2017: The year of the CPU

Before I begin, this is an involved topic, and I assume you are well versed with the semiconductor industry. It's no detailed analysis though. I shall avoid pictures to avoid clutter and achieve a better flow of commentary.

Context

It has been a fairly stagnant decade for the CPU industry. The battle was heating up in the early 2000s, with AMD firing on all cylinders. Intel made many mistakes, to make matters more interesting. The result was, by 2005, AMD's Athlon 64 line was superior to Intel's Pentium in every way.

Of course, as is often the case with AMD, better products never equal market share. Thanks to extremely shady tactics by Intel, they continued to have much higher revenues and thus development budgets. Later, Intel was deemed to have broken anti-trust laws across the world, and were imposed the maximum possible fines by EU EC and US FTC alike. It finally settled with AMD half a decade later. But by then, the damage as done. Intel may have paid close to $10 billion in fines, but it was a strategic expense - it has probably yielded over $100 bn in the long run. Such is the folly of capitalism.

Intel gave up on their Pentium line, and transitioned their mobile architectures to high-performance desktop. In 2006, they released a revolution - Core 2. Immediately, Core 2 outperformed AMD's now aging Athlon 64 X2. AMD took a massive gamble next - to integrate 4 native cores and an IMC on to a single die. The Phenom X4 was an engineering feat, but fell behind Intel's more practical Core 2 Quad - which was basically two Core 2 Duos glued together. Furthermore, AMD's manufacturing division was slipping behind Intel's fabs - them alone backed by R&D budgets several times the entirety of AMD's market cap.

A year after Phenom, in late 2008, Intel basically copied everything Phenom introduced, but this time with a manufacturing process that could handle it. Thus was born the Core i line. For the next decade, there was basically no competition. Intel commanded the prices they wanted, segmented the market as they liked. AMD took another gamble at doing something different with Bulldozer, and well, failed again. By the early 2010s, everyone had written off AMD as imminently fighting for bankruptcy. Meanwhile, the Intel juggernaut continued rolling on, relentlessly.

A glimmer of Zen

Not only did AMD not go bust, around 2012-14, they quietly started hiring top CPU designers from around the world. Jim Keller joined AMD, back from Apple, to lead development of Zen. Mark Papermaster took over AMD's CTO role - also previously at Apple. On paper, that sounds insane - why would the top CPU engineers leave the biggest company in the world to join a dying one?

Occam's razor - there was something exciting going on at AMD. A once-in-a-lifetime opportunity for the ultimate underdog to shove it to the ultimate goliath.

AMD starting talking about Zen in 2015, claiming unprecedented gains from Bulldozer. Granted, that's a low starting point, but the numbers made people sit up and take notice. Of course, most dismissed it as nonsense, it was simply not possible AMD could bridge an insurmountable gap with a fraction of the resources. Yet, AMD was not doing anything different this time - they were designing something to compete with Intel head on. No over ambitious innovations, or forward thinking designs - just something reasonable that'll work for the present.

On the other hand, Intel was getting complacent. The first Core i7 920 was a 4 core / 8 thread CPU at $300 in 2008. Core i7 7700 was a 4 core / 8 thread $300 CPU in 2017.

Return of the Jedi

The above title is an in-joke - I'll offer a hefty upvote if you get this reference and comment about it :)

In March 2017, Ryzen 7 released doing exactly what a seemingly crazy Jim Keller saw coming when none of us could. The ultimate underdog beating an abusive, dominant goliath.

At the same price, Ryzen 7 offered 8 cores / 16 threads - a near 2x bump over Core i7 7700. Amazingly, it did so at similar power consumption as well.

This is a milestone, and a turning point for the entire CPU industry. What came next was an absolute whirlwind of a year.

AMD goes for the jugular with Threadripper and Epyc

Remember how Intel beat AMD in 2007 by gluing two Core 2 Duos instead of going for a native four core die? Well, it's the exact reverse this time around. Intel has enormous 28 die CPUs for the server market that cost a fortune to manufacture. AMD, however, simply glued 4 of their Ryzen dies to create a behemoth 32 core CPU at a fraction of Intel's cost. Better still, they extended this concept to the desktop, releasing a 16 core / 32 thread CPU for $999. That is insane, considering Intel was peddling a 10 core CPU for $1800 at the time!

The result was Threadripper cleaned house, and was for the first time in over a decade, the fastest desktop CPU in the world.

The server and professional market was more challenging, with Intel's corrupt practices having firmly entrenched corporates. Even so, Epyc offered remarkable value, often offering twice the performance of Intel for the same price. The game was well and truly on!

Intel's knee-jerk reaction

It was clear that Intel had been caught well and truly napping with their pants down. Intel had no answer to Threadripper, so they did something unthinkable. They brought their massive server Xeon dies to the Core market!

Thus came a barrage of Core i9s, rushed to the market. The top i9 managed to take the lead back from Threadripper, but at a 80% higher cost, and using more power. Even so, the lead was iffy at best. It was a power play, one Intel had to make.

