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

Hmm, 7790 has more transistors but in the end shows either lower or on par performance with the 6870.
The 6870 has a lot more bandwidth due to the 256-bit but. It would be pretty much impossible to fit that on a chip as small as Bonnaire, regardless of transistor count (which is not particularly meaningful anyway), so this comparison is meaningless.

Have to agree it's laughably overpriced when you look at the 7850.
In the long term, Bonnaire has a far lower cost floor than Pitcairn. It looks overpriced only because it's new.

Cape Verde was already clearly beating GK107, and Bonnaire takes 128-bit performance to another level.
 
Note the 7790 also handles the 6870 in some newer titles like sleeping dogs and bf3. In a couple years I wouldn't be surprised if the 7790 looks much stronger.
Or in a couple of years they may stop optimizing for the chip. We don't know if they optimize drivers for VLIW chips anymore. I get the vibe that they don't do much other than make sure they work. Still my 6950 has visual issues with Crysis 3.
 
It's not better for their grand scheme to compete with NVIDIA CUDA of course.

My point though was comparing VLIW and GCN may not be clear cut because we don't know that they bother to make sure the cards perform as well as they could. Or even fix visual problems for that matter.
 
It's not just cuda. Engines are using more compute.
We've had compute stuff in games for awhile. Just Cause 2 has quite a bit. But if TressFX is any indication of the future, everyone would be best off waiting for new GPU architectures because the current cards can't take the heat.

Not that it's unprecedented. Crysis didn't run well until I got the 6950. :D
 
We've had compute stuff in games for awhile. Just Cause 2 has quite a bit. But if TressFX is any indication of the future, everyone would be best off waiting for new GPU architectures because the current cards can't take the heat.

Not that it's unprecedented. Crysis didn't run well until I got the 6950. :D

TressFX runs fine on a 7970 and even some lesser parts.

And I doubt it makes any sense for AMD to carry multiple architectures forward even if a VLIW part could edge out newer parts in some cases.
 
The 6870 has a lot more bandwidth due to the 256-bit but. It would be pretty much impossible to fit that on a chip as small as Bonnaire, regardless of transistor count (which is not particularly meaningful anyway), so this comparison is meaningless.

The 6870 uses memory modules with much lower speed. Does this mean that if you throw at it the memory modules from the 7790, the performance will scale accordingly? Or it will stop somewhere?
 
It's basically like, Barts was a very balanced part and Bonaire much less so. Throwing more bandwidth at Barts would give far lesser returns than throwing more at Bonaire will.

AMD found the easiest (most profitable) way to fill the gap between the 7770 and 7850. This has been their strategy for a while now, but I believe such a hefty front-end tagged on a weakish rear-end is new for them (not sure). It's not anything like as bad as APU levels of starvation, but Bonaire is surely one of their most bandwidth-limited chips of recent times.
 
Last edited by a moderator:
Does anyone see them releasing a double up of bonair as a product to fill the gap between Pitcairn and Tahiti? I personally don't think the gap is big enough, and unless they are confident of being able to clear the rest of their 7800 and 7900 stock, it doesn't make sense. On the other hand, they really could use such a card to take the light of infallibility away from GTX680.

On 20nm they should be aiming for something like this with a few more compute units. I guess it makes more sense for that time frame so that it's not cannibalising 7800/7900, it's just a shame they didn't make such chip from the very start of this generation.
 
From a design PoV, Bonaire seems about perfect. It does very well vs its green competition.

But... as good as it is, it strangely feels almost like a product without a proper place in the world. It definitely could have made use of 2GB of DDR5, so it is a little odd it isn't there. Then the 7850 is just above it in price, while the 7770 is apparently going to continue? I don't really see where they can fit another SKU from salvage chips either...

It is a great chip, but it seems like AMD is mostly just competing with themselves. Why not call it the 8790 and have salvage dies be the 8770 and 8750?
 
I agree that the positioning would have been a lot more ideal if they had released an entire 8000 series range. This would have allowed for a double-Bonair chip to replace Pitcairn and salvage parts to fill the gaps below that. The only issue I have here is what to do with Tahiti... Triple Bonair seems too big and probably too power hungry to hit high clocks. I guess Titan does well with lower clocks, so such a chip could have a place. Without a thriving professional market I can't see a lot of viability in AMD producing a 450mm^2+ chip personally.
 
You have to wonder at what the other parts were like. Did they stick with the 256-bit bus and just increase shaders only to find that the pitcairn replacement was heavily bandwidth bound?

Or was it simply a case that the extra mm2 wasn't worth the performance? The midrange is pretty competitive already and AMD may have believed that the Pitcairn replacement wasn't worth the extra area. Nvidia could surely scythe prices on GK104 also if required, which could have beaten this upgraded part.

The more I think about it the more it seems reasonable that both companies have called some kind of truce in this war of attrition.
 
From a design PoV, Bonaire seems about perfect. It does very well vs its green competition.

