AMD to launch Volcanic Islands GPU (HD8000s) in 2013

[vitchilo]

Then it's not 4096sp.

In frequencies... (in mhz)

[LordEC911]

Is the 20 nm rumor one of them?

Yes but I personally think it is somewhat unlikely, at least for the highend GPU.

Over a month? Well, either AMD is telling me the wrong thing or your source is exaggerating 2 weeks old is exactly what I was told (not a rumor, absolute proof haha). But not a huge difference.

I don't think we are talking about the same ASIC.

[BababooeyHTJ]

FYI- They are/were targeting a ~60% increase over Tahiti.

How do they plan on doing that without considerably higher power consumption? As is Tahiti is no small die and Ghz edition draws a lot and runs hot.

They had a lot more room to work with Cypress as far as die size and power consumption goes and didn't come close to a 60% improvement with Cayman.

Are they taking up Nvidia's massive die strategy all of a sudden?

[NEOAethyr]

I think honestly they are going for bankruptcy, on purpose.
It seems to be a fad these days.
You go bankrupt so you can ask for more money from the bankers.
I got a hint of it looking up the euro crisis, somewhere I heard on the news that amd was talking to certain bankers.
I don't remember where or anything but at the time, I thought they was going towards that direction because of who they were talking to.

I'm betting that they will try to count as to big to fail, being in all the new consoles, and having the rights to x86-64.
I mean really, x86-64, how could you even get a patent for something that is so simple lol.
Eh whatever though ^^.

The next gen of consoles are a joke anyways, I personally wouldn't back those.
But whatever floats your boat.

[PedantOne]

How do they plan on doing that without considerably higher power consumption? As is Tahiti is no small die and Ghz edition draws a lot and runs hot.

They had a lot more room to work with Cypress as far as die size and power consumption goes and didn't come close to a 60% improvement with Cayman.

Are they taking up Nvidia's massive die strategy all of a sudden?

LordEC911 - blah blah, it is absolute nonsense. On 28nm is impossible to make a bigger chip then Tahiti. it would not be cost-effective

[Sushi Warrior]

Yes but I personally think it is somewhat unlikely, at least for the highend GPU.


I don't think we are talking about the same ASIC.

I can guarantee you Hawaii was first tested 2 weeks ago... Maybe it was in transit from TSMC for 2 more weeks

I wonder if they are going to take the high-end "superchip" route that Titan took. I'm not sure if they're going to do the same thing as they did with Intel and focus on the low end.

[vitchilo]

I think honestly they are going for bankruptcy, on purpose.
It seems to be a fad these days.
You go bankrupt so you can ask for more money from the bankers.

Get real. AMD is not a bank. AMD hasn't most of government on their payroll, the scam can go on without AMD.

[tajoh111]

AMD can't afford to make a super high end big chip. They don't have a big enough foothold in the super computing/professional market to make it financially feasible. That being the payoff from such a heavy r and d move and the number of wafers at 20nm these take up, would be wasted unless they sell a lot more professional cards.

With the cash reserves so close to the breaking point, its a move AMD can't afford. AMD will not take the huge monolithic chip route unless they have more guarantees in the professional market. The only way AMD is getting a 60 percent performance increase is through 20nm. It isn't happening at 28nm.

It seems convenient for AMD that these rumors pop up just when the gtx 7xx series is coming out.

[NEOAethyr]

Get real. AMD is not a bank. AMD hasn't most of government on their payroll, the scam can go on without AMD.

I never said they were a bank lol.
http://www.reuters.com/article/2012/...8AC14Z20121113

Again I'll repeat myself.
I'm only guessing, I don't know, but this is my guess.
They'll run the company into the ground, and a bank will bail them out.

Increase sales in there professional line lol..
Yeah right, not gonna happen yo.
No one is gonna buy amd for professional use.
They make consumer parts, there professional parts, all though labeled as such they are not that's for sure.
That's where I would say get real lol...

I don't mean no offense dudes, I don't think I'm pissing in the wind either.

[BababooeyHTJ]

The only way AMD is getting a 60 percent performance increase is through 20nm. It isn't happening at 28nm.

Exactly, is Tahiti even 60% faster than Cayman?

I know that it wasn't at launch but maybe now.

[LordEC911]

How do they plan on doing that without considerably higher power consumption? As is Tahiti is no small die and Ghz edition draws a lot and runs hot.

They had a lot more room to work with Cypress as far as die size and power consumption goes and didn't come close to a 60% improvement with Cayman.

Are they taking up Nvidia's massive die strategy all of a sudden?

