AMD Radeon HD 9970 Specifications Leaked – Twice as fast as GTX 780 (ChipLoco rumor)

[insertcarehere]

Some news about partners reading monstrous colling(Gigabyte Windforce new cooling seems to be 50% stronger than their 7970 cooler) solutions arrived these days. Has AMD gone big to fight the Titan?

Assuming the newest news are not false... new bet from me:
- 9970 beat Titan by the exact margin that GTX 580 bested HD 5870 in 2010(Its possible even on 28nm, assuming titan don't was made to gaming(the GK110 performance/mm² is horrible...)). Beats GTX 880 by about the same margin but will use double of the power. Of course will beat GTX 690 in some games....
- double-fan cooling(Like 7990 cooling tech) on reference model to make 9970 temperatures and noise impressive.
- Power efficiency for the 9970 will be better than the power efficiency of 7970. But 9970 power consumption will be par to GTX 690.
- No-overvolting overclocking 9970 will make the performance past 770 sli performance by 10%. Max OCed 9970(will max out at something about 1250Mhz) will be 50% faster than 780HOF.
- Frametimes distribution will be better than Nvidias relatively and ever(in every VI vs Kelper) crossfire scaling(FPS) will be greater by about 8%. Hardware frame pacing will come in PI or next graphics architecture. This time VI crossfires will be good on frame latencies like 7990 was or a little better, but not more.
512 bit interface - 4GB Vram ~@6,5GHZ;
3328sps (52 CUs) - 1,6 times increased.

According to TPU, the GTX580 was faster than the 5870 by 33% in 1200p on average, keep in mind that cypress was not that big on 40nm and how massive GF110 was in comparison. If 9970 is to prove to beat titan by the same margin, even with a slightly more area efficient architecture, a 512bit, 52CU gpu will be at the absolute edge of what is acheivable on 28nm (>=600mm^2), to keep power consumption under control, it would also have to be clocked very low. Given that AMD hasn't pursued this all-out strategy for years, I don't see why they would change so abruptly now.

 

[SolMiester]

deleted

 

[Keysplayr]

If you think that AMD isn't going to release a faster than GTX780/ single-GPU card on 28 nm, that's fine. They would still need a faster-than-7970Ghz chip to make the dual-GPU card you've been talking about. This thread is about that faster-than-7970Ghz chip and a card/s based on it(not that it means anything by now). While a dual-GPU card is likely, there is no reason to think it's going to be released first or even soon.

It's like going into a Titan rumour thread and saying "Nvidia needs GTX690 to beat a 7970Ghz, so they're going to make a smaller GK104 and the GK110 isn't going to happen".

Well, not exactly. Nvidia has a huge die history. Not so much with AMD. So, in a Titan rumour thread, people saying it's a single huge die would probably be met with some modicum of collective agreement.
People in here want to see AMD succeed. I can't blame them for having hopes that AMD will release a HUGE die Titan killer. While I am keeping an open mind about it, I just don't see 9970 or whatever arrive at 450+mm2 huge die. This isn't just a blind dismissal of such a thing happening, but taking several pieces of data into consideration which I have explained (repeatedly) in several posts throughout this thread.

AMD could even be releasing a faster GPU that is smaller than Tahiti. It could be Pitcairn or Bonaire, either one extremely updated, doubled or tripled specs with as high of a transistor density possible on 28nm and can be more efficient and faster than Tahiti. And two of these can be placed on a single PCB, with corrected Xfire multi-monitor drivers. To me, this is a far more likely scenario than anything else being thrown out there.

One reason I'm so reluctant to believe any AMD rumours is because of their last FUD campaign. Releasing false information all over to throw off competition and to prevent/ferret out partner leaks. Probably even to their own staff. Just look at what happened to JFAMD.

 

[Leadbox]

Well, not exactly. Nvidia has a huge die history. Not so much with AMD. So, in a Titan rumour thread, people saying it's a single huge die would probably be met with some modicum of collective agreement.
People in here want to see AMD succeed. I can't blame them for having hopes that AMD will release a HUGE die Titan killer. While I am keeping an open mind about it, I just don't see 9970 or whatever arrive at 450+mm2 huge die. This isn't just a blind dismissal of such a thing happening, but taking several pieces of data into consideration which I have explained (repeatedly) in several posts throughout this thread.

