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

So, Juniper was a little too big for the notebook segment it was in then? Those chips do run quite hot in notebooks, although that's usually because of a barely adequate cooling design sometimes even in big notebooks (ie G73JH at >100C).

Huh? I don't see how Juniper size has even the slightest bit of relevance on anything (and it's been used on laptops, just like Barts is, so I'm not sure what you mean anyway)
 
Huh? I don't see how Juniper size has even the slightest bit of relevance on anything (and it's been used on laptops, just like Barts is, so I'm not sure what you mean anyway)
Uhm well I was putting together Dave's comment about people being too desktop-centric when considering Cape Verde's design goals and that it's smaller than Juniper and this could indicate that they want lower power consumption in the segment Juniper is/was in. Make it cheaper to build a notebook of that class.
 
Uhm well I was putting together Dave's comment about people being too desktop-centric when considering Cape Verde's design goals and that it's smaller than Juniper and this could indicate that they want lower power consumption in the segment Juniper is/was in. Make it cheaper to build a notebook of that class.

Yep, that's how I see it too. Juniper was fine for high-end gaming laptops, but I don't recall seeing it in many mainstream SKUs, which is probably where most of the money is made.
 
Uhm well I was putting together Dave's comment about people being too desktop-centric when considering Cape Verde's design goals and that it's smaller than Juniper and this could indicate that they want lower power consumption in the segment Juniper is/was in. Make it cheaper to build a notebook of that class.
I think you also have to consider there isn't really any other chip slotting in below Cape Verde for notebooks, except old (40nm) ones (I missed that line of thought earlier too). Which was quite different before, as notebooks could use Turks if they didn't quite have the power envelope for Juniper (I'm quite sure there was actually performance overlap between these chips, since mobile Juniper also shipped with ddr3 and with quite low clocks, which is almost certain to lose against gddr5 equipped higher clocked Turks). In any case I suspect (at least for HD6xxx series) a lot more notebooks used Turks rather than Juniper. So Cape Verde with disabled units and low clocks might be quite popular for mobile I guess.
 
Of course it was built with laptops in mind now, and Kaveri in a years time. I thought that was fairly obvious - the problem is it might leave a real problem for AMD if Nvidia's lower parts can actually surpass it in desktop performance.

It's a dangerous game AMD is playing here on desktop and it could end up a big mistake, however I can see the reasoning behind it. Right now, giving up a lead like what Juniper had seems to be extremely ill-advised though. We'll see what the result is once Nvidia actually releases something I guess.
 
Well from the looks of it the Kepler competitor is going to be a chip with 384SPs (GK107). That should most likely have similar performance to a GTX550Ti per clock, and seeing how high nvidia pushed that chip already there probably won't be massive clock gains without sacrificing (just like GTX550Ti) power efficiency completely. That chip should also be very small imho.
 
But, GTX550Ti has half that many SPs (192). How could a new Kepler chip with 384 SPs not blow that away? [or did you mean 560?]

Regarding the risk of underspeccing Cape Verde - sure it could end up looking like RV410 and RV530 did against NV43 and G76 in a few months. (I had to verify those codenames with Google just to be sure I remembered right ;))
 
Is anyone able to provide some reasoning for going 4-3-3 with CUs?
I just don't see any gain for non-symetrycal units by design.
The transistor&power cost for 4-4-4 should be minimal, 2-3, maybe 5(duh) mm2?
Makes no sense.
 
Is anyone able to provide some reasoning for going 4-3-3 with CUs?
I just don't see any gain for non-symetrycal units by design.
The transistor&power cost for 4-4-4 should be minimal, 2-3, maybe 5(duh) mm2?
Makes no sense.

Do we know that there actually are 4-3-3 units? When I saw the numbers the first and only thought I had was "two units disabled for better yields".
 
Do we know that there actually are 4-3-3 units? When I saw the numbers the first and only thought I had was "two units disabled for better yields".

The small chipshot pretty much confirms there isn't anything disabled/hidden
 
It seems CV size was defined by the size, which was needed for the 4,5Gbps GDDR5 pads and the other IO-stuff.

Maybe they will fill the probably huge gap between CV and Pitcairn in 2012:
14-16CUs + 192-Bit@4,5Gbps or 128-Bit@5,5Gbps on ~160-170mm²
 
It seems CV size was defined by the size, which was needed for the 4,5Gbps GDDR5 pads and the other IO-stuff.

Maybe they will fill the probably huge gap between CV and Pitcairn in 2012:
14-16CUs + 192-Bit@4,5Gbps or 128-Bit@5,5Gbps on ~160-170mm²
My quick analysis of the die concludes that there's quite a lot of vacant space along the perimeter. The two 64-bit memory pad arrays are very well defined and separated. AMD could, on theory, fit another 64-bit GDDR5 interface in there with minimum die size increase.
The Tahiti die, on the other hand, reveals pretty exact and tight fit of the memory pads and the misc interface (dispaly I/O, host phy and fuses). I could bet that a higher resolution shot would show us a very well crafted layout.
 
But, GTX550Ti has half that many SPs (192). How could a new Kepler chip with 384 SPs not blow that away? [or did you mean 560?]
No I meant 384 SPs, just no hotclock (and if I'd have to guess no SFUs) as that is what all the rumors say. In comparison to GF116/106 probably only 128bit bus too, but that shouldn't actually hurt (unless you'd equip it with ddr3 ram).
 
VR-Zone reports a verified price of $199 for HD 7870. Either Kepler is coming in the same time-frame or its just rumor without double confirmation.:LOL:

Rest of specification are the old and probably wrong rumors from Lenzfire, which were wrong with HD 7950 and HD 7700-series.
 
Last edited by a moderator:
VR-Zone reports a verified price of $199 for HD 7870. Either Kepler is coming in the same time-frame or its just rumor without double confirmation.:LOL:

I dont know what they're smoking but its appears to be quite potent.. :LOL:

Here we have the HD 7770 1 GB at $159, and VRZ is saying HD 7850 1 GB rumoured at $139 and 2 GB at $169 :rolleyes:

I dont know if they somehow mistakenly wrote 1 instead of 2, because $239, $269 and $299 sounds somewhat likely at least
 
VR-Zone reports a verified price of $199 for HD 7870. Either Kepler is coming in the same time-frame or its just rumor without double confirmation.:LOL:

Rest of specification are the old and probably wrong rumors from Lenzfire, which were wrong with HD 7950 and HD 7700-series.

It's very unlikely, as is the $139 price they mention for the HD 7850 1GB: that's $20 cheaper than the HD 7770!
 
Back
Top