What is going to happen to AMD?

[frozentundra123456]

Hopefully if hbm is more efficient, 4gb might be enough and the market would recognize that. They certainly need more memory than that for APUs though.

 

[Ajay]

Hopefully if hbm is more efficient, 4gb might be enough and the market would recognize that. They certainly need more memory than that for APUs though.

The problem is that AMD needs a single CPU GFX board that beats the Titan X. AMD and Nvidia traded market share when they were leapfrogging each other in terms of performance. To get back marketshare, AMD needs to leapfrog Nvidia - actually, they need to hold the performance lead for a couple of generations to get back the share they've lost. It's hard to imagine that scenario when AMD, at the recent AD, indicated that it expects perf/watt advances in the next generation (2016) to come primarily from the new process node. So, how is AMD going to beat NV in perf & perf/watt when the architecture isn't pursuing higher efficiency?

It's sounding like AMD doesn't have the R&D $$s to create a new ground up architecture and they we are only likely to see incremental improvements to GCN. Nvidia isn't nearly so limited in GPU R&D budget - hence they are going to be increasingly difficult for AMD to compete with.

 

[mrmt]

It's sounding like AMD doesn't have the R&D $$s to create a new ground up architecture and they we are only likely to see incremental improvements to GCN. Nvidia isn't nearly so limited in GPU R&D budget - hence they are going to be increasingly difficult for AMD to compete with.

This is very evident from the product stack, where Nvidia has basically two uarchs, one optimized for GPGPU and the other being a cost effective graphics chip, while AMD has to live with a jack of all trade, master of nones, on top of the usual AMD power consumption/noise tax. Deficiencies in capital structure that impacts R&D take time to show up, but once it does it has highly deleterious effects. This is part of the the bill generated by the cuts mandated by Rory Read circa 2012.

 

[jpiniero]

The problem is that AMD needs a single CPU GFX board that beats the Titan X.

I don't agree with this. Why should they try to chase the extreme high end, especially when it requires these massive dies? They are getting killed by Maxwell's efficiency in mobile... that is a bigger deal.

 

[erunion]

I don't agree with this. Why should they try to chase the extreme high end, especially when it requires these massive dies? They are getting killed by Maxwell's efficiency in mobile... that is a bigger deal.

It's what they've chosen to do.
Fiji is new, everything else is rebadges.

 

[ShintaiDK]

Hopefully if hbm is more efficient, 4gb might be enough and the market would recognize that. They certainly need more memory than that for APUs though.

It will be the same as any 4GB card. The only benefit will be the GCN 1.2 improvements. But thats irrelevant to GDDR5 or HBM.

Its quite clear AMD got caught on the wrong leg. the 8 high HBM1 stacks to get 8GB via 4 stacks never appeared.

 

[monstercameron]

This is very evident from the product stack, where Nvidia has basically two uarchs, one optimized for GPGPU and the other being a cost effective graphics chip, while AMD has to live with a jack of all trade, master of nones, on top of the usual AMD power consumption/noise tax. Deficiencies in capital structure that impacts R&D take time to show up, but once it does it has highly deleterious effects. This is part of the the bill generated by the cuts mandated by Rory Read circa 2012.

Even with all those debatable points, amd still won the apple business, so they must be doing something right...ans don't say giving away their chips, that's Intels thing

 

[ShintaiDK]

Even with all those debatable points, amd still won the apple business, so they must be doing something right...ans don't say giving away their chips, that's Intels thing

Its going so right the company keeps shrinking.

 

[mrmt]

Even with all those debatable points, amd still won the apple business, so they must be doing something right...

As if the Apple business was the decisive business success factor... The truth is that AMD consumer business is bleeding and shrinking since 2011, Apple or no Apple.

ans don't say giving away their chips, that's Intels thing

To be able to give away chips for free one must make a lot of money elsewhere in order to subsidize the money-losing business. As AMD cannot do that, all they can do is to downsize.

 

[2is]

Even with all those debatable points, amd still won the apple business, so they must be doing something right...ans don't say giving away their chips, that's Intels thing

What Apple business would you be referring to? The small subset of products that have a dGPU? What about the much larger subset that doesn't?

 

[2is]

Hopefully if hbm is more efficient, 4gb might be enough and the market would recognize that. They certainly need more memory than that for APUs though.

What do you mean by efficient? It may use less power than GDDR5 but if you need 5GB of vram you need 5GB of vram whether it's HBM, GDDR5, GDDR3, etc etc.

