If I were the CEO of AMD... (rate)

[BenchPress]

As I just posted, Intel has access to Nvidia's patents as part of that agreement. So are they taking advantage of it? Not so far.

You seem to be under the impression that there's much left to improve on the hardware side. That's not the case. Integrated graphics is limited by bandwidth. Just have a look at how tiny the 16 GPU EU's are on Ivy Bridge. They can easily increase that number and beat AMD in theoretical GFLOPS. No NVIDIA technology required. But the only way that Haswell GT3 can have 40 EUs is by adding a big chunk of eDRAM to provide additional bandwidth for frequently used data. Again, I doubt any of NVIDIA's patents were required for that.

There's room for improvement on the driver side, and they've made good progress over the last few years, but there's no need for them to increase their investment into it. They are doing incredibly well with what they have, and that's because the average consumer only plays games casually.

x86 into everything because they have exclusive rights to it. Intel wanted to push x86 into graphics because that would help extent their virtual monopoly. Ill conceived yes, but that was a major motivating force for them.

Again on the question of why Intel doesn't go out and make a GCN clone, it is not so easy. You need hundreds of engineers with the proper experience, a strategical challenge that takes years.

It's not easy, but it would be easy enough for a company like Intel. But again, hardware is not the problem here. Using a GCN clone would not end up making a difference in their bottom line.

You're right about x86 into everything though. And they've not abandoned that idea. Larrabee got cancelled but much of that technology went straight into AVX2. Aside from being twice as wide, LRBni has a lot of overlap with AVX2. Larrabee has also evolved into Xeon Phi, and they changed the encoding format to MVEX, which is very similar to the VEX encoding used for AVX. That can't be a coincidence.

With AVX-512, the CPU cores' computing density would be equivalent to that of a GPU. And unlike a GPU, you don't need an API and layers of drivers to make use of it. It's highly generic computing power available to any developer. So Intel doesn't need hundreds of highly skilled software engineers in-house to implement drivers for all these different APIs before application developers can do something with their hardware. Application developers themselves can use AVX directly by simply flipping a compiler switch.

Unified computing also benefits from the vibrant software eco-system. Developers can create x86 libraries and frameworks that use AVX2+ and sell or share them with other developers. In contrast, anything developed to run on the GPU is much less likely to be directly interchangeable between projects.

 

[Vesku]

I agree it doesn't make much sense on the surface, but what it says is making a good GPU is not so easy. As I just posted, Intel has access to Nvidia's patents as part of that agreement. So are they taking advantage of it? Not so far.

Actually the HD series of graphics utilizes some of that Nvidia know how Intel licensed. The reason I say this is that HD X000 graphics are strong in the same games as Nvidia and weak in the ones AMD did better in. Also it's GPGPU behavior is more like Nvidias than AMDs. All indications are that Intel has been using the IP gained from their Nvidia deal.

Imo, the quickest way for AMD to go completely bust would be if they sold off their GPU IP. As much as some big iron CPU lovers would like, there is no do-over possible on the ATI acquisition and GloFo spin off.

 

[Phynaz]

If Intel really did have access to all of AMD's GPU IP do you really think they'd not use any of it?

That they'd keep plodding along with their existing iGPU tech?

It just doesn't stand to reason.

Intel wasted zero time implementing every other patent of relevance to their CPU product line - from ISA extensions to x64 - that they could gleen from AMD's cross-licensed IP.

That they would sit on their hands, having access to leading edge GPU technology but do nothing with it, while continuing to develop an unquestionably inferior GPU microarchitecture is silly if true.

If Intel was willing to dump a billion bucks into Larrabee, there is no way you can convince me they wouldn't dump a billion bucks into creating a GCN clone and eating AMD's lunch the same they have done with CPUs. I don't buy it.

Let me throw this at you...Maybe Intel does have access to AMD graphics patents and that's why they have on-board graphics to begin with. What if AMD has such basic GPU IP that you can't build a GPU without it?

 

[Idontcare]

Let me throw this at you...Maybe Intel does have access to AMD graphics patents and that's why they have on-board graphics to begin with. What if AMD has such basic GPU IP that you can't build a GPU without it?

Sure, I'll buy that. But there is a rather large performance gap between the two's integrated GPUs at this time, and Intel is spending a small fortune to close that gap with each new generation. If they had access to all the IP necessary to build their own GCN-clone then why wouldn't they?

What did AMD do with all their access to Intel's x86 IP? Did they make an ARM processor to compete with Intel or did they go and make an x86-clone to compete with Intel's x86 chips?

I have no choice but to accept what you guys are saying to be true, but it just doesn't make sense to me is all.

If AMD's tech is the best, and you have unhindered licensing rights to it, why would you not be beating down the doors to use every last bit of it that you could? Unless it isn't really the best, or perhaps you don't really have unhindered access to it.

