From Skylake to Coffee Lake: New E3, E5, and E7 Chips
In the mainstream Intel-dominated enterprise chip market, most devices shipping today are in the Kaby Lake class and will soon transition to Coffee Lake. If you are shopping for servers for VMware or other bare-metal applications in the data center, consider dropping a few detailed high-end server requirements and leverage more affordable gold-level Skylake processors like the 6000 series.
When discussing enterprise server chip architectures, it’s common to get caught up in the technical specs — core processing threads (cores), CPU clock rates, memory, SmartCache, and thermal design power.
But I promised myself I would not blog about those everyday details, especially when I have recently been receiving more questions from clients, colleagues, and other IT procurement professionals about microprocessor architectures.
Essentially, they want advice, configuration details, sizing recommendations, and for me to explain the shift from the tick-tock model to the process, architecture, optimization model.
Code Names for CPU Releases
Intel uses these code names for their chipset releases. I’ll explain in a bit, but visually, it looks something like this:
- Westmere was a tick
- Sandy Bridge was a tock
- Ivy Bridge (tick)
- Haswell (tock)
- In 2016, we could say that Broadwell was both a tick and the start of a process generation in the new model
- Skylake (tock/architecture)
- By 2018, the tick-tock model ended, and Kaby Lake was simply optimization
- Coffee Lake starts a new process cycle
Every tick represented a shrinking process technology relative to the previous new microarchitecture (tock), with some exceptions. For example, released in late 2014, Broadwell introduced new instructions.
Every tock designated a completely new microarchitecture, again, with some rare exceptions. For example, in 2014, Intel created a tock refresh by releasing a smaller update to the microarchitecture that was not considered a new generation.
Generally speaking, tocks introduced new features and improved architectural performance while ticks improved clock speed or consumed less power.
Without debating the precise timing of Moore’s Law, we have generally expected either a tick or a tock every 12 to 18 months.
Then, in March 2016, Intel announced in a Form-10K report that it was replacing the tick–tock cycle with a new three-step process, architecture, optimization model. The new model would produce three generations of processors under a single unified manufacturing paradigm that would shift its focus from process to architecture and then to optimization.
Just when we thought the model was solid, Intel changed the game a bit by releasing two optimizations of its 14-nm Skylake architecture — Kaby Lake and Coffee Lake. Instead of three steps and three generations, Skylake represented a total of four generations.
E3, E5 and E7 Chipsets
Behind each code-named release, deep within the nanoarchitecture, is another world powered by the following chipsets:
E3 Chips
These are first to market and are generally one or two versions ahead of all chips. In today’s nomenclature, E3 chips are now considered bronze-level and exhibit the following characteristics:
- Same 14-nm architecture shared by Skylake, Kaby Lake, and Coffee Lake
- Very low core count of just 2-6 cores per socket
- Lowest cost, non-enterprise grade
- Not available in HP, Dell, or Cisco rack mounted or blade servers
- Typically found in consumer market laptops (as i3 processors) and small NAS appliances
E5 Chips
This is your workhorse in the data center. Since Skylake, these chips offer silver, gold, and even platinum levels of performance. Other highlights:
- Moderately priced
- Commonly deployed in VMware or bare-metal server applications that don’t require more than two sockets
- Ideal for high-end workstations and CAD applications
- Enterprise-grade chips with 6-28 cores
- Generally, 2-3 years behind the E3 chips and often skip versions completely
- Latest versions are Skylake in most OEM models with the next version of this chipset unofficially codenamed Cascade Lake
E7 Chips
These are the big dogs for dense CPU processing, enterprise applications, and large databases. Major characteristics include:
- High-end performance
- Highest cost
- Found in quad-socket applications
- Typically reach the market one year behind the E5
- Currently Skylake in most OEM models
Final Thoughts
In today’s blade server and rack mount datacenter market, the last major chipset release was Skylake in a new process model that superseded the legacy tick-tock model. Most devices are already at Kaby Lake and very soon will be moving on to Coffee Lake. Coffee Lake CPUs promise to put up to 6 cores and 12 threads inside servers, workstations, and high-end notebook PCs for the first time in history.
While we wait for newer 10-nm architectures (which some say are experiencing delays in manufacturing), we still have reliable 14-nm architectures with improvements, patches, updates, and revisions. If you do not have very specific high-end requirements, we see a good opportunity right now to leverage the gold-level Skylake processors (specifically the 6000 series) for VMware or bare-metal applications in the data center.