SemiAccurate: Southern Islands, Kepler, and Apple’s A6 process puzzle outed

[Chiropteran]

I'm not even going to give him the page-hit....

That is okay, I refreshed the page once for you.

I don't know why anyone does.

And once again for you.

I love the site. So what if he is wrong half the time? It gives us something to talk about. How boring would computer technology be if there were never any rumors?

Posters here are wrong a lot more often, some even tell outright lies, yet you obviously read the forums. A rumor is a rumor, don't expect it to be a guaranteed fact and you can have some enjoyment out of it.

 

[jvroig]

Do we have to turn every rumor thread into a partisan meta-discussion about how reliable / unrelieable a rumor site is?

That was rhetorical. Because we don't.

You hate SA (or Fudzilla, or BSN, or whatever other rumorsite), good for you. But nobody is pointing a gun at your head forcing you to enter this thread solely to derail and crap on it. If you are not interested in rumors from a particular source (like SA) and a thread is about it, then please continue living your life and not crap on the thread. It's a rumor. People will want to talk about it, even if they know deep down it can probably turn out so far from the truth, and even if you don't like it or the source.

If we wanted to hear about your personal gripes about site/author, we'll check your facebook wall, not a tech forum. You already know from experience that such posts will only start more heated partisan arguments, and at the very least derail a thread into a meta-discussion of the site, instead of the rumor.

All posts after this should be about the RUMOR, and not solely about the RUMORSITE. Let the people interested in the related issues here have a place to discuss it without being harassed or having to read your derails.

Moderator jvroig

 

[sandorski]

People want more than a 10%ish performance improvement, but if this improves Power Requirements and Thermals then it should be a success. If another 5-6 month period exists without a similar response from Nvidia, AMD will once again be skimming the cream.

 

[Kenmitch]

Do we have to turn every rumor thread into a partisan meta-discussion about how reliable / unrelieable a rumor site is?

That was rhetorical. Because we don't.

You hate SA (or Fudzilla, or BSN, or whatever other rumorsite), good for you. But nobody is pointing a gun at your head forcing you to enter this thread solely to derail and crap on it. If you are not interested in rumors from a particular source (like SA) and a thread is about it, then please continue living your life and not crap on the thread. It's a rumor. People will want to talk about it, even if they know deep down it can probably turn out so far from the truth, and even if you don't like it or the source.

If we wanted to hear about your personal gripes about site/author, we'll check your facebook wall, not a tech forum. You already know from experience that such posts will only start more heated partisan arguments, and at the very least derail a thread into a meta-discussion of the site, instead of the rumor.

All posts after this should be about the RUMOR, and not solely about the RUMORSITE. Let the people interested in the related issues here have a place to discuss it without being harassed or having to read your derails.

Moderator jvroig

Guess you stated that rather nicely

Back on topic.

It's possible AMD could use tsmc's 28nm lp process and release the first quad-sli single card ever....Hell they seemed to get all the kinks out of cross-fire scaling! I doubt they did it just to see if they could do it.

 

[Grooveriding]

Good news if accurate! If so, AMD is once again on track like they were with the 5870, to get a timely release and new levels of performance delivered to the consumer.

I am ready to drop my coin on 3 7970s! There has been nothing exciting or groundbreaking released since the 5870 back in September 09.

 

[tviceman]

Good news if accurate! If so, AMD is once again on track like they were with the 5870, to get a timely release and new levels of performance delivered to the consumer.

I am ready to drop my coin on 3 7970s! There has been nothing exciting or groundbreaking released since the 5870 back in September 09.

Could you please explain again why you bought a gtx480 over an hd5870?

 

[Arkadrel]

short version:
-tsmc 28nm hp hkmg is probably going to be late/buggy and cause delays in next gen.
-tsmc's 28nm lp (lower power transistors) are yielding just fine for xilinx and apple.
-amd's architecture isnt dependent on high frequency/high power, so they have the option to use 28 lp on southern islands now and switch to 28 hp when process matures.
-nvidia cant use 28 lp because they need higher clocks.

= amd can potentially deliver this year at 28 nm, while nv may be late or very late.

AMD on 28nm lp = performance/watt skyrockets with these new cards?

