AMD marketing Steamroller before Vishera launch. Thoughts?

[happysmiles]

Rory Read didn't admit defeat for AMD. What he said was that he was going to differentiate AMD's products from Intel's products. Also, the particular market he was talking about was the mobile one.http://www.businessweek.com/news/2012-05-15/amd-ceo-read-says-flexibility-trumps-speed-in-sales-race

Flexibility > Speed
= more sales for AMD.

This isn't aimed at the enthusiast market as the enthusiast market is based on the server chips not the mainstream chips for AMD.
This is aimed at the enthusiast market.

well said, although we still haven't fully benefited from this plan yet

 

[ShintaiDK]

VIA said the same thing. You just cant charge premium prices for mediocore products. We can all argue the rethorics, but after the day its still what happens.

 

[Ajay]

VIA said the same thing. You just cant charge premium prices for mediocore products. We can all argue the rethorics, but after the day its still what happens.

I'm not quite sure what you are saying here, but AMD is in a better position than VIA was, IMHO. AMD doesn't need premium prices (though that helps), but if they can turn a profit based on the strategy they have, buy targeting only a couple of areas were they can be successful, they may turn out O.K. And Read may actually be able to lead them there, he has a reasonable plan. As always, it depends on how well AMD executes on it's designs and how well it's fab partners perform, we'll see. The next two or three years should prove whether AMD has taken the right approach or not. VIA's failures are just water under the bridge at this point, IMO.

I realize the problem of the rising cost of new nodes, but that will mainly affect the number of available fab partners for AMD. If AMD can get their unit volume sales higher under their current plans, then they will still have options and there will still be Fabs that want their business.

 

[Fanatical Meat]

I don't know much about processor design but I do know about managing & budgets. I can't think of anyone who cut employees, products and spending to success, survival yes success no.
To me AMD needs to keep launch dates, stop hyping performance and become real good at something. In our current low competition environment I see more expensive Intel chips coming soon.

 

[Ajay]

I don't know much about processor design but I do know about managing & budgets. I can't think of anyone who cut employees, products and spending to success, survival yes success no.

GE? IBM? Ford? These companies seem to have been able to re-invent themselves and hence remain thriving businesses. Maybe Read's innovating thinking, such as not trying to compete tit for tat with Intel will lead to that kind of re-invention. Maybe not, but it's not out of the realm of possibility that AMD could stumble into a culture of innovation that changes them forever. That, or they crash and burn and live on only in the pages of history.

 

[ShintaiDK]

I'm not quite sure what you are saying here, but AMD is in a better position than VIA was, IMHO. AMD doesn't need premium prices (though that helps), but if they can turn a profit based on the strategy they have, buy targeting only a couple of areas were they can be successful, they may turn out O.K. And Read may actually be able to lead them there, he has a reasonable plan. As always, it depends on how well AMD executes on it's designs and how well it's fab partners perform, we'll see. The next two or three years should prove whether AMD has taken the right approach or not. VIA's failures are just water under the bridge at this point, IMO.

I realize the problem of the rising cost of new nodes, but that will mainly affect the number of available fab partners for AMD. If AMD can get their unit volume sales higher under their current plans, then they will still have options and there will still be Fabs that want their business.

As I see it, AMD is in the exact same situation as VIA. Downsizing, selling off, layoffs etc. Its the slow turtlepath towards the unavoidable. AMD keeps giving up segemnts to stay alive. This time it was the performance parts.

Processnodes for AMD is a huge problem. TSMC and GloFo falls further and further behind, not to mention the increased cost. Dont be surprised when node types expand to 3 or 4 years for those 2 companies due to the ROI. Plus the cost per transistor increases rapidly. Unlike with nVidia, AMD is against the most efficieny foundry company in the world besides all the design and volume issues. Design costs also keeps going up and its also about volume for the ROI. You might see Intel with 10nm before TSMC/GloFo hits 22/20nm. TSMC/GloFo also uses generic process nodes unlike Intel who can optimize design and processnode to go hand in hand.

 

[piesquared]

Seems like some people are having a hard time remembering that AMD is and has been developingHSA and the fusion architecture for 5 years and the foundations includes ARM, Samsung, TI, etc. Heterogeneous computing is the future whether the shills wish to drive their head up their ass or not. The industry seems to agree.

 

[AtenRa]

You might see Intel with 10nm before TSMC/GloFo hits 22/20nm.

You exaggerating, TSMC will go 20nm in 2014 and FinFet at 16nm in 2015-2016. GloFo will go 20nm 2014-2015.

TSMC/GloFo also uses generic process nodes unlike Intel who can optimize design and processnode to go hand in hand.

What that suppose to mean ??