The Empire Strikes Back

Intel had to respond in the mainstream market too. For the first time ever in Core i history, the mainstream $300 Core i7 saw a boost in core count. Coffee Lake came with 6 cores / 12 threads, and with a significant clock speed advantage, the Core i7 8700K is a formidable product - Intel's best in many years. It competes well with Ryzen 7 in multi-threaded workloads, while being significantly faster in lightly threaded workloads.

The new Core i3 at $120 is basically the same as the old Core i5 7600, which cost double the price! An incredible change for Intel.

Yet, while all of this is impressive on the surface, cracks are starting to appear. Firstly, there's nothing new about this Coffee Lake. Indeed, the Core i3 is basically the old Core i5, and the Core i7 is the old Core i7 with 2 extra cores slapped on.

It was clearly rushed, too, with affordable motherboards still missing and not due till Q1 2018.

Mobile sees a revolution too

The 15W TDP mobile CPU has the highest volumes in the industry. It finds home in a vast majority of laptops today, save for some high-performance niche products which go with 45W instead.

As with the desktops, the 15W line of Core i CPUs had been stuck at 2 core / 4 thread since the very beginning. The 8th gen saw an incredible boost to 4 core / 8 threads. Mind you - this is not a 100% uplift, as the 4 cores have to throttle to stay within the 15W budget. Still, this so an unprecedented uplift, never seen before by Intel.

Make no mistake, though, this is not Intel being consumer friendly. Nope, this is very much being scared shitless by Ryzen Mobile. Ryzen 7 and Ryzen 5 for notebooks brings the same 4 core / 8 threads as Intel to the 15W TDP. Incredibly, the early benchmarks show Ryzen 5 actually beats equivalent Core i5 while costing less. While there are no Ryzen 7 laptops out yet, it's likely it will be in the same ballpark as Core i7 laptops.

However, it is in the GPU side of things where Ryzen 5 with Vega 8 thoroughly trounces even Core i7. The performance differentials are alarming - often as high as twice the performance. Games that were totally unplayable on Intel are now suddenly fluid on AMD. It's night and day. Ryzen 7 with Vega 10 will open that gap further.

I don't expect AMD to gain much market share in notebooks - Intel is far too deep in bed with the OEMs - but this has surely shaken things up.

If you can't beat 'em, join 'em

There have been rumours about this for years, but no one actually believed them. The unthinkable, that with partner with AMD.

Yet, it is happening. Intel knows they cannot compete with AMD on the GPU front, so they have no licensed AMD GPUs to go along their CPUs in a single package. This will no doubt find its way into Dell XPS 15 or MacBook Pro 15 next year.

There was another stunning announcement - Intel had poached AMD's GPU head Raja Koduri away, to create new GPU architectures in-house for Intel. That AMD-Intel collaboration is seemingly a ticking time bomb. Though Raja Koduri - again, of Apple and AMD fame - is a tour de force in the industry, building a high-performance GPU from the ground up has proven to be insanely hard. Yet, Intel is now serious about pouring billions into it. It may just happen.

ARM joins the party

To make matters worse for Intel, Qualcomm Snapdragon now officially finds its way into Windows PCs. Of course, Microsoft tried this with Windows RT, and Google has had ARM Chromebooks too. However, those early efforts were utter failures due to compatibility issues. Now, ARM processors can run legacy Windows apps through an emulation layer. Of course, performance will be far superior on Intel and AMD, but perhaps ARM can differentiate with lower costs and long battery lives.

Indeed, the first ARM powered laptops claim a ridiculous battery life of over 20 hours in ultra thin form factors.

Intel will fall further behind

Coffee Lake may seem like a stunning success, but it's only a temporary one. Intel has reached the end of line with 14nm processors. Their 10nm process has been an absolute disaster, and by their own admission, they won't release a desktop part on 10nm till 10nm+. That's 2019 at the earliest, more likely 2020.

Meanwhile, for AMD, the first generation Ryzen 7 was a worst-case baseline. It was made on a borrowed 14nm process from Samsung, which is designed for mobile chips - not high-performance CPU. Intel has pretty much played their hand, for now.

In early 2018, AMD is set to release Ryzen+ at 12nm. This is only a minor evolution, but should be enough to leapfrog Coffee Lake. However, things get interesting in 2019 with Ryzen 2 and 7nm. This will be the first Ryzen made on a high-performance process, that takes all the learnings from the first generation. Without any doubt, AMD is going to command a significant lead throughout much of 2019, till Intel finally responds with a 10nm Ice Lake.

On the other side, Qualcomm's focusing on more powerful ARM SoCs to take on Intel.

Don't worry about Intel though, they'll do what they do best. Brute force their way into shady deals, and continue raking in the cash with inferior products. In addition, rampant consumer ignorance means they'll continue to thrive. Over time, through sheer amount of resources, they'll catch back up to AMD.

---

It's been an insane year for the CPU industry. More has happened in 2017 than the last ten years combined. I have deliberately edited out the mobile stuff, as this was getting too long and unwieldy.

What do you think? Will Intel fight back and crush the AMD rebellion?

PS: Hint about the Star Wars reference. A certain iconic technology publisher used to make Star Wars themed reviews for CPUs.