[Motley Fool] 3 Reasons AMD is Falling Apart Before Our Eyes

[Exophase]

You are welcomed to disagree, but, when I see 4 generations of CPU's (Sandy, Ivy, Haswell, Broadwell) only about 20% is apart in benches, it makes me think that yes, they are stagnating.

How about AMD's last four generations? Has peak performance changed much more between Bulldozer, Piledriver, Steamroller, and Excavator? AMD is starting from further behind and they have more of a pressing need to try to catch up. They're definitely not resting on a monopoly.

There are many technical factors that have caused CPU performance improvements to slow down. Intel keeps cranking out manufacturing processes but Dennard scaling has hit a wall. Frequency scaling has topped out due to higher wire resistance offsetting shorter wire length. Uarch improvements hit diminishing returns trying to extract more ILP while nearing fundamental limits of branch prediction accuracy, cache/memory latency, and intrinsic parallelism in the code. Highly threaded applications are not considered prevalent enough in mainstream code so core counts aren't increasing in the mainline CPU lines, but the server and enthusiast lines continue to increase core counts.

Meanwhile, power efficiency is a bigger deal because at this point people are more interested in longer lasting laptops and tablets than more performance on desktops. The increased power efficiency inevitably results in a trickle down of increased performance throughout the thermally-constrained product lines (and AMD has certainly focused the hardest here), just not at the highest end for desktops. Despite that, Intel still released an enthusiast grade product tweaked to perform better with higher clocks (Devil's Canyon) so it's hard to say they're completely ignoring this segment either.

If anything, Intel has been including more CPU uarch improvements in their ticks (Ivy Bridge and Broadwell both have them) which were previously only die-shrinks. Because previously a die-shrink meant more clock speed and cache, now that isn't viable or isn't worth it. The uarch improvements, for whatever they're worth, are more work. This is not to mention the substantial IGP improvements made each generation.

 

[sm625]

If you really look at what has been happening, its clear that the node shrinks are not bringing the future that everyone thought. This is not an intel issue, it is industry wide. Intel is actually working thru the quicksand where other fabs have completely cancelled multibillion dollar nodes. But I just look at the products and you can see there is a clear issue. Focusing on intel alone, they hit a wall. Their last impressive new node was 32nm. It had a large drop in power consumption and really moved the bar in max clocks. As a matter of fact, 22nm not only failed to move the bar on max clocks, it actually seems to have dropped the ceiling. When it comes to what your average overclocker should expect, the bar did drop. Its clear that intel's 22nm node failed to move the max clocks, it is not even debatable. But now we have broadwell to examine. It is basically a shrink of haswell, and after so so much of a delay............its looking like the upper tier products won't be making it at all. Do you really think this is because of no competition from AMD? It is the node. Their 14nm is even worse than the 22nm results. Its a real issue and one that I find extremely scary.

Were you around when 65nm came out? That one was pretty bad too. 65nm only got you 20% more clock speed for the same amount of power. But then Conroe came around and got you about 100% more IPC for the same amount of power, on the same 65nm node. Every node shrink has produced similar gains. You cant measure those gains at the top end though, you have to look about halfway down the range, like around 2.5GHz, to see it.

It's all about architecture, not process.

 

[ShintaiDK]

SB to HW in the server went from 8 to 18 cores. Mobile is like 1/3rd the power usage with better turbos.

I think people are just grumpy because desktop CPUs isnt the prime crop anymore. And we see GPUs suffer the same in terms of nodes.

 

[pm]

I was going to post up a reply to keysplayr, but Exophase pretty much wrote exactly what I was going to say. Great post, Exo.

Beyond his excellent comments, I don't see the current situation as stagnation, as much as it's the market is moving away from what we Anandtech'ers generally want. I'm a desktop gaming/coder/computing enthusiast - and this market segment is generally a dying breed. I personally don't care much about IGP performance because I'm going to stick a card in regardless. I don't care much about power dissipation because I have a massive case which has enough fans to make a mini-tornado and a 850W power supply. All I really care about is raw compute performance. Meanwhile, Intel is focused on tablets, and laptops and in this area power and IGP performance are important and raw compute power isn't so much. And Intel is focused on servers, but the key metric here is performance per watt, not raw performance. So what Intel is innovating on is not really the thing that I'm interested in - or the metric that keysplayr quoted with his 20% performance delta - and so if you look at this one particular thing then it all looks like stagnation, but really it's stagnation on a feature that isn't really the focus because it's not where the money is....

