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

[hackerballs]

darn darn darn, now I have to trade in my 780 lol and wait for the GTX990 on a 7 nm die that is 18 x faster than the 9970 I have the specs and it's called the UberTitan for $3000 lol

 

[hawtdawg]

Nope

 

[Timorous]

Overall at TPU is a useless metric. Who buys a $600 GPU and games at 1600x900?

1600P
Bonaire = 31%
780 = 94%
MSI 780 = 100%

3x Bonaire won't beat after-market 780s in demanding games/high resolution gaming.

Well to be honest the 7790 only has 1 GB of ram so at 1600p it is very likely to be ram limited in at least some scenarios bringing down that number.

 

[Shmee]

I will be pleased with a competitive product, especially if current prices drop.

 

[SiliconWars]

Overall at TPU is a useless metric. Who buys a $600 GPU and games at 1600x900?

1600P
Bonaire = 31%
780 = 94%
MSI 780 = 100%

3x Bonaire won't beat after-market 780s in demanding games/high resolution gaming.

Yes it will, easily. The reason Bonaire falters at 1600p is because at that resolution memory and bandwidth is king. I find it hard to believe you don't know this. Tripling the bandwidth of the card would help massively, as would tripling memory.

If you just take 3x Bonaire chips, you have a DP crippled flagship. Once you add/enable those double precision transistors, do you think your die size/power consumption won't be adversely affected?

3x Bonaire would crush Titan with 15% smaller die.

Bonaire peaks at 77W. Either way, you missed the point that double precision transistors are not free - not in the form of power consumption or die size. Therefore, even if you triple Bonaire's peak power you still have to account for DP hit, unless AMD will cripple its flagship.

Why not have a DP switch like Titan?

The longer 9970 is delayed, the less impressive it will be because every month we are that much closer to 20nm GPUs.

There is no delay.

 

[Jacky60]

We've been through this Raghu. It is FAR more likely that AMD will utilize multi-GPUs to compete at the top end than it is for AMD to go big die at this stage of their financial and technological (engineer firings) game. So please, unless you have undeniable proof that what you claim is truth, please don't ask me if I'm joking when I'm the one being realistic here.

Why would AMD release another multi GPU card to compete at the top end when the 7990 is already stomping all over Nvidia's ass. It's as cheap as the cheapest 780 and faster than the TITAN by a good margin.

 

[seitur]

The longer 9970 is delayed, the less impressive it will be because every month we are that much closer to 20nm GPUs.

There is no delay. Besides first series of NVIDIA Maxwell cards will be 28nm as well.

 

[Keysplayr]

Why would AMD release another multi GPU card to compete at the top end when the 7990 is already stomping all over Nvidia's ass. It's as cheap as the cheapest 780 and faster than the TITAN by a good margin.

Multi GPU drivers need more work. AMD made a step in the right direction. Still needs to work on multi-monitor. And the 7990 is old school now. Time for new GPU generation. 7xxx series is what 21 months old now? And wasn't 7990 scrapped by AMD because of TechReport and others using FCAT? Really hurt sales for multi GPUs. Price went from a grand, to what? 500 to 600 and change?

 

[raghu78]

Why would AMD release another multi GPU card to compete at the top end when the 7990 is already stomping all over Nvidia's ass. It's as cheap as the cheapest 780 and faster than the TITAN by a good margin.

because keys says so. :biggrin: don't take keys seriously. you know what he is doing. I don't need to explain it.

We've been through this Raghu. It is FAR more likely that AMD will utilize multi-GPUs to compete at the top end than it is for AMD to go big die at this stage of their financial and technological (engineer firings) game. So please, unless you have undeniable proof that what you claim is truth, please don't ask me if I'm joking when I'm the one being realistic here.

you wish it were so. honestly everybody who knows a bit about GPUs knows that Hawaii is a single big die GPU. Also AMD's past financial problems have little bearing on its future. Their future is secure. For he next 6 - 7 years their PS4 / XB1 semi-custom business is going to generate USD 1.5 billion per year and roughly USD 225 - 300 million profit per year. I wouldn't be surprised if AMD's semi custom business revenue in H2 2013 is more than Nvidia's Tegra division revenue for the full year of 2013. :biggrin:

 

[Keysplayr]

because keys says so. :biggrin: don't take keys seriously. you know what he is doing. I don't need to explain it.

