMD is aggressively binning chips or heavily under-clocking to maximize efficiency. 1440 MHz with an additional 10% factory OC (typical) makes 1584 MHz.

Pure napkin math, however.

Well you are looking at a higher end parts. Reason we are seeing low supply may be due to not all chips running at those stated frequencies and may be nvidia cutting those down to go in 1060's later.

Regardless though. GTX 1070 is 1680 boost. So lets go by that we cant straight go to 1080. Most Polaris cards are hitting 1400 and even 1500+ so does look like they clocked it lower and let AIBs fuck with it. we shall see. That is still a 50% increase from their previous architecture if it does clock to 1500 and 40% if it clocks to 1400 on the cards with 6pin connector. Because most amd cards average around 1000. That is not bad a similar increase in clock speed that nvidia got. Now Nvidia always had more headroom due to their architecture but it won't be bad for amd if they can pull off 1400-1500 stable.
 
Speculation is fun, this has been a great few days in these threads. I really wish the NDA was up the 24th.
 
you know for a fact they don't have binned chips for 300?

How about AIB cards coming out at 1500 stock and that may clock up to 1600? At that point we are talking about with in striking distance of 1070.

Yes we will see if that happens, but you can't say they don't have binned chips any more than I can say they will because rumors point towards that, when we already have rumors or AIBs coming out with 1500mhz stock cards. If that happens they are using binned chips most likely or simply it needs more juice, either way I will wait to make a judgement on that until after I see those cards at 300. Those might be the ones coming creeping up to 1070.


They definitely don't have it for reference boards. I am going to say that right now 100% for sure.

The thing is to get into the performance bracket price range they don't need to get to the1070 lol, because the pricing of the 1070 is at 379 bucks, AMD only needs to get to 300 bucks to be in the performance range. Even though people are saying they are getting crazy overclocks, there must be something in there that doesn't allow AMD to clock that high for their reference boards, if they were able to, I think they could have a P10 that would be compelling in the performance range of cards.

And the clocks don't matter much its the end performance, I'm thinking the power usage wouldn't have looked good for a reference card to demand 300 bucks.

They would have gotten in the same issue as the 3xx and Fury Lines, good performance just too much power consumption compared to their nV counterparts.
 
Last edited:
They definitely don't have it for reference boards. I am going to say that right now 100% for sure.

The thing is to get into the performance bracket price range they don't need to get to the1070 lol, because the pricing of the 1070 is at 379 bucks, AMD only needs to get to 300 bucks to be in the performance range. Even though people are saying they are getting crazy overclocks, there must be something in there that doesn't allow AMD to clock that high for their reference boards, if they were able to, I think they could have a P10 that would be compelling in the performance range of cards.

Ofcourse I am not expecting 1500oc with reference unless you get a really nice chip. 1400 should be do able may be little over. You are right if they can have 1500mhz reported card under 300 and they can get to 1600 for more power(which is obivous since you will be pushing the chip, which I don't care about) That would be intriguing option to say the lease. I am also interested in seeing how far the 6 pin card can be pushed.
 
I disagree. The card is just too small, and I just don't agree that Polaris 10 would have been the first die ever to fill the 3 slots of 490,480,470. There is no room for such a card in the lineup given Vega 10 and Vega 11. P11 would be entry-level (and compared to the 950), but suddenly P10 would be Fury level?

It's not too small. The 1070 is a 75% part of the 1080. The 1080 is at 314mm2. 75% of that is at 235mm2. And guess what, the 1070 is Titan X/GTX 980Ti level performance.

So why is it hard to believe that AMD, with a full 232mm2 die that's more dense, wouldn't at least make Nano/Fury X level performance? And you don't need Pascal clocks to do it, either.

Also, let's assume that Polaris will only support GDDR5. That will further tilt in its favour, because the competition uses a memory controller that needs to support both GDDR5X and GDDR5, making it more complex and takes up more space on the die, while with Polaris they can get away with a smaller, less complex memory controller. That allows for more die area used for performance improvements in comparison with Pascal.

