[SA] GK110 aka GTX 680 release date: Late Q3 '12

[LOL_Wut_Axel]

But that's 2013 according to AMD's FAD slides: http://www.anandtech.com/show/5491/amds-2012-2013-client-cpugpuapu-roadmap-revealed

They can easily launch ahead of time. It's based on the same architecture and it'll be on the 28nm process, which by Q3 or so should be giving AMD noticeably better yields than currently.

 

[96Firebird]

You are very, very optimistic...

 

[blackened23]

They can easily launch ahead of time. It's based on the same architecture and it'll be on the 28nm process, which by Q3 or so should be giving AMD noticeably better yields than currently.

Also, sea islands is a new architecture. Who knows what naming convention a Tahiti refresh would have? 2013 is sea islands, that is correct. However, as we have seen Tahiti has a lot of headroom for higher clocks so I think it is very likely we'll get another super tahiti later this year. Similar to what NV did with the Fermi refresh - GTX 480 to 580. Same architecture, refined.

So lets summarize. Sea Islands is 2013. Tahiti refresh could happen whenever AMD Feels like it, and they could easily do it well before GTX 680 is released.

The 7970 scales ridiculously well and at 1250+ is honestly like a different GPU. It leaves a stock 7970 in the dust - 3dmark11 gains something ridiculous like 4000 points and framerates are astronomically higher.. So if AMD were to release a "refresh" clocked at 1300 or higher with better TDP, which is EXACTLY what NV did with Fermi, I could see AMD doing this. In fact, they would be crazy NOT to. Cayman was not refreshed because the architecture did not scale well at all with higher clock speeds, that was the nail in the coffin for it. That is not the case with the tahiti.

 

[LOL_Wut_Axel]

So you think AMD is going to refresh their Cayman Islands GPU's this year, on the same 28nm node, then come out with their new architecture - Sea Islands - next year, again, on the same 28nm node?

4870 series debut - 6/25/08
5870 series debut - 9/23/09
6970 series debut - 12/15/10

Hmmmm.....15 months between each new top end GPU. Something tells me AMD is not going to bring out a new lineup of cards anytime soon.

It's not a new architecture. And even if they don't release a whole new lineup by the end of the year, they can release a card with faster speeds and put a higher number on it--maybe HD 7980?

They know the HD 7970 alone isn't gonna be anywhere near enough to counter whatever NVIDIA will cook up with GK110. Just like in the past, NVIDIA is gonna be looking at making a new card that's 50-65% faster than the one it replaces. Even if it's hot and loud, the fastest NVIDIA can bring WILL be 30-40% faster than the HD 7970 (but no more). But as we've also seen, Tahiti has quite a lot of clocking headroom left in it and AMD can make a small respin with more compute units. We'll see what happens by then.

 

[Ajay]

They can easily launch ahead of time. It's based on the same architecture and it'll be on the 28nm process, which by Q3 or so should be giving AMD noticeably better yields than currently.

Well, that's not what AMD is saying - officially. And we don't know how easy it will be, there have been staff cuts and we don't know how ambitious Sea Islands design is.

 

[blackened23]

Well, that's not what AMD is saying - officially. And we don't know how easy it will be, there have been staff cuts and we don't know how ambitious Sea Islands design is.

AMD doesn't need to release Sea Islands this year because they have 5-6 months of competition free sales, and in addition to that they can release a refreshed Tahiti with better TDP and higher clock speeds. That is exactly what NV did with Fermi, and it seems likely that AMD will do something similar with their part since it scales very well.

 

[LOL_Wut_Axel]

Also, sea islands is a new architecture. Who knows what naming convention a Tahiti refresh would have? 2013 is sea islands, that is correct. However, as we have seen Tahiti has a lot of headroom for higher clocks so I think it is very likely we'll get another super tahiti later this year. Similar to what NV did with the Fermi refresh - GTX 480 to 580. Same architecture, refined.

So lets summarize. Sea Islands is 2013. Tahiti refresh could happen whenever AMD Feels like it.

