[RumorTT] AMD Radeon HD8970 a quick refresh product

[Ken g6]

I personally don't believe it. 2048 is a nice round number. Unless the chip already has one or more deactivated areas, I don't expect that it would be easy to add another one.

 

[videoclone]

I personally don't believe it. 2048 is a nice round number. Unless the chip already has one or more deactivated areas, I don't expect that it would be easy to add another one.

It does, check the anandtech review of the design.

 

[Arzachel]

It does, check the anandtech review of the design.

You got a link for that? Everything I've read has confirmed, that there are no disabled/fused off parts on the 7970.

 

[3DVagabond]

You got a link for that? Everything I've read has confirmed, that there are no disabled/fused off parts on the 7970.

I believe he meant that it's easy (relatively) to add to. Not that it has deactivated areas.

 

[videoclone]

You got a link for that? Everything I've read has confirmed, that there are no disabled/fused off parts on the 7970.

Sorry my mistake i ment it does have extra room for more SP's going by the info on the reviews...

 

[Joseph F]

I hope that if AMD does a quickie refresh, they don't call it the Radeon 8970. I would prefer that they call it a 7980, since it's not supposed to be a major departure from the 7970. Also, IMO, the GTX 580 should be called the GTX 485.

 

[3DVagabond]

I hope that if AMD does a quickie refresh, they don't call it the Radeon 8970. I would prefer that they call it a 7980, since it's not supposed to be a major departure from the 7970. Also, IMO, the GTX 580 should be called the GTX 485.

GTX 385, actually.

 

[Quantos]

I hope that if AMD does a quickie refresh, they don't call it the Radeon 8970. I would prefer that they call it a 7980, since it's not supposed to be a major departure from the 7970. Also, IMO, the GTX 580 should be called the GTX 485.

This is where AMD and Nvidia's marketing departments don't agree with you! ^_^

It's a thousand more, it has to be much better.

 

[blackened23]

Welp...either way its going to be a great year for PC gamers/hobbyists/enthusiasts. Been a while since we had interesting competition.

 

[AtenRa]

I dont expect a quick refresh before Q4 2012

 

[PrincessFrosty]

Based on the refreshes usually being about 12 months apart and the simply massive amount of over clocking headroom on the 7970 I'd say that's entirely possible.

If you're ahead of the game why flex all your power in one go, two wins are better than one, if you can pull it off and especially these days where almost no games benefit from the added graphics horsepower. I suspect AMD have a reasonable idea what to expect from Nvidia so could be holding out, certainly dual GPU parts based on this technology is really going to shine.

 

[Nintendesert]

Based on the refreshes usually being about 12 months apart and the simply massive amount of over clocking headroom on the 7970 I'd say that's entirely possible.

If you're ahead of the game why flex all your power in one go, two wins are better than one, if you can pull it off and especially these days where almost no games benefit from the added graphics horsepower. I suspect AMD have a reasonable idea what to expect from Nvidia so could be holding out, certainly dual GPU parts based on this technology is really going to shine.

Good point on the two wins with one chip. They take the crown now, charge the premium they are right now with the 7970 and when Nvidia hits the market AMD just tweaks the 7970 and releases it higher clocked and better performing with little real engineering work. So they can then take the crown and keep charging the same high premium for what's effectively the same chip they released 6-12 months earlier.

 

[MrTeal]

This is where AMD and Nvidia's marketing departments don't agree with you! ^_^

It's a thousand more, it has to be much better.

It still won't be over 9000 though. I'm holding out.

 

[Kenmitch]

Good point on the two wins with one chip. They take the crown now, charge the premium they are right now with the 7970 and when Nvidia hits the market AMD just tweaks the 7970 and releases it higher clocked and better performing with little real engineering work. So they can then take the crown and keep charging the same high premium for what's effectively the same chip they released 6-12 months earlier.

Two wins with one chip? You gotta remember that GCN is a new architecture and nobody get's it right the first time. I think this even holds true for the almighty Intel.

The second win as you guys are calling it could come with drivers alone. How much performance can still be tapped with drivers on the 7970 as it sits is an unkown. You have to remember all the benchmarks shown so far are pretty much with first release drivers vs some heavily optimized drivers for the 580. Ther performance gap between the 7970 and 580 will only get larger as drivers mature. Not sure how much more the 580 will improve with drivers but I'd say it's pretty much tapped out except for optimization for newer games.

The hypothetical chip being discussed will have a huge advantage as it's gonna be born with more mature and optimized drivers for the new architecture. So if the hypothetical increase from the 7970 to the 8970 as your guys are calling it is 20% then in the end it could relate to 40-50% more than the 580.

The hypothetical performance of a gpu/cpu is kinda unpredictable without looking at all the variables.

 

[PrincessFrosty]

But what we're trying to account for (at least in part) is the seemingly deliberate under clocking of the 7970, the card has a lot more to offer even within its current revision they could have put the stock clocks a lot higher and so far there doesn't seem to be a reasonable explanation for that.