It would've been amazing and unbelievable given the perf/watt of Grenada. AMD are already claiming close to 3x perf/watt from their previous arch.
According to your performance numbers where would P11 end up? An entry level 50Watt card at 390 performance?


And it would mean that P10 would clock ~1.8ghz/1.9ghz. I don't see it. They would also have to use HBM2 or GDDR5x. It doesn't make sense.

I have nothing more to add regarding this or anything new to add. I don't really get the "it is not unreasonable" debate. There is no evidence that P10 should've been enthusiast or anywhere close to Fury X, except some misunderstood (indeed, maybe even intentionally misleading and vague) slides and quotes (and maybe also some wishful thinking). No direct AMD quotes, no nothing. Perhaps AMD were hoping for a faster Polaris 10 card, but not as much as people are insinuating. They explicitly state (multiple times) how they're not going for performance but for perf/watt. They give a talk about Polaris in January about how Polaris is all about perf/$ (i.e. no HBM, no costly memory), explicitly marking the 970 if I recall correctly. Also, again, the card is just too small.

We shall see how P10 compares to GP106 in the future.


I've already stated it a few times. Only someone uninformed would assume that 1080=980ti. The 1080 had to be at least 25% better than the 980ti (stock), as AIBs were already releasing 20% faster 980tis.

Going by TPU's review of the Fury X it's only 20% faster than the 390x. That means theoretically you only need clock speeds around 1500 to be Fury X level performance, assuming linear scaling. Account in some of the architectural changes and you can get that number down further. I don't see why you'd need 1.8GHz, unless you want to compare pure FLOPs, which is largely useless outside of specific gaming scenarios.

Also,

This is Polaris 10 and that’s Polaris 11. In terms of what we’ve done at the high level, it’s our most revolutionary jump in performance so far. We’ve redesigned many blocks in our cores. We’ve redesigned the main processor, a new geometry processor, a completely new fourth-generation Graphics Core Next with a very high increase in performance. We have new multimedia cores, a new display engine.

How would you interpret most revolutionary jump in performance so far?
 
It's not too small. The 1070 is a 75% part of the 1080. The 1080 is at 314mm2. 75% of that is at 235mm2. And guess what, the 1070 is Titan X/GTX 980Ti level performance.

So why is it hard to believe that AMD, with a full 232mm2 die that's more dense, wouldn't at least make Nano/Fury X level performance? And you don't need Pascal clocks to do it, either.

Also, let's assume that Polaris will only support GDDR5. That will further tilt in its favour, because the competition uses a memory controller that needs to support both GDDR5X and GDDR5, making it more complex and takes up more space on the die, while with Polaris they can get away with a smaller, less complex memory controller. That allows for more die area used for performance improvements in comparison with Pascal.



Going by TPU's review of the Fury X it's only 20% faster than the 390x. That means theoretically you only need clock speeds around 1500 to be Fury X level performance, assuming linear scaling. Account in some of the architectural changes and you can get that number down further. I don't see why you'd need 1.8GHz, unless you want to compare pure FLOPs, which is largely useless outside of specific gaming scenarios.

Also,



How would you interpret most revolutionary jump in performance so far?


That's true you won't need 1.8ghz to get to fury x level lol. I think you will start to see it match it at 1500 or trade blows atleast.
 
That's true you won't need 1.8ghz to get to fury x level lol. I think you will start to see it match it at 1500 or trade blows atleast.


I think it has to be higher than 1500...to get to Fury X, I'm thinking 1500 will get you to nano or in the middle of nano and furyX
 
It's not too small. The 1070 is a 75% part of the 1080. The 1080 is at 314mm2. 75% of that is at 235mm2. And guess what, the 1070 is Titan X/GTX 980Ti level performance.

So why is it hard to believe that AMD, with a full 232mm2 die that's more dense, wouldn't at least make Nano/Fury X level performance? And you don't need Pascal clocks to do it, either.

Your calculation is wrong. the 1070 is not 75% of the 1080. It has 75% amount of the shaders and texture units, yes. However, they also have the same amount of ROPs, they both have a 256bit memory bus, and they also both have the video encoder/decoder. I assume that there are also other things that they share, and the "effective" die is bigger than 75% of the 1080.