AMD could even be releasing a faster GPU that is smaller than Tahiti. It could be Pitcairn or Bonaire, either one extremely updated, doubled or tripled specs with as high of a transistor density possible on 28nm and can be more efficient and faster than Tahiti. And two of these can be placed on a single PCB, with corrected Xfire multi-monitor drivers. To me, this is a far more likely scenario than anything else being thrown out there.

One reason I'm so reluctant to believe any AMD rumours is because of their last FUD campaign. Releasing false information all over to throw off competition and to prevent/ferret out partner leaks. Probably even to their own staff. Just look at what happened to JFAMD.

There's only so much density you can have on any given node and if you're going to double/triple specs... well, sounds like a big die to me

 

[Keysplayr]

There's only so much density you can have on any given node and if you're going to double/triple specs... well, sounds like a big die to me

Really? How big is bonaire? 160mm2 Pitcairn? 212mm2

Taken from PCPerspective:
http://www.pcper.com/reviews/Graphi...eview-Sea-Islands-and-Bonaire-Make-Appearance



I know what you're thinking right off the bat. Double Pitcairn and you get 424mm2. This however does not account for not needing every aspect of the GPU doubled or increased transistor density. You say there is only so much density on a given node. While I agree with you, what is the maximum transistor per mm2 for TSMC's 28nm High performance process? Is there a maximum written anywhere? I'll look.

EDIT: Sheesh. Any engineers in here? I've looked but it's mostly Cuneiform to me.

 

[MeldarthX]

Really? How big is bonaire? 160mm2 Pitcairn? 212mm2

Taken from PCPerspective:
http://www.pcper.com/reviews/Graphi...eview-Sea-Islands-and-Bonaire-Make-Appearance



I know what you're thinking right off the bat. Double Pitcairn and you get 424mm2. This however does not account for not needing every aspect of the GPU doubled or increased transistor density. You say there is only so much density on a given node. While I agree with you, what is the maximum transistor per mm2 for TSMC's 28nm High performance process? Is there a maximum written anywhere? I'll look.

Right Keys - you continue to ignore the fact where AMD has stated on record - the next target they want to go after is HPC market.

They can't do that with double a stripped chip of DP - not going to happen. Second they will be restructuring GCN int 2.0 - They will be going bigger this time no matter how you try to argue they can't or won't.

To go after HPC market they are going to need to fix issues in Tahiti - rebalance it - and continue the work on GCN - which is what they are doing.

Will it beat Titan? I believe it will; will it be say a Titan Ultra; that I don't know - now how conserv AMD goes with their clocks I don't know.

Now for their 7800 series double of Pitcairn is very much the way they could go and it would a fast chip for gaming; but also still have damn good chip in Tahiti in the 7970.....

AMD will go bigger than they have been for last few generations - how big? I've already said most likely 450m-480m range; yes its a guess but its logical guess from what AMD has been saying along with the other leaks.

Now we'll get to see soon - AMD learned how to be masters of misdirection when it comes to their video hardware.

 

[Leadbox]

Bonnaire is probably about as dense as AMD will go on 28nm
If you look at Bonnaire vs Cape Verde its:

Doubled frontend (2ACE and 2 geometry vs 1 each for CV)
40% more SPs (896 vs 640)

All in a 33% larger die for roughly the same 33% performance increase and some ~5W more power.
A 33% larger Tahiti would be ~485mm^2

 

[Keysplayr]

Bonnaire is probably about as dense as AMD will go on 28nm
If you look at Bonnaire vs Cape Verde its:

Doubled frontend (2ACE and 2 geometry vs 1 each for CV)
40% more SPs (896 vs 640)

All in a 33% larger die for roughly the same 33% performance increase and some ~5W more power.
A 33% larger Tahiti would be ~485mm^2

All probably due to Bonaire's 128-bit memory system.

But how do you figure Bonaire is about as dense as AMD will go? What is the maximum transistor per mm2 on this process? Are they there? Are they even close? Or, given as much time as they've had with 28nm, has it matured enough for even higher densities?

