XBitlabs: Advanced Micro Devices Set to Unveil New Strategy Next Week

[Rifter]

Well not necessarily. There is a new market for very power efficient, not as powerful, server CPUs. In 5-10 years, the really powerful servers may become a much smaller part of the marketplace, required for only the top corporations. It's quite feasible that small and medium-sized businesses will choose to move towards more efficient servers at some sacrifice in performance. Even the top 3 social network companies, as well as HP are interested in these extremely efficient servers.



In desperate times, companies can commit to what at the time sounds like an all-or-nothing strategic change.

If this is true it doesnt matter anyways as AMD is getting absolutly RAPED by intel in perf/watt.

 

[Cerb]

AMD needs to figure out a way to get the folks at Khronos working harder on OpenCL.

No, they needed support in public driver packages, just as nVidia has done. Not doing that for so long, and not having public development and release timeline (IE, "it will be ready for users by about MM/YY."), lost them momentum and credibility.

AMD needed Stream and OpenCL support a long time ago. They dragged their feet, not offering support until 2011, despite trying to sell their technology by its capability to use such tools. While AMD was hyping vaporware, NVidia developed support for OpenCL, polished it, and rolled it into their drivers.

OpenCL GPU support coming so late was 100% AMD dropping the ball, IMO.

I was very disappointed today when I checked and found out the latest stable release was the same version 1.1 release from June, 2010.

OpenGL does not move quickly, and neither will OpenCL. Vendor extensions do the extra work. They are designed not only by committee, but members on the committee are competitors.

 

[Cerb]

AMD needs to figure out a way to get the folks at Khronos working harder on OpenCL.

No, they needed support in public driver packages, just as nVidia has done. Not doing that for so long, and not having public development and release timeline (IE, "it will be ready for users by about MM/YY."), lost them momentum and credibility.

AMD needed Stream and OpenCL support a long time ago. They dragged their feet, not offering support for OpenCL until 2011, despite trying to sell their technology by its capability to use such tools. While AMD was hyping vaporware, NVidia developed support for OpenCL, polished it, and rolled it into their drivers.

OpenCL GPU support coming so late was 100% AMD dropping the ball, IMO.

I was very disappointed today when I checked and found out the latest stable release was the same version 1.1 release from June, 2010.

OpenGL does not move quickly, and neither will OpenCL. Vendor extensions do the extra work. They are designed not only by committee, but members on the committee are competitors.

 

[cbn]

OpenGL does not move quickly, and neither will OpenCL. Vendor extensions do the extra work. They are designed not only by committee, but members on the committee are competitors.

How do these Vendor extensions work?

EDIT: Are these extensions similar to the extensions added to CPU instruction sets?

 

[cbn]

No, they needed support in public driver packages, just as nVidia has done. Not doing that for so long, and not having public development and release timeline (IE, "it will be ready for users by about MM/YY."), lost them momentum and credibility.

AMD needed Stream and OpenCL support a long time ago. They dragged their feet, not offering support for OpenCL until 2011, despite trying to sell their technology by its capability to use such tools. While AMD was hyping vaporware, NVidia developed support for OpenCL, polished it, and rolled it into their drivers.

OpenCL GPU support coming so late was 100% AMD dropping the ball, IMO.

I forgot about the issue of AMD and its OPEN CL drivers. I am glad to hear they finally got that fixed!

 

[Cerb]

How do these Vendor extensions work?

Well, turns out OpenGL actually has a good description that covers more than I was going to smartassedly remark about . OpenCL's website goes straight into the docs, with no sugar coated about page.

http://www.opengl.org/resources/features/OGLextensions/

 

[boxleitnerb]

Info from our beloved Charlie:

The rumor de jour is that AMD will announce an ARM license tomorrow, and move the company in the direction of cell phones/tablets/widgets and other zero margin niches. Given the competition in those markets is much greater than AMD has ever faced before, Intel included, and AMD has zero wireless experience, not to mention patents, we think this is a monumental mistake.
AMD has already set the groundwork for ARM integration, but they would be much much better off as an IP provider than a chip maker. Short of buying an established player, we can’t see this working out too well. Buying an established player, if it is financially possible, would be a really messy integration challenge, something AMD is ill prepared for now.
What will happen? Will things all work out in the end? Will all of this be a distant but painful memory when the Q1/2012 numbers are released? Who knows, but the first official clues about the new direction will come out tomorrow morning, and it will be interesting. Where it goes from there is anyone’s guess. Either way, our condolences to all those who lost their jobs, SemiAccurate hopes things work out well for you.S|A

http://semiaccurate.com/2011/11/08/looking-at-the-amd-amid-layoffs-and-new-directions/

 

[Vesku]

Whelp, I'll be getting some cruel laughter tomorrow if that is what is announced. Why not beg Qualcomm for Imageon IP back?

 

[cbn]

Well, turns out OpenGL actually has a good description that covers more than I was going to smartassedly remark about . OpenCL's website goes straight into the docs, with no sugar coated about page.

http://www.opengl.org/resources/features/OGLextensions/

Thank you for the link. I really appreciate that.

As far as AMD proposing OPEN CL extensions, maybe they would need to acquire a higher market share in order to justify the costs?

At the moment, AMD does have the majority of the discrete video cards....but Intel still controls the majority of the total x86 market with its IGP.

 

[cbn]

AMD needs to be bought by Texas Instruments.

Just for the fun of it I decided to see what kind of SOC acronym I could make out of the following pool of 15 letters:

TI
OMAP
AMD
ARM
APU

Here is what I come up with:

ARM'D UP + ATI AP "Application Processor" AMMO.

Yes, I know its super cheesy....but I just couldn't resist

 

[formulav8]

Why not beg Qualcomm for Imageon IP back?