Going by TPU's review of the Fury X it's only 20% faster than the 390x. That means theoretically you only need clock speeds around 1500 to be Fury X level performance, assuming linear scaling. Account in some of the architectural changes and you can get that number down further. I don't see why you'd need 1.8GHz, unless you want to compare pure FLOPs, which is largely useless outside of specific gaming scenarios.
It appears that I have exaggerated a bit with my clocks. I did not check exactly what the delta is. From the current benchmarks (leaks) which place the 480 a bit under the 390x, it would seem that you would need around 1.6ghz clocks to equate a Fury X @ 1440p (assuming no memory bottlenecks). Still quite a distance from the ~1266 clocks, and definitely much higher than the previous gen's 1050mhz of the 390x. I can believe 1350/1400 clocks as the target (as I already said, suppose that they're targeting vanilla Fury), but it would still be some distance from a Fury X.

Either way, you said "this part meant that they planned on beating existing GPUs from NVIDIA, such as Titan X and GTX 980Ti.". As I already mentioned in a previous post, the Fury X is not really close to any decent AIB 980ti. That 690$ MSRP (at the time) 980ti is 30% faster than the Fury X at 4k or 28% faster at 1080p. It is inconceivable that anyone was seriously considering this small die as a 980ti killer.

How would you interpret most revolutionary jump in performance so far?
I would interpret it in the manner of the whole conversation? Raja explicitly stated that they weren't going after the performance crown but after perf/watt. Isn't a 3x improvement of perf/watt revolutionary?

Indeed, from January 4th:
Radeon%20Technologies%20Group_Graphics%202016-page-017.jpg


"a historic leap in performance per watt for Radeon GPUs"

I will only link the next images, as I do not want to spam the page:
http://images.anandtech.com/doci/9886/Radeon Technologies Group_Graphics 2016-page-014.jpg
http://images.anandtech.com/doci/9886/Radeon Technologies Group_Graphics 2016-page-009.jpg
http://images.anandtech.com/doci/9886/Radeon Technologies Group_Graphics 2016-page-013.jpg

They all mention perf/watt and how this is the best perf/watt jump for amd in their history. The only mention of performance is at the low end. Nothing about Fury X performance.

Is there any inherent reason you keep ignoring my Vega and P11 questions?
 
linkgoron

There would be no way in a million years, nV or AMD would specifically target only the midrange and low end markets and leave the performance segment alone for an entire 2 quarters, they would end up loosing too much, just 1 quarter for not having a performance card to go up against the gtx 970 and 980 they lost a sizeable chuck of marketshare lol, you think they want to do that again? And you can't tell me that they lost in other segments because nV didn't have any new cards for 1 quarter after they released the 970 and 980. Something must of have forced AMD's hand to do something like that, and we will KNOW exactly what that is with the review of the rx480.
 
the rx480 will be on shelves on the 29th. Perhaps the 24th if AMD does decide to move up the NDA

There will be no non-reference cards at launch day. Most you will see is a bit altered clocks, and a backplate. Example at the 290/290x launch the XFX version of the 290 Black edition had a clock of 980mhz which was above the base clock of 947mhz. You can expect the same thing with the RX series cards.

I will also confirm clock speeds can be in excess of 1400mhz on the reference card when overclocked.
 
Last edited:
linkgoron

There would be no way in a million years, nV or AMD would specifically target only the midrange and low end markets and leave the performance segment alone for an entire 2 quarters, they would end up loosing too much, just 1 quarter for not having a performance card to go up against the gtx 970 and 980 they lost a sizeable chuck of marketshare lol, you think they want to do that again? And you can't tell me that they lost in other segments because nV didn't have any new cards for 1 quarter after they released the 970 and 980. Something must of have forced AMD's hand to do something like that, and we will KNOW exactly what that is with the review of the rx480.
Well if you plan to get rid of your previous high end stock that would be a good time to do it for AMD. Come out with your high end first as in Nvidia's case preserves you previous generation and allows them to be sold off. AMD coming out low to Mainstream throws everything into chaos. If AMD can deliver the cards in numbers, perform at next tier up levels - what is Nvdia going to do with all the 980's and down cards on the shelves :D?