Seems like AMD is chaseing perf/watt more and more, atleast with their GPUs.

 

[AtenRa]

From the latest TSMC Brochure

http://www.tsmc.com/download/brochures/2011_28%20Nanometer Process Technology.pdf

The 28nm high performance (HP) process is in risk production and will be the first process from TSMC to use high-k metal
gate technology. Featuring superior speed and performance, the 28HP process targets CPU, GPU, FPGA, PC, networking,
and consumer electronics applications. The 28HP process supports a 45 percent speed improvement over the 40G process
at the same leakage per gate.

TSMC says that 28nm HP will be the first with HKMG and not HLP

And from the next paragraph

TSMC also supports a high performance technology with extremely low leakage that is also in risk production. This process
adopts the HP technology gate stack while meeting more stringent low leakage requirements and a trades off speed for
performance. With extremely low leakage, 28HPL is best suited for cellular baseband, application processors, wireless
connectivity, and programmable logic. The 28HPL process reduces standby power by more than 40% compared to the 40LP
process.

So AMD will use HLP in order to achieve what ?? same performance of 69xx with lower power ??

No way in hell, they both will use 28nm HP (Edit: For their High End Designs)

 

[v8envy]

So AMD will use HLP in order to achieve what ?? same performance of 69xx with lower power ??

No way in hell, they both will use 28nm HP (Edit: For their High End Designs)

Smaller process could also mean higher yield and lower cost. If you could get a 6970 for the same price as a 460 would that still not be a win? How about that scenario and the chip costing AMD 25% less than a 460 costs NV? Think 4870 repeat.

And high end has been going multi-gpu for a while. A 6990 equivalent with the same form factor and power consumption as a 6970 would sell extremely well considering you can't buy a 6990 at all -- they're sold out!

People with current high end cards probably wouldn't upgrade, but moving existing high end into mainstream price brackets would be more than good enough for 6 months.

 

[sm625]

To make this change would mean putting a lot of work into producing a one-off part that can't be much bigger than 100mm2 (which is small for a GPU), which would be in the neighborhood of Turks in size.

Couldnt they fit about 10 SIMDs into 100mm^2 at 28nm? I would hardly call that small for a gpu. It is well beyond what anyone needs to run any game at 1920x1080.

 

[ViRGE]

Couldnt they fit about 10 SIMDs into 100mm^2 at 28nm? I would hardly call that small for a gpu. It is well beyond what anyone needs to run any game at 1920x1080.

Assuming linear scaling, they'd be able to get something between Juniper and Barts (assuming it was VLIW5). Fewer SPs if it's VLIW4 or GCN.

 

[cusideabelincoln]

Could you please explain again why you bought a gtx480 over an hd5870?

He had 5870s. He needed more VRAM. He got some 480s, which were also slightly faster. He probably doesn't get excited over slightly faster.

 

[Jionix]

From the latest TSMC Brochure

http://www.tsmc.com/download/brochures/2011_28 Nanometer Process Technology.pdf



TSMC says that 28nm HP will be the first with HKMG and not HLP

And from the next paragraph



So AMD will use HLP in order to achieve what ?? same performance of 69xx with lower power ??

No way in hell, they both will use 28nm HP (Edit: For their High End Designs)

I think the implications here are that, it's not a good fit for the high end, but AMD's mid range (current 6800 series) might work with some tweaks --- tweaks that I trust the Radeon engineers will be able to solve.

I think the timing right now is deja vu all over --- 7870/7850 first to come, just like last generation, but the performance delta will be much large then the percieved increase that was supposed to happen between 5800 to 6800.

 

[ZimZum]

I think the implications here are that, it's not a good fit for the high end, but AMD's mid range (current 6800 series) might work with some tweaks --- tweaks that I trust the Radeon engineers will be able to solve.

I think the timing right now is deja vu all over --- 7870/7850 first to come, just like last generation, but the performance delta will be much large then the percieved increase that was supposed to happen between 5800 to 6800.

Makes sense. Neither AMD nor nVidia have had much success recently when releasing flagship parts on a new process.

