Will AMD ever be able to compete with Intel again?

[Mand]

and without the fabs.

This, I think, is one of the more important aspects. Intel's fabs are an ENORMOUS benefit to them. Not having them stings.

Now, you could say ah but look, AMD and Nvidia are doing just fine competing with each other, and neither has fabs of their own! Sure, but I'm pretty sure if one of them did start a dedicated GPU fab, they'd be able to do some rather amazing things. Not going to happen given the capital investment required, but the impact would be substantial.

 

[witeken]

I dont expect 10nm products earlier than Q4 2016-Q1 2017. And if you actually want me to predict the future, i would say no 10nm products before Q2 2017.
Also since we are talking Desktop here, at least Im, first 14nm Intel Desktop will not release before mid 2015. So 14nm Desktop will carry on for 2016 as well. That leaves a nice opportunity for AMD in 2016 to come back.

You are way off, about 12 months, to be precise. Production will start in 2015.

 

[witeken]

GloFos 20nm is Planar not FinFet. Samsung 14nm is FinFet.

It are 20nm transistors, but marketing calls it 14nm. That doesn't suddenly make it a 14nm transistor. If you take a 20nm process and replace the planar transistor with a 3D one, you still have a 20nm process.

 

[classy]

Anything is possible, but at this point from a performance perspective, its not likely. They are just so far behind.

 

[raghu78]

AMD will compete with Intel in traditional high end CPU performance with its new x86-64 core due in 2016. AMD made a failed design with Bulldozer. But they have proven with their Cat family cores that they can compete with Intel even with a process node disadvantage.

 

[Ajay]

The near term limiter for AMD is financial. AMD is not increasing it's top line, is still making WSA payments and and is still servicing allot of debt. Now recently AMD restructured some of it's debt pushing out payoff dates an slightly decreasing quarterly debt payments: http://seekingalpha.com/news/1584993-amd-refinancing-590m-worth-of-debt.

But this is just a stop gap measure. I just can't see AMD pulling off any miracles without the miracle of finding a minority partner with deep pockets.

A case in point, uArch overhauls typically take 5 years, yet the new x86 uArch being talked about will have been done four years after Keller took over as Chief Architect. Maybe this has been achieved with almost exclusive reliance on EDA tools - if so, how does AMD address it's weaknesses in memory and cache performance with this kind off TTM and budget pressure? Basically, this will be AMD's best shot at advancing their x86 perf/watt in the shortest time possible, with assuredly less R&D dollars than they had for Bulldozer. It's not going to best Intel's top of the line desktop CPU and it will cost more per die to produce. It will help them keep loyal fans, but again, it'll take a miracle for them to get more design wins with OEMs.

An important question I have: can AMD continue to support at least 2 x86 design teams with their declining budget? If they have to move to a serial design organization, their x86 CPUs will be dead in the water.

 

[AtenRa]

You are way off, about 12 months, to be precise. Production will start in 2015.

That link is from 2013, you know very well that even if Intel will keep the two years cadence they will start production of 10nm in Q1 2016 the earliest. I bet they will not, they didnt the last two times(22,14) and i dont believe they are going to do on the next one.

So if you see the first 14nm products in Q3-Q4 2014, then the earlier you could see 10nm products would be Q3-4 of 2016 IF they would start production in Q1 2016.

 

[AtenRa]

It are 20nm transistors, but marketing calls it 14nm. That doesn't suddenly make it a 14nm transistor. If you take a 20nm process and replace the planar transistor with a 3D one, you still have a 20nm process.

I dont care how they call them, one is Planar the other is FinFet. They are different with different specs and characteristics.

 

[AtenRa]

AMD will compete with Intel in traditional high end CPU performance with its new x86-64 core due in 2016. AMD made a failed design with Bulldozer. But they have proven with their Cat family cores that they can compete with Intel even with a process node disadvantage.

I would like to see you talk about the BD architecture after Mobile Kaveri release. 😉

 

[Ajay]

I dont care how they call them, one is Planar the other is FinFet. They are different with different specs and characteristics.

