Why they are hyping more on AMD GCN and Kaveri APU

[advt.naveen]

Hi,
Why they are hyping more on AMD GCN and Kaveri APU, on every AMD post i can see some one taking about AMD kaveri apu. whether the performance increase will be on cpu or on gpu
Thnx in advance

 

[preslove]

http://forums.anandtech.com/forumdisplay.php?f=5

 

[xanis]

wut

 

[Third_Eye]

Hi,
Why they are hyping more on AMD GCN and Kaveri APU, on every AMD post i can see some one taking about AMD kaveri apu. whether the performance increase will be on cpu or on gpu
Thnx in advance

First time both HSA (Heterogeneous System Architecture) and UMA (Unified Memory Access between CPU and Graphic cores) will be available to the general x86 world. Opens a whole world of possibilities.

So this will be the first true justification of "fusion" as envisioned by AMD when they spent 5.4 billion buying ATI.

 

[frozentundra123456]

It does open up the first area where AMD can potentially have an advantage over intel, except in having a better igp, which so far hasnt increased their market share.

It is dependent of the software being written (re-written) to take advantage of it however. When, how efficiently, and in how many applications this will be useful remains to be seen.

Both the PS4 and new x-box have fusion APUs though, so I am somwhat more optimistic than originally about the capabilities being utilized.

 

[podspi]

What else are people supposed to talk about? Trinity/Richland is already a known quantity, and unfortunately hasn't done much in the market. They seem relegated to bargain-junk laptops, which is a shame.

People are hoping that Kaveri will be enough of an advancement that OEMs will take it seriously. It certainly seems like that could be the case. I was very surprised/impressed at the Samsung Kabini unit. Could just be the fruits of their new CEO making nice with OEMs too.

As for me, my feeling is that Kaveri won't change much. AMD will win on the iGP side, and lose on the CPU side. I'm hoping that Kaverni will bring what I expected from Bulldozer in the first place (Nehalem-level IPC). If they can do that and not lose too much in clocks they will be in a much better competitive situation than they currently are.

Broadwell is going to bring a lot of pain to AMD, though. I really hope AMD has post 28nm plans they just aren't publicizing yet. I just don't see how anyone is going to continue competing with such a process gap, especially something like IGPs where throwing transistors at the problem is the solution.

 

[Ajay]

As for me, my feeling is that Kaveri won't change much. AMD will win on the iGP side, and lose on the CPU side. I'm hoping that Kaverni will bring what I expected from Bulldozer in the first place (Nehalem-level IPC). If they can do that and not lose too much in clocks they will be in a much better competitive situation than they currently are.

Broadwell is going to bring a lot of pain to AMD, though. I really hope AMD has post 28nm plans they just aren't publicizing yet. I just don't see how anyone is going to continue competing with such a process gap, especially something like IGPs where throwing transistors at the problem is the solution.

The problem with post 20nm plans on the desktop is GFL which is running very late - so there is nothing to talk about. I don't know if next gen Jaguar will be made at TSMC, but that just came out so I wouldn't expect to hear much about it yet.

 

[Arachnotronic]

What else are people supposed to talk about? Trinity/Richland is already a known quantity, and unfortunately hasn't done much in the market. They seem relegated to bargain-junk laptops, which is a shame.

People are hoping that Kaveri will be enough of an advancement that OEMs will take it seriously. It certainly seems like that could be the case. I was very surprised/impressed at the Samsung Kabini unit. Could just be the fruits of their new CEO making nice with OEMs too.

As for me, my feeling is that Kaveri won't change much. AMD will win on the iGP side, and lose on the CPU side. I'm hoping that Kaverni will bring what I expected from Bulldozer in the first place (Nehalem-level IPC). If they can do that and not lose too much in clocks they will be in a much better competitive situation than they currently are.

Broadwell is going to bring a lot of pain to AMD, though. I really hope AMD has post 28nm plans they just aren't publicizing yet. I just don't see how anyone is going to continue competing with such a process gap, especially something like IGPs where throwing transistors at the problem is the solution.

20nm (planar) for AMD should be at the earliest a 2015 event, possibly even mid 2015, which won't give AMD too much time to blink as Intel transitions to 10nm in 1H 2016.

Do people understand why the hopes that AMD will just "come back" in the "big core" space are misguided?

 

[Rvenger]

Same can be said to your Intel predictions too. Haswell wasn't what everyone dreamed of.

 

[Arachnotronic]

Same can be said to your Intel predictions too. Haswell wasn't what everyone dreamed of.