Hmmm, yes you said this last time to. "We all know what you're doing keys".

I am expressing my opinion JUST as it seems you are. You have no facts to back up your claim of a huge AMD die. Nobody does. So it is a fair assumption that until that happens, and one looks at the current state of AMD, that it isn't at all likely that would happen and AMD will continue it's fight with a multi-GPU solution.

So, expressing my opinion. What was it that you "think" I'm doing and are trying to convince others that I am doing?

 

[raghu78]

Hmmm, yes you said this last time to. "We all know what you're doing keys".

I am expressing my opinion JUST as it seems you are. You have no facts to back up your claim of a huge AMD die. Nobody does. So it is a fair assumption that until that happens, and one looks at the current state of AMD, that it isn't at all likely that would happen and AMD will continue it's fight with a multi-GPU solution.

So, expressing my opinion. What was it that you "think" I'm doing and are trying to convince others that I am doing?

AMD is releasing a flagship chip from the Volcanic Islands family manufactured at 28nm, most probably TSMC. This chip is called Hawaii. architecture is GCN 2.0.

When has a AMD flagship of a new architecture or family not been a single die GPU. never

If you have the guts to take a bet I dare you to take it. if Hawaii is not single die I will close this account and stop posting at anandtech. if Hawaii is single die you will close your anandtech account and stop posting here. willing to take the challenge. :biggrin:

 

[Keysplayr]

AMD is releasing a flagship chip from the Volcanic Islands family manufactured at 28nm, most probably TSMC. This chip is called Hawaii. architecture is GCN 2.0.

When has a AMD flagship of a new architecture or family not been a single die GPU. never

If you have the guts to take a bet I dare you to take it. if Hawaii is not single die I will close this account and stop posting at anandtech. if Hawaii is single die you will close your anandtech account and stop posting here. willing to take the challenge. :biggrin:

When has AMD ever been in this position before? Slender work force, new CEO, restructuring, focusing mostly on APU/console wins. ??

This isn't about a manly game of who's got more guts, Raghu. Offering an ultimatum wager isn't the way to win a debate. It just shows you want me out of your hair as quickly as possible.

And single die is certainly possible. Just not a single HUGE die, but smaller ones and doubled up for Xfire performance.

"AMD is releasing a flagship chip from the Volcanic Islands family manufactured at 28nm, most probably TSMC. This chip is called Hawaii. architecture is GCN 2.0."

Yeah? So? This tells me AMD is releasing a new GPU. Raise the roof.

 

[LegSWAT]

Smaller, simpler and less complex SPs can get you 4096 on 28nm (think GF110 512 CCs vs GK110 2880 CCs
IO of the memory bus is notoriously difficult to shrink so 28nm large chip makes even more sense here for a 512 bit bus

Smaller, simpler and less complex SPs are on the complete other side of the spectrum of what AMD is going for with their HSA strategy.

3x Bonaire is still slower than a stock 780 and after-market 780s.

http://www.techpowerup.com/reviews/MSI/GTX_780_TF_Gaming/26.html

No, not even by your own metric (3 * 37% = 111%, relative performance, all res)

Also, DP performance will be undermined which goes against AMD's strategy. If you enable full DP, how can you have 255W as it'll use more power than Bonaire x3.

Let's guess: by doing it the nV Way it's meant to be designed and power/clock gating DP relevant transistors independently?

 

[Genx87]

What do you mean especially the 20nm? If it's not 20nm how would they get 4096sp, 512bit MB, etc. into a 28nm chip @ ~255W? The only way this can even begin to make sense is if it's on 20nm.

Or clocked at 200mhz

 

[Keysplayr]

Xbit on the state of TSMC's 20nm status. A few weeks old.

http://www.xbitlabs.com/news/other/...ng_20nm_Process_Technology_to_Early_2014.html

States volume production of 20nm won't happen until Feb 2014. 20nm Risk production has been underway since Feb. 2013.

 

[Silverforce11]

Xbit on the state of TSMC's 20nm status. A few weeks old.

http://www.xbitlabs.com/news/other/...ng_20nm_Process_Technology_to_Early_2014.html

States volume production of 20nm won't happen until Feb 2014. 20nm Risk production has been underway since Feb. 2013.