 

[nonameo]

Seems like some people are having a hard time remembering that AMD is and has been developingHSA and the fusion architecture for 5 years and the foundations includes ARM, Samsung, TI, etc. Heterogeneous computing is the future whether the shills wish to drive their head up their ass or not. The industry seems to agree.

The problem is that even if AMD has success with HSA, it will only keep them alive and not make them grow. They'd basically have to catch intel with their pants down, which is a lot less likely to happen now than it was 10 years ago(in my unscientific opinion ).

Even if they did, think about the P4. Iintel got caught with their pants down, AMD had something great that bumped their stock prices to 40+. That was the heydays of AMD, but even then Intel still had a majority of the market. AMD needs to find something other than CPUs to get their fingers in. GPU would almost be good except that seems to be hitting a limit too, and nVidia is almost like Intel, but not nearly to that level. AMD can compete with nVidia. Oh yeah, and they already have their fingers in it

HSA may keep the boat afloat but it isn't going to be revolutionary. Again, in my unscientific opinion

It really seems like things are getting absorbed into phones. PDAs and the palm pilot got sucked into them, and now it seems like handheld game consoles are getting sucked into them too. I wonder what other things will get eaten up? Probably the wallet will be next XD People already take credit card payments with their cellphones.

Okay, I digressed. I'll quit now

 

[MLSCrow]

The problem is that even if AMD has success with HSA, it will only keep them alive and not make them grow. They'd basically have to catch intel with their pants down, which is a lot less likely to happen now than it was 10 years ago(in my unscientific opinion ).

HSA has the potential to catch intel in that position again.

HSA may keep the boat afloat but it isn't going to be revolutionary. Again, in my unscientific opinion

On the contrary, it absolutely is revolutionary, it just needs to be taken advantage of. It will change the entire CPU industry. If HSA takes off the way AMD is hoping for, it will become the future of all CPU's, which will of course then all be APU's.

 

[Pilum]

On the contrary, it absolutely is revolutionary, it just needs to be taken advantage of. It will change the entire CPU industry. If HSA takes off the way AMD is hoping for, it will become the future of all CPU's, which will of course then all be APU's.

Intels mainstream CPUs have been "APU"s for 1.5 years. IIRC GPGPU abilities did increase considerably between SNB and IVB. There's no reason to believe that Intel will not continue to improve these abilities in future generations. Future Atoms are supposed to use IVBs GPU architecture, so should be GPGPU-capable as well. GPGPU abilities will be used with industry-standard languages, so AMD will not be able to take any advantage on that front, like Nvidia managed for some time with CUDA.

Revolution? No. Continuance of status quo? Yes.

 

[MLSCrow]

Intels mainstream CPUs have been "APU"s for 1.5 years. IIRC GPGPU abilities did increase considerably between SNB and IVB. There's no reason to believe that Intel will not continue to improve these abilities in future generations. Future Atoms are supposed to use IVBs GPU architecture, so should be GPGPU-capable as well. GPGPU abilities will be used with industry-standard languages, so AMD will not be able to take any advantage on that front, like Nvidia managed for some time with CUDA.

Revolution? No. Continuance of status quo? Yes.

I never nor will I ever argue whether or not Intel has IGP's built into their CPU's, so APU's are not what I was referring to as revolutionary. What will be revolutionary is HSA inspired hardware executing HSA inspired code. One of the main goals of the HSA Alliance is to create a standard that they together will be able to execute along with AMD, first and ahead of Intel. An AMD APU, if things go the way HSA intends, will outperform an Intel APU until Intel is able to adjust to the standard (it will be open after all). So, again, if things go according to HSA's plan, then yes, it will be a revolution both in terms of how CPU's are designed, but also in terms of how code is written to take advantage of the design, which is not the status quo now and only time will tell if it will be in the future.

 

[ShintaiDK]

You exaggerating, TSMC will go 20nm in 2014 and FinFet at 16nm in 2015-2016. GloFo will go 20nm 2014-2015.




What that suppose to mean ??

I think you confuse factory roadmaps with shipping. Do i even have to show you this?

Did TSMC and GloFo ship 28nm in 2010? No? Good.

And it means the processnode and design is developed for one another. Generic foundry nodes works relatively well with all designs, but not optimal.

 

[sefsefsefsef]

And it means the processnode and design is developed for one another. Generic foundry nodes works relatively well with all designs, but not optimal.

That last part doesn't make any sense. Every chip layout is made specifically for the target process. Each process node has its own rules about the placement of silicon, vias, and metal layers, so there's no such thing as a universal design that works with more than one foundry/node.

Having said that, it's clear that process designers absolutely influence gate/layout designers, but it doesn't work the other way around. Circuit designers don't get to tell process developers what the rules of the process should be.

 

[CTho9305]

And it means the processnode and design is developed for one another. Generic foundry nodes works relatively well with all designs, but not optimal.

