AMD CPUs Became 70% Faster Per Core With Ryzen; Intel Stagnant Since 2015’s Skylake

[DKT27]

Ever since AMD released the Zen-based Ryzen CPUs, their fortunes have overturned. The company sent rival Intel into a panic, resulting in price cuts and unplanned product launches that made a mess of their product lineup. But before Ryzen, things weren’t so “green” for AMD. The infamous Bulldozer architecture and its reiterations in the form of Steamroller, and Excavator were far from successful. On the other hand, Intel’s Core architecture and its successors kept on building a formidable lead over team red’s processors. This finally resulted in the pre-Ryzen scene where octa-core AMD CPUs were equal to quad-core Intel chips and even dual core at times. Now, things have gotten much better for Dr. Lisa Su and Co, but let’s go down memory lane and see how CPU architectures have improved over the past decades.

 

AMD CPU Architectures from 2003 to 2018: Single Threaded Performance or IPC

 

 

AMD was much better off back in the early 2000s when the 64-bit version of the x86 instruction set came out. The Sledgehammer and Opteron server chips were quite competitive if not groundbreaking (well, definitely not the latter). After the K8 architecture, things started going south, when Intel released its Core microarchitecture, popularly marketed as the Next-Generation Micro-Architecture. Bulldozer only made things worse, with the IPC taking a dive to pre-K10 times. This wasn’t because Bulldozer wasn’t a new design, it’s just that it was a moronic new design. They decided to go with higher core counts, but with shared logic. These “cores” weren’t cores, but in-fact just ALU clusters. Traditional CPU cores have their own frontend, cache, and floating point units, but AMD’s Bulldozer had two cores/Integer Clusters sharing the frontend, cache, and floating-point logic. This made the CPUs easier to build, but also severely handicapped their single-threaded performance due to the limited resources available to each thread or “core”. This was known as Clustered Multi-Threading (CMT).

 

And then came Zen, otherwise known as the Ryzen series which drastically improved the CPU IPC (by almost 70% while Intel’s Core architecture was reaching its limits). The rest happened in the last few years and is history.

 

Intel CPU Architectures from 2003 to 2018: Single Threaded Performance or IPC

 

 

Intel’s story is the exact opposite. Before the Core microarchitecture came up, team blue was rather deep in **** with the Prescott processors being a major failure, both in terms of performance as well as efficiency. However, thanks to the new Core architecture, and at the same time AMD’s Bulldozer being a massive flop, things just kept getting better and better for Intel. That is until Skylake, since then Intel has abandoned its Tick-Tock design model and has been stuck on the 14nm node. AMD, on the other hand, has regained much of its lost ground and is ready to transition to the 7nm node with Zen2.

 

There’s really not much to say about the present situation of the CPU market. Intel is struggling to migrate to the 10nm node, and if the recently leaked roadmaps are legit, then that’ll continue for the time being. AMD, on the other hand, has regained its lost market share thanks to the efficiency of the Zen micro-architectures and is looking to take the fight to Intel in the server territory as well. Zen2 might just make up for the blunder that was the Bulldozer design and in the process give its competitor a thorough pummelling. I suppose we’ll know soon enough.

 

View: Original Article.