Yeah, but the scaling reduction is only around 15% (IIRC) vs Intel's ~2x scaling from 22nm to 14nm. So GFL (Samsung) 14nm will mainly be a play on lower power whereas Intel's 14nm will be a play on performance and density (particularly, $$s/xtor). So Intel's 14nm is in a different class which certainly will matter a great deal for x86.

It's a step in the right direction in perf/watt for AMD, but not in step with $s/xtor - which ultimately may even be more important than perf/watt. Sadly, AMD has not control over $s/xtor (which is a manufacturing cost).

 

[AtenRa]

Yeah, but the scaling reduction is only around 15% (IIRC) vs Intel's ~2x scaling from 22nm to 14nm. So GFL (Samsung) 14nm will mainly be a play on lower power whereas Intel's 14nm will be a play on performance and density (particularly, $$s/xtor). So Intel's 14nm is in a different class which certainly will matter a great deal for x86.

It's a step in the right direction in perf/watt for AMD, but not in step with $s/xtor - which ultimately may even be more important than perf/watt. Sadly, AMD has not control over $s/xtor (which is a manufacturing cost).

Intel 22nm to 14nm has 2x transistor density.

GloFo 28nm to 20nm has 2x transistor density.

GloFo/Samsung 14nm FinFet +15% transistor density over 20nm.

GloFos 28nm is close to Intels 22nm in density, Intel has a small lead.

So at the end, both GloFos/Samsung (i will say TSMC also) and Intel 14nm processes will have almost the same density and comparable electrical characteristics(Intel could probable have a small lead).
That will only hold until Intel will launch its 10nm products sometime in 2017.

I believe AMD will have a chance in 2016 by using the 14nm FinFet process. If they will not, then there is no possible way to directly compete in CPU performance alone.

One more thing, even if they will not get the faster performing product, they could have a better Performance per Watt SKU by using the 14nm FinFet process.

 

[raghu78]

I would like to see you talk about the BD architecture after Mobile Kaveri release. 😉

AMD has started to change the architecture of Bulldozer. Already decode has been decoupled and is no longer shared. The rumour is that AMD is going to widen the execution pipes to 4 ALU and 4 AGU per integer cluster. The move to a 256 bit FPU for Excavator is confirmed. AMD could quickly close up on Haswell IPC. I think Excavator could be a surprise in the good sense. I am a bullish supporter of AMD but the original Bulldozer was an unmitigated disaster as said by Andrew Feldman of AMD.

 

[Ajay]

Intel 22nm to 14nm has 2x transistor density.

GloFo 28nm to 20nm has 2x transistor density.

GloFo/Samsung 14nm FinFet +15% transistor density over 20nm.

GloFos 28nm is close to Intels 22nm in density, Intel has a small lead.

So at the end, both GloFos/Samsung (i will say TSMC also) and Intel 14nm processes will have almost the same density and comparable electrical characteristics(Intel could probable have a small lead).
That will only hold until Intel will launch its 10nm products sometime in 2017.

I believe AMD will have a chance in 2016 by using the 14nm FinFet process. If they will not, then there is no possible way to directly compete in CPU performance alone.

One more thing, even if they will not get the faster performing product, they could have a better Performance per Watt SKU by using the 14nm FinFet process.

My bad, forgot the 20nm base. I already made the same comment about perf/watt. $s/xtor will be much better than I thought. AMD needs to hit it out of the park on 14nm x86 in order to have a chance of keeping their R&D pipeline moving forward, IMHO. They need a value proposition that doesn't crater their margins.

 

[ninaholic37]

There always seems to be around a 50/50 ratio of AMD and Intel laptops and desktops being sold in stores (and advertised in flyers) in the three different cities I've lived in. I think that's quite impressive, considering Intel's size and budget is ~10x larger. So whether or not I agree with their products, they still seem quite competitive/popular here.

 

[AtenRa]

AMD has started to change the architecture of Bulldozer. Already decode has been decoupled and is no longer shared. The rumour is that AMD is going to widen the execution pipes to 4 ALU and 4 AGU per integer cluster. The move to a 256 bit FPU for Excavator is confirmed. AMD could quickly close up on Haswell IPC. I think Excavator could be a surprise in the good sense. I am a bullish supporter of AMD but the original Bulldozer was an unmitigated disaster as said by Andrew Feldman of AMD.