 

[monstercameron]

What Apple business would you be referring to? The small subset of products that have a dGPU? What about the much larger subset that doesn't?

dont know the ratios but the mac pro, imac and mac book pro all -will- have amd gpus.

 

[2is]

dont know the ratios but the mac pro, imac and mac book pro all -will- have amd gpus.

That's inaccurate, only the higher end ones do and only the 15" at that. All 13 and most 15" ones are all intel.

http://store.apple.com/us/buy-mac/macbook-pro

1 out of 6 configuration have AMD

And only the iMac Retina 5K has AMD, the others are either intel or nVidia

http://store.apple.com/us/buy-mac/imac
http://store.apple.com/us/buy-mac/imac-retina

 

[Ajay]

This is very evident from the product stack, where Nvidia has basically two uarchs, one optimized for GPGPU and the other being a cost effective graphics chip, while AMD has to live with a jack of all trade, master of none, on top of the usual AMD power consumption/noise tax. Deficiencies in capital structure that impacts R&D take time to show up, but once it does it has highly deleterious effects. This is part of the the bill generated by the cuts mandated by Rory Read circa 2012.

Well said. You hit the nail on the head. I expected a contraction in redesigned product offerings, but this is worse than I thought. Given AMD's already weakened financial situation, the 20nm debacle probably hurt them worse than their competitors. They simply had no elasticity in their budget.

Some additional thoughts...

Fiji looks to be an enhanced Hawaii design and the next gen after that looks to be an enhanced Fiji design. So no major changes to the GPU architecture for three years is bad news considering the leap in efficiency found in Nvidia's Maxwell. Now it become apparent why AMD is going with HBM to squeeze out some extra performance in Fiji and why they are counting on 14FF to improve their perf/watt ratio. AMD's GPUs are now stuck in a cycle similar to that of the construction core lineup.

Further more...

It appears as though AMD focused a large portion of their R&D budget on Zen - basically trying to repeat the success of the K8. If executed perfectly, they will have a product that competes with Intel's i7 line up. This in turn means that they will have a viable core to serve as a foundation for a drive back into the server market - but there are major obstacles especially for the latter.

The server market today requires a solid foundation of in-house support for chipsets and other motherboard components for network fabrics, storage support, etc. The server market also requires very strong partnerships with OEMs to bring server products to market. I'm concerned that AMD doesn't have the depth anymore to offer a rich component infrastructure on which OEMs can build fully featured world class server systems. I'm also feel like AMD will be hobbled financially again when it comes to customer acquisition costs in this high competitive market.

Now, it sounds like AMD's server products won't start hitting the market till a year after Zen debuts. This does give AMD more time to partner with component suppliers to produce some exemplars in terms of motherboard solutions - and more time to develop relationships should Zen be a success. The bad news is that time is not on AMD's side.

Lastly...

My last concern relates to my previous mention of perfect execution. AMD cannot afford releasing a substandard version of Zen. Expectations are high and another Bulldozer like release would all but end AMD's CPU business. AMD wisely shortened Zen's time to market by eliminating an iGPU. Nonetheless, my main concern is that AMD could delay Zen's release to match expectations as closely as possible (either because of respins or shortcoming in GloFo's 14FF process). The risk factors vis-à-vis Zen are sky high.

So in conclusion (aka, tl;dr) ...

- The APU business is weak, due to inadequate investment in new tech.
- The GPU business is weak, due to inadequate investment in new tech.
- The ARM business is non-existent, due to inadequate investment in new tech.
- Semi-custom is stagnant due to the long lead time in developing prospects and products.
- The CPU & Server business is all but dead, although AMD is making significant investments in new technology.

In closing, AMD's recovery is dependent on a complete turn around in their lowest performing business sectors ( their Enterprise business and the HEDT part of their consumer business). Now I can see why AMD's chance for recovery is so unlikely. It's very said to think about, but reality is what it is. One good product here an there isn't going to save AMD - not when their primary competition are two well funded, successful and very aggressive companies (Intel & Nvidia).

 

[MiddleOfTheRoad]

Its going so right the company keeps shrinking.

Saying it as if the entire desktop PC industry wasn't also shrinking along with it. There are several OEM's that have lost even more marketshare than AMD -- Heck, Sony completely shut down their entire PC business.

 

[mrmt]

- Semi-custom is stagnant due to the long lead time in developing prospects and products.

Semi-custom is a very hard value proposition when compared to off-the-shelf solution, both in terms of costs and TTM. That AMD could only get two relevant customers for its business is a testament on how bad their judgement were when betting that this market would be a growth engine for the company.

