S/A: AMD kills Kaveri, Steamroller, Excavator

[cbn]

No, it's not likely. But we'll have to wait and see how Jaguar does.

Isn't the claim something like 10% better frequency and 15% better IPC?

If so, that would put it on par with an Opteron 165 or 3800 x 2 from the socket 939 days.

 

[Abwx]

Isn't the claim something like 10% better frequency and 15% better IPC?

If so, that would put it on par with an Opteron 165 or 3800 x 2 from the socket 939 days.

Piledriver has already higher IPC/core than an athlon64.


For Steamroller , according to AMD , claim is 30% more operations/clock
for a whole module.

Granted it is in comparison to BD that would still put him at 20% higher
IPC than Piledriver.

http://techreport.com/review/23485/amd-cto-reveals-first-steamroller-details

 

[cbn]

Piledriver has already higher IPC/core than an athlon64.


For Steamroller , according to AMD , claim is 30% more operations/clock
for a whole module.

Granted it is in comparison to BD that would still put him at 20% higher
IPC than Piledriver.

http://techreport.com/review/23485/amd-cto-reveals-first-steamroller-details

I know the Piledrive core is stronger, but some of the Zacate Mini-ITX boards on Newegg are just so cheap.

I'll bet a dual core Jaguar Mini-ITX would be cheap too.

Speaking of these little AMD cores, what plans does AMD have for the future after Jaguar? Do they plan on designing a successor or is the plan to move completely to ARM?

 

[phenomkid7]

Speaking of these little AMD cores, what plans does AMD have for the future after Jaguar? Do they plan on designing a successor or is the plan to move completely to ARM?

Nothing is known past 2013...I think the company is not sure if they will survive 2013. Kaveri/Steamroller was supposed to launch 2013. It's dead now. Regardless of the denial by AMD, thats just PR ****.

The only thing known about 2014 is the ARM x64 SoC, and Volcanic Islands GPU architecture.

Intel will be making 14nm chips in 2014. AMD simply cannot compete. TSMC/GloFlo will be rolling out 20nm in 2014...probably with **** yields, and a lot of f ups.

It makes perfect sense for AMD to quit now and not waste money competing with Intel in x86, where they own 83% of the market....

 

[AtenRa]

Nothing is known past 2013...I think the company is not sure if they will survive 2013. Kaveri/Steamroller was supposed to launch 2013. It's dead now. Regardless of the denial by AMD, thats just PR ****.

The only thing known about 2014 is the ARM x64 SoC, and Volcanic Islands GPU architecture.

Intel will be making 14nm chips in 2014. AMD simply cannot compete. TSMC/GloFlo will be rolling out 20nm in 2014...probably with **** yields, and a lot of f ups.

It makes perfect sense for AMD to quit now and not waste money competing with Intel in x86, where they own 83% of the market....

So, just explain to me how they will keep the 1.3 Billion in revenues after Q3 2013 with only ARM SoCs for SERVER and ATI Graphics Cards when NVIDIA only now have just passed the 1 Billinion mark in Revenue having 90% of HPC cards, 60% of Discrete Graphics Cards and 3 years in the SoC market with their own SoC design??

AMD CANNOT leave the x86 market because 60-70% of its revenue comes from that as of now.

 

[phenomkid7]

So, just explain to me how they will keep the 1.3 Billion in revenues after Q3 2013

I dont know 😕

Kaveri was supposed to launch in Q3 2013. Now its not on any recent roadmap.

You have to be realistic. In 2014 Intel will be on 14nm. How can AMD hope to compete? Engineering innovations are incremental. 28nm SteamRoller won't do jacksquat...and there is absolutely NO innovation happening at AMD right now.

The only thing you can be certain of is the ARM server SoC, and the Volcanic Islands architecture. The on-die gfx in APUs is always a generation behind.

Why do they need 1.3 Bn in revenues? Welcome to the new reality. Yea i wanna believe in AMD, but its over dude. AMD is done.


Bulldozer ironically bulldozed this company to the ground.

 

[AtenRa]

I dont know 😕

Kaveri was supposed to launch in Q3 2013. Now its not on any recent roadmap.

