Now that 7790 has been out, where is 7890 and 7980?

[Wall Street]

Now that Bonaire has been out for quite some time, where are the 256-bit and 384-bit versions?

This doesn't make much sense to me, AMD can compete vs. nVidia GPUs of the same size and can compete on memory frequency with this architecture but they aren't moving this technology up the product chain with any additional product releases beyond the 7790. In fact, isn't the new 7990 based on the older GCN chips? I have a few theories of what it could be:

1. nVidia's release 700 series is imminent and they want to play the spoiler this time rather than showing their pricing first.

2. Too much stock of older chips left to sell.

3. Low 28 mm wafer availability at TSMC (although this shouldn't be an issue by now).

 

[wand3r3r]

It's pretty easy to misread that code name.

4. PS4 is taking resources
5. Lack of funds is delaying stuff.
6. the real 2.0 on 20nm is being expedited.

I hope they actually release said chips to counter the 780 price hike. Even if 2.0 hits in December, it's so far out they should have something out 1.5 years after the initial 79xx. It's almost like the competition is slowing or they are just milking out everything (both sides).

 

[VulgarDisplay]

I see the 7790 as a test of some of the features they are going to implement in Volcanic Islands. They see that it works and they are now pushing full steam ahead for 20nm GPU's at the end of the year. Nvidia isn't really pressuring them to release new GPU's at all. Especially if the rumors of the gtx780 being around $700 are true. There is zero reason for AMD to try and get out a refresh on 28nm.

I think they are diverting all their resources towards hitting a home run on 20nm, and thus the 28nm refresh is not happening.

 

[sublime420ec]

I see the 7790 as a test of some of the features they are going to implement in Volcanic Islands. They see that it works and they are now pushing full steam ahead for 20nm GPU's at the end of the year. Nvidia isn't really pressuring them to release new GPU's at all. Especially if the rumors of the gtx780 being around $700 are true. There is zero reason for AMD to try and get out a refresh on 28nm.

I think they are diverting all their resources towards hitting a home run on 20nm, and thus the 28nm refresh is not happening.

i like your thinking

hope they are :biggrin:

 

[Arkadrel]

I think they are diverting all their resources towards hitting a home run on 20nm, and thus the 28nm refresh is not happening.

I hope thats why.

Some part of me wishes they had put out more parts though...
the 7770->7790, saw a 30-35%+ increase in performance.

If the same could be true on a 7870, that would be a pretty decent 256bit bus card.
A 7890 could potentially be a 680 rival if it had been made.

Amd could have stopped makeing their 79xx series.
A 384bit version would probably have been too big a chip to make.
That would be AMD makeing their version of a GTX Titan.

 

[Wall Street]

I would think that once you have one card, making larger versions wouldn't be that much work. Besides, 7790 came out in March, if 20 nm cards launch in December, that would be nine months which still should have been plenty of time to pay off the small fixed cost of taping out the other chips, given that these chips are so much better on a performance per mm2, performance per memory bus width and performance per watt basis.

Maybe this is a pipecleaner and when Bonaire taped out they didn't know if it would work, or come to market this soon or be as fast as it is.

 

[raghu78]

good question. the Hainan and Curacao code names were supposed to be significant updates to Pitcairn and Tahiti . Hainan was supposed to offer GTX 680 kind of perf on a sub USD 300 chip, basically a HD 7950 on a 256 bit memory bus with 6 ghz memory. perf/watt optimized part. Curacao was supposed to be the update to Tahiti. should have been around 25 - 30% faster than HD 7970 Ghz.

http://videocardz.com/39041/meet-ar...d-bonaire-the-codenames-radeon-hd-8000-series

http://videocardz.com/34981/amd-radeon-hd-8870-and-hd-8850-specifiation-leaked

But it looks like AMD skipped this 28nm refresh and pulled in the 20nm Volcanic Islands based on GCN 2.0 architecture. HD 9970 seems to be the chip which might launch in Q4 2013.

 

[iMacmatician]

My guess is that Curacao and Hainan were either canceled or merged with Hawaii and whatever the Pitcairn/Hainan successor is.