I will agree the first BD products and not the architecture were lacking the performance they should of had. The architecture is fine, what AMD was lacking was a good and competitive manufacturing process. It seams that 28nm can be good enough to compete against Intel's 22nm by using a better design/architecture (im talking about APUs not CPU alone).

 

[witeken]

Yeah, but the scaling reduction is only around 15% (IIRC) vs Intel's ~2x scaling from 22nm to 14nm. So GFL (Samsung) 14nm will mainly be a play on lower power whereas Intel's 14nm will be a play on performance and density (particularly, $$s/xtor). So Intel's 14nm is in a different class which certainly will matter a great deal for x86.

It's a step in the right direction in perf/watt for AMD, but not in step with $s/xtor - which ultimately may even be more important than perf/watt. Sadly, AMD has not control over $s/xtor (which is a manufacturing cost).

+ 14nm will already be Intel's second generation Tri-Gate, while GlobalFoundries doesn't even have FinFET yet.

 

[witeken]

Intel 22nm to 14nm has 2x transistor density.

GloFo 28nm to 20nm has 2x transistor density.

GloFo/Samsung 14nm FinFet +15% transistor density over 20nm.

GloFos 28nm is close to Intels 22nm in density, Intel has a small lead.

So at the end, both GloFos/Samsung (i will say TSMC also) and Intel 14nm processes will have almost the same density

Here are some real facts (14nm, for example, has more than 2x scaling):

Intel has a massive 35% density lead until FinFETs arrive in H1 2016. If 10nm doesn't have any problems, it will launch only a short time later, substantially increasing its density advantage. The competitor's 14nm FinFETs (marketing: 10nm) will just arrive 2+ years later, which will still have a lower density than even Intel 14nm, while Intel's already launching 7nm.

Anyway, you will now probably say that there will be FinFET Plus, but we don't know when those arrive, it might be a few months or a year, and they will partly narrow gap; to about 1.3x lower density.

and comparable electrical characteristics(Intel could probable have a small lead).

You think that the competition's first generation FinFETs, which might have been rushed somewhat since they don't 2 full nodes after HKMG, but just 1 node later, will easily compete against Intel second generation Tri-Gates?

That will only hold until Intel will launch its 10nm products sometime in 2017.

I don't think you have any substantial information that confirms a 2017 launch while Intel stated a 2015 ramp up of 10nm and 2017 for 7nm?

I believe AMD will have a chance in 2016 by using the 14nm FinFet process. If they will not, then there is no possible way to directly compete in CPU performance alone.

One more thing, even if they will not get the faster performing product, they could have a better Performance per Watt SKU by using the 14nm FinFet process.

I don't think the 14nm process will be any different than 28nm vs 22nm, at least not in AMD's advantage.

 

[VirtualLarry]

There always seems to be around a 50/50 ratio of AMD and Intel laptops and desktops being sold in stores (and advertised in flyers) in the three different cities I've lived in. I think that's quite impressive, considering Intel's size and budget is ~10x larger. So whether or not I agree with their products, they still seem quite competitive/popular here.

And how many of those AMD laptops, are "worthless" Kabini models (OK, the A4-5000 isn't bad), and how many are mobile Richland or Kaveri?

I see tons of Kabini models of desktops around here too, and AIOs. (Granted, I bought one.) But, man are they slow. I'm not even sure that the E1-2500 is faster than my E-350 CPU, even though it is several generations newer. (The E-350 is 1.6Ghz Brazos, the E1-2500 is 1.4Ghz Kabini.)

 

[frozentundra123456]

The microcenter in my area has 8 laptops listed with AMD cpus and 150 with intel. The weekly add for Best Buy has 10 intel laptops and only one AMD. It may be different in different areas of the country or in different countries, but in my area, AMD has virtually no presence in laptops except for kabini models.

