Richland & Kabini rumours

[Nemesis 1]

Great find almost reads like the BD stuff right up till the ugly truth raised its head .

 

[mrmt]

Wow man, nice find! :thumbsup:
Kabini and Temash were expected to bring some massive perf/watt gains over previous low power Brazos offerings,but Richland and 40% increase?.

A move from VLIW to GNC wouldn't yields a 40% increase in GPU compute power in the same TDP? I find this hypothesis far more plausible than some magic dust that would allow 40% increase in clock speed at the same power level.

 

[inf64]

Indeed it could be just a figure for GPU perf. increase. We just don't know what it is yet. If it's done on 28nm it can be new GPU+clock bump.

 

[SiliconWars]

I'd assume it was 40% gpu performance increase, which is pretty amazing anyway.

 

[boxleitnerb]

We don't know, because there are no 40nm GCN products to my knowledge. GCN might ensure that additional units scale better or that available computing power is used more efficiently, but the majority of the improvements seem to come from the process shrink.

 

[mrmt]

Indeed it could be just a figure for GPU perf. increase. We just don't know what it is yet. If it's done on 28nm it can be new GPU+clock bump.

I'm talking about GPGPU specifically, not GPU overall.

 

[Olikan]

We don't know, because there are no 40nm GCN products to my knowledge. GCN might ensure that additional units scale better or that available computing power is used more efficiently, but the majority of the improvements seem to come from the process shrink.

nop, under the same shaders/clocks the GCN is generaly faster than VLIW-4

 

[FlanK3r]

Up to 4.4GHz? I thought GloFo's 28nm process netted lower clocks (but much lower TDPs)

I thought that this was part of the reason SR was moved to 2014. It doesn't matter if you get a ~30% IPC boost if you lose the benefit due to lower clocks (that and $$s for back end OPs).

Dont forget, the big size of die (and about half of power consumption in load) making UNB+iGPU part. Not core size+L2. ANd this iGPU is much perform than anything else in integreted PC systems.

 

[mrmt]

Dont forget, the big size of die (and about half of power consumption in load) making UNB+iGPU part. Not core size+L2. ANd this iGPU is much perform than anything else in integreted PC systems.

http://www.anandtech.com/show/5261/amd-radeon-hd-7970-review/25

You just look have to look at those benchmarks. There is a huge difference between GNC and VLIW4 in GPU compute power. With that data it isn't hard to fathom a 40% increase from Trinity to Richland even at the same die size and with same clocks.

On top of that to expect a 40% clock increase from a half-node shrink is wishful thinking, or believe that GLF 32nm SOI is botched high time, but then you would have to ask why they went for such a process.

The 40% number also matches AMD policy of cherry-picking specific benchmarks and selling them to the public as a general performance increase.

 

[boxleitnerb]

nop, under the same shaders/clocks the GCN is generaly faster than VLIW-4

I didn't say it couldn't be. I said, the majority of the performance improvement is due to 28nm.

 

[Olikan]

I didn't say it couldn't be. I said, the majority of the performance improvement is due to 28nm.

ops!
IMO, since the last drivers the bump is more split in half

 

[SiliconWars]

On top of that to expect a 40% clock increase from a half-node shrink is wishful thinking, or believe that GLF 32nm SOI is botched high time, but then you would have to ask why they went for such a process.

I'm sure I wouldn't have to look very hard to find a post of yours telling us how bad GF's 32nm is. So is it or not?

 

[mrmt]

I'm sure I wouldn't have to look very hard to find a post of yours telling us how bad GF's 32nm is. So is it or not?

It's the same GLF that delivered 32nm SOI that will deliver 28nm bulk, so what's your point?

 

[inf64]

We haven't seen bulk node from GloFo so we cannot judge it yet.

 

[SiliconWars]

It's the same GLF that delivered 32nm SOI that will deliver 28nm bulk, so what's your point?

Yes and the same intel delivered 90nm and 45nm. The same GF delivered 65nm and 45nm. The same TSMC delivered 40nm and 28nm.

That would be my point.

 

[AtenRa]

AMD also introduced the new “Richland” APU, which is currently shipping to OEMs, which delivers a 40 percent increase in performance over the previous generation of A-Series APUs. “Richland” is expected to come bundled with new software for consumers such as gesture and facial recognition to dramatically expand and enhance consumers’ user experiences. The follow-on to “Richland” will be the 28nm “Kaveri” APU with revolutionary heterogeneous system architecture (HSA) features and is expected to begin shipping to customers in the second half of 2013.