 

[DooKey]

good question. the Hainan and Curacao code names were supposed to be significant updates to Pitcairn and Tahiti . Hainan was supposed to offer GTX 680 kind of perf on a sub USD 300 chip, basically a HD 7950 on a 256 bit memory bus with 6 ghz memory. perf/watt optimized part. Curacao was supposed to be the update to Tahiti. should have been around 25 - 30% faster than HD 7970 Ghz.

http://videocardz.com/39041/meet-ar...d-bonaire-the-codenames-radeon-hd-8000-series

http://videocardz.com/34981/amd-radeon-hd-8870-and-hd-8850-specifiation-leaked

But it looks like AMD skipped this 28nm refresh and pulled in the 20nm Volcanic Islands based on GCN 2.0 architecture. HD 9970 seems to be the chip which might launch in Q4 2013.

Hopefully they'll have their crossfire issues worked out by then. I wouldn't mind a pair of Titan killer 20nm gpus.

 

[parvadomus]

My guess is that Curacao and Hainan were either canceled or merged with Hawaii and whatever the Pitcairn/Hainan successor is.

They are not canceled, they are waiting for GTX700 parts. I bet that when GTX780 is released, they will follow in less than a month, at least Curacao, maybe as a 7980 or something. Hainan may be canceled.. it all depends on current stock of 7950/7970.

 

[FalseChristian]

I've just got this feeling that AMD is gonna bring out something soon that will smoke the Titan. Let's hope so. $1,049.00 + tax for a GPU. Get real.

 

[toyota]

I've just got this feeling that AMD is gonna bring out something soon that will smoke the Titan. Let's hope so. $1,049.00 + tax for a GPU. Get real.

well they did bring out the 7990 and guess what...same price as Titan. lol

 

[3DVagabond]

I would think that once you have one card, making larger versions wouldn't be that much work. Besides, 7790 came out in March, if 20 nm cards launch in December, that would be nine months which still should have been plenty of time to pay off the small fixed cost of taping out the other chips, given that these chips are so much better on a performance per mm2, performance per memory bus width and performance per watt basis.

Maybe this is a pipecleaner and when Bonaire taped out they didn't know if it would work, or come to market this soon or be as fast as it is.

I'm not too sure you are correct on your assumptions. I think making larger versions of Bonaire isn't easy. It's easier to make smaller chips, harder to make big chips. Also 9 months is very likely not enough time to recoup the associated costs of bringing a line of cards to market.

Look at what nVidia is doing. They aren't releasing a lineup of new chips. They are taking GK110 and cutting it down further (This is the easier route. Not scaling it up.). They are than taking the GK104 and moving it down in their product stack. No new chips at all. This is why I don't think that 9 months (Or even a year anymore.) is enough to recoup the costs of a new product stack. Neither one of them are attempting it.

The 7790 was intended to fill the performance gap between the 7770 and 7850 in a bread and butter spot in the market, which is why it was released. It did that, but nVidia was very aware of it's performance target and price point and were easily able to counter it with the 650ti Boost.

 

[iMacmatician]

The 7790 was intended to fill the performance gap between the 7770 and 7850 in a bread and butter spot in the market, which is why it was released. It did that, but nVidia was very aware of it's performance target and price point and were easily able to counter it with the 650ti Boost.

I wouldn't be surprised if Bonaire stuck around for a while (depending on when AMD goes 20 nm).

 

[Wall Street]

I'm not too sure you are correct on your assumptions. I think making larger versions of Bonaire isn't easy. It's easier to make smaller chips, harder to make big chips. Also 9 months is very likely not enough time to recoup the associated costs of bringing a line of cards to market.

Look at what nVidia is doing. They aren't releasing a lineup of new chips. They are taking GK110 and cutting it down further (This is the easier route. Not scaling it up.). They are than taking the GK104 and moving it down in their product stack. No new chips at all. This is why I don't think that 9 months (Or even a year anymore.) is enough to recoup the costs of a new product stack. Neither one of them are attempting it.