Overall I see Lisa doing what Rory was supposed to do when assumed the company, albeit from a much more precarious resource position than in 2012.

 

[Ajay]

Semi-custom is a very hard value proposition when compared to off-the-shelf solution, both in terms of costs and TTM. That AMD could only get two relevant customers for its business is a testament on how bad their judgement were when betting that this market would be a growth engine for the company.

Why semi-custom is allowed to continue (beyond fulling the Xbox & PS4 agreements) is beyond me.

Overall I see Lisa doing what Rory was supposed to do when assumed the company, albeit from a much more precarious resource position than in 2012.

I hope so, but AMD's position is very precarious. Unless they get some kind of cash infusion, they seem destined to shrink to a nearly irrelevant point. Sadly, Papermaster is still there and so is his ARM fantasy. I wish they the best, but I surely wouldn't want to be Lisa Su right now.

 

[jpiniero]

Why semi-custom is allowed to continue (beyond fulling the Xbox & PS4 agreements) is beyond me.

Well, it's not like what AMD is doing now is working either.

 

[mrmt]

I hope so, but AMD's position is very precarious. Unless they get some kind of cash infusion, they seem destined to shrink to a nearly irrelevant point. Sadly, Papermaster is still there and so is his ARM fantasy. I wish they the best, but I surely wouldn't want to be Lisa Su right now.

I think that the semi-custom and ARM "dreams" are in fact, just one. ARM is AMD strategy to reduce R&D spending in order to make their semi-custom design a good business proposition. The problem isn't about pursuing ARM, but about pursuing ARM in a non-existent market (semi-custom).

 

[Ajay]

I think that the semi-custom and ARM "dreams" are in fact, just one. ARM is AMD strategy to reduce R&D spending in order to make their semi-custom design a good business proposition. The problem isn't about pursuing ARM, but about pursuing ARM in a non-existent market (semi-custom).

I've wondered about the same thing - so it makes sense to me that AMD might be thinking that way. In which case, as you point out, there is no market for AMD's products since there are cheaper alternatives from companies with an existing track record (ASICs, FPGA, Mips cores, etc.).

 

[ShintaiDK]

Saying it as if the entire desktop PC industry wasn't also shrinking along with it. There are several OEM's that have lost even more marketshare than AMD -- Heck, Sony completely shut down their entire PC business.

Thats no excuse, its actually a terrible bad excuse for the failures of a company.

Try compare AMDs YoY revenue in its PC and graphics division with Intels.

Or look at the JPR numbers.

 

[superstition]

Corporations are buying PCs, but they have always ignored AMD...

You mean Intel has bullied their suppliers into not stocking AMD products.

Intel paid Dell up to $1 billion a year not to use AMD chips

AMD has complained since the 1980s about the perks Intel gives customers. Computer makers that place big orders get discounts. They may also get "market development funds," or money for featuring Intel in advertisements. That's why so many computer ads mention "Intel inside." The funds are substantial, sometimes totaling 70% of a computer maker's ad budget, says IDC PC analyst Roger Kay.

Intel is no longer allowed to pay off (either directly, or through unit discounts) OEMs to exclusively carry Intel CPUs or to not carry competitor Advanced Micro Devices' CPUs. Likewise, it can no longer retaliate against OEMs who opt to offer competitive products.

Intel is also banned from specifically redesigning its chips to harm its competitors. Specifically it will be forced to not limit the performance of rivals' GPU chips for at least the next six years. Also, it must publish clearly that its compiler discriminates against non-Intel processors (such as AMD's designs), not fully utilizing their features and producing inferior code.

From 2003 to 2006 AMD produced CPUs that were widely considered to outperform similarly priced Intel designs. While other issues also hampered AMD, it alleges that it largely failed to gain market share during this era thanks to Intel's questionable business practices -- and legal settlements seem to back up this claim.

Paying off and/or bullying OEMs
Sabotaging its compilers
Sabotaging benchmarks

What else?

Will the company survive and challenge Intel again? Or will it shrivel? I find it hard to believe that Intel would allow AMD to fold, just like MS rescued Apple.

Just for those who don't know... MS didn't rescue Apple with its tiny cash infusion. However, it did rescue it by promising more releases of its cash cow: Office.

Had MS said no more Office... it would have been no more Apple most likely.

It's rather amusing to see people talk about a company rescuing another by telling consumers they can spend hundreds of dollars for a product that has no competition in the market.