The HPC market is a key one for Intel’s Xeon E5 family of processors, and that is perfectly evident as the chip maker rolls out the “Haswell” Xeon E5-2600 v3 processors for servers and workstations at its Intel Developer Forum today in San Francisco.
Many of the new features in the Haswell Xeon E5s are designed to boost the performance and efficiency of calculations that are commonly done in modeling and simulation applications. This is not a surprise, since HPC is now a big – and growing – business for Intel.
Sales of processors and other components to HPC customers grew faster than Intel expected in the past couple of years, as we reported in EnterpriseTech back in November when Intel revised its revenue projections for its Data Center Group. An earlier forecast pegged HPC-related revenue growth at 14 percent compounded annually between 2011 and 2016, and Intel actually delivered 20 percent growth from 2011 through 2013 and went back to redo the projections going forward. Spending by HPC customers as well as hyperscale and cloud customers were much larger than expected, and enterprises did not spend as much as anticipated, and because enterprises are still the biggest part of the base, Intel’s overall growth was lower than it had planned in the Data Center Group. But, HPC was a bright spot, with the market for both traditional and commercial HPC rising, Intel making gains against competitive architectures, and customers tending to buy more powerful and more expensive processors than might be expected based on past trends.
In a briefing ahead of the Haswell Xeon E5 launch, Matt Kmiecik, segment marketing manager for the Technical Computing Group at Intel, said that the traditional scientific modeling and simulation workloads at supercomputer centers run by governments and academia – climate, physics, neuroscience, and genomics – as well as simulation and modeling software in the manufacturing, oil and gas, finance, and media businesses were driving the growth for Intel in the HPC sector. Data-driven analytics, used for fraud detection, marketing, and social science research among other areas, was also expanding the HPC pie, and so is a growing trend to use HPC tools in the cloud. High-fidelity visualization is also pushing growth.
The plan now is for Data Center Group to grow revenues by 15 percent per year at a compound rate between 2013 and 2017, with sales to HPC, cloud/hyperscale, and telecommunications firms growing in excess of 20 percent and compensating for the lower 8 percent growth rate for sales into the enterprise space.
Given the stakes, it is not hard to see why Intel is once again significantly boosting the core count with the Haswell Xeon E5s and doing other things to dramatically improve the floating point performance and memory bandwidth of the new processors, which are designed mostly for two-socket servers but which do have some single-socket variants for workstations and overclocked servers. We have gone through the detailed specs and price analysis of the entire line of Xeon E5-2600 v3 chips over at EnterpriseTech, and here, we will drill down into the specific features that will matter to HPC customers.
There are 29 chips in the Haswell Xeon E5 v3 family, and not all of them are aimed at HPC. Here is the general breakdown of the chips by type in the prior “Ivy Bridge-EP” Xeon E5-2600 v2 processors, on the left, compared to the product categories for the Haswell-EP Xeon E5-2600 v3 chips, on the right. This is not a complete SKU listing, but the ones that Intel picked out as representative:
The Xeon E5-2600 v2 processors topped out at 12 cores per die, but the new Haswell chips, which are implemented in the same 22 nanometer processes as their predecessors, can have as many as 18 cores per die. So that is up to a 50 percent increase in the number of cores, which drives per-system performance and which allows HPC customers to cram a lot more performance into the same space. The two top-bin parts with high core counts, the E5-2699 v3 with 18 cores and the E5-2698 v3 with 16 cores, run at 2.3 GHz and have 45 MB and 40 MB of L3 cache across those cores, respectively. They also run a little hot – in part due to the integration of the voltage regulator on the die on all Haswell Xeon E5s. And what also makes these two top-end Haswell Xeon E5 processors unique in Intel chip history is that they do not come with an official list price. These are not customized chips and will be made available to all customers, but they are, according to an Intel spokesperson, “unique offerings that fall outside of our traditional, publically available 2S product offering” and were created for HPC, virtualization, and cloud customers looking for maximum performance. The 18-core chip is obviously the most interesting one for a lot of HPC workloads that can use threads and are not as sensitive to clock speeds, and it is also the one that Intel is using to show off relative performance compared to prior Xeon E5s.
Other big changes that come with the Haswell Xeon E5 chip include faster QuickPath Interconnect (QPI) bus speeds, which are raised by 20 percent to 9.6 GT/sec. This extra bandwidth between the two sockets in an E5-2600 v3 server will help balance out the performance of systems with considerably more cores. That bandwidth is also necessary headroom for DDR4 main memory. Intel is the first chip maker to support DDR4 memory, and in the Haswell Xeon E5s Intel is supporting four DDR4 channels per socket and two memory DIMMs per channel when memory runs at 1.33 GHz, 1.6 GHz, or 1.87 GHz speeds. Only one DIMM per channel is supported at the fastest 2.13 GHz speed, although Intel says that in some cases server makers will certify their systems to support two DIMMs per channel at that higher speed. (This is not recommended by Intel at this time.) The DDR4 memory runs at 1.2 volts compared to 1.5 volts for standard DDR3 memory, and therefore better power efficiency. Intel estimates the shift to DDR4 will save about 2 watts per memory stick, and on early STREAM memory bandwidth benchmark tests, a two-socket Haswell machine with 2.13 GHz DDR4 memory delivered about 14 percent higher performance than an Ivy Bridge machine using 1.87 GHz DDR3 memory.
Inside the Haswell core is where a lot of HPC goodies have been woven in. First, Intel has added more execution units, deeper buffers, better branch prediction, and an entirely new front end to the core to boost the instructions per clock (IPC) for the Haswell Xeon E5 by around 10 percent. This is a big jump in IPC performance, and is akin to that with the “Nehalem” Xeon cores in March 2009 and the “Sandy Bridge” Xeons in March 2012.
The L1 and L2 cache memory bandwidth on the Haswell cores has been doubled up, and one of the reasons why is that the second generation of Advanced Vector Extensions (AVX) integer and floating point math units have a lot more performance than the AVX1 units etched into the prior Sandy Bridge and Ivy Bridge cores. The AVX1 unit had eight 256-bit registers for floating point (four for AVX add and four for AVX multiply) and could double peak floating point operations per second (flops) in these two chips compared to their many Xeon predecessors that have had 128-bit SSE math units to 8 flops per clock. The Haswell core has 256-bit registers with its AVX2 unit, based on two Fused Multiply Add units, which doubles the peak performance to 16 flops per clock at double precision and 32 flops per clock at single precision.
Kmiecik explained to HPCwire that the new FMA instructions in the AVX2 unit would potentially increase performance for structural analysis, computational fluid dynamics, and electromagnetic field and cosmology simulations. The AVX2 feature also supports full 256-bit wide integer calculations, rather than the 128-bit width for the prior AVX1, which will be useful to accelerate image and signal processing, genomics, and cryptographic workloads.
The trick is adding up the effects of the increased core counts, better single-threaded performance, memory bandwidth, and AVX2 math units as a whole running real workloads. And here is what the initial test results look like on a variety of HPC applications:
The tests above compare a two-socket server with the 18-core Haswell Xeon E5-2699 v3, which runs at 2.3 GHz, against a machine using the 12-core Xeon E5-2697 v2 processor, which runs at 2.7 GHz.
In case you are wondering, on the Linpack Fortran benchmark test using the AVX2 features, this Haswell machine has twice the peak floating point performance as the Ivy Bridge machine; the exact floating point numbers have not been divulged by Intel as yet.