The 1070 DX12 benchmarks are around the Fury level DX12 benchmarks - is that about right?
 
If AMD can deliver the cards in numbers, perform at next tier up levels - what is Nvdia going to do with all the 980's and down cards on the shelves :D?
Its not NVidias problem.
Prices will drop to shift inventory. That will dent AMDs sales a bit.
 
Well if you plan to get rid of your previous high end stock that would be a good time to do it for AMD. Come out with your high end first as in Nvidia's case preserves you previous generation and allows them to be sold off. AMD coming out low to Mainstream throws everything into chaos. If AMD can deliver the cards in numbers, perform at next tier up levels - what is Nvdia going to do with all the 980's and down cards on the shelves :D?

The 1070 DX12 benchmarks are around the Fury level DX12 benchmarks - is that about right?


Yes that is what I stated, they don't need to get to FuryX or 1070 performance to hit the performance market lol. At 300 bucks and Nano performance would be a great card but for a reference design they won't be able to get that, and we will know why soon enough. If they did that they would have the performance market and the mid range and the low end, no segment untouched and being alone in the mainstream and value for how ever time.

Also Fiji right now is pretty much unsalable with the 1070 well if it was ever in stock.

nV has always been pretty good about inventory levels, so I don't foresee any problems on their end. Very few times have they miscalculated on managing their inventory and at when they did the calculation wasn't too far off, only about 10 days. If I remember correctly the highest inventory numbers they have had in recent years after the fx series pretty much, its been around 70 days, they like to keep inventory at around 60 days.
 
Last edited:
Your calculation is wrong. the 1070 is not 75% of the 1080. It has 75% amount of the shaders and texture units, yes. However, they also have the same amount of ROPs, they both have a 256bit memory bus, and they also both have the video encoder/decoder. I assume that there are also other things that they share, and the "effective" die is bigger than 75% of the 1080.


It appears that I have exaggerated a bit with my clocks. I did not check exactly what the delta is. From the current benchmarks (leaks) which place the 480 a bit under the 390x, it would seem that you would need around 1.6ghz clocks to equate a Fury X @ 1440p (assuming no memory bottlenecks). Still quite a distance from the ~1266 clocks, and definitely much higher than the previous gen's 1050mhz of the 390x. I can believe 1350/1400 clocks as the target (as I already said, suppose that they're targeting vanilla Fury), but it would still be some distance from a Fury X.

Most 390(x) hit around 1150MHz or more, not 1050. Performance-wise it's up for debate, but it's plausible they targeted the 390 instead of 390x, which is a 30% gap. Assuming 1266 gives you 390 performance, theoretically you will need minimum 1.65GHz to get to Fury X levels. Give or take performance scaling vs clocks, and arch improvements, and I'm guesstimating around 1.55-1.6GHz for Fury X performance.

That would've been par for the course. Shift 1 tier performance down, compete with 980Ti replacement from NVIDIA with lower MSRP.

Either way, you said "this part meant that they planned on beating existing GPUs from NVIDIA, such as Titan X and GTX 980Ti.". As I already mentioned in a previous post, the Fury X is not really close to any decent AIB 980ti. That 690$ MSRP (at the time) 980ti is 30% faster than the Fury X at 4k or 28% faster at 1080p. It is inconceivable that anyone was seriously considering this small die as a 980ti killer.


I would interpret it in the manner of the whole conversation? Raja explicitly stated that they weren't going after the performance crown but after perf/watt. Isn't a 3x improvement of perf/watt revolutionary?

I will only link the next images, as I do not want to spam the page:
http://images.anandtech.com/doci/9886/Radeon Technologies Group_Graphics 2016-page-014.jpg
http://images.anandtech.com/doci/9886/Radeon Technologies Group_Graphics 2016-page-009.jpg
http://images.anandtech.com/doci/9886/Radeon Technologies Group_Graphics 2016-page-013.jpg

They all mention perf/watt and how this is the best perf/watt jump for amd in their history. The only mention of performance is at the low end. Nothing about Fury X performance.