Hes joking right ??

Richland 40% over Trinity and is currently shipping to OEMs ?? and then Kaveri will be on 28nm and expected to begin shipping in the second half of 2013 ???? they will release both of them in the same year ?? WTF

 

[Hitman928]

Hes joking right ??

Richland 40% over Trinity and is currently shipping to OEMs ?? and then Kaveri will be on 28nm and expected to begin shipping in the second half of 2013 ???? they will release both of them in the same year ?? WTF

I don't see where the info is coming from. The website is written as if AMD already had their press conference releasing all this info, but it doesn't start for another 2 hours and change. It seems like it is someone writing an article on what they think AMD will announce rather than on any valid (i.e. pr approved) information. I guess we'll find out soon.

 

[AtenRa]

Hondo based Tablet
11.6-inch Vizio tablet taps AMD APU

 

[Nemesis 1]

Well I just watched the pod cast Silvermont will be out in the stores by holidays of 2013 . From the horses mouth. So that takes care of that made up story . Windows 8 is actually looking pretty good . I like the presentation . Haswell also is no longer 10 watt . Its been moved to 7 watts . I figured 8watts based on last years demo . Intel managed 1 extra watt. So there you have in the tablet space intel will dominate the high end all others are cream puffs . Silvermont will likely also beat the others giving us to GREAT intel choices. Core size means nothing other than more dies per 300m on smaller dies since intel has its own fabs its not hurting intel so much as those who have to pay foundries for dies. Samsung being intels fab nemesis

 

[SiliconWars]

Well I just watched the pod cast Silvermont will be out in the stores by holidays of 2013 . From the horses mouth. So that takes care of that made up story . Windows 8 is actually looking pretty good . I like the presentation . Haswell also is no longer 10 watt . Its been moved to 7 watts . I figured 8watts based on last years demo . Intel managed 1 extra watt. So there you have in the tablet space intel will dominate the high end all others are cream puffs . Silvermont will likely also beat the others giving us to GREAT intel choices

That's :thumbsup: GREAT NEWS :thumbsup: now why don't you go make a thread about it so the people who give a shit can post in it?

 

[Nemesis 1]

NA you know about it . That good enough for me

 

[inf64]

Name of the thread: Richland and Kabini rumors. Atom is not Richland and Atom is not Kabini. Thank you.

 

[MightyMalus]

Well I just watched the pod cast Silvermont will be out in the stores by holidays of 2013 . From the horses mouth. So that takes care of that made up story . Windows 8 is actually looking pretty good . I like the presentation . Haswell also is no longer 10 watt . Its been moved to 7 watts . I figured 8watts based on last years demo . Intel managed 1 extra watt. So there you have in the tablet space intel will dominate the high end all others are cream puffs . Silvermont will likely also beat the others giving us to GREAT intel choices. Core size means nothing other than more dies per 300m on smaller dies since intel has its own fabs its not hurting intel so much as those who have to pay foundries for dies. Samsung being intels fab nemesis

Offtopic response: Haswell no longer being 10w means its a weaker Haswell. Core sizes means ALOT! Less room for batteries! omg...and other equipment! Sheesh. Atom actually looks better than Haswell in that market! This guy eats Intel marketing like its nothing.

On topic, why is the APU's so sensitive to RAM? More so than Intel's iGPU. And do you guys see this "trend" improving even more?

I just saw some G.skill RAM OC'ed to 3900MHz, and made me wonder.

 

[Nemesis 1]

Hes joking right ??

Richland 40% over Trinity and is currently shipping to OEMs ?? and then Kaveri will be on 28nm and expected to begin shipping in the second half of 2013 ???? they will release both of them in the same year ?? WTF

Actually go read the whole thing it say up to 40% depending on what ever.

 

[NTMBK]

On topic, why is the APU's so sensitive to RAM? More so than Intel's iGPU. And do you guys see this "trend" improving even more?

I just saw some G.skill RAM OC'ed to 3900MHz, and made me wonder.

Because the actual graphics execution units are more numerous and powerful than their Intel counterpart. If the graphics can do more work per second, then it needs faster access to memory in order to a) get more work to do and b) return its results. It's why you'll still see DDR3 memory on low-end discrete GPUs, but GDDR5 exclusively on anything faster.