(Note: Intel is my employer)

 

[boozzer]

Look at Intels R&D budget.

Also dont confuse a dynamic market with a static. Not to mention this market requires a huge cashflow. No innovation? No sales. High prices? Low sales. Both would put your monopoly in the track for bankruptcy. Funny enough we never had cheaper CPUs since the competition dissapeared. Its all about volume/profit ratio.

The CPU side of things actually still have great advances and innovation. R&D budgets have never been higher. However, the focus is just not desktop. The 99% crowd wants performance/watt. Not 500W CPUs.

The marketleader on GPUs set the price that people are willing to pay for many years. So dont expect any changes there.

yea, R&D in anything but the desktop cpus. I think my ivy bridge cpu would probably serve me for 3 more years judging by the 5 to 10% increase with every intel cpu gen unless amd delivers and forces intel's hand. I know about the explosive growth in mobile and tablets. But I also know 5% to 10% per new gen = me not buying any new cpu at all.

I do not want gpu to become the same as this. do you?

if every new gen of cpu or gpu can come close to 40% performance increase, I would personally upgrade almost every gen.

 

[ShintaiDK]

yea, R&D in anything but the desktop cpus. I think my ivy bridge cpu would probably serve me for 3 more years judging by the 5 to 10% increase with every intel cpu gen unless amd delivers and forces intel's hand. I know about the explosive growth in mobile and tablets. But I also know 5% to 10% per new gen = me not buying any new cpu at all.

I do not want gpu to become the same as this. do you?

if every new gen of cpu or gpu can come close to 40% performance increase, I would personally upgrade almost every gen.

The GPU obviously still got some large design benefits to be reaped, plus the benefit of a shrink is the same as always, tho there may be longer in between. But thats simply because GPU workloads are parallel in nature unlike CPU workloads. Hence the reason why you dont buy an 8 core atom over a dualcore Haswell.

What you and I want may not be what the market wants. We just have to accept its us who are the dinosaurs and not everyone else being wrong.

 

[el etro]

A lot of people on SA own shares obviously. Doesn't mean they (or he) can't do good analysis.

For me, it looks like is the cause of much of the partiality here.


Forum members should mandatorily disclose here if they have shares of any company, as they nowdays need to disclose if they have influencer board membership participation or not.

 

[ShintaiDK]

My stocks isnt any secret. Since I am too lazy and just use an investment fund for the time. Quite the generous return and dividends tho.

Nordea Invest USA
Nordea Invest Europe
Nordea Invest India

Note the danish, so you have to click "Vis alle beholdninger". Then you can see I am indirectly an evil Intel shareholder with 1.42% in the USA one. But obviously I am a bigger Apple fanboy with 6.55%.

 

[Silverforce11]

I am not really sure that intel is sitting on tbese huge improvements in cpus and just not bringing them to market because of lack of competition from AMD. Nobody really knows except intel insiders, but i think cpus are so fast and efficient now that it may not really be possible to make much larger improvements than what we are seeing, at least at reasonable cost.

Look at the die size of their CPUs and compare generation to generation.

Note with die shrinks, their dies for each segment gets smaller. It means they are NOT taking advantage of die shrinks for performance. If they did, their die size would stay relative, but have vastly more transistors dedicated for performance.

This is why GPUs have been leaping massively with die shrinks. A shrink often results in doubling of transistor count (or more, due to bigger dies).

Intel is taking advantage of die shrinks in two ways: 1) Slightly better power use and 2) Much reduced die size, therefore, more CPUs per wafer, increasing their output and potential margins/profit.

You can see their priority is NOT on performance gains. They don't need it because nobody pushes them. Now compare their CPUs designed to compete against ARM... they've been pushed by the best so they have to chase perf and perf/w.

Competition is good for innovation, lack of it causes stagnation. This applies to many disciplines, not just business or tech.

 

[Enigmoid]

Look at the die size of their CPUs and compare generation to generation.

Note with die shrinks, their dies for each segment gets smaller. It means they are NOT taking advantage of die shrinks for performance. If they did, their die size would stay relative, but have vastly more transistors dedicated for performance.

This is why GPUs have been leaping massively with die shrinks. A shrink often results in doubling of transistor count (or more, due to bigger dies).

Intel is taking advantage of die shrinks in two ways: 1) Slightly better power use and 2) Much reduced die size, therefore, more CPUs per wafer, increasing their output and potential margins/profit.