Larger cards are "harder to make" from a yield standpoint, I was trying to suggest that it wouldn't be hard to make from a design standpoint, since they just need to add more of identical parallel parts.

Also, the GTX 770 is rumored to be based on a GK104-425 die instead of a GK104-400 that the original 670/680 had, which suggests that nVidia probably did indeed pay to tape out new chips for this refresh.

Finally, yes nVidia reacted to AMDs pricing of the 7790, but what is AMD supposed to respond with when nVidia refreshes. The HD7970 Ghz edition is a very big expensive card to be competing against a 256-bit sub-200 watt GTX 770.

 

[3DVagabond]

I wouldn't be surprised if Bonaire stuck around for a while (depending on when AMD goes 20 nm).

True. I wouldn't be surprised if we had a mixed process generation. 20nm used on the high end where volume is low and margins are high and 28nm used in the lower tier price and margin sensitive models.

 

[3DVagabond]

Larger cards are "harder to make" from a yield standpoint, I was trying to suggest that it wouldn't be hard to make from a design standpoint, since they just need to add more of identical parallel parts.

Also, the GTX 770 is rumored to be based on a GK104-425 die instead of a GK104-400 that the original 670/680 had, which suggests that nVidia probably did indeed pay to tape out new chips for this refresh.

Finally, yes nVidia reacted to AMDs pricing of the 7790, but what is AMD supposed to respond with when nVidia refreshes. The HD7970 Ghz edition is a very big expensive card to be competing against a 256-bit sub-200 watt GTX 770.

I understood what you said. I just don't think you're right about how it's not hard to design a larger chip. Or, more to the point that it's not expensive and that 9 mos. is enough time to recoup the cost.

 

[raghu78]

Larger cards are "harder to make" from a yield standpoint, I was trying to suggest that it wouldn't be hard to make from a design standpoint, since they just need to add more of identical parallel parts.

Also, the GTX 770 is rumored to be based on a GK104-425 die instead of a GK104-400 that the original 670/680 had, which suggests that nVidia probably did indeed pay to tape out new chips for this refresh.

Finally, yes nVidia reacted to AMDs pricing of the 7790, but what is AMD supposed to respond with when nVidia refreshes. The HD7970 Ghz edition is a very big expensive card to be competing against a 256-bit sub-200 watt GTX 770.

The only reasons AMD would have skipped the 28nm refresh of Pitcairn and Tahiti is that they are strained on financials and resources. Also AMD would have prioritized resources for PS4 and Xbox next release in Q4 2013 instead of the 28nm refresh. another reason which could have aided the decision is the pulling in of the 20nm Volcanic islands chip HD 9970.

Hainan's die size would have been around 270 - 280 sq mm and the perfect product to go against a GTX 760 Ti and GTX 770. Maybe AMD might release Hainan after Nvidia releases GTX 760 Ti. there will be no yield issues in producing this part.

Curacao would be around 420 - 440 sq mm and should be easier on a very mature 28nm process even though yields would be a slight problem. But if Nvidia can manufacture Titan which is 560 sq mm then AMD should have no problems manufacturing Curacao. The high end curacao will go against a GTX 780 aka Titan LE at USD 550. So AMD won't sell it for cheap.

So as I said the reasons for skipping must be purely one of resource constraints / financials and maybe expediting the 20nm HD 9970 product launch. But still as far as I can see skipping Curacao makes sense from a financial standpoint if AMD are looking at a 20nm HD 9970 in Q4 2013. but skipping Hainan seems silly. It would be a cheaper part to produce than Tahiti, consume less power than Tahiti and compete well with GK104. Also a true 20nm mid range part update to Hainan cannot release until the TSMC 20nm production volumes are high which could be as late as Q2 2014.

 

[3DVagabond]

The only reasons AMD would have skipped the 28nm refresh of Pitcairn and Tahiti is that they are strained on financials and resources. Also AMD would have prioritized resources for PS4 and Xbox next release in Q4 2013 instead of the 28nm refresh. another reason which could have aided the decision is the pulling in of the 20nm Volcanic islands chip HD 9970.