The 7970 scales ridiculously well and at 1250+ is honestly like a different GPU. It leaves a stock 7970 in the dust - 3dmark11 gains something ridiculous like 4000 points and framerates are astronomically higher.. So if AMD were to release a "refresh" clocked at 1300 or higher with better TDP, which is EXACTLY what NV did with Fermi, I could see AMD doing this. In fact, they would be crazy NOT to. Cayman was not refreshed because the architecture did not scale well at all with higher clock speeds, that was the nail in the coffin for it. That is not the case with the tahiti.

Well, it'll be based on GCN, so it won't be a new architecture like going from the HD 5870 to the 6970. But that doesn't mean there can't be improvements; things like compute units and clock speeds can be easily upgraded as yields get better. With that they can easily get a ~30% improvement over the HD 7970.

 

[Arzachel]

You know that nVidia has Kepler-GPUs since months? That they know how to optimise their process tools? That's the reason why every Fermi Chip after GF100 came with A1 (tape out) to the market.

Because they had smaller dies and the legwork was already done with GF100. You know why both AMD and Nvidia usualy release the biggest chips before smaller ones? Because bugs, that can slip by unnoticed with a small chip, are much more likely to be an issue for a larger one. GK104 isn't that much of a help and even less so, if GK110 has additional compute features that need debugging.

 

[Will Robinson]

So you think AMD is going to refresh their Cayman Islands GPU's this year, on the same 28nm node, then come out with their new architecture - Sea Islands - next year, again, on the same 28nm node?

4870 series debut - 6/25/08
5870 series debut - 9/23/09
6970 series debut - 12/15/10

Hmmmm.....15 months between each new top end GPU. Something tells me AMD is not going to bring out a new lineup of cards anytime soon.

:whiste:

 

[evilspoons]

Damnit, my poor GTX 285 looks like it's got another six months in "the brig".

 

[Stuka87]

All AMD would really have to do is release a 7980, or basically an upclocked 7970. There is plenty of headroom for the 7970 to be sold at a higher clock. They then adjust prices of the already released models down some. This would be extremely easy to do, would not require a new series of cards, and would allow them to help combat whatever the 680 ends up being.

 

[wahdangun]

yup just like i predicted before, nvdia will not release their flagship GPU on new process. because they don't want to repeat their FX 5800 mistake, so i was expecting nvdia to release some die-shrink fermi, just like GT240 before GTX480.

 

[tviceman]

It's not a new architecture. And even if they don't release a whole new lineup by the end of the year, they can release a card with faster speeds and put a higher number on it--maybe HD 7980?

They know the HD 7970 alone isn't gonna be anywhere near enough to counter whatever NVIDIA will cook up with GK110. Just like in the past, NVIDIA is gonna be looking at making a new card that's 50-65% faster than the one it replaces. Even if it's hot and loud, the fastest NVIDIA can bring WILL be 30-40% faster than the HD 7970 (but no more). But as we've also seen, Tahiti has quite a lot of clocking headroom left in it and AMD can make a small respin with more compute units. We'll see what happens by then.

It does have a large headroom, yes. But I also remember the same things were being said about the 5870 prior to gtx480's release. A "5890" would come out and essentially negate all performance advantages Nvidia had.... but it never happened.

 

[tviceman]

:whiste:

I have no idea what this means or how, in any way, this post or any other post of yours contributes anything useful to any thread you have participated in.

 

[Lonyo]

I have no idea what this means or how, in any way, this post or any other post of yours contributes anything useful to any thread you have participated in.

Well your post conveniently missed out the HD7970, which was 1 year and 2 weeks after the 6970.
Maybe because it broke your 15 month cycle.
1 year after HD7970 would be January 2013.
Late Q3 is August-September sort of time, which means potentially only 3 months later. Which is an acceptable time if you consider the time lag between the 5870 and 480, or the 7970 and 6xx.

 

[blackened23]

It does have a large headroom, yes. But I also remember the same things were being said about the 5870 prior to gtx480's release. A "5890" would come out and essentially negate all performance advantages Nvidia had.... but it never happened.

