AMD: Southern Islands (7*** series) Speculation/ Rumour Thread

[swaaye]

New, agenda/purchase justification-free, entirely friendly query :

If NVIDIA can put Thermi and "Little" Kepler in a notebook, why hasn't AMD used Cayman or Tahiti? Yeah I suppose they might just not see the value in bothering but still I find it curious. Of course those 100W notebook solutions are a bit crazy too... But still it seems like AMD skipping a segment.

 

[Topman]

1. You cannot compare clock speeds between different architectures like that.
2. The 680 is GK104, the successor to GF114. GF114 was the 660Ti. The 6970 tied the 570 last generation. This generation, the 7970 tied (or close enough) what is essentially the 660 Ti of this generation.

AMD lost considerable ground this generation in terms of gaming performance. Need more proof? Look at the die sizes: 294mm² for the 680 and 352mm² for the 7970.

Nvidia's execution this generation has been dreadful, yes, but that is irrelevant to the argument you had presented earlier.

On a side note, your maturity level and knowledge about GPUs are incredibly contrasting to that of the rest of the community here...

Well...now... talk me about Cape Verde vs GK107.
Talk me about Pitcairn (and why is so close to GK104 in 'mobility' and far cheaper).
Talk me where's GK106.

And if nVidia launchs a GK104 after three months 7970, why AMD can't reacts with a 7970 OC (= GTX680) three months after?

Prove to me than US$ 349-999 cards sells more than US$ 149-349 cards.

Prove to me than you can extrapolates performance from high end to low end. And justifies why this not occurs always.

Thank you.

PS: sorry for the engRish

 

[Gipsel]

If NVIDIA can put Thermi and "Little" Kepler in a notebook, why hasn't AMD used Cayman or Tahiti? Yeah I suppose they might just not see the value in bothering but still I find it curious. Of course those 100W notebook solutions are a bit crazy too... But still it seems like AMD skipping a segment.

The HD7970M is a Pitcairn based solution with a TDP of 100W, iirc (typically it uses a bit less). So they are offering something in that range now. Underclocking a Tahiti to reach the same power consumption probably doesn't result in a higher performance and carries the burden of a significantly larger die (+70%) and therefore higher cost. And first tests appear to demonstrate that a HD7970M ties a mobile GK104 (GTX680M?) on performance at the same TDP.

 

[Silent_Buddha]

And a lot of that is due to the fact that nVidia schooled AMD's top dog with a mid range chip while laughing at the bank.
Last gen nVidia struggled to come up with a dual GPU solution and now things seem to be reversed and so on.

Not really. GK104 and Tahiti are roughly the same speed. Only Tahiti arrived almost 3 months earlier.

If you consider not actually being able to launch a chip due it to being so bad that they can't bring it to market (GK100) as schooling AMD, then, uh...what?

Or AMD's Tahiti being schooled by GPU (GK110) that is likely to appear 9-12 months AFTER it, then again, what?

Fact is that currently GK104 and Tahiti are the fastest GPUs that either company can produce. If Nvidia could make and release a faster GPU they would have done so, but they cannot. Using some chip that they cannot currently produce as a basis for claiming AMD got schooled this generation is ridiculous.

Sure, GK104 is the successor to GF114. But it is still the best GPU they can currently make, which makes it irrelevant what past chip it compares to. It is and will remain Nvidia's single best manufacturable GPU for the next 5-9 months.

In other words, there is currently parity between the two companies. Although GK104 certainly has a die size advantage but also consequently is at a feature disadvantage when compared to Tahiti. But that is purely a consequence of GK104 targetting the consumer space only (but once Nvidia couldn't manufacture a proper BigK, it had to be shoehorned into the professional space as a stop gap).

It is certainly a great chip. The only thing I dislike about it is the random nature of the automatic overclock and not being guaranteed of attaining the same speeds as "review" cards.

Likewise both Pitcairn (similar chip design philosophy to GK104) and Tahiti (similar design philosophy to a non-existant BigK) are great chips.