I have heard multiple possibilities...
Cypress->Cayman isn't the same situation.
No.

LordEC911 - blah blah, it is absolute nonsense. On 28nm is impossible to make a bigger chip then Tahiti. it would not be cost-effective

What?

I can guarantee you Hawaii was first tested 2 weeks ago... Maybe it was in transit from TSMC for 2 more weeks

I wonder if they are going to take the high-end "superchip" route that Titan took. I'm not sure if they're going to do the same thing as they did with Intel and focus on the low end.

Again... I don't think we are talking about the same ASIC.

AMD can't afford to make a super high end big chip. They don't have a big enough foothold in the super computing/professional market to make it financially feasible. That being the payoff from such a heavy r and d move and the number of wafers at 20nm these take up, would be wasted unless they sell a lot more professional cards.

With the cash reserves so close to the breaking point, its a move AMD can't afford. AMD will not take the huge monolithic chip route unless they have more guarantees in the professional market. The only way AMD is getting a 60 percent performance increase is through 20nm. It isn't happening at 28nm.

It seems convenient for AMD that these rumors pop up just when the gtx 7xx series is coming out.

The only way is through 20nm? That isn't the only option...

Exactly, is Tahiti even 60% faster than Cayman?

I know that it wasn't at launch but maybe now.

~30-40% on average at launch, 50-60% back in 2H '12, GHZ edition is 60-70% faster during the same time period.

Edit- I just want to point out that I am probably more confused with all the stuff I have heard compared to the rest of you. I am just trying to make sense of it and not just throw out BS rumors.

[tajoh111]

I have heard multiple possibilities...
Cypress->Cayman isn't the same situation.
No.

Again... I don't think we are talking about the same ASIC.

The only way is through 20nm? That isn't the only option...


~30-40% on average at launch, 50-60% back in 2H '12, GHZ edition is 60-70% faster during the same time period.

Edit- I just want to point out that I am probably more confused with all the stuff I have heard compared to the rest of you. I am just trying to make sense of it and not just throw out BS rumors.

What?

Only way to get 60% at 28nm is for AMD to go monolithic while throwing out a bunch of compute features to lower power consumptions which is paradoxical for a large chip.

Depending on how well it actually performs in compute tasks, Titan is actually efficient with its size and compute performance. Titan is actually 100 percent faster than a gtx 580(at 2560*1600), which hasn't been said about any flagship over the prior in a while, even with a new architecture, even on a new node. I think the last time this was actually the case was the gtx 8800, because 5870, 6970, gtx 280, gtx 480, gtx 580 don't fullfill that requirement.

And even still AMD needs another 15-25 percent faster card than titan to reach 60 percent faster than tahiti.

I remember you saying similar things about the 6970 over the 5870 and we all know how that turned out.

[LordEC911]

Only way to get 60% at 28nm is for AMD to go monolithic while throwing out a bunch of compute features to lower power consumptions which is paradoxical for a large chip.

Well at least you are thinking in the right direction now... though not making the right calculations.

Depending on how well it actually performs in compute tasks, Titan is actually efficient with its size and compute performance. Titan is actually 100 percent faster than a gtx 580(at 2560*1600), which hasn't been said about any flagship over the prior in a while, even with a new architecture, even on a new node. I think the last time this was actually the case was the gtx 8800, because 5870, 6970, gtx 280, gtx 480, gtx 580 don't fullfill that requirement.

LoL, Nvidia got two tries to get close and in truth 100% over GTX580 is stretching a bit. More like 70-80% which has been done.
So what is the 4870?

I remember you saying similar things about the 6970 over the 5870 and we all know how that turned out.

I apologize for getting performance target data from old designs...
Though to toot my own horn I was right on the money with almost everything about Tahiti.

[zalbard]

LoL, Nvidia got two tries to get close and in truth 100% over GTX580 is stretching a bit.

You were saying?

[tajoh111]

Well at least you are thinking in the right direction now... though not making the right calculations.


LoL, Nvidia got two tries to get close and in truth 100% over GTX580 is stretching a bit. More like 70-80% which has been done.
So what is the 4870?


I apologize for getting performance target data from old designs...
Though to toot my own horn I was right on the money with almost everything about Tahiti.

tech powerup tends to reduce averages(e.g the 7970 originally tested only 10 percent faster than the gtx 580) because of the amount of games they use and older games and even they got 85 percent at the highest resolution. look at the comparison chart between the 4870 vs its predecessors

Vs titan vs its competitors and you will just how dramatic of a improvement titan is vs earlier cards. It beats the pants off the dual cards of last generation and spanks the big monolithic chips of last generation. Compared this to the 4870 and the 4870 results are just not as dramatic. Although the comparison isn't perfect because during the 4870 the only chart there is the combined chart, take the combined chart from titan and the difference between titan and old cards is still alot larger than the 4870s.