That last part doesn't make any sense. Every chip layout is made specifically for the target process. Each process node has its own rules about the placement of silicon, vias, and metal layers, so there's no such thing as a universal design that works with more than one foundry/node.

Having said that, it's clear that process designers absolutely influence gate/layout designers, but it doesn't work the other way around. Circuit designers don't get to tell process developers what the rules of the process should be.

Fabs can tweak processes, but if you don't own the fab, they may be unwilling to make certain changes (or they might charge extra for it). In addition, there are sometimes things that a fab will learn about its process that it won't disclose if it doesn't have to; a better chip can be designed with that knowledge than without it. Some of this stuff is only discovered after a chip tapes out and they have large sample sizes to analyze, so it's not a matter of simply designing to the process. You actually design to the model of the process, and the model isn't 100% accurate. Idontcare probably has first-hand knowledge of a fab changing things for designers.

 

[frozentundra123456]

I never nor will I ever argue whether or not Intel has IGP's built into their CPU's, so APU's are not what I was referring to as revolutionary. What will be revolutionary is HSA inspired hardware executing HSA inspired code. One of the main goals of the HSA Alliance is to create a standard that they together will be able to execute along with AMD, first and ahead of Intel. An AMD APU, if things go the way HSA intends, will outperform an Intel APU until Intel is able to adjust to the standard (it will be open after all). So, again, if things go according to HSA's plan, then yes, it will be a revolution both in terms of how CPU's are designed, but also in terms of how code is written to take advantage of the design, which is not the status quo now and only time will tell if it will be in the future.

That is a lot of ifs. Amd always seems to be chasing something in the future while trailing in the present. Maybe HSA will change this, maybe not.

 

[MLSCrow]

That is a lot of ifs. Amd always seems to be chasing something in the future while trailing in the present. Maybe HSA will change this, maybe not.

:biggrin: I won't deny that at all. It is indeed a lot of "if"s, but that's the whole point. HSA is taking a big chance on if they'll be able to pull this off and to be honest, with Samsung having recently joined the Foundation, things are looking rather good, at least in my opinion. A huge company like Samsung, who along with IBM, Electronics, Co., Ltd., and Globalfoundries, which together apparently formed the world's largest chip-making consortium as of this year, jumping on the HSA bandwagon is potentially extremely good news for HSA's future as it draws attention, which may lead to additional members, and ultimately and hopefully (at least for us consumers) a successful and popular standard.

So, while yes, it's just a possibility, an idea in the making, an "if", the signs are looking good for HSA. They are gaining support, APU's are already here (innovation award to AMD with Intel tipping it's hat by following suit), and with the vague, but promising advancements made with the up-and-coming Steamroller, all that's really left is for the code to align and honestly, it seems like an HSA standard will be the easiest route for that even with CUDA having been around for some time already.

As we all saw, CUDA started off slow and it hasn't gained that much traction since. It wasn't until June of last year that NVidia released multi-cpu x86 CUDA compilers that could run CUDA code on Intel and AMD processors, so they really don't have that much of a lead as it is and from articles I've read about responses to CUDA, it has been frequently criticized for being more difficult to deploy than Nvidia claims.

It's just my opinion, but I think (and hope for our sake as consumers) that HSA takes off in a big way and that Steamroller is a great APU. In the meantime, I'll be settling for an FX8350, if the damn thing is ever released.

Oh, and by the way, I've been looking and reading some things here and there, but could someone confirm for me or even provide some links to show just how much CPU-NB overclocking affects Bulldozer performance? I know that with Stars it was huge, especially with Thuban (CPU-NB overclocking on Thuban netted better gains than frequency overclocking), and so I'm interested if it will benefit Vishera performance much and since Vish is basically Bulldozer, that info should already be out there somewhere, but I just haven't found much about it yet. Any help would be appreciated.

My particular problem is that my motherboard is rather crappy when it comes to CPU-NB overclocking, it won't even boot at anything past 2200MHz, but it can OC frequencies just fine, but if CPU-NB OC'ing will make much of a difference on Vish, then I'm probably going to have to get a new motherboard and if that's the case, then I may just forget Vish altogether and get, and delid, a 3570K.

 

[guskline]

MLSCrow: I doubt a newer 990FX mb will close the gap even with a 8350. If you are already going to buy a new mb, go Z77 and 3570 for the fastest combo. I have a thread on my 8150 experience and I bought it because I already owned an Asus Sabertooth 990FX purchased in anticipation of the Bulldozer release. I kept the mb due to the low offers to buy it and when the 8150 dozer fell to a low enough price I jumped on it. If you have, what you describe as a "crappy" mb for OCing, I'm not sure the 8350 will help. If I've learned anything, these 8150s need POWER to close the gap with the I series chips.