 

[AnandThenMan]

I think it's a combination of Intel having the vast majority of the market locked up so not much incentive to make anything better, and the fact that they simply have questionable expertise in graphics processors.

On the question of what exactly Intel has access to that information is not public. But it's certainly logical that they have access to more than enough ATI/Nvidia tech to get them well on the way to making a modern GPU. There is another factor as well, AMD and Nvidia have until recently released new product at a break neck pace, Intel would have to get extremely serious about things to keep pace. With the slowing of new products maybe this will give Intel an in to close the gap.

 

[Phynaz]

If AMD's tech is the best, and you have unhindered licensing rights to it, why would you not be beating down the doors to use every last bit of it that you could? Unless it isn't really the best, or perhaps you don't really have unhindered access to it.

Or, one that you and I have both have had personal experience with - "Not invented here".

Didn't Intel get access to Nvidia IP in a settlement a couple of years ago? Intel promised to not remove PCIe support in their chips and also wrote a check if I recall correctly.

 

[mrmt]

For those wanting to give a good laugh, Rory Read tapdancing at Morgan Stanley Technology, Media & Telecom Conference:

http://cc.talkpoint.com/morg007/022513a_mc/?entity=126_6W1Y5HO

 

[Idontcare]

Or, one that you and I have both have had personal experience with - "Not invented here".

Didn't Intel get access to Nvidia IP in a settlement a couple of years ago? Intel promised to not remove PCIe support in their chips and also wrote a check if I recall correctly.

Good point!

For those wanting to give a good laugh, Rory Read tapdancing at Morgan Stanley Technology, Media & Telecom Conference:

http://cc.talkpoint.com/morg007/022513a_mc/?entity=126_6W1Y5HO

Listening now, thanks for the link

 

[Vesku]

Wouldn't mind a brief summary, highlights of the 'tapdancing'. heh

 

[Idontcare]

For those wanting to give a good laugh, Rory Read tapdancing at Morgan Stanley Technology, Media & Telecom Conference:

http://cc.talkpoint.com/morg007/022513a_mc/?entity=126_6W1Y5HO

Just finished listening to the entire audio clip and I gotta say I didn't get the impression that he was tapdancing to any extent not expected (meaning of course he can't divuldge certain specifics about roadmap timelines and customers).

I actually walked away from that audio clip with a renewed sense that this guy knows what he is doing and it is just going to take time to get there before the dividends will be clearcut and visible.

 

[mrmt]

T

Just finished listening to the entire audio clip and I gotta say I didn't get the impression that he was tapdancing to any extent not expected (meaning of course he can't divuldge certain specifics about roadmap timelines and customers).

I actually walked away from that audio clip with a renewed sense that this guy knows what he is doing and it is just going to take time to get there before the dividends will be clearcut and visible.

I didn't see a strategy for their PC business beyond kabini, and non-kabini chips are 60% of AMD shipments. He didn't provide a good explanation on why they are building arm chips that will compete with kabini chips, much less tam for microservers, and his explanation abput trinity performance on the market isn't based on reality.

Too much talk, but in truth amd will have to live with kabini/temash, and they won't make money on fast growing markets alone (meaning low sales), they need embedded to survive.

I don't think this guy knows what he is doing,he's just pressing ahead with a simple strategy (kabini) and seeing if something else sticks (arm, microservers, etc)

 

[Arg Clin]

1. Sell off every piece of AMD not directly related to CPU engineering. Fire almost everyone in marketing, accounting, and middle/upper management.

This includes the Radeon business and the server business. I would restructure AMD so that it's only about CPUs again, nothing else. The sale would hope to raise 3-4 billion dollars in cash.

IMO that would be a huge mistake. The APU is a pretty good idea for mainstram computing and with browers slowly beginning to adapt GPU accelleration I think this might the one ace AMD has on their hand.

AMD has never been able to go head-to-head with Intel in the CPU business alone. Even when AMD has the technological upper hand for years it moved surprisingly little in terms of market share. Intel just has that good a grip on the market. Takes more than a few anti trust convictions to shake that.

With the ATI purchase AMD did the one thing they could to differentiate themselves from Intel. A smart move also because it seems unlikely that authorities would accept Intel buying up nVidia, so Intel would not just swoop down and do the same if/when the demand for capable iGPU's begin to really pick up.

 

[pablo87]

For those wanting to give a good laugh, Rory Read tapdancing at Morgan Stanley Technology, Media & Telecom Conference:

http://cc.talkpoint.com/morg007/022513a_mc/?entity=126_6W1Y5HO

Thanks. Tried to listen but gave up after 10 minutes - too much hot air. Interviewer was getting annoyed at the 3 minute soap box to every straightforward question.

Did find interesting the comment on die size and die consumption...--> WSA.