If nvidia didn't price titan as it did and it came earlier, it would have been the second coming of the gtx 8800. The more impressive thing about titan is it demolishes similar sized monolithes from last gen. Not tiny cards like the 3870. Basically for the 4870 to do something similar, it would have to double the performance of the gtx 8800. This is from a pure performance point of view(the most impressive thing about the 4870 was the value, the least impressive thing about titan).

http://www.techpowerup.com/reviews/N..._Titan/27.html

And

http://www.techpowerup.com/reviews/D...D_4870/23.html

[Iconyu]

This is probably the best time for AMD to gamble big.

Pros:

1) They know the 700 series will be slower than Titan, even with rumours of a Titan ultra on the horizon AMD can predict it's specs and work beyond them
2) They've already got developers on their side due to their work on next gen Consoles, thus games will be optimised more for AMD and they can insure their drivers are good for these games before launch. (A position nvidia used very well)
3) With access to these developers they can release their new cards with nice game bundles that take the sting out of the price.
4) Nvidia are distracted by the mobile market.
5) The fastest card on the block sells, the slower card has to take a price cut.

Cons:

1) Yields: On 20nm it'll be risky to go all out with a huge chip this early, on 28nm it would be crazy to go with what would be GF100 sized monolith. Possible but crazy.
2) Timing: Any delay would allow nvida to react with a chip outside of AMD's plans, the 8970 would have to land early enough so nvidia would effectively stay silent about the 800 series to protect 700 series sales.
3) Drivers: The driver team would have to be spot on, right out of the gates, it took them too long to realise the potential of the 7970.
4) Demand: Titan is selling because people think that the 780GTX will slower, thus they think there won't be a faster card on the market until Maxwell. It's an assuring investment, a super fast but super expensive 8970 will have Maxwell on the Horizon just as the 7970 had the presence of the 680GTX looming behind it.

It's tight, but I think the Pro's outweigh the cons this time around. It's rare to see nvidia unfocused on the GPU market, but to be honest if any company needed to diversify it would be nvidia. Right now the PC market is shrinking, every idiot wants to make a phone, and too many people just do facebook, and twitter. And Nintendo has been outshone by Gungho entertainment with the release of a single mobile game.

[Russian]

This is probably the best time for AMD to gamble big.

Pros:

1) They know the 700 series will be slower than Titan, even with rumours of a Titan ultra on the horizon AMD can predict it's specs and work beyond them
2) They've already got developers on their side due to their work on next gen Consoles, thus games will be optimised more for AMD and they can insure their drivers are good for these games before launch. (A position nvidia used very well)
3) With access to these developers they can release their new cards with nice game bundles that take the sting out of the price.
4) Nvidia are distracted by the mobile market.
5) The fastest card on the block sells, the slower card has to take a price cut.

Cons:

1) Yields: On 20nm it'll be risky to go all out with a huge chip this early, on 28nm it would be crazy to go with what would be GF100 sized monolith. Possible but crazy.
2) Timing: Any delay would allow nvida to react with a chip outside of AMD's plans, the 8970 would have to land early enough so nvidia would effectively stay silent about the 800 series to protect 700 series sales.
3) Drivers: The driver team would have to be spot on, right out of the gates, it took them too long to realise the potential of the 7970.
4) Demand: Titan is selling because people think that the 780GTX will slower, thus they think there won't be a faster card on the market until Maxwell. It's an assuring investment, a super fast but super expensive 8970 will have Maxwell on the Horizon just as the 7970 had the presence of the 680GTX looming behind it.

It's tight, but I think the Pro's outweigh the cons this time around. It's rare to see nvidia unfocused on the GPU market, but to be honest if any company needed to diversify it would be nvidia. Right now the PC market is shrinking, every idiot wants to make a phone, and too many people just do facebook, and twitter. And Nintendo has been outshone by Gungho entertainment with the release of a single mobile game.

Not a bad analysis, however, I would not say that Nvidia is unfocused on GPU Technology or their GPU business. They're full aware that they need to keep it rolling and rolling fast if they're going to maintain the profitability they have. Without professional graphics and gaming graphics, nvidia is a company that loses money.

[SKYMTL]

AMD is getting into a bit of a pickle these days. They publicly stated that 2013 wouldn't see a new high end card introduced with the HD 7970 GHz remaining in the flagship position. So unless they were intentionally misdirecting the press and their customers, I can't see anything that would compete against GK110 this year.