 

[bryanW1995]

I quite literally have that domain blocked at my router. Could not see it if I tried. But I'm guessing that Charlie is playing the old "Lets see how much sh*t I can throw against the wall" game again. He can't go wrong!!!

The key is to say "I predict X" and "I predict Y" in two different articles, then accidently lose the incorrect article once the facts emerge. Or, even better, write something so vague that it has to be right on some level.

If he'd been born 100 years earlier, Charlie would have made a great fortune teller.


This is a thread derail, and is not acceptable. To quote the poster directly below you, there has already been a "giant bolded mod warning".

Moderator jvroig

 

[gorobei]

The key is to say "I predict X" and "I predict Y" in two different articles, then accidently lose the incorrect article once the facts emerge. Or, even better, write something so vague that it has to be right on some level.

If he'd been born 100 years earlier, Charlie would have made a great fortune teller.

somewhat ironic post given the giant bolded mod warning above.



so back to topic....

Clearly amd has heard the cries and pleas of the bitcoin miners and are releasing a 500 Mhash/s card that uses 75w.

The question is how ready were they with GCN and when. If northern islands was supposed to be GCN with vliw4 simd's and all of it was derailed by the abandonment of 32nm, then they could be very close to having it refined enough to not be affected by having to use 28nm LP and the attendant tweaks/compromises.

While it would be unlikely for 7970/7950 to be LP, having 7870/7850 on LP wouldnt be out of line. Traditionally the mainstream part has been targeted for roughly the equal of last gen performance part(4770 = 3870, 5770 = 4870, 6870 = 5870). If 7870 is close to 6970 at lower power then it fits. It also keeps SI from cannibalizing the still relatively high priced cayman sales.

Or they could come out early with a 7830 and 7930 on LP with the normal naming cards on the later arriving 28 HP.

It all depends on how committed amd is to their "delivering on time/show up to the fight" business plan.

 

[pw38]

So because I'm not the sharpest knife in the drawer when it comes to the intricacies of graphics card chip sets let me try to sum up what I figure has been said here: They plan on releasing a refresh of the 6X series with a smaller die (work out the kinks, reduce heat and power draw, etc.) and then release the 7K series not long after that once yields are more satisfactory? Am I close or way off? Heck I just bought my 6950 a few months ago, I don't see myself upgrading for another year.

 

[gorobei]

So because I'm not the sharpest knife in the drawer when it comes to the intricacies of graphics card chip sets let me try to sum up what I figure has been said here: They plan on releasing a refresh of the 6X series with a smaller die (work out the kinks, reduce heat and power draw, etc.) and then release the 7K series not long after that once yields are more satisfactory? Am I close or way off? Heck I just bought my 6950 a few months ago, I don't see myself upgrading for another year.

kinda off.

the OP article has no indication of what architecture(gcn or vliw) will be used. only that amd has the option of using 28 LP and delivering this year Q4. Some people here have posited that it will be an up and down die shrink of cayman/barts. Others are hoping they just release an early pipe cleaner card like the 4770 that gets amd primed to release the main line on 28 HP.

Charlie does indicate that if amd did go 28 LP now, then they would skip 28 HP this gen and use it on HD89x0/88x0 as it would be too expensive to do the same gen in 28 LP and HP(most likely due to die mask costs i would think, if 28 needs double patterning).

 

[pw38]

Ah like I said this isn't my area of expertise, so forgive my ignorance. I do find it fascinating though and learn more every day.

 

[gorobei]

no problem, happy to clarify.

 

[AtenRa]

For low power footprint designs at less than 100W (40-60W) the 28HLP is better, but for more than that (say 200W), the 28HP process is clearly superior.

HD68xx replacement will have more than 100W TDP and i dont believe AMD will use 28HLP for it, not even for the HD67xx (~100W).

 

[Lonyo]

For low power footprint designs at less than 100W (40-60W) the 28HLP is better, but for more than that (say 200W), the 28HP process is clearly superior.

HD68xx replacement will have more than 100W TDP and i dont believe AMD will use 28HLP for it, not even for the HD67xx (~100W).

There's a hell of a gap between 60w and 200w.

 

[formulav8]

That is okay, I refreshed the page once for you.

ha ha ha

Will see hopefully before this Fall what will be coming out from AMD...