A recent Microway blog post offers a detailed look at how the new Intel Xeon E5-2600v3 “Haswell-EP” processors perform for HPC applications.
“The Xeon E5-2600v3 processors introduce the highest performance available to date in a socketed CPU. For the first time, a single CPU is capable of more than half a TeraFLOPS (500 GFLOPS),” notes the post. “This is made possible through the use of AVX2 with FMA3 instructions.”
Here’s a refresher on some of the most significant changes:
- Up to 18 processor cores per socket (with options for 4-, 6-, 8-, 10-, 12-, 14- and 16-cores)
- Support for Quad-channel ECC DDR4 memory speeds up to 2133MHz
- Advanced Vector Extensions version 2.0 (AVX2 instructions)
- Improved energy efficiency with Per Core P-States and independent uncore frequency control
Microway reports that the new Haswell chips are faster in nearly every metric, such that users can expect at least a 10 percent increase in performance per core, excluding the new instructions. The chart below shows the theoretical peak performance of the new Haswell-EP CPUs with the new instructions.
Although these are theoretical peak speeds, the SPEC CPU2006 benchmarks, also reviewed in the blog, reflect a suite of real world applications. With the new memory and instructions, moving from v2 to v3 SKUs saw net improvement gains in the 20-30 percent range.
Of all the improvements in the new line, clock speeds were only modestly boosted. It’s worth noting that Turbo Boost in tandem with non-AVX instructions can enable 100MHz to 200MHz increase in the highest clock speed. However, since AVX2 allows 256-bit wide operations for both integer and floating-point numbers, HPC users will likely be running in AVX mode most of the time.