Is there any inherent reason you keep ignoring my Vega and P11 questions?

AIB cards are not used for marketing purposes. You're being disingenuous by including those in the comparison. Even NVIDIA compares the 1070 to stock Titan X, not overclocked Titan X.

Interpretation is personal, so let's agree to disagree.

As for Vega, I'd say those are Fury replacements. So AMD's lineup could've gone like this:
  • P11: 460 470 + mobile chips, countering GP106/108
  • P10: 480 490 + enthusiast mobile chips, countering GP104
  • Cut Vega Nano/Fury replacement, countering GP102
  • Full Vega Fury X replacement, countering GP100.
But the current situation looks to be like this:
  • P11: competes against GP108
  • P10: competes against GP106
  • Cut Vega: competing against cut GP104
  • Full Vega: competing against full GP104
OR
  • P11: competes against GP108
  • P10: competes against GP106
  • Cut Vega: competing against GP102 (Ti Part)
  • Full Vega: competing against GP100 (Titan Part)
In either case it leaves a pretty big hole in AMD's lineup. Either they don't compete with GP104 at all due to Vega being much more powerful or they don't compete with GP102 because Vega is too weak.

It's possible AMD can go with this lineup:
  • P11 - GP108
  • P10 - GP106
  • Castrated Vega: Full GP104
  • Cut Vega: GP102
  • Full Vega: GP100
That'll still leave the 1070 to dominate that segment. It's highly unlikely that AMD will introduce yet another cut of Vega to compete with the 1070 simply because the margins will be too small. Vega will probably compete >$500, and P10 <$300. Then what's gonna compete $300-$500?
 
no one is going to want a 980/970 anymore. Either they will wait for the GTX 1060, or buy a RX-480/470.
Price and performance will dictate that.
They will shift them to clear inventory space.
 
Hoping for some rather good deals with the Polaris at the lower end and Pascal at the higher end making everything older obsolete.

the 1070 pretty much decimates AMD line above $350 as well as Nvidia 980Ti and TitanX. Price/performance
The 1080 pretty much decimates Nvidia 980Ti and TitanX - Pure Performance

The 480/470/460 Pretty much decimates anything less than $350.
 
linkgoron

There would be no way in a million years, nV or AMD would specifically target only the midrange and low end markets and leave the performance segment alone for an entire 2 quarters, they would end up loosing too much, just 1 quarter for not having a performance card to go up against the gtx 970 and 980 they lost a sizeable chuck of marketshare lol, you think they want to do that again? And you can't tell me that they lost in other segments because nV didn't have any new cards for 1 quarter after they released the 970 and 980. Something must of have forced AMD's hand to do something like that, and we will KNOW exactly what that is with the review of the rx480.

What would have forced amd's hand? Like all of us knowing that hbm2 wont be ready for another quarter? Common now, this conspiracy theory is getting kind of old. if the card is at 110w while gaming and gives you 390/390x performance. That is historic leap in performance per watt compared to amd's last gen cards that were space heaters and bunch of pieces of shit when it came to power usage.

Why would amd decide not to make a card with 40CUs or 44CUs nothing is stopping them unless they dont want to. Seriously just tell me why? Are they stupid that they intentionally decided to make a 2300 shader card first? May be its a new process and they wanted to make as many as they can first and then ramp it up to higher end die. Remember 28nm, shit was horrible. Look at 1070s no standard models to be found anywhere and no stock, I see founders edition in stock may be one or two here and there. May be amd didn't wanna take a chance with bigger die and horrible yields on this process until it gets more refined.

I really believe AMD didn't want to go straight to a big die given the fuck ups global foundries has done in the past. So they probably went with what they were comfortable with at first and giving them the best yields after all they are supplying this chip to oem's so they probably decided to focus on one die size for all in the beginning and make as many as they can and then slowly build out bigger chips. Makes perfect sense.
 
Plus look at the prices of 390 and 390x looks like amd is clearing them out fast.

lowest 390 on newegg is already at 249.99 and lowest 390x is at $319

The prices have literally tanked in the last month. Looks like they are on their way out.
 
Status
Not open for further replies.
Back
Top