But... as good as it is, it strangely feels almost like a product without a proper place in the world. It definitely could have made use of 2GB of DDR5, so it is a little odd it isn't there. Then the 7850 is just above it in price, while the 7770 is apparently going to continue? I don't really see where they can fit another SKU from salvage chips either...

It is a great chip, but it seems like AMD is mostly just competing with themselves. Why not call it the 8790 and have salvage dies be the 8770 and 8750?

It probably was going to be 8770 when they still planned full HD8-lineup and not just OEM-rebrands (if they ever did?)

2GB versions are coming from at least some AIBs.

Anyway, it'll be sitting comfortably at GTX 650 Ti price beating it, so there clearly is a place for it in the markets
 
You will need to recover your source ( or maybe if it is right i should do it myelf, so i let the door open if i was wrong ), but it is not a native 192bits interface .. The 660TI use the same sku of the 670-680 ( GK104 ) and the same chip and memory controller with some part disabed.
The very fact that there are SKUs with MCs disabled is enough to conclude that they don't use monolithic 256bit wide controllers, but an integer number of 64bit controllers. 104 has 4 of them. 106 has 3.

There is nothing weird about having a non-power of 2 number of controllers. All you need is some way to distribute the memory locations evenly to make sure one controller doesn't get more than the other because that would imbalance things. (How they do this for non-uniform DDR sizes per controller is anybody's guess.)

The arguments against a monolithic wide MC are many, but the most important one is undoubtedly that it would result in ridiculously large memory atoms, with severe performance degradation as an inevitable result.

I honestly don't understand why anyone could have a beef with 192 wide (aggregate) memory busses.
 
Kaotik said:
2GB versions are coming from at least some AIBs.
I kind of figured that... but if they are going to be priced above $149, there is only $159 and $169 before you start to hit 7850s at $179. I just think it would have made more sense for the 2GB version to be the default SKU.

Kaotik said:
Anyway, it'll be sitting comfortably at GTX 650 Ti price beating it
Well, that was kinda my whole point.... AMD is competing more with itself than with Nvidia. Just seems like an awful lot of trouble for a single SKU. If they were/are going to EOL the 7770 and 7750, it makes sense, but that doesn't appear to be the case (given the 7790 name).
 
And I doubt it makes any sense for AMD to carry multiple architectures forward even if a VLIW part could edge out newer parts in some cases.
Of course. VLIW had to go but I think mainly because it was poorly suited for getting into markets besides games.

GCN in a APU will be interesting because of the higher transistor count per performance level than VLIW. In the mobile market with phones and such NV is even still using non-unified architectures because of transistor count AFAIK.
 
swaaye said:
GCN in a APU will be interesting because of the higher transistor count per performance level than VLIW
I don't think there is any doubt at this point that GCN is worth it. Sure transistor count per flop may be slightly up, but so is real world performance. When GCN first launched, the issue was less clear but as the drivers have matured I'd say it has pretty well established its dominance over their prior architectures.

With GCN, AMD has a great foundation to build upon. And if they focus on "good enough," power efficient CPUs with GCN GPUs providing a qualitatively different user experience in comparison to Intel, they can be successful.
 
Also I don't know if AMD could partition into non-power-of-two sizes easily (that goes for both ROPs and MCs). All their designs disable ROPs so the ROP number stays a power of two and I can't remember what the last card was which disabled memory controllers at all.
The HD 7870 Boost is a Tahiti disabled to 256-bit, and before that, the only one I can think of is the HD 2900 GT or so that was a R600 disabled to 256-bit.

Does anyone see them releasing a double up of bonair as a product to fill the gap between Pitcairn and Tahiti? I personally don't think the gap is big enough, and unless they are confident of being able to clear the rest of their 7800 and 7900 stock, it doesn't make sense. On the other hand, they really could use such a card to take the light of infallibility away from GTX680.
I also don't think there's much of a gap between Pitcairn and Tahiti, and with the 7870 Boost there's essentially no gap. That being said, I think Hainan, which is rumored to be a double Bonaire (except for the same 2 primitives/clock), could easily replace a number of 7000 series parts. 1792 SPs at ≥1 GHz and a 256-bit bus at 6 Gbps plus a cut-down part would essentially replace both 7870s and the 7950. The 7850 still has a space but they may push 2 GB 7790 OCs to slightly close on it from below. They could even keep both Pitcairn parts and release just the full Hainan as a "7790," which would slot in in place of the 7950. So I think Hainan makes sense only when they want to replace a number of existing parts, or if they want a dual-GPU part within 300 W.

I kind of figured that... but if they are going to be priced above $149, there is only $159 and $169 before you start to hit 7850s at $179. I just think it would have made more sense for the 2GB version to be the default SKU.
Does that include the 1 GB 7850s?
 
Yeah I don't see the gap between Pitcairn and Tahiti either.

Double Bonair is supposed to achieve what that AMD doesn't already have? That's my problem with it basically. Would they just be substituting a close fight with a small win at a bigger die? That's no good to them.
 
Back
Top