AMD Kaveri OC On Planet Neptune

[NaroonGTX]

I would be surprised if there is any big core from AMD after Excavator.

I wonder why people keep saying this. HSA requires potent CPU perf, it can't be universally replaced with small, weak cores like Jaguar across their product stack or anything. If they are to push HSA so hard and it takes off, I don't see why they would suddenly halt x86 big cores post-Excavator.

 

[ShintaiDK]

I wonder why people keep saying this. HSA requires potent CPU perf, it can't be universally replaced with small, weak cores like Jaguar across their product stack or anything. If they are to push HSA so hard and it takes off, I don't see why they would suddenly halt x86 big cores post-Excavator.

R&D budget to start with. Their lack of big core performance, yet big dies. High TDPs compared to performance etc. Excavator is 65W TDP. Its essentially just a shrink with semistagnant performance. And AMD is awfully quiet about what happens after Excavator. As quiet as they was for AM3+.

HSA is about as useful as OpenCL, CUDA and so on. Its simply another dream about a killer feature that dies in its track.

 

[JDG1980]

I wonder why people keep saying this. HSA requires potent CPU perf, it can't be universally replaced with small, weak cores like Jaguar across their product stack or anything. If they are to push HSA so hard and it takes off, I don't see why they would suddenly halt x86 big cores post-Excavator.

There are quite a few people on this board who, for whatever reason, seem to want AMD to fail. I don't know what their motives are, if it's just fanboyism or if they have some kind of financial stake, but I wouldn't take what they say too seriously.

 

[Ajay]

R&D budget to start with. Their lack of big core performance, yet big dies. High TDPs compared to performance etc. Excavator is 65W TDP. Its essentially just a shrink with semistagnant performance. And AMD is awfully quiet about what happens after Excavator. As quiet as they was for AM3+.

HSA is about as useful as OpenCL, CUDA and so on. Its simply another dream about a killer feature that dies in its track.

What are sales and margins on big core vs. small core x86? I can't recall, but from the last con call, it seems AMD sees more future in the desktop - which is probably true and sad at the same time.

CUDA netted Nvidia some real sales and profits in professional and HPC markets and was a good play for them. Their problem is that the 800 lbs gorilla just walked onto their playground (in HPC). It looks like Intel is putting more effort into OpenCL (again - for compute markets), so that will boost the languishing standard - partly to back NV into a corner.

All that said, if we don't start hearing about AMD's next gen after EX by the middle of next year, I would seriously start to believe that big core x86 is dead. AMD may indeed be gun shy wrt to PR after the BD fiasco, but not giving your customers a clues of what to expect in the near future isn't a good plan, IMO.

 

[JDG1980]

CUDA netted Nvidia some real sales and profits in professional and HPC markets and was a good play for them. Their problem is that the 800 lbs gorilla just walked onto their playground (in HPC). It looks like Intel is putting more effort into OpenCL (again - for compute markets), so that will boost the languishing standard - partly to back NV into a corner.

I think that AMD's design win on the new Mac Pro might have some impact here, too. Macs are always popular with the graphics design crowd, and every new Mac Pro will have two Pitcairn or Tahiti AMD GPUs. This is going to put some pressure on application vendors to make sure that OpenCL is implemented well, and that it scales well for multiple-GPU systems. That, in turn, may help to boost the performance of AMD APUs on some of these benchmarks as well.

With both Intel and AMD lined up against it, CUDA is going to have a very hard time going forward.

 

[NaroonGTX]

R&D budget to start with. Their lack of big core performance, yet big dies. High TDPs compared to performance etc. Excavator is 65W TDP. Its essentially just a shrink with semistagnant performance. And AMD is awfully quiet about what happens after Excavator. As quiet as they was for AM3+.

HSA is about as useful as OpenCL, CUDA and so on. Its simply another dream about a killer feature that dies in its track.

R&D budget? They're already only focusing on things that will actually get them some money, rather than wasting it on something like another big die successor to the PD-based Opterons.

