How do server-grade CPUs compare to consumer-grade ones?

Post Reply
User avatar
Edge100x
Founder
Founder
Posts: 12235
Joined: Thu Apr 18, 2002 11:04 pm
Location: Seattle
Contact:

How do server-grade CPUs compare to consumer-grade ones?

Post by Edge100x » Sun Oct 20, 2019 10:28 pm

We use Xeon E-series CPUs here at NFO instead of the consumer-side chip equivalents for a few reasons:

- They use ECC memory, which is important for preventing crashes (both application and OS crashes). Random bit flips in non-ECC memory are suprisingly common -- Google found that 8% of DIMMs were affected by memory errors in a year, for instance -- and ECC corrects most of these types of errors.
- Server-grade CPUs are typically binned higher, which means that the manufacturer (Intel, in this case) tests them to run faster and cooler than their equivalent consumer procesors. This also means that server-grade chips are more likely to be able to run at their listed turbo speeds. (Intel knows that customers who want the server-grade CPUs also expect a higher-quality, longer-lasting product.)
- Server-grade CPUs are supported by higher-quality server-grade hardware. Server-grade CPUs require server-grade motherboards inside specially-designed cases and are meant to be racked in a datacenter environment and run 24/7. Consumer-grade equipment is not built to the same standards, and those machines are often placed inside standard tower cases and put on breadracks in the datacenter.
- Server-grade CPUs can be efficiently packed into datacenter cabinets, so we can afford to place them into the high-end datacenters that we use inside the US.

One concern that we hear from customers is performance. How does the performance of an E-series server CPU compare to a consumer-oriented i7 CPU? It is relatively straightforward to compare the two sets of chips, as long as you compare them within the same generation, because Intel uses the same physical cores for the two types. Intel performs different tests on the two different types of chips, offers some different options for them (such as ECC on server chips, and sometimes different turbo tables), but otherwise, they are nearly identical.

Here are some examples of how the single-threaded performance of Intel CPUs compares.

an E-2286G is the same generation (Coffee Lake) and faster than an i9-9700K
an E-2186G is the same generation (Coffee Lake) and faster than an i7-8700K
an E3-1275v6 is the same generation (Kaby Lake) and similar to a i7-7700K (in part because the Xeon uses faster RAM)
an E3-1270v5 is the same generation (Skylake) and between an i7-6700 and an i7-6700K in speed
an E3-1270v3 or E3-1271v3 is the same generation (Haswell) and faster than an i7-4770K

In general, a later generation's processors will outperform a previous generation's processors on a clock-for-clock basis. In the last 10 years, Intel has seen a 5-15% performance increase for each generation.

Some hosts also advertise "overclocked" consumer processors. This means that they turn off turbo -- the built-in processor feature that automatically monitors processor load, voltages, and temperatures in order to dynamically raise the clock speed as needed -- and override it by always running the processor at something close to its maximum turbo rate. This can squeeze a small amount of additional performance out of the processor in exchange for lower reliability and higher power consumption. Overclocking an i7-7700K processor might make it run a few hundred MHz faster and closer to the speed of an E-2186G, for instance -- though not quite getting there (the i7 would still have slower RAM, and the processor generation difference is significant).

User avatar
Edge100x
Founder
Founder
Posts: 12235
Joined: Thu Apr 18, 2002 11:04 pm
Location: Seattle
Contact:

What about NVMe?

Post by Edge100x » Sun Oct 20, 2019 10:44 pm

NVMe is a faster interface between SSD storage and the processor/memory. An SSD running on the fastest SATA interface will top out at a random write rate of around 500 MB/s, while a similar drive, from the same manufacturer and running on an NVMe interface, might write at 2,500 MB/s.

Since both types of SSDs support 10s of thousands of IOPs (I/O Operations Per Second) at minimum and I/O is not the limiting factor for most applications at that level, both interfaces work well for game servers and most other applications, and most customers will not notice a difference between the two in production. On the provider side, NVMe drives can be a little cheaper than regular SSDs, and NVMe drives are typically a bit harder to replace (not being hot-swappable).

Since there are pros and cons to each, and the end result is similar, we don't recommend basing a purchasing decision on the type of SSD.

We use standard 2.5" SATA SSDs in most machines and M.2 NVME drives in our E-2286G and above offerings.

Post Reply