Plus, after the disastrous HD 7990 launch, their driver team needs to focus on a number of key issues before they even think about announcing anything.

Personally I think this is damage control time since GK110 is pretty much untouchable right now.

First batch of Hawaii is already out of the oven btw.

"Already"? If they hope for a 2013 release, the first batch of ES cards would have to be "out of the oven" so to speak a good 4 months ago, if not earlier. For example, NVIDIA had working Kepler cards about a year before the GTX 680 was officially announced.

They have had silicon in house for over a month now.
There are at least 3 plausible rumors floating around out there right now but with a key source keeping quieat it might be awhile until things get semi-confirmed.

FYI- They are/were targeting a ~60% increase over Tahiti.

If the increase is only 60% over Tahiti, Maxwell will walk all over it from day one.

[Tao~]

Product placement (which depends on the design) was decided almost 2 years ago, now is only the question of getting it to the market at the appropriate price. The question is, how soon will TSMC 20nm be ready to produce 300-something mm^2 chips with >20% yields ?

As for Maxwell, AMD will either try to get a 6-9 month headstart (unlikely) or hope Nvidia don't do another GK104 with Maxwell, preferably both. I wonder if AMD themselves work on 2 designs this time around.

[LordEC911]

If the increase is only 60% over Tahiti, Maxwell will walk all over it from day one.

We will see what happens. There might be something else in the works to compete with Maxwell...

[SKYMTL]

We will see what happens. There might be something else in the works to compete with Maxwell...

There's always something new in the works. However, I don't see AMD and NVIDIA taking parallel paths in the graphics market for all that much longer. Especially as AMD transitions to APUs and NVIDIA moves towards dynamic virtualized memory.

[hurleybird]

I could see AMD going with a really big die that doesn't move away from the current GCN architecture all that much. AMD's GT200 if you will. All AAA devs will be targeting that architecture anyway, and the cost savings they would get from slowing down architectural development could go into making larger dies.

There's always something new in the works. However, I don't see AMD and NVIDIA taking parallel paths in the graphics market for all that much longer. Especially as AMD transitions to APUs and NVIDIA moves towards dynamic virtualized memory.

I think with AMD locking up next generation consoles it forces Nvidia to focus on making GCN's current strengths their own for future generations. Forward+ rendering, for example, is something they very badly to catch up on.

[highoctane]

I think with AMD locking up next generation consoles it forces Nvidia to focus on making GCN's current strengths their own for future generations. Forward+ rendering, for example, is something they very badly to catch up on.

AMD has already had 2/3 the console market for years now with no appreciable impact on PC gaming. Before next gen consoles even hit the shelves we will be seeing next gen PC hardware being put to use, the console hardware is locked 5+ years until the next gen.

If anything new PC hardware from both AMD and Nvidia will be hamstrung by yet to be released console hardware for the next half decade as far as console ports are concerned.

In the end I don't see a tangible performance advantage or disadvantage for either brand because the hardware needed to run console ports will have the bar set much lower than pc hardware is capable of already today. We already have PC hardware more than capable of handling anything next gen consoles would need to run.

[Rollo]

AMD is getting into a bit of a pickle these days. They publicly stated that 2013 wouldn't see a new high end card introduced with the HD 7970 GHz remaining in the flagship position...Plus, after the disastrous HD 7990 launch, their driver team needs to focus on a number of key issues before they even think about announcing anything.

I have not seen a post by anyone on the forums about their 7990, it may be the worst selling high end launch of all time. So I checked at newegg, all out of stock, but only about 20 reviews for all five cards.

Then I went to Amazon, not selling them. So I went to Tiger Direct, one model, no reviews.

I tried Google Shopping, got the one on TigerDirect, and a couple more on Ebay.

The 7990 seems to truly be The Card That Time Forgot.

You'd think AMD would have lots of these on hand as the chips are 17 months old, but no one has them and no one seems to be buying them the few that exist.

Contrast this with NV's last two $1000 chip launches: Lots of vendors had the cards, product going in and out of stock daily, lots of people on the forums posting about their new cards.

AMD could learn something about timing from this. If they put out a quad 7970 card next year as their high end, people will give it the same reception.

Don't know if they were duped by TSMC on 20nm process availability, but two whole years without new high end parts is disastrous for them. About the only way they'll be selling 7970s at any decent price is if they throw in a FX8350.

[kromosto]

amd never needed a card like 7990. my guess is they just launched it for getting attention but if they really thought it will sell well then this means they have a really big problem in their way of thinking.