Where did you get "semistagnant performance" from, lol? No one knows anything about Carrizo/Excavator's performance besides the obvious fact that they'll be faster and more efficient than SR/Kaveri.

AMD is already quiet right now. They don't talk about anything nowadays besides some general ideas of what their vision is and some basic things about whatever product it is they're hyping up. Kaveri is launching next month and we still don't know anything about how it will perform. We know nothing of any architectural improvements besides what they showed on SR ver1 back at Hot Chips '12, which could be partially invalid now with the move to SR ver2.

As for AM3+, they've been quiet yeah, but we already know what the deal is with that socket anyway. What is there to say?

 

[inf64]

R&D budget to start with. Their lack of big core performance, yet big dies. High TDPs compared to performance etc. Excavator is 65W TDP. Its essentially just a shrink with semistagnant performance.

You are wrong 😉.
SR to Excavator is bigger change than PD to SR. But none of this will matter if process side of things is a failure...

 

[krumme]

I wonder why people keep saying this. HSA requires potent CPU perf, it can't be universally replaced with small, weak cores like Jaguar across their product stack or anything. If they are to push HSA so hard and it takes off, I don't see why they would suddenly halt x86 big cores post-Excavator.

Why do AMD need big x86 cores for HSA?

 

[inf64]

Why do AMD need big x86 cores for HSA?

They cannot push APUs to mainstream desktop segment with arm or jaguar since they will be up against broadwell+ cores.
HSA is still not mainstream and AMD cannot rely on it to sell desktop parts. They need relatively potent x86 cores and they will need it for years to come.

 

[Ajay]

You are wrong 😉.
SR to Excavator is bigger change than PD to SR. But none of this will matter if process side of things is a failure...

Exactly!

Some think EX will be on 28 nm. If that's true, I don't have high hopes for Carrizo, since the more complex logic could lead to even lower clock speeds (maybe not though, since 28nm will have matured).

IMHO, AMD desperately needs EX on 20nm if they hope to compete in low to mid-ranged desktop. They should get lower TDP and better switching speeds than the current 28nm process.

Skylake will be out around the same time as Carrizo, so Intel will own the high-end. AMD has a chance in mid-range only because of Intel's pricing and only if Carrizo really raises the bar for AMD (and even that's not a sure thing since Intel will compete more on price for large OEMs if the need to).

 

[Ajay]

R&D budget? They're already only focusing on things that will actually get them some money, rather than wasting it on something like another big die successor to the PD-based Opterons.

AMD's R&D budget is tight. While there is little to no work on a PD successor (not sure if Warsaw changes anything); AMD has added ARM based Opterons. So they still have a heavy load for the size of their engineering staff. What really hurts, is that AMD has no budget for process node R&D - so they are stuck with whatever GF's standard high performance node is (which likely has higher leakage and lower clocks than AMD would prefer).

 

[SiliconWars]

Intel has owned the high-end for how long already? There is no way AMD is ever getting it back through silicon, they knew this years ago.

AMD is committed to HSA and Mantle and their future chips will reflect that. Expect 8 cores as standard after excavator - whether or not they'll count as true "big cores" or not I dunno - even bigger IGP's and more effort and money going towards software than actual silicon.

Moore's law is on the way out, software is what will make the difference over the next 10 years. 20nm CPU design will cost over $100 million each, that same $100 million buys 100 Mantle-enabled games. It's as close to a no-brainer as you'll ever see. x86 performance is just not on their radar any longer.

 

[videogames101]

This is LN2. BD was overclocked on LHe (-269 degrees Celsius), IIRC. Now, liquid Helium is definitely exotic!

90% sure Si CMOS freezes out before -269 C, sounds fishy

 

[NostaSeronx]

90% sure Si CMOS freezes out before -269 C, sounds fishy

They aren't freezing the chip, they are freezing the stuff around the chip.

 

[mrmt]

You are wrong 😉.
SR to Excavator is bigger change than PD to SR. But none of this will matter if process side of things is a failure...

I don't think AMD problem is the performance of the node alone. AMD is selling chips twice the die size of Intel's, and that is before 14nm. I can see why they need HDL much more than they need higher performance from Globalfoundries node.