I'm not saying you're wrong. I'm asking how you know. That is all.

 

[Keysplayr]

Right Keys - you continue to ignore the fact where AMD has stated on record - the next target they want to go after is HPC market.

They can't do that with double a stripped chip of DP - not going to happen. Second they will be restructuring GCN int 2.0 - They will be going bigger this time no matter how you try to argue they can't or won't.

To go after HPC market they are going to need to fix issues in Tahiti - rebalance it - and continue the work on GCN - which is what they are doing.

Will it beat Titan? I believe it will; will it be say a Titan Ultra; that I don't know - now how conserv AMD goes with their clocks I don't know.

Now for their 7800 series double of Pitcairn is very much the way they could go and it would a fast chip for gaming; but also still have damn good chip in Tahiti in the 7970.....

AMD will go bigger than they have been for last few generations - how big? I've already said most likely 450m-480m range; yes its a guess but its logical guess from what AMD has been saying along with the other leaks.

Now we'll get to see soon - AMD learned how to be masters of misdirection when it comes to their video hardware.

D: I'm not "ignoring" anything dude. You are the one believing anything you hear, not me. You are ignoring the voice of reason and totally dismissing anything that suggests AMD can't or won't go big die. Aren't you? I am at least willing to accept that miracles do happen, but aren't likely to.

Wanting to go after the HPC market, and actually having the financial and technical capability to do so are two entirely different things. I believe AMD "wants" to do a lot of things. Like release a superior CPU to Intel for example.

 

[Leadbox]

All probably due to Bonaire's 128-bit memory system.

But how do you figure Bonaire is about as dense as AMD will go? What is the maximum transistor per mm2 on this process? Are they there? Are they even close? Or, given as much time as they've had with 28nm, has it matured enough for even higher densities?

I'm not saying you're wrong. I'm asking how you know. That is all.

I don't know, but I pretty sure they're at or around their limits. It's like a cup, there's only so much water you can put in there, in the end, it's easier to get a bigger cup if you want more water than it is to try and rearrange the water molecule to get more in the same old cup

 

[dac7nco]

your shoes size + your house number = hd 9970 specifications

521

lol

 

[raghu78]

D: I'm not "ignoring" anything dude. You are the one believing anything you hear, not me. You are ignoring the voice of reason and totally dismissing anything that suggests AMD can't or won't go big die. Aren't you? I am at least willing to accept that miracles do happen, but aren't likely to.

Wanting to go after the HPC market, and actually having the financial and technical capability to do so are two entirely different things. I believe AMD "wants" to do a lot of things. Like release a superior CPU to Intel for example.

rubbish. You dwell in the past and keep idolizing Intel and Nvidia. But even in the past AMD has beaten Intel twice at their own game - Athlon K7 and Athlon 64 K8. This is inspite of having much lesser resources and revenues which are 10% of Intel. Intel in fact is a company which abused its monopoly and has been fined across the world.

http://www.nytimes.com/2009/05/14/business/global/14compete.html?pagewanted=all&_r=0

So stop talking up Intel. If anything AMD needs to be given credit for fighting and surviving against an abusive monopoly. As for Nvidia atleast they are not an abusive monopoly. They are the leader in the graphics industry but AMD competes quite well.

AMD is definitely capable of producing a massive die GPU which competes with Nvidia's best. The same goes for their CPU division. Over the next 2 - 3 years AMD will re-establish itself as a leader in both the GPU and CPU markets. The semi-custom wins and Hawaii are the start of a long term revival of AMD. As usual you don't need to have any optimism towards AMD. But I would like to see the tone of your posts come mid-Oct.

 

[Zanovar]

Raghu as you seem to know so much about these upcoming card/s will you be buying one?dont guess as you know the specs already*Laughs*.