 

[AtenRa]

AMD has a serious problem in mobile and that is not because of the product performance and/or power consumption. Quad Core Richland was not existent here and im willing to bet Kaveri will have the same fate.
Also, i was looking for plain iGPU only Quad Core Richland Laptops and i couldnt find more than one or two. And those were too expensive making the Intel Core i5 + dGPU look like they were on sales 😛
It was frustrating to see dual cores with low end dGPUs and not the much better Quad Cores with iGPUs. There were millions of Intel dual Core Celerons and Pentiums and Core i3 with Intel HD graphics when Quad Core Richlands would be way better but nowhere to be found.

 

[jpiniero]

The microcenter in my area has 8 laptops listed with AMD cpus and 150 with intel. The weekly add for Best Buy has 10 intel laptops and only one AMD. It may be different in different areas of the country or in different countries, but in my area, AMD has virtually no presence in laptops except for kabini models.

It was not long ago (ie: well past Conroe) that AMD had 50-55% retail market share. It's probably still the 40ish range. Of course the retail market has shrunk so much that it's hard to just live on that, and that's kind of been AMD's fundamental problem.

 

[Enigmoid]

I will agree the first BD products and not the architecture were lacking the performance they should of had. The architecture is fine, what AMD was lacking was a good and competitive manufacturing process. It seams that 28nm can be good enough to compete against Intel's 22nm by using a better design/architecture (im talking about APUs not CPU alone).

The architecture was crap. With a full node shrink efficiency was worse in tasks without BD specific extensions (such as AVX). BD had a full node shrink and it used more power than PII x6 for similar performance. Imagine how bad it would have been on the same 45nm process.

Cache on the FX was crap and what was worse was that the chip required so much of it (L2 not shared among cores).

 

[chrisjames61]

It will be very difficult for AMD's 2016+ K12/x86 equivalent to go up against Skylake, let alone Cannonlake. Intel's manufacturing lead is formidable, its cost structure is much better (which allows it to more profitably put in more features/performance), and its physical design teams are simply larger/more capable.

Look at the damage that Bay Trail-M is doing to AMD's computing solutions group...the chip that so many have mocked is winning almost all of the attractive low-cost notebook sockets. And, no, there's no "contra-revenue" here -- these are extremely profitable for Intel.

Once Intel moves to Braswell with Gen 8 GPU + 14nm + a focus on lowering system BoM, it really will be more or less lights out for AMD across the spectrum of PCs. And while AMD will hype K12 until it is blue in the face all throughout 2015 (AMD is doing the NVIDIA project denver thing..."high performance ARM core"), it ultimately won't save them.

AMD's best bet is to more or less put all of its energies into becoming an NVIDIA competitor. Semi-custom + dGPU should serve the business well, but any hopes that AMD will -- in an environment where Intel, Apple, Qualcomm, etc. all have soaked up the best engineers -- regain former glory is just misguided at best.

The odds are certainly stacked against AMD. Hopefully they can carve out a profitable niche. I would like to see them stay in the X86 game.

 

[AtenRa]

The architecture was crap. With a full node shrink efficiency was worse in tasks without BD specific extensions (such as AVX). BD had a full node shrink and it used more power than PII x6 for similar performance. Imagine how bad it would have been on the same 45nm process.

Cache on the FX was crap and what was worse was that the chip required so much of it (L2 not shared among cores).

As i have said, first Gen product was not performing as it should. Second gen products like FX8350 using the derived Bulldozer mArchitecture (Vishera) are performing as they should off. Third gen (Kaveri) products have higher IPC and MT performance using the Bulldozer derived architecture (steamroller).
All four products including next year Carrizo are using the Bulldozer derived architecture. So, i wouldn't judge the mArchitecture from a first gen product only and dismiss the other 3.

 

[TreVader]

AMD doesn't currently have the Fabs to beat intel. I think intel has way more to fear from ARM than any x86 challengers.


AMD is better at engineering. It's manufacturing and money that got intel here, not intelligent design. They would have ridden the sinking ship of net burst straight to bankruptcy if they hadn't payed of Dell and others to buy their processors.


AMD is a the superior processor engineering company. From a business perspective, they just don't have intels killer instinct. They aren't willing to bribe their way into devices which is how stuff works now.

This thread has gone surprisingly well so far. However intentional trolling/thread crapping is and always will be against the rules.
-ViRGE