 

[mrmt]

I wonder why people keep saying this. HSA requires potent CPU perf, it can't be universally replaced with small, weak cores like Jaguar across their product stack or anything. If they are to push HSA so hard and it takes off, I don't see why they would suddenly halt x86 big cores post-Excavator.

People are saying this because AMD big core business is a former shadow of itself:

- They seem to have forfeit big cores on the servers, and big core has been bleeding share on mobile and desktop for the last few quarters, both to Intel and their own internal competitor Kabini.

- AMD went from having entire fabs devoted to their big core lines to not being able to afford even a specific node for it.

- From being a 4 billion/year business to being a 1.3/1.5 billion business.

To me, it's quite a surprise that they are still running this business. I can't think of any other reason besides the WSA to make their worth on the big core business.

 

[jpiniero]

It would make sense to finish Excavator even if it ends up being the last of the big core products because of the sunk cost already spent to complete it.

 

[AtenRa]

- From being a 4 billion/year business to being a 1.3/1.5 billion business.

I dont want to believe you did this on purpose and missed those numbers big time. They are publicly available and easy to optain. 🙄

They were a 6-7 Billion company and are now at 4-5 Billion. Even in 2013, worst ever year will be above 4 Billion.

http://quarterlyearnings.amd.com/phoenix.zhtml?c=74093&p=quarterlyearnings

2010 = 6,49 Billion
2011 = 6,56 Billion
2012 = 5,43 Billion
2013 (up to Q3) = 3,71 Billion.

 

[ShintaiDK]

I dont want to believe you did this on purpose and missed those numbers big time. They are publicly available and easy to optain. 🙄.

He talks about the big core part. Not the entire company...

 

[krumme]

They cannot push APUs to mainstream desktop segment with arm or jaguar since they will be up against broadwell+ cores.
HSA is still not mainstream and AMD cannot rely on it to sell desktop parts. They need relatively potent x86 cores and they will need it for years to come.

I thought amd investment in java development was because of a mobile push of hsa and using mobile as a platform for hsa. Makes far more sense to me than some x86 big core ??

 

[AtenRa]

He talks about the big core part. Not the entire company...

Even in 2013 BigCore FX and APUs are more than 50-60% of that 4-5 Billions. His numbers are completely wrong.

 

[ShintaiDK]

Even in 2013 BigCore FX and APUs are more than 50-60% of that 4-5 Billions. His numbers are completely wrong.

Are they? Considering that AMD wants to sell quadcore Kabinis from 70$ and up. Its hard to see that the failed big core lines can turn a majority revenue. When they are not even the majority product in volume. Specially when the small cores already makes up the wast majority of AMDs mobile sale. The area with higher revenue than desktop for products.

But again, it kinda speaks for itself when there doesnt seem to be a big core life for AMD after Excavator. Even Streamroller is hardly getting any love compared to the small cores.

 

[mrmt]

Are they? Considering that AMD wants to sell quadcore Kabinis from 70$ and up. Its hard to see that the failed big core lines can turn a majority revenue.

Big core is not the majority of AMD CPU sales.

We know that the only thing AMD acquires from Globalfoundries is their big core chips, and that's roughly 1.1 billion per year, if they were able to fulfill the commitment. Apply a 35% gross margins and then you'll reach 1.7 billion. Problem is that AMD is below the expected target to reach the WSA, so their big core business must be somewhat smaller than this optimal case, let's say 1.3/1.4 billion, or 325/350 million per quarter. With AMD CPU sales being around 800 million per quarter, that means that the cat cores are selling more than the big core family.

 

[ShintaiDK]

No wonder the big cores are dead for AMD.

 

[NTMBK]

But again, it kinda speaks for itself when there doesnt seem to be a big core life for AMD after Excavator. Even Streamroller is hardly getting any love compared to the small cores.

And remind me how much Intel has announced about Skylake? We have only a handful of leaks, for something which will arrive much sooner than any theoretical post-XV big core.

Not saying that necessarily XV isn't the final big core, but we don't have enough information yet.