You can see their priority is NOT on performance gains. They don't need it because nobody pushes them. Now compare their CPUs designed to compete against ARM... they've been pushed by the best so they have to chase perf and perf/w.

Competition is good for innovation, lack of it causes stagnation. This applies to many disciplines, not just business or tech.

Haswell is smaller but its definitely not the smallest from a historical perspective. Otherwise I agree.

Broadwell and Skylake will be significantly smaller than Haswell. With BW-U GT2 around 80 mm^2, I expect 4C GT2 BW to be around 110 mm^2.

Kinda sad that we don't get an 8 core not igp Broadwell chip on the consumer line. Probably be just a tiny bit bigger than the 4C GT2 die considering the graphics are around the size of 3.5 cores.

 

[frozentundra123456]

Look at the die size of their CPUs and compare generation to generation.

Note with die shrinks, their dies for each segment gets smaller. It means they are NOT taking advantage of die shrinks for performance. If they did, their die size would stay relative, but have vastly more transistors dedicated for performance.

This is why GPUs have been leaping massively with die shrinks. A shrink often results in doubling of transistor count (or more, due to bigger dies).

Intel is taking advantage of die shrinks in two ways: 1) Slightly better power use and 2) Much reduced die size, therefore, more CPUs per wafer, increasing their output and potential margins/profit.

You can see their priority is NOT on performance gains. They don't need it because nobody pushes them. Now compare their CPUs designed to compete against ARM... they've been pushed by the best so they have to chase perf and perf/w.

Competition is good for innovation, lack of it causes stagnation. This applies to many disciplines, not just business or tech.

Well, i was primarily talking about per core performance, not simply using the die shrinks to add more cores.

 

[TurtleCrusher]

My stocks isnt any secret. Since I am too lazy and just use an investment fund for the time. Quite the generous return and dividends tho.

Nordea Invest USA
Nordea Invest Europe
Nordea Invest India

Note the danish, so you have to click "Vis alle beholdninger". Then you can see I am indirectly an evil Intel shareholder with 1.42% in the USA one. But obviously I am a bigger Apple fanboy with 6.55%.

I now work for one, interact with the other and still don't bag on the competition.

Weird concept, I know.

 

[Exophase]

This is why GPUs have been leaping massively with die shrinks. A shrink often results in doubling of transistor count (or more, due to bigger dies).

GPUs work with massively parallel problems that can still benefit from even more parallelism. People with desktops and laptops don't usually. Increasing single threaded performance is a much harder problem. For the small fraction that does seek more throughput on highly threaded tasks they sell enthusiast CPUs. With Haswell they have moved from 6 cores to 8 cores in the enthusiast line (and the 6 core variants can now be had for much less) For those who REALLY benefit from more throughput there are Xeons with very high core counts.

I'm sure if they went all out they could dedicate a huge amount of die area or a huge amount of extra power consumption for some single threaded performance gain, but it's just not worth it everything considered.

Now compare their CPUs designed to compete against ARM... they've been pushed by the best so they have to chase perf and perf/w.

But they're still also aggressively chasing perf/mm^2 there. Like when their flagship smartphone chip only had two cores. They've corrected it since then, but they don't exactly have class leading CPU or GPU performance with it.

 

[grimpr]

OP that is a rubbish article by none other than Ashraf Eassa. He loves Intel/Nvidia and cannot stop writing such trash about AMD. Nothing new from what he has done in the past. :thumbsdown:

here is another trash article when he was writing at SeekingAlpha, incidentally his last article at SeekingAlpha . He was probably booted out for the rubbish he spewed.

http://seekingalpha.com/article/2099473-amd-looks-finished
http://seekingalpha.com/author/ashraf-eassa/articles

Isn't that Ashraf Easssa guy our very own "Intel17" user on anandtech forums? Certainly looks very familiar.

 

[CHADBOGA]

OP that is a rubbish article by none other than Ashraf Eassa. He loves Intel/Nvidia and cannot stop writing such trash about AMD. Nothing new from what he has done in the past. :thumbsdown:

here is another trash article when he was writing at SeekingAlpha, incidentally his last article at SeekingAlpha . He was probably booted out for the rubbish he spewed.

http://seekingalpha.com/article/2099473-amd-looks-finished
http://seekingalpha.com/author/ashraf-eassa/articles

This was another wonderful article by Ashraf, it is a pity that some people are too immature to appreciate it.

 

[piesquared]

Shouldn't this and other threads like it be:

Moving to General Hardware.
-- stahlhart