Regards,
SB

 

[trinibwoy]

Not really. GK104 and Tahiti are roughly the same speed. Only Tahiti arrived almost 3 months earlier.

Yeah those are the facts. The common perception is still that nVidia is ruling the roost with a midrange part. The lineage of big, power hungry, compute oriented flagships that preceded GK104 are to blame for that.

If AMD wants to change that perception they need a bigger Pitcairn that can put the 680 in its place.

 

[Silent_Buddha]

Yeah those are the facts. The common perception is still that nVidia is ruling the roost with a midrange part. The lineage of big, power hungry, compute oriented flagships that preceded GK104 are to blame for that.

If AMD wants to change that perception they need a bigger Pitcairn that can put the 680 in its place.

Perceptions are hard to change. I don't think it's going to be that easy for AMD. They can't just be on par, they'll need to be significantly better (that bigger Pitcairn example) for a significant period of time I think.

They've been able to manage the first part from time to time (5870 for about 6 months, for example) but not the second part.

And despite AMD investing more into developer relations, I'm pretty sure Nvidia are still outspending them there. That's also something that indirectly impacts that perception.

Regards,
SB

 

[DuckThor Evil]

Not really. GK104 and Tahiti are roughly the same speed. Only Tahiti arrived almost 3 months earlier.

If you consider not actually being able to launch a chip due it to being so bad that they can't bring it to market (GK100) as schooling AMD, then, uh...what?

Or AMD's Tahiti being schooled by GPU (GK110) that is likely to appear 9-12 months AFTER it, then again, what?

Fact is that currently GK104 and Tahiti are the fastest GPUs that either company can produce. If Nvidia could make and release a faster GPU they would have done so, but they cannot. Using some chip that they cannot currently produce as a basis for claiming AMD got schooled this generation is ridiculous.

I already said what I meant by schooling and it had nothing to with GK110.

The schooling part was that nVidia came in with a successor to the GF114 while having a smaller 294mm2 die size, 2GB of memory on a 256bit wide bus and a small PCB seeing great demand at $499 and being very competitive in performance vs Tahiti and it's corresponding more expensive to make metrics selling at 450$-479$ (for most regular models) and apparently not selling as good at retail.

It is certainly a great chip. The only thing I dislike about it is the random nature of the automatic overclock and not being guaranteed of attaining the same speeds as "review" cards.

The automatic overclocking is a great feature. The only problem is locked voltage control for most models.

As far as I know the review cards haven't had very high boost capabilities compared to retail cards. I've heard of few stinkers that don't really boost much past their advertised rating, but most go well past that.

 

[Homeles]

Well...now... talk me about Cape Verde vs GK107.
Talk me about Pitcairn (and why is so close to GK104 in 'mobility' and far cheaper).
Talk me where's GK106.

And if nVidia launchs a GK104 after three months 7970, why AMD can't reacts with a 7970 OC (= GTX680) three months after?

Prove to me than US$ 349-999 cards sells more than US$ 149-349 cards.

Prove to me than you can extrapolates performance from high end to low end. And justifies why this not occurs always.

Thank you.

PS: sorry for the engRish

1. Pitcairn and Cape Verde are far superior to Tahiti in terms of performance per watt and performance per mm². AMD realized after designing Tahiti that a 4-3-3 configuration was superior to a 4-4-4 configuration.
2. I don't know where GK106 is -- like I said, Nvidia's execution has been awful. It's still irrelevant to his initial argument.
3. They certainly can. Your implication that I said otherwise is a straw man.
4. ???
5. ???

 

[3dilettante]

Pitcairn and Cape Verde are far superior to Tahiti in terms of performance per watt and performance per mm². AMD realized after designing Tahiti that a 4-3-3 configuration was superior to a 4-4-4 configuration.

