Figure 3 - Intel Xeon processor E7 family scalability to support 2- to 256-sockets3 White Paper Intel® Xeon® Processor E7 Family Performance
processor SKU (70); which indicates same core frequency of 2.4 GHz, the same Intel® QuickPath Interconnect speed of 6.4 GT/s, the same last-level cache (LLC) of 30 MB, and the same number of cores at 10 per processor.
So the only difference is in the first product family number represents wayness (2, 4, or 8) capability indicating that the Intel Xeon processor E7-4xxx and E7-8xxx models can scale natively beyond just 2-sockets (see Figure 2 below). It is common IT practice to buy “headroom” by purchasing a larger server but only initially partitioning a portion of the processor sockets for today’s level of requirements allowing for future compute power expansion as the number of users, transactions, or problem fidelity increases. Ideally, with perfect scaling, you can double the number of users, for example, when doubling the number of processor compute power (assuming storage, memory, and I/O are scaled as to not be the bottleneck). However, when any of these otherwise identical processors are populated in 2-sockets only though, performance throughput should be expected to be the same.
Figure 2 - Intel® Xeon® Processor E7-8800/4800/2800 Product Family Numbering2
PERFORMANCE IMPACT
For the purposes of demonstrating the impact of model numbers on performance, the top of the advanced capability levels of each product family is compared below (Intel Xeon processor E7-8870/4870/2870). Figure 3 below illustrates the options original equipment manufacturers (OEMs) have in designing an Intel Xeon processor E7 family-based server.
Looking at the first number in the Intel Xeon processor E7 family, -8xxx, -4xxx or -2xxx, which represents the number of processors natively supported in a server, the processors can scale to support the increased number of users, transactions or throughput as additional sockets are tested in performance benchmarks. The typical example of this can be found while using the SPECint*_rate_base2006
benchmark that is fairly representative of typical integer-based, compute-intensive server applications to test the 2