You have to be realistic. In 2014 Intel will be on 14nm. How can AMD hope to compete? Engineering innovations are incremental. 28nm SteamRoller won't do jacksquat...and there is absolutely NO innovation happening at AMD right now.

The only thing you can be certain of is the ARM server SoC, and the Volcanic Islands architecture. The on-die gfx in APUs is always a generation behind.

Why do they need 1.3 Bn in revenues? Welcome to the new reality. Yea i wanna believe in AMD, but its over dude. AMD is done.


Bulldozer ironically bulldozed this company to the ground.

I was never in to the BS that AMD will go 28nm Bulk. If you understand a little about xtors you would know that going from 32nm HKMG SOI to 28nm HKMG Bulk will not give you any substantial benefit in die size or power consumption since GloFos 28nm is just a 32nm half node (talks speaking about 10% in size reduction only).

I have said this before but i will again, im expecting AMD to go with GloFos 20nm HKMG (SOI or not) in 2014, that was the original plan years ago when construction in FAB 8 in New York started.

The 1.3B in revenue is the target for Q3 2013 as it was explained in Q3 2012 Earnings Call.

I believe AMD, Intel and NVIDIA will adjust their new products launches to be every 18 months and not every 12 months as they currently are. This is because of the softer market and the higher manufacturing cost that comes with the smaller nodes.

 

[OBLAMA2009]

if you axe me, if you axe me, they would be smart to get out of x86. x86 isnt a hip product anymore, people have stopped buying new desktops. the future is all this mobile stuff. theyre better off getting into arm now and if that leads back to arm desktops then so be it but for sure they need to be in arm phones, tablets and laptops.

 

[phenomkid7]

AMD CANNOT leave the x86 market

To be clear I didn't say or imply they're leaving x86. They're leaving high perf. 'Big Core' should die at AMD.

http://www.techpowerup.com/175751/APUs-Make-Up-Nearly-75-of-AMD-s-Processor-Sales.html

posted by someone earlier today. Too lazy to click: 75% of all revenue is APU.
Big core doesn't earn AMD anything. Of course they should still do R&D work on the architecture side, but get out of the manufacturing business for big core/stand alone desktop x86.

No one wants a 8 core piledriver piece of garbage. I wanna meet all these millions of video encoding PC enthusiasts that have need for heavily multi-threaded compute...and will save AMD...lol.

The software ecosystem doesn't exist for octa core...and I don't give a rats behind if it will in the *future*. Piledriver will be outdated in said *future*.

Tech companies need to have long term strategies, but need to execute and deliver products that consumers need in the present.

 

[AtenRa]

To be clear I didn't say or imply they're leaving x86. They're leaving high perf. 'Big Core' should die at AMD.

http://www.techpowerup.com/175751/APUs-Make-Up-Nearly-75-of-AMD-s-Processor-Sales.html

posted by someone earlier today. Big core doesn't earn AMD anything. Of course they should still do R&D work on the architecture side, but get out of the manufacturing business for big core/stand alone desktop x86.

No one wants a 8 core piledriver piece of garbage. I wanna meet all these millions of video encoding PC enthusiasts that have need for heavily multi-threaded compute...and will save AMD...lol.

The software ecosystem doesn't exist for octa core...and I don't give a rats behind if it will in the *future*. Piledriver will be outdated in said *future*.

Tech companies need to have long term strategies, but need to execute and deliver products that consumers need in the present.

Back in 2010, Intel's Socket 1366 was only 1% of Intel's market share, i will estimate that Socket 2011 has the same percentage today.

Desktop High-End is not where the large volume is, it was like that all along. I dont see why people believe that Bulldozer/Piledriver is the problem with AMD.

The problem is they haven't took any advantage of their APU to gain market share and increase their revenue. They have a better product than Intel and they still cant take any advantage out of it. They spend 5bill 5 years ago and they havent flooded the market both in Desktop and Mobile with APUs.

That is the problem not the 10% of Bulldozer

They need to be able to make the OEMs make cheap Mobile(Tablets & Laptops) and Small Form factor Desktops with AMD APUs raging from single core low power Brazos for Tablets all the way up to Trinity A10-5800K.