The only other thing I can think of it power draw, but they are under 300W TDP to meet PCI-E specs so there's room there as well. Unless they were extremely selective with the samples sent to reviewers, I doubt it though.

 

[Kenmitch]

But what we're trying to account for (at least in part) is the seemingly deliberate under clocking of the 7970, the card has a lot more to offer even within its current revision they could have put the stock clocks a lot higher and so far there doesn't seem to be a reasonable explanation for that.

The only other thing I can think of it power draw, but they are under 300W TDP to meet PCI-E specs so there's room there as well. Unless they were extremely selective with the samples sent to reviewers, I doubt it though.

AMD is shaken up after the bulldozer fiasco. They could just be playing it safe as GCN and 28nm are a new thing. It's hard to predict how far you can push the clocks without doing alot of sampling of the yields. I think it's more of a " We don't want another bulldozer fiasco " than anything. I think it's safer to lauch with conservative clocks than to push GCN to the limit out of the gate. Once yields are tested better and it's found that they can bump the clocks up more AMD will. Card makers will launch cards at higher clocks it's a given. AMD just wants to play it safe it looks like to me.

 

[Vesku]

It's not just about staying within PCIe spec, they would have designed the reference cooler based on a desired TDP as well as given power and expected TDP characteristics to their AIB partners. I don't think they would necessarily know whether the retail batches of 7970 dies would come out just acceptable or cherry.

I think it's a good luck for both AMD and 7970 buyers that it seems to have ~20-30% headroom in a somewhat random sampling.

 

[lifeblood]

I tend to agree with appopin's thoughts that AMD felt it needed to release something now before the 28nm process was as ready as they wanted. AMD felt they need to ship something new for Christmas (see Anands 'Showing up for the fight ' article) so the 7970 is released and wins the performance crown while proving the GCN architecture and makes AMD money. Developers can now start writing apps for GCN. Its not as fast as maybe AMD wants, but its still the fastest.

I fully expect a slightly tweaked and higher clocked version of the 7970 to come out later as the process matures. Hopefully, since it would only be slightly tweaked, they'd call it the 7975 or something; to call it an 8XXX would be disingenuous.

My bigger question is, wheres nVidia? I have to wonder if nVidia isn't losing interest in GPU's, or at least x86 capable GPU's? As they cannot create an x86 APU, they may not be putting the money into the graphics anymore. That would explain the delay in Kepler. nVidia engineers are just as smart and capable as AMD engineers so their is no reason why AMD can release a totally new architecture (GCN) on a new process (28nm) long before nVidia launches its Fermi update. That is, unless nVidia is less motivated, or they keep having really bad luck. Kepler is a evolution of Fermi and shouldn't be that much more difficult to do.

nVidia must change with the times, and if they feel they have no future in x86 graphics then maybe they need to focus elsewhere. JHH is a shrewd businessman and will take his company where he thinks the money is. But this would be a calamity for us.

 

[LiuKangBakinPie]

What happened to the XDR2 ram

 

[Arkaign]

Serious post? I would have been pissed if I pass on the 5XXX in favor of the 6XXX series.

Hahah true. I should have gotten a real 58xx to begin with, but got a 5770. I waited until 6xxx dropped and picked up a 6950, which honestly is no better, and sometimes worse, than the 5850/5870, aside from the 2GB memory.

 

[toyota]

What happened to the XDR2 ram

you mean what happened to some random rumor?

 

[PrincessFrosty]

It's not just about staying within PCIe spec, they would have designed the reference cooler based on a desired TDP as well as given power and expected TDP characteristics to their AIB partners.

I don't see this being a problem, the 5970 had a reference GPU cooler built to handle up to 400W of thermal power, this is pullng 250W TDP (? not sure) and probably 300 or so overclocked.

My bigger question is, wheres nVidia? I have to wonder if nVidia isn't losing interest in GPU's, or at least x86 capable GPU's?

Being that AMD won the next gen console gpu contracts and Nvidias move towards marketing the PC as the lead platform, it makes me think we should be seeing the exact opposite.

 

[Vesku]

I don't see this being a problem, the 5970 had a reference GPU cooler built to handle up to 400W of thermal power, this is pullng 250W TDP (? not sure) and probably 300 or so overclocked.

The 7970 reference cooler is it's own beast, to get the high overclocks on the reference boards the fan has to run at quite a clip from what I have read.

 

[at80eighty]

if it's targeted for this year, i just may hold out. hope there's a game to keep the new class of cards on their toes

 

[Grooveriding]

if it's targeted for this year, i just may hold out. hope there's a game to keep the new class of cards on their toes

Kind of sad we have so much GPU horsepower now that we are reduced for hoping for some game, any game to please come out so we can justify our high end purchases. lol