So, So true. Can't blame Dirk at the time, because that division was majorly bleeding money. But if AMD could have held on for just a couple more years they would be very competitive right now. I think thats one of the reasons he's gone, because he was probably the one to push the sell. And all they got was 12 million or something? In light of today, that is downright nutty. The IP now is worth much much more. I do think the board majorly regrets selling off that division....

 

[sm625]

AMD needs to announce a new chip with 2 bobcat cores, two ARM cores, 80 SP, 2GB of stacked DDR3, and an 8 channel integrated SSD controller with 64 GB of stacked NAND. All for $100. Wake me up when its 2014.

 

[cbn]

Maybe we will see some Synergy between AMD and ARM?

ARM is already pretty strong with Cortex A15, why not use Bobcat's Radeon GPU with some kind of future ARMv8?

This would increase the Radeon Installation base compared to the current situation where it is isolated to a small percent of the x86 market.

 

[wlee15]

Xbitlabs and other sites seem to suggest that there will be emphasis on the Bobcat architecture.

http://www.xbitlabs.com/news/cpu/di...sses_Employees_Following_Massive_Layoffs.html

 

[cbn]

If AMD announces ARM, Will it be:

1. Licensing stock ARM cores
2. Obtaining an ARM Architectural license

 

[exar333]

If AMD announces ARM, Will it be:

1. Licensing stock ARM cores
2. Obtaining an ARM Architectural license
3. AMD ceases to exist as we know it

added the third step there....

 

[Arachnotronic]

I'll bet anyone $100 it's not going to be an ARM thing.

 

[tweakboy]

This has more to do with companies, then the consumer.. imo ^_^

 

[Tuna-Fish]

How about bobcat "sea of cores" to compete directly with the new arm server projects?

A soc with 4 quadcore bobcats would be an interesting competitor to the just announced arm servers. Basically, it should have similar integer performance, but with the advantage of x86 compatibility.

 

[cbn]

Isn't TI trying to sell their main performance ARM line? Don't see why they'd be looking to buy if that is true. Then again it is a SEMIaccurate rumor, http://semiaccurate.com/2011/07/29/texas-instruments-selling-their-omap-division/

That article mentions AMD buying OMAP!

I wonder how large the OMAP group is compared to Nvidia's SOC division? (Something tells me OMAP is not small and would be quite expensive!)

For a point of reference this article mentions Nvidia employing 500 chip designers and 1000 software engineers in order to bring Tegra to market.

"In order to build these system-on-chips, we need 500 chip designers and 1,000 software engineers to bring them to market. Two or three years each, so not exactly economic to build 'em for just the tablet, but some versions are perfect for tablets, some for high-end smartphones, somewhere between two segments.

 

[cbn]

If you are saying that AMD can't attain higher ST performance than Intel w/ x86, but might be able to beat the ARM hoards, I concur.

Hypothetically speaking, I just wonder what conditions would need to arise in order for AMD to again justify "hand optimized" CPU designs (x86 or maybe even ARM someday!) or will they always use automated tools for future CPU designs?

This article mentions a 20% larger die size and 20% performance drop when using automated tools.

The reason why performance of the long-awaited Bulldozer was below expectations is not only because it was late, but because AMD had adopted design techniques that did not allow it tweak performance, according to an ex-AMD engineer.

Cliff A. Maier, an AMD engineer who left the company several years ago, the chip designer decided to abandon practice of hand-crafting various performance-critical parts of its chips and rely completely on automatic tools. While usage of tools that automatically implement certain technologies into silicon speeds up the design process, they cannot ensure maximum performance and efficiency.

Automated Design = 20% Bigger, 20% Slower

 

[Idontcare]

That article mentions AMD buying OMAP!

I wonder how large the OMAP group is compared to Nvidia's SOC division? (Something tells me OMAP is not small and would be quite expensive!)

For a point of reference this article mentions Nvidia employing 500 chip designers and 1000 software engineers in order to bring Tegra to market.

Does AMD even have the financial means to buy out TI's OMAP?

Being "asset lite" they hardly have the assets to put up as collateral for more debt.

AMD is not what you'd consider to be ripe for M&A activity with them in the driver's seat, they are ripe for M&A in terms of someone else coming along and buying them up.

As Ellison noted of SUN, AMD's management has destroyed the company so well that its now a steal.

 

[bononos]

.... As Ellison noted of SUN, AMD's management has destroyed the company so well that its now a steal.

Do you have a link or search phrases I could use to read about it?

 

[cbn]

TI has an existing x86 license (so AMD's x86 license agreement being nullified by a buyout would be irrelevant), they have resources, and they have a massive entrenchment in the ARM space but they lack a superior contender to Nvidia's GPU tech. (i.e. they are faced with Intel's predicament, albeit in ARM space for TI)

TI might not care to get back into the x86 race though. We had a good run selling 486 chips, I worked in one of the fabs that was manufacturing them at the time and the clockspeed race was a rather exciting one from and engineering process development standpoint. Competition was good. But the economics were bad, so TI bailed on x86 when the pentium-era came of age.

Well I certainly think it would make for an interesting product and a great battle with Nvidia!

1. AMD Engineered Custom ARM CPU

2. Radeon GPU

All contained within the OMAP SOC. I like it! Quite frankly it would probably be my future dream product if it really had that spec!

I just wonder how TI would compete against Nvidia's CUDA on the HPC level If they got rid of x86?

 

[wlee15]

There are a few things that I can see AMD doing.

1) Develop a power-optimized GPU design where third parties can license, in the similar vein of PowerVR or ARM's Mali.
2) Convert the professional graphics card from a hardware-based one to a subscription based one. (i.e purchase an Radeon graphic card and gain access FirePro features and drivers by purchasing a software subscription)
3) Produce a scale-up design based on the Bobcat Architecture.