This is the reason they not doing well, not Bulldozer.

 

[zlatan]

Not sure I'd ever try to relate something that they're trying to sell based on speed and efficiency with java.... 😉

I'm no said that. But if you want to understand how HSA works, than Java is the best example now. Of course there are differences. HSA is focusing on speed, but the concept behind it is really Java-like.

Do you have anything to confirm that? My specific understanding is that GCN is an absolute requirement for HSA.

I think it's a communication problem when peoples think this. HSA don't really have any hardware requirements.
You may read this: http://developer.amd.com.php53-23.ord1-1.websitetestlink.com/wordpress/media/2012/10/hsa10.pdf
This is a more consumer-friendly article, but it doesn't analyze HSA deeply: http://developer.amd.com/resources/...hat-is-heterogeneous-system-architecture-hsa/

You need a GPU with the ability to handle pointers, which VLIW4 can't do.

There are extensions for some future hardware. An HSA-aware complier will detect this, and the HSA program will run more optimal in these systems.

 

[phenomkid7]

Desktop High-End is not where the large volume is, it was like that all along. I dont see why people believe that Bulldozer/Piledriver is the problem with AMD.

Ahem, BD is not desktop high-end. You cannot compare BD to Socket 2011...anyway off topic. BD problem because R&D costs (they also had to make the chipset for BD). Manufacturing costs. Engineering resources wasted...etc.

The problem is they haven't took any advantage of their APU to gain market share and increase their revenue. They have a better product than Intel

The x86 cores in the APU are crap. So its not an all out "better" product. Intel HD 4000 (IVB) is good enough for web browsing, youtube, 1080p etc.

And the APU gfx is NOT good enough for realistic gaming. The APU either serves a need that doesn't exist, OR its a marketing problem. Either way, its AMDs fault.

And actually according to the link i posted, 75% of AMDs revenue is from APUs. So i think they have been fairly successful in terms of utilizing the APU to generate revenue.. Its just too bad that they didn't gain market share. Either the APU is **** or no one cares enough to buy one.

 

[NTMBK]

And the APU gfx is NOT good enough for realistic gaming.

Yes it is. And that's at 1080p; on typical laptop screens it'll do far better.

 

[AtenRa]

Ahem, BD is not desktop high-end. You cannot compare BD to Socket 2011...

Both are High-End like it or not.

The x86 cores in the APU are crap. So its not an all out "better" product. Intel HD 4000 (IVB) is good enough for web browsing, youtube, 1080p etc.

Why every time the APU x86 cores are Crap they Suck etc, and Intel iGPU is Good enough ???

And the APU gfx is NOT good enough for realistic gaming. The APU either serves a need that doesn't exist, OR its a marketing problem. Either way, its AMDs fault.

And now APUs graphics is NOT good and nobody needs one.


Being a little bias are we ??? 🙄


ps: What is realistic Gaming ??

 

[mrmt]

So, just explain to me how they will keep the 1.3 Billion in revenues after Q3 2013 with only ARM SoCs for SERVER and ATI Graphics Cards when NVIDIA only now have just passed the 1 Billinion mark in Revenue having 90% of HPC cards, 60% of Discrete Graphics Cards and 3 years in the SoC market with their own SoC design??

AMD CANNOT leave the x86 market because 60-70% of its revenue comes from that as of now.

As a matter of fact they are not planning to leave the x86 market completely, there will be Jaguar.

Brazos sales should be around 250-300 million per quarter, according to their plan they aim to bring it to around 500 million (Check the Q312 EC), which added to 300+ million from the ATI business leaves a gap of just 200 million for APU/ARM. That's not unfeasible at all.

Edit: I'm assuming an winding down of the Trinity/Bulldozer line up during 2013/2014. From 700 million today for half of that by the end of the next year, and zero by the end of 2014.