Just because a few enthusiasts had unrealistic expectations of Haswell's OC'ing abilities doesn't mean that Intel didn't deliver on the Haswell promise.

Check out the battery life and power draw for those new Haswell ULT parts and get back to me on how Intel missed the mark with that part

 

[ShintaiDK]

Same can be said to your Intel predictions too. Haswell wasn't what everyone dreamed of.

Besides (early?) OC (that nobody hyped.), what did Haswell run short of?

 

[frozentundra123456]

Besides (early?) OC (that nobody hyped.), what did Haswell run short of?

Higher clockspeed maybe??? Six cores maybe??? Lower temperatures maybe??

I dont ascribe the the "its intels bulldozer" at all, but it was a minimal improvement, made more disappointing by the fact that it came after another generation that was a minimal improvement as well. Yes, it know it is designed for mobile, etc, etc. but on the desktop, I was hoping for more.

 

[mrmt]

Hi,
Why they are hyping more on AMD GCN and Kaveri APU, on every AMD post i can see some one taking about AMD kaveri apu. whether the performance increase will be on cpu or on gpu
Thnx in advance

AMD has nothing to show now, what do you expect them to do? They have to spin something, and this something is now Kaveri.

 

[Plimogz]

For my part, the reason I'm waiting on Kaveri is to finally get a look at Steamroller and the front-end improvements that it's supposed to bring with it.

My layman's hope is that something was fundamentally unbalanced with the bulldozer/piledriver design which will be addressed with steamroller. AMD's improvement with Piledriver were pretty good, and if only it turns out that there was something particularly wrong with Bulldozer's shared fetch and decode hardware which can be remedied with Steamroller, this might turn out to be a better than average step up in performance.

 

[Ajay]

Besides (early?) OC (that nobody hyped.), what did Haswell run short of?

I actually think that is a legitimate complaint from enthusiasts. Intel did point out with some hype the new overclocking features in Haswell - the implication being they'd be a bit less bound by heat than they are.

 

[mrmt]

For my part, the reason I'm waiting on Kaveri is to finally get a look at Steamroller and the front-end improvements that it's supposed to bring with it.

My layman's hope is that something was fundamentally unbalanced with the bulldozer/piledriver design which will be addressed with steamroller. AMD's improvement with Piledriver were pretty good, and if only it turns out that there was something particularly wrong with Bulldozer's shared fetch and decode hardware which can be remedied with Steamroller, this might turn out to be a better than average step up in performance.

If the problem was just the decoder, don't you think AMD would have addressed it when they canned the first Bulldozer circa 2009?

 

[Plimogz]

If the problem was just the decoder, don't you think AMD would have addressed it when they canned the first Bulldozer circa 2009?

Dunno.

Although neither do I know which "canned first Bulldozer, circa 2009" you are referring to. AFAIK there is no such. I think the big AMD chips go something like Agena/Deneb/Thuban/Zambezi/Vishera -- and I don't remember any big cancellations in the meantime. What's your comment in reference to?

But as I said, I don't at all have a good grasp on how difficult or not a problem "fixing" the decoder may be. My impression from reading Anand's articles for years is that this sort of thing is a pretty significant change, which for some reason usually precedes positive benchmark results. FWIW (which is not a lot, I realize).

And to clarify my previous statement: if unfortunate design choices were indeed made when designing their first CMT chips which they are set on rectifying by beefing up the parts in question on the next iteration, it could turn out that they themselves set out some low hanging fruit in the front end of the design which -- though they may have hamstrung current parts -- could be more easily than otherwise be picked in order to reap relatively good benefits in Steamroller.

 

[mrmt]

Dunno.

Although neither do I know which "canned first Bulldozer, circa 2009" you are referring to. AFAIK there is no such. I think the big AMD chips go something like Agena/Deneb/Thuban/Zambezi/Vishera -- and I don't remember any big cancellations in the meantime. What's your comment in reference to?

This:

http://www.xbitlabs.com/news/cpu/display/20090305165448_AMD_Delays_Bulldozer_Processors_to_2011.html

Bulldozer was supposed to hit the market in 2009 in 45nm, but it was probably so broken that it wouldn't even match Thuban at the time, so the thing was canned and they went back to the drawing board to give us what we got in 2011 (this was probably the decision that cost Dirk his job).

It strikes me deeply that you and some other people have AMD engineers in such a low contempt. If Bulldozer problem was only a matter of doubling the decoders, AMD engineers would have thought of it during the 7 years they worked on the chip, no?