AMD’s market share has been on a steady upward trend since Ryzen debuted in 2017 and the most recent market share estimates from Mercury Research bear out this continued improvement.
According to Mercury, AMD’s overall x86 share is 18.3 percent, an increase of 3.5 share points quarter-over-quarter and 1.2 share points year-on-year. The implication of this is that Intel’s market share had been higher in Q1 2020 than it was in Q2 2019, showing that there’s moderate fluctuations between the two companies on an ongoing basis.
AMD’s share of the desktop market, excluding IoT, hit 19.2 percent, an increase of 0.6 percent quarter-on-quarter and 2.1 percent year-on-year. Notebook share was 19.9 percent, an increase of 2.9 percent QoQ and 5.8 percent YoY. AMD has grown notebook share for the previous twelve consecutive quarters and broke its previous record for notebook share (19.4 percent, achieved in Q4 2006).
Overall client (consumer) market share is 19.7 percent, an increase of 2.2 share points QoQ and 4.7 share points YoY. This represents AMD’s highest overall share of the market since Q1 2012.
AMD says it’ll report its server share details later in the year when IDC reports, but the question of AMD’s relatively slow ramp in servers comes up practically every conference call, and I’m sure it’ll come up again when they eventually publish. What everyone remembers is this graph:
What this graph shows is that somewhere between January 2005 and June of 2006, AMD’s server market share went from ~5-7 percent to ~22 percent in just 18 months. There are several reasons why AMD isn’t repeating that ramp this time around.
First, the chart’s starting position is incorrect. AMD didn’t begin trying to break into server markets with Opteron, it started with K7 and Athlon MP. The surge at the end of 2005 was the result of over four years of work and careful generational improvement. The market was so wary of adopting AMD parts, even the launch of Opteron only won the company a small bump. The reason it’s important to include those first two years is because it establishes that even the leap from Athlon MP to Opteron wasn’t a game-changer for AMD’s server roadmap.
Two things changed from 2003 – 2005. First, AMD’s success with 64-bit Opteron forced Intel to completely tear up its Itanium roadmap and to pivot back towards building a 64-bit x86 CPU. The PR win for AMD on this issue was enormous and it helped to favorably position Opteron as the server CPU of the future. Companies were starting to think about 64-bit support at this point in time, and AMD looked like the vendor with the better overall roadmap. AMD also had a specific and particular lead in s0-called “glueless” architectures that Intel couldn’t match (at the time), and this helped it win space in the small-but-profitable 4S server space during this era.
Second, the dual-core Opterons typically destroyed the Xeons Intel was fielding, by fairly large gaps. Intel still dominated overall server shipments, but AMD had enough dramatic wins to make it the server of choice for certain vendors and applications.
AMD has claimed a wide number of wins relative to Intel in recent benchmarks, but many of these have been in systems where AMD can leverage high core counts. Back in 2005, AMD was decisively beating Intel at the dual-core and quad-core system level, at a time when upgrading to new server motherboards for the purpose of improving rack density was also very popular. AMD took a leadership position in server at a time when the benefits of adopting new hardware with higher CPU core counts was very clear, and its market share benefited tremendously.
What we see happening with Epyc is more of a conservative ramp-up as companies like what they see and deploy more AMD hardware. Intel’s competitive positioning is stronger lower in the product stack than at the top, where AMD’s core counts can outstrip Intel’s absolute performance in all but the most AVX-512 optimized tests. Markets like AI and ML are both hot topics right now, and AMD’s presence in those spaces is weak compared to the amount of work Intel has poured into them on the hardware and software side.
None of this says anything bad about Epyc or AMD’s server business. The difference in their growth rates, in my opinion, have more to do with relative positioning between AMD and Intel and the difference in the types of products that are selling the best. Even Intel has said that it expects AMD to be a more robust competitor, and we’re definitely seeing that.
I would expect to see a more aggressive shift towards AMD in the server space if Intel continues to struggle with node ramps and is forced to delay future server products. So long as the company continues to deliver on its yearly cadences and can keep Xeon competitive through IPC improvements, price cuts, and increased core counts over the long term, Intel can limit the damage AMD does. That’ll translate into a slower, consistent quarter-on-quarter market share improvement, but less chance of a huge leap over 12-18 months.