Hainan's die size would have been around 270 - 280 sq mm and the perfect product to go against a GTX 760 Ti and GTX 770. Maybe AMD might release Hainan after Nvidia releases GTX 760 Ti. there will be no yield issues in producing this part.

Curacao would be around 420 - 440 sq mm and should be easier on a very mature 28nm process even though yields would be a slight problem. But if Nvidia can manufacture Titan which is 560 sq mm then AMD should have no problems manufacturing Curacao. The high end curacao will go against a GTX 780 aka Titan LE at USD 550. So AMD won't sell it for cheap.

So as I said the reasons for skipping must be purely one of resource constraints / financials and maybe expediting the 20nm HD 9970 product launch. But still as far as I can see skipping Curacao makes sense from a financial standpoint if AMD are looking at a 20nm HD 9970 in Q4 2013. but skipping Hainan seems silly. It would be a cheaper part to produce than Tahiti, consume less power than Tahiti and compete well with GK104. Also a true 20nm mid range part update to Hainan cannot release until the TSMC 20nm production volumes are high which could be as late as Q2 2014.

They say here their reasons for doing it. The only thing that I've seen that changed is they are pulling VI (volcanic islands) forward. There was also a statement that releasing new cards isn't as big of a priority as changing the mindset that their cards aren't better than nVidia's. As long as they have to sell better cards for less money to be competitive they are fighting a losing battle. They have to be seen on par with nVidia to be able to sell faster cards for more money. Their biggest downfall is the public's perception of poorer drivers and budget value products. I don't think anyone percieves that their cards are of lesser quality than nVidia's.

As far as PS4, they worked with Sony on that. That means they didn't foot the whole bill. I don't think that's made them resource constrained, per se. Their negative financial situation is being caused by other things. Mainly poorer than anticipated CPU/APU sales.

 

[raghu78]

They say here their reasons for doing it. The only thing that I've seen that changed is they are pulling VI (volcanic islands) forward. There was also a statement that releasing new cards isn't as big of a priority as changing the mindset that their cards aren't better than nVidia's. As long as they have to sell better cards for less money to be competitive they are fighting a losing battle. They have to be seen on par with nVidia to be able to sell faster cards for more money. Their biggest downfall is the public's perception of poorer drivers and budget value products. I don't think anyone percieves that their cards are of lesser quality than nVidia's.

As far as PS4, they worked with Sony on that. That means they didn't foot the whole bill. I don't think that's made them resource constrained, per se. Their negative financial situation is being caused by other things. Mainly poorer than anticipated CPU/APU sales.

i agree that AMD might have wanted to go directly to 20nm HD 9970 instead of 28nm HD 8970 aka Curacao. but skipping the Hainan chip aka HD 8870 would be very bad from a sales point of view. AMD needs a GK104 competitor at USD 300. on perf, perf/watt and price. Otherwise GTX 760 Ti aka a faster GTX 670 will eat market share from AMD at the USD 300 price point. Thats one of the price points where AMD is doing very well with HD 7950 boost. a 1.1 Ghz HD 8870 would do very well against a similarly clocked GK104, not to mention improve AMD margins because of being a smaller chip. But not showing up to the fight would be painful for the next 6 - 9 months. AMD needs a Hainan which is 270 - 280 sq mm and competes well with GK104.

Drivers for single GPU have improved to the state where they are now on par. multi GPU not yet. would take atleast till July. changing mindset is going to take time. AMD needs to show over a period of time that they can compete with Nvidia on a lot more metrics than just performance. perf, perf/watt, drivers especially on launch day for major titles, CF vs SLI on frametimes and robustness of game support.