 

[lopri]

Well the thing is that APUs have never satisfactorily resolved the cache sharing and memory bandwidth between CPU cores and GPU cores. So it ended up being jack of all trade, a mediocre one at that. Instead of offering synergy, they end up hurting each other competing for resources. Sure it's better than nothing (compared to HD 4000), but it's not a happy combination at all.

 

[AtenRa]

Sorry but 500 from Brazos + 300 from the GPUs = 800, where are they going to find 500 more to get to 1.3B ???

 

[boxleitnerb]

Why does AMD need to be so large anyway? They can become smaller and more agile. Work on fewer really good products and when they are "stable" as a company, they can begin to grow again.

 

[AtenRa]

Why does AMD need to be so large anyway? They can become smaller and more agile. Work on fewer really good products and when they are "stable" as a company, they can begin to grow again.

I thought it was obvious that's what they doing the last year or so.

 

[lopri]

You know what I think is similar to APUs.. ? Windows 8. :biggrin:

x86's days (in consumer market) are numbered, unless MS pulls off a miracle.

 

[boxleitnerb]

I thought it was obvious that's what they doing the last year or so.

By doing what exactly? Getting yet another product line with ARM without giving up anything else?
If they ditch Bulldozer and concentrate on ARM, GPU and Jaguar, they could be better off. They can still sell Piledriver and Trinity but skip expensive improvements on the large cores, because they will never catch up with Intel anyway. Having to sell a 315mm2 CPU for less than 200 bucks and still loose - at times badly - is just not good financially. They have a better standing in other market segments.

 

[mrmt]

The problem is they haven't took any advantage of their APU to gain market share and increase their revenue. They have a better product than Intel and they still cant take any advantage out of it. They spend 5bill 5 years ago and they havent flooded the market both in Desktop and Mobile with APUs.

Did you bother reading Mercury numbers and Q312 EC numbers? Where their revenues are falling is exactly on APU and mainstream desktop chips. They even got inventory build up with Llano. In other words, they tried the strategy you are describing, flood the market with APU, but failed.

Trinity in the third quarter wasn't being able to sustain Llano levels of shipments from the first quarter. On the desktop, same thing, but even worse. One year after Bulldozer launch the most shipped product is still the Athlon X2.

So your opinion that Trinity is a better than Intel chips isn't really representative. Customers aren't buying and OEMs don't like the thing. As for your other opinion of Bulldozer not being the problem, you should review your sentence. The fact you have a chip supposedly directed to the mainstream market has less than 10% of your product mix is a problem in itself. It means that consumers/OEM like Bulldozer even less than Trinity.

 

[mrmt]

Sorry but 500 from Brazos + 300 from the GPUs = 800, where are they going to find 500 more to get to 1.3B ???

Phasing out of the Bulldozer/Trinity line up, royalties and seamicro boxes.

Edit: Edited the original post.

 

[Lonbjerg]

I thought "pulling a Pentium M" meant dropping your current inefficient architecture, going back to what you had before, and die-shrinking it,. In Intel's case, they dropped P4 and went back to PIII. In AMD's case, that would mean dropping Bulldozer and going back to...K10. They already die-shrunk it to 32nm and it seemed to work well, after all.

I think you suffer from bad memeory.

Intel's Core Architechture has both PIII and P4 stuff in it....it's prefetch is from the P4...not the PIII.
Same thing with HT.



It's like the annoying FUD about NetBurst not being competitive...sloppy memory..or ignorance

 

[beginner99]

I think what we are all coming to acknowledge is that The Peter Principle

Dirk was billed as a genius at DEC, I personally wouldn't know if that was true (didn't work with anyone who worked with Dirk when he was at DEC) but he was also billed as a genius at AMD in creating the original Athlon K7 (something I can attest too based on first-person accounts of coworkers at TI who came from AMD).

Sadly it looks like AMD promoted him to the point where he not only was no longer able to utilize his genius, but it was also a liability at that point in the decision chain.

The problem is that people get promoted to management instead of offering a technical career path too. There is often no way to get promoted (= get better salary) if you stay on the "technical side". This is the main issue.
being a good designer tells nothing about management skills, nothing. Personally I think technical and research folks are just too down-to-earth to deal with the buzzword using bull-shiters in management. 😉