So they can do a paper launch on 20nm with the risk productions soon... interesting.

 

[Arzachel]

Multi GPU drivers need more work. AMD made a step in the right direction. Still needs to work on multi-monitor.

1. AMDs multi-GPU drivers need work.
2. Instead of investing into their driver team they should bet the farm on a new dual GPU despite point 1.

And the 7990 is old school now. Time for new GPU generation. 7xxx series is what 21 months old now?

3. AMD can't build a single GPU solution to compete with GTX780, goes ahead with an arch update that doesn't increase performance more than 15% instead of just upping the clocks and calling it a day like Nvidia did with GK104, which would be magnitudes cheaper.

And wasn't 7990 scrapped by AMD because of TechReport and others using FCAT? Really hurt sales for multi GPUs. Price went from a grand, to what? 500 to 600 and change?

4. AMDs previous multi-GPU product sold and reviewed poorly, so they should make a very slightly faster one.

Your arguments are dumb.

 

[raghu78]

When has AMD ever been in this position before? Slender work force, new CEO, restructuring, focusing mostly on APU/console wins. ??

keys contrary to what you may believe AMD's position has never been better for a long time. The hard work for consoles has been done in the past 3 - 4 years. AMD will reap the benefits for 6 - 7 years. Also AMD's semi-custom designs build from the existing IP blocks like Jaguar CPU cores and GCN GPU cores and combine to form APUs. AMD is reusing their existing IP. There is significant SOC level integration which is co-architected with Sony and Microsoft but still there is a lot of leveraging the existing IP blocks.

And single die is certainly possible. Just not a single HUGE die, but smaller ones and doubled up for Xfire performance.

You are setting yourself for massive disappointment or surprise. whichever way you want to look at it.

 

[Keysplayr]

So they can do a paper launch on 20nm with the risk productions soon... interesting.

If that's how you'd choose to interpret it. Sure it can. :hmm:

 

[MeldarthX]

Most are forgetting what has been said - *yea its been while for me posting* AMD wants to go after HPC market - that is one of the next markets they are going to go agressively after.

Do you honestly think small die will work there? No - they will go bigger this time round; big as NV? No; but its going to be most likely 450mm-480mm range....

 

[raghu78]

So they can do a paper launch on 20nm with the risk productions soon... interesting.

TSMC 20nm volume production is just not happening in Q1 2014. in fact even Q2 2014 is looking unlikely. Q3 2014 seems to be the realistic timeframe.

http://forums.anandtech.com/showpost.php?p=35405030&postcount=133

During risk production the chips are not even finalized and will be undergoing changes to get the designs working as expected and to improve yields. Only at start of volume production are the designs fully functional with economically viable yields. there is a meaning why "risk production" is called so. people need to understand.

 

[raghu78]

Most are forgetting what has been said - *yea its been while for me posting* AMD wants to go after HPC market - that is one of the next markets they are going to go agressively after.

Do you honestly think small die will work there? No - they will go bigger this time round; big as NV? No; but its going to be most likely 450mm-480mm range....

somebody who understands AMD's long term strategy. go after Nvidia's cash cows- Tesla and Quadro.

AMD will go at it from two directions. HPC clusters built from APU chips like Berlin (based on Kaveri) and from Firepro compute boards with Intel or AMD server CPUs. HSA and simplifying programmability is a key component in that strategy.

 

[Ibra]

somebody who understands AMD's long term strategy. go after Nvidia's cash cows- Tesla and Quadro.

AMD will go at it from two directions. HPC clusters built from APU chips like Berlin (based on Kaveri) and from Firepro compute boards with Intel or AMD server CPUs. HSA and simplifying programmability is a key component in that strategy.

How many APUs will be used to get 499 place from Top500? 7000000000 APUs? :sneaky:

 

[Nintendesert]

I do love the frenzy season. As a member of the 3dfx Focus Group I'd like to add there's no way AMD/ATI can possibly add enough GPUs to their mega-card to satisfy me.

 

[LegSWAT]

3. AMD can't build a single GPU solution to compete with GTX780, goes ahead with an arch update that doesn't increase performance more than 15% instead of just upping the clocks and calling it a day like Nvidia did with GK104, which would be magnitudes cheaper.

They just did. Why else release a new GPU!