Are you looking at the CU count per L1 Icache?
For what it's worth Tahiti is a 4-4-4-4 (x2), and there's been no sign that this, of all the things different between Tahiti and the smaller variants, is worthy of note.

 

[Homeles]

Are you looking at the CU count per L1 Icache?
For what it's worth Tahiti is a 4-4-4-4 (x2), and there's been no sign that this, of all the things different between Tahiti and the smaller variants, is worthy of note.

Yes. Your second sentence is hard to understand, come again?

 

[3dilettante]

The arrangement of CUs relative to the shared L1 is what I think you are referring to with the 4-3-3 numbering.
At a high level, this is one the least significant differences between the designs. No evidence has been presented that this matters.

 

[swaaye]

The HD7970M is a Pitcairn based solution with a TDP of 100W, iirc (typically it uses a bit less). So they are offering something in that range now. Underclocking a Tahiti to reach the same power consumption probably doesn't result in a higher performance and carries the burden of a significantly larger die (+70%) and therefore higher cost. And first tests appear to demonstrate that a HD7970M ties a mobile GK104 (GTX680M?) on performance at the same TDP.

Ah I see. Thanks for the info. I hadn't considered that Pitcairn would make sense for AMD, whereas NV doesn't have a smaller GPU at the moment that would be competitive.

 

[ECH]

1st review I've seen where the clock rates of the GPU are provided.
review

 

[DuckThor Evil]

1st review I've seen where the clock rates of the GPU are provided.
review

Excuse me??

edit: The chart wasn't visible on my iPad...Thought you were only talking about the Sapphire 7970.

 

[lanek]

aiee, 12.4 was a bad performer in many games ( bad scaling in BF3 and skirym ) Hopefuly they have not test CFX. This driver is full of bug for games ( but one good for 3Dmark11 Ln2 bench )

 

[ECH]

Arch differences become moot when you have benchmarks with gpu clock rates at or above 1200MHz. Which is why we need to see more reviews showing the gpu clock rate in games. If you can't control the gpu clock rate then all gpus tested should be overclocked similarly.

 

[Rangers]

Yeah those are the facts. The common perception is still that nVidia is ruling the roost with a midrange part. The lineage of big, power hungry, compute oriented flagships that preceded GK104 are to blame for that.

If AMD wants to change that perception they need a bigger Pitcairn that can put the 680 in its place.

nah, easier way to change perception is to release higher clocked tahiti part that shows longer bars compared to competition in reviews. which they are doing.

Or have we already forgotten "people dont care about die size".

I dont think people care if it's 294mm or 365mm (not huge difference anyway, much less than fermi/caymen) they care about speed and cost.

By the same token you have AMD pitcairn 210mm schooling what, 500mm fermi parts 580/570/560ti 448 at that price range? where is public perception there?

Anyways the only people I see bringing up the "Nvidia schooling AMD with it's mid range part" on forums are pretty rabid partisan fanboys, so I doubt those guys are ripe fruit to be swayed to the other side regardless.

 

[Homeles]

I dont think people care if it's 294mm or 365mm (not huge difference anyway, much less than fermi/caymen) they care about speed and cost.

You argument is flawed, considering a 294mm² die is cheaper to manufacture than a 365mm² die on the same process.

 

[UniversalTruth]

You argument is flawed, considering a 294mm² die is cheaper to manufacture than a 365mm² die on the same process.

Assuming yield is one and the same and the agreed wafer cost too.

 

[trinibwoy]

Or have we already forgotten "people dont care about die size".

Consumers don't care about die size. However, is it really that hard for you to grasp the impact on people's perceptions by nVidia no longer relying on a big power hungry chip to beat AMD's best. Or are you just pretending?

Anyways the only people I see bringing up the "Nvidia schooling AMD with it's mid range part" on forums are pretty rabid partisan fanboys, so I doubt those guys are ripe fruit to be swayed to the other side regardless.

The number of rabid partisan fanboys must have gone up recently. Or maybe you're so far off the other end of the scale everybody else seems like one to you.