 

[videogames101]

rubbish. You dwell in the past and keep idolizing Intel and Nvidia. But even in the past AMD has beaten Intel twice at their own game - Athlon K7 and Athlon 64 K8. This is inspite of having much lesser resources and revenues which are 10% of Intel. Intel in fact is a company which abused its monopoly and has been fined across the world.

http://www.nytimes.com/2009/05/14/business/global/14compete.html?pagewanted=all&_r=0

So stop talking up Intel. If anything AMD needs to be given credit for fighting and surviving against an abusive monopoly. As for Nvidia atleast they are not an abusive monopoly. They are the leader in the graphics industry but AMD competes quite well.

AMD is definitely capable of producing a massive die GPU which competes with Nvidia's best. The same goes for their CPU division. Over the next 2 - 3 years AMD will re-establish itself as a leader in both the GPU and CPU markets. The semi-custom wins and Hawaii are the start of a long term revival of AMD. As usual you don't need to have any optimism towards AMD. But I would like to see the tone of your posts come mid-Oct.

normally i can't stand keys and his rhetoric, but here i'm gonna have to defend him - man from what it sounds like you just don't want to dump your amd stock - come off it

 

[Keysplayr]

I don't know, but I pretty sure they're at or around their limits. It's like a cup, there's only so much water you can put in there, in the end, it's easier to get a bigger cup if you want more water than it is to try and rearrange the water molecule to get more in the same old cup

True, but you can pour an awful lot of water into a cup that's already filled to the brim with sand.

 

[el etro]

According to TPU, the GTX580 was faster than the 5870 by 33% in 1200p on average

Hey, im not talking about 33%, i saw 22% there! 33% faster will be par to 7990!

 

[Keysplayr]

n/m

 

[rgallant]

Some news about partners reading monstrous colling(Gigabyte Windforce new cooling seems to be 50% stronger than their 7970 cooler) solutions arrived these days. Has AMD gone big to fight the Titan?

Assuming the newest news are not false... new bet from me:
- 9970 beat Titan by the exact margin that GTX 580 bested HD 5870 in 2010(Its possible even on 28nm, assuming titan don't was made to gaming(the GK110 performance/mm² is horrible...)). Beats GTX 880 by about the same margin but will use double of the power. Of course will beat GTX 690 in some games....
- double-fan cooling(Like 7990 cooling tech) on reference model to make 9970 temperatures and noise impressive.
- Power efficiency for the 9970 will be better than the power efficiency of 7970. But 9970 power consumption will be par to GTX 690.
- No-overvolting overclocking 9970 will make the performance past 770 sli performance by 10%. Max OCed 9970(will max out at something about 1250Mhz) will be 50% faster than 780HOF.
- Frametimes distribution will be better than Nvidias relatively and ever(in every VI vs Kelper) crossfire scaling(FPS) will be greater by about 8%. Hardware frame pacing will come in PI or next graphics architecture. This time VI crossfires will be good on frame latencies like 7990 was or a little better, but not more.
512 bit interface - 4GB Vram ~@6,5GHZ;
3328sps (52 CUs) - 1,6 times increased.

I hope so , then maybe nv will put their small die maxwell back into a filing cabinet and come out with a sizable die[400+nm] on 20nm.
-just have a feeling how the gtx780 has become king , the first high end maxwells would be less nm than the 680's to save cost while super high end buyers would wait for the larger dies , leaving most $550.00 buyers with smaller mid range chips launched as high end.

 

[el etro]

Well, this is the news about the cooling of Hawai's. Hoping monster dies this time!
http://www.pcgameshardware.de/AMD-R...-warum-AMD-auch-nur-mit-Wasser-kocht-1085977/

 

[Saylick]

Well, this is the news about the cooling of Hawai's. Hoping monster dies this time!
http://www.pcgameshardware.de/AMD-R...-warum-AMD-auch-nur-mit-Wasser-kocht-1085977/

Looks like single die.

 

[Saylick]

Also, rumors of new naming scheme.

 

[Keysplayr]

Looks like single die.

Looks like a hsf.

 

[el etro]

Maybe this is the original Windforce cooler(is that i understood). The new cooler will be more potent. Could the 9970 be a overclocking champion like Titan is?

 

[Saylick]

Looks like a hsf.

Nah man, it's clearly more than that. It's obviously a video card.

 

[Keysplayr]

Nah man, it's clearly more than that. It's obviously a video card.

You could be onto something.