 

[3DVagabond]

i agree that AMD might have wanted to go directly to 20nm HD 9970 instead of 28nm HD 8970 aka Curacao. but skipping the Hainan chip aka HD 8870 would be very bad from a sales point of view. AMD needs a GK104 competitor at USD 300. on perf, perf/watt and price. Otherwise GTX 760 Ti aka a faster GTX 670 will eat market share from AMD at the USD 300 price point. Thats one of the price points where AMD is doing very well with HD 7950 boost. a 1.1 Ghz HD 8870 would do very well against a similarly clocked GK104, not to mention improve AMD margins because of being a smaller chip. But not showing up to the fight would be painful for the next 6 - 9 months. AMD needs a Hainan which is 270 - 280 sq mm and competes well with GK104.

They have a $300 chip that competes with GK104. It's the 7950. IMO, it's actually better than the GK104. Problem is that most people don't see the 7950 actually being the best card, everything considered, over the GTX670, GTX680, and the HD7970 (Some people do, obviously. Which is why so many on these boards own 7950's.). That's what AMD has to fix. Trust me, if the GTX670 was $299 against the rest of the other cards, $300 7950, $400 7970, and $450 680's would sit on the shelf and grow mold. the 7950 overall is marginally better than the 670, but people don't see the same value in it. This is what AMD needs to fix. Badly!

Drivers for single GPU have improved to the state where they are now on par. multi GPU not yet. would take atleast till July. changing mindset is going to take time. AMD needs to show over a period of time that they can compete with Nvidia on a lot more metrics than just performance. perf, perf/watt, drivers especially on launch day for major titles, CF vs SLI on frametimes and robustness of game support.

AMD is in the position now that they have to outperform nVidia in the driver department. A game comes out that performs bad on nVidia and people say, "Don't worry, nVidia's driver team rocks, they'll fix it." Or, "AMD cheated and that's why their cards are faster." When a game comes out that's faster on nVidia then people say, "AMD's drivers suck." See the difference?

They can release all of the new cards they want to, but until they fix this perception they are treading water at best.

 

[raghu78]

They have a $300 chip that competes with GK104. It's the 7950. IMO, it's actually better than the GK104. Problem is that most people don't see the 7950 actually being the best card, everything considered, over the GTX670, GTX680, and the HD7970 (Some people do, obviously. Which is why so many on these boards own 7950's.). That's what AMD has to fix. Trust me, if the GTX670 was $299 against the rest of the other cards, $300 7950, $400 7970, and $450 680's would sit on the shelf and grow mold. the 7950 overall is marginally better than the 670, but people don't see the same value in it. This is what AMD needs to fix. Badly!

The problem is the HD 7950 boost is based on the larger Tahiti chip. also AMD need to clock HD 7950 at 1 Ghz to match a GTX 670 at 1.1 ghz (which is what GTX 760 Ti is going to be). also the overclocking headroom of HD 7950 is rarely appreciated by tech sites. make no mistake when the GTX 760 Ti launches all the sites would give accolades to the GTX 760 Ti aka faster clocked GTX 670. some would also directly go so far as to say AMD needs to price cut HD 7950 boost. this is the problem. Even though the HD 7950 has good OC headroom and at average OC of 1150 mhz can compete with a GTX 670 at 1250 Mhz, the fact is the HD 7950 boost would be judged on the lower stock performance.

AMD is in the position now that they have to outperform nVidia in the driver department. A game comes out that performs bad on nVidia and people say, "Don't worry, nVidia's driver team rocks, they'll fix it." Or, "AMD cheated and that's why their cards are faster." When a game comes out that's faster on nVidia then people say, "AMD's drivers suck." See the difference?

They can release all of the new cards they want to, but until they fix this perception they are treading water at best.

true to some extent but that will take a longer while to change.

 

[parvadomus]

Hainan lives?
http://www.phoronix.com/scan.php?page=news_item&px=MTM3MTA

 

[wand3r3r]

Whoa, interesting. Surprising if they actually release 20nm at years end. Probably deserves it's own speculation thread.

 

[omeds]

A 384bit version would probably have been too big a chip to make.
That would be AMD makeing their version of a GTX Titan.

They already did make their version of Titan, its called 7970.

Hainan lives?
http://www.phoronix.com/scan.php?page=news_item&px=MTM3MTA

Interesting news!