 

[Idontcare]

The biggest concern I have for the viability of HSA is that AMD needs HSA to pan out, but all the other big names associated with it will do just fine if HSA nosedives or never gains traction.

Samsung doesn't need HSA, but they aren't fools and they'll keep their fingers in the pot just in case it goes somewhere. Same with everyone else I've seen throw their name into the HSA hat.

But AMD appears to be setting themselves up to live or die based on the market's adoption of HSA. Didn't work out for Rambus, not sure it is going to work out here either.

 

[piesquared]

Everyone in the HSAfoundation agree that heterogeneous computing is the future. They aren't just 'hedging their bets'.

 

[Idontcare]

Everyone in the HSAfoundation agree that heterogeneous computing is the future. They aren't just 'hedging their bets'.

That is like saying everyone in the memory market agrees that memory is the future. But that doesn't mean the NAND guys think NAND is the future, or FRAM, or PRAM, etc.

AMD needs a specific microarchitecture to be the future of HSA. Everyone else could care less what microarchitecture ends up being the future of heterogeneous computing.

Same thing for Rambus, Rambus needed JEDEC to not only decide the future of computing would involving ram but also that it involved a very specific type of ram (one that Rambus made money on).

Getting a consortia of businesses to agree that they all need a future is not the challenge, but usually there is one player involved who desperately needs everyone else to play the game in a way that preferentially benefits them more than everyone else in the consortia.

This is AMD's situation, be it for heterogeneous computing or "freedom fabric".

 

[AtenRa]

I think you confuse factory roadmaps with shipping. Do i even have to show you this?

Did TSMC and GloFo ship 28nm in 2010? No? Good.

I can play the same game,
Did Intel shipped 32nm in 2009 ?? No. Did Intel shipped 22nm in 2011 ?? No ?? Good

Let me quote what you said once again,

You might see Intel with 10nm before TSMC/GloFo hits 22/20nm.

AMD started the HD58xx production at TSMC 40nm process in 2009 (HD58xx launched September 09)
AMD started the HD79xx production at TSMC 28nm process in 2011 (HD79xx launched January 2012)
TSMC 20nm process will be ready in 2013 with volume production in 2014(1)

TSMC Confirms Plans to Start Low-Volume 20nm Manufacturing Next Year.

Global Foundries 32nm process started production in late 2010 with Llano shipping for revenue in April(H1) 2011.
Glofos 28nm is ready with products shipping for revenue in H1 2013 probably with AMDs Kaveri APU.
GloFos 20nm will be ready in 2013 with volume production(SHP) in 2014. (2)

Globalfoundries to Extend Production Capacities of Fab 8.

Intel will be at 22nm in the entire 2013 with HasWell and they will be moving to 14nm in 2014. . Intel's 10nm will be ready in 2015 with shipping products in 2016.(3)

(1), (2), (3) Once again you have over exaggerated.

 

[MLSCrow]

MLSCrow: I doubt a newer 990FX mb will close the gap even with a 8350. If you are already going to buy a new mb, go Z77 and 3570 for the fastest combo. I have a thread on my 8150 experience and I bought it because I already owned an Asus Sabertooth 990FX purchased in anticipation of the Bulldozer release. I kept the mb due to the low offers to buy it and when the 8150 dozer fell to a low enough price I jumped on it. If you have, what you describe as a "crappy" mb for OCing, I'm not sure the 8350 will help. If I've learned anything, these 8150s need POWER to close the gap with the I series chips.

Right, that's exactly what I said in that if I have to get a new MB, I may skip Vish altogether anyway, however, it all depends on whether or not CPU-NB OC'ing improves Vish performance the way it did with Stars. If Vish only truly see's performance improvements from frequency OC'ing alone, I'm fine, my board handles frequency OC'ing just fine. Only time will tell and even with Intel's little show today, I still haven't seen anything from AMD...GRRRR!

 

[ShintaiDK]

TSMC 20nm process will be ready in 2013 with volume production in 2014.

Keep believing that. TSMC and GloFo sofar never meet their roadmaps. They always gets modified and delayed.

Once again you have over exaggerated.

Thats your area of expertice, not mine. By your logic we should have had 28nm GPUs in 2010, or early 2011 as latest.

Plus as I said before. You confuse factory roadmaps with actual production output capabilities.

 

[AtenRa]

Keep believing that. TSMC and GloFo sofar never meet their roadmaps. They always gets modified and delayed.



Thats your area of expertice, not mine. By your logic we should have had 28nm GPUs in 2010, or early 2011 as latest.

Plus as I said before. You confuse factory roadmaps with actual production output capabilities.

You are the one to believe that TSMC and GloFo will still be at 28nm until 2016 when Intel will be in production at 10nm, are you or are you not over exaggerating ??