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

so that shifting the goal posts, the statement was:
When I said a few watts, I meant closer to 10 than 2.


also under load a bobcat core can consumer way more then 2W more like around 5 watts per core.
That doesn't sound right. In the die shots, the CPU core take far less area than the the GPU core. Then on a low leakage process, I would expect them to consume less or equal power then the GPU, much less 10W out of an 18W budget. Unless, of course, these numbers are fudged.
dumb question: what does peak power usage have to do with power gating? its not like it disapates the enwegy.
Because your power gate needs to be able to shut down the power network provisioned for full load.
 
Because your power gate needs to be able to shut down the power network provisioned for full load.
I always thought shutting it down isn't much of a problem. Letting all the current through your power gate transistors is. ;) After all, it's a few Amps for a few Watts.
 
That doesn't sound right. In the die shots, the CPU core take far less area than the the GPU core. Then on a low leakage process, I would expect them to consume less or equal power then the GPU, much less 10W out of an 18W budget. Unless, of course, these numbers are fudged.
it hard to get an accurate measure but bobcat cores on E-350 when maxing both cores go well above 2 watts, thats for sure.
http://www.xbitlabs.com/articles/cpu/display/amd-e-350_12.html#sect0
just like llano and sandy bridge when maxing both cpu and gpu you get above TDP, but the E-350 only looks to just be over where the only two especially sandy bridge can be world apart.
 
it hard to get an accurate measure but bobcat cores on E-350 when maxing both cores go well above 2 watts, thats for sure.
http://www.xbitlabs.com/articles/cpu/display/amd-e-350_12.html#sect0
Using some crude approximation, estimating roughly 2W of the difference between idle and load power will be due to MC and memory and assuming 0W at idle for a cpu core, that would be ~3W max for a core :).
Still, a quad-CU will be quite a bit more (should probably be around 10-20W).
 
http://www.xbitlabs.com/articles/cpu/display/amd-e-350_12.html#sect0
just like llano and sandy bridge when maxing both cpu and gpu you get above TDP, but the E-350 only looks to just be over where the only two especially sandy bridge can be world apart.

Looking at that article, I'm pretty sure the E-350 isn't going above TDP.

power4.png


That's system power consumption, so to the 18W TDP E-350 you still have to add both DDR3 modules, the Hudson southbridge, cooling fan, hard drive, etc.
I wouldn't be surprised if the E-350 is actually spending a lot less than 18W in that CPU+GPU stress test.
 
I haven't been following this thread muc so I'll ask: right now is there any idea if Southern Islands will indeed be delayed until 2012 like Kepler ?
 
I haven't been following this thread muc so I'll ask: right now is there any idea if Southern Islands will indeed be delayed until 2012 like Kepler ?

Current news say no. However some suggest that only VLIW4 parts would come this year and GCN parts 2012
 
And some suggest there will be no VLIW4 discrete parts at all :devilish:.

i keep asking myself what is the return of investment of enginering two different new architecture so close
ok, the vliw4 is providing a boost and probably is more suitable for fusion, but why spend in r&d and rewrite the driver if it must stay only a year?
why not stick with with the old architecture until gcn is ready?
 
VLIW-4 was developed for 32nm process, which was cancelled. AMD decided to use (already finished) VLIW-4 design for HD6900. It doesn't make sense to use it for current mainstream and low-end, which isn't focused on HPC segment (VLIW-5 offers better perf./mm² in games).
I doubt more and more, that there will be any single VLIW-4 graphics card in the Sounthern Islands family. AMD can pair Trinity with GNC discrete GPU, so they have no reason to keep VLIW-4 alive. Why they decidet to use VLIW-4 for Trinity, is another question. Maybe VLIW-5 would appear to be too obsolete (2 gens behind GCN), but GCN wasn't mature enough at the time of Trinity's development.
 
TBH I suspect that VLIW4 was supposed to be used for both SI and NI, but cancellation of 32nm and maybe getting GCN designs finished earlier than thought moved VLIW4 to be as short lived as it was/is
 
i keep asking myself what is the return of investment of enginering two different new architecture so close
ok, the vliw4 is providing a boost and probably is more suitable for fusion, but why spend in r&d and rewrite the driver if it must stay only a year?
why not stick with with the old architecture until gcn is ready?

To an extent, GCN builds upon VLIW4. There are 64 shaders per SIMD/CU, the ALUs themselves are probably very similar, and I bet a lot of improvements made to Northern Islands (e.g. geometry) were built upon for GCN.
 
EX Deux Revolution will be available with some AMD selected hardware coming at some future date. I wonder if that will include 7000 series? Or is that with current gen hardware?
Starts at 5:19
 
That doesn't seem like a good game to demo new hardware on. The graphics are kind of shitty.
 
Back
Top