Server Refresh Can Reduce Total Cost of Ownership

Snackable-MoreBangforYourBuck.png

More bang for your buck. Essentially that is the driving force behind my team in Intel IT. Our IT department is on a tight budget, just like most enterprise IT departments. Therefore, return on investment and total cost of ownership are important considerations for deciding when to upgrade the servers that run our silicon design workloads. As a principal engineer in infrastructure engineering, I direct the comparison of the various models of each new generation of Intel® CPU to those of previous generations of processors. (We may re-evaluate the TCO of particular models between generations, if price points significantly change.) We evaluate all the Intel® Xeon® processor families – Intel Xeon processor E3 family, Intel Xeon processor E5 family, and Intel Xeon processor E7 family – each of which have different niches in Intel’s silicon design efforts.

We use industry benchmarks and actual electronic design automation (EDA) workloads in our evaluations, which go beyond performance to address TCO – we include throughput, form factor (density), energy efficiency, cost, software licensing costs, and other factors. In many cases over the years, one of the models might turn out better in terms of price/watt, but performance is slower, or the software licensing fees are triple those for a different model.

In silicon design, back-end design jobs are time critical and require servers with considerable processing power, large memory capacity, and memory bandwidth. For these types of jobs, the bottleneck has historically been memory, not CPU cycles; with more memory, we can run more jobs in parallel. The Intel Xeon processor E7-8800 v3 product family offers new features that can increase EDA throughput, including up to 20% more cores than the previous generation and DDR4 memory support for higher memory bandwidth. A server based on the Intel Xeon processor E7-8800 v3 can take either DDR3 (thereby protecting existing investment) or DDR4 DIMMs – and supports memory capacity up to 6 TB per 4-socket server (with 64 GB DIMMs) to deliver fast turnaround time for large silicon design jobs.

We recently completed an evaluation of the Intel Xeon processor E7-8800 v3 product family, as documented in our recent brief. According to our test results, the Intel Xeon processor E7 v3-based server delivers excellent gains in performance and supports larger models, faster iterations, and greater throughput than was possible with the previous generation of the processor. These improvements can accelerate long-running silicon design jobs and shorten the time required to bring new silicon design to market. These improvements can also reduce data center footprint and help control operational and software licensing costs by achieving greater throughput using fewer systems than were necessary with previous generations of processors.

Our tests used a large multi-threaded EDA application operating on current Intel® silicon design data sets. The result shows an Intel Xeon processor E7-8890 v3-based server completed a complex silicon design workload 1.18x faster than the previous-generation Intel Xeon processor E7-4890 v2-based server and 17.04x faster than a server based on Intel® Xeon® processor 7100 series (Intel Xeon processor 7140M).

The Intel Xeon processor E7-8800 v3 product family also supports the Intel® Advanced Vector Extensions 2 (Intel® AVX2) instruction set. Benefits of Intel AVX2 include doubling the number of FLOPS (floating-point operations per second) per clock cycle, 256-bit integer instructions, floating-point fused multiply-add instructions, and gather operations. While our silicon design jobs do not currently use AVX2 – mostly because the design cycles can take over a year to complete and during that time we cannot modify the plan of record (POR) EDA tools and infrastructure for those servers – we anticipate that Intel AVX2 can provide a performance boost for many technical applications.

I’d like to hear from other IT professionals – are you considering refreshing? If you’ve already refreshed, can you share your observed benefits and concrete ROI? What best-known methods have you developed and what are some remaining pain points? If you have any questions, I’d be happy to answer them and pass on our own best practices in deploying these servers. Please share your thoughts and insights with me – and your other IT colleagues – by leaving a comment below. Join our conversation here in the IT Peer Network.