I don't recall any such thing being said - IIRC the 5870 scaled horribly. The 6970 did as well- 6970 in particular scaled maybe 1-2% with 20% higher gpu clock speeds.... The performance doesn't even really increase with higher clock speeds. I think its pretty obvious why the cayman architecture wasn't refreshed. The architecture didn't allow for it.

We all know this isn't the case with Tahiti, it performs like a different gpu at 1200 and up. AMD could easily do a Fermi type refresh, the question is will they? Who knows. I personally think they would be stupid not to because clearly tahiti allows for it (while previous architectures didn't) Just as a basis of comparison, judging from benchmarks at H a 7970 clocked at 1260 performs roughly 35% faster than a 7970 clocked at stock clocks. That kind of scaling is ridiculous and unheard of, which is why I think they will release another part down the road.

 

[Quantos]

I agree that the 7970 almost begs to be refreshed. However, we can only speculate on that, as nothing from AMD has thus far even hinted at such a refresh (unless I'm wrong, but I don't recall anything).

The potential definitely is there, but maybe AMD knowing that Kepler is so late won't bother spending time working on a refresh while they can work on Sea Islands instead? Maybe working on a refresh of the 7970 would cause Sea Islands to come out later in 2013? If that's the case, perhaps they would rather have Sea Islands come out as soon as possible, fully knowing and accepting the fact that Kepler would reign on the market for a bit?

 

[Despoiler]

OMG, you have a card already? Did you get an engineering sample, or do you work for one of the vendors? Very interested to hear more...

If you want to know more you should look to Fermi's launch.....

 

[tviceman]

Well your post conveniently missed out the HD7970, which was 1 year and 2 weeks after the 6970.
Maybe because it broke your 15 month cycle.
1 year after HD7970 would be January 2013.
Late Q3 is August-September sort of time, which means potentially only 3 months later. Which is an acceptable time if you consider the time lag between the 5870 and 480, or the 7970 and 6xx.

I honestly didn't mean to leave it out. In fact I had it pulled up and meant to put it, but I just now realized I closed it out on accident without putting it in.

Anyways, 12 months or 15 months, the point I was trying to make is that AMD will probably not have an immediate leapfrog answer to high end Kepler. 6 months later sure, that seems reasonable, but nothing immediate.

 

[AtenRa]

Nice, GK110 was taped out. That means it will be ready for Xmas time 2012. :biggrin:

Now lets talk about GK100 launch in April 2012

 

[Silverforce11]

They dont even have to do a "refresh", as process matures the clock head room should go up. The cards aren't stuck on reference designs, this opens up AIBs to make special high OC cards and price it accordingly.

Or heck, they can do a 7970 XT with an extra 200-300 mhz at default settings.

 

[blackened23]

They dont even have to do a "refresh", as process matures the clock head room should go up. The cards aren't stuck on reference designs, this opens up AIBs to make special high OC cards and price it accordingly.

Or heck, they can do a 7970 XT with an extra 200-300 mhz at default settings.

This sounds pretty likely. I could see an XTX with a 1200-1300mhz clock. Thats still a huge performance leap over a stock clocked 7970, 30-40% higher as stated earlier.

 

[OILFIELDTRASH]

I was told in the "nvidia will win big" Charlie article thread that Charlie was full of it and can't be trusted. Now that Charlie is saying kepler will be extremely late he's now legit.

 

[OILFIELDTRASH]

They need to release 7970's at 1300mhz since all of them can do that. Strange though that many can't get above 1050mhz once they reach us.

 

[blackened23]

I was told in the "nvidia will win big" Charlie article thread that Charlie was full of it and can't be trusted. Now that Charlie is saying kepler will be extremely late he's now legit.

What I gather is that he saw tailored benchmarks provided by NV that were for physx only games. When he received complete benchmarks (across a wider spectrum of games) thats when the whole hardware ageia part came into play - and that kepler will have "peaky" performance. High in phsx games, not so high in others.

Anyway, he had a pretty good - actually flawless IIRC - track record with Fermi leaks. We won't know anything for sure unless NV speaks up, they're being curiously quiet...which is strange. They were all too eager to speak about Fermi even when it wasn't close to release.