Intel Etches HPC Niche with Xeon E7 V2

By Nicole Hemsoth

February 18, 2014

Intel has finally pulled the lid off their Westmere-followup processors with the new Xeon E7 v2 series, which while generally targeting enterprise big data and analytics workloads, offers far more for HPC than might meet the eye.

Despite the range of targeted application areas, the 22 nanometer process, memory capabilities, robust AVX instruction capabilities, and unique reliability features make this well suited for HPC. Intel has managed to slice a sizable piece of performance to create a 120-core “HPC cluster in a box” that while higher in price than the previous Ivy Bridges that rolled out last September, could be a suitable fit for a wide swath of technical computing workloads.

As Intel’s Joe Curley told us today, “The E7s offer a ton of density in a node and this is indeed a great HPC story–as well as a great big data and enterprise story. Any type of application where you need a lot of memory and reliability and you’re doing a long set of calculations, that’s where the E7 set of products is going to do really well for you.” You end up with 120 cores in a node, 12 terabytes of memory maximum, you can fit a lot of density and compute there and there are HPC and many other workloads that this will be ideal for.”

Before we dive into the specifics, it’s worth acknowledging that one of the most fascinating aspects of this more general E7 v2 announcement is how Intel was able to balance out a multitude of goals and targets for a wide range of potential workloads via a segmentation model. It would have been perfectly feasible for this same news to span across several versions of the same essential chip sporting variations in cache or clock speeds—but these are actually well-optimized for the workloads being targeted and can be broken down along multiple lines according to priorities (power draw, memory and cache requirements, clock speed, etc.). In other words, the marketing for the E7 v2 is that there’s almost guaranteed to be something for everyone, in all price ranges and across nearly all workloads.

For context, take a look below at the two comparison charts that compare the Westmere predecessor (the E7-4800) with against SPEC (at the top) and then according to LINPACK performance just underneath.

 slide70And for LINPACK

slide71

The AVX capabilities are another noteworthy feature in these new processors—a point to consider with the LINPACK slide above. The 256-bit AVX vector units in the Ivy Bridge core can do eight double-precision floating point calculations per clock cycle via two AVX units per core (4 per AVX unit). This is good news for HPC on this front since the Westmere didn’t have any AVX at all.

The thing to notice here is that if you look below at the figures with 15 cores operating on this level, you’re looking at 1.246 teraflops—which is right on par with the Xeon Phi. Theoretically, this means that for those who aren’t concerned with the price (and the difference is quite hefty) it’s possible to circumvent the Xeon Phi and the programming talent required to get code up to snuff entirely and stock a cluster with these parts instead. Whether or not shops will consider this is up in the air—especially since this comparison won’t be valid by next year if rumors are correct and a three-teraflop Phi chip actually emerges.

According to Curley, “The products complement each other. Each Xeon Phi coprocessor has greater peak FLOPS and memory bandwidth per device, but lower memory capacity. The Xeon E7 Family provides very large memory capacity per processor, and allows up to 8 processors per node.”

The products address different application spaces for the most part, but as part of a portfolio produce an outstanding economic return for application developers, said Curley. “The Intel Xeon E7 Family is optimized for computing with memory density per node, superior results when compared to RISC and mainframe alternatives, and server consolidation markets. Xeon Phi addresses highly parallel applications that can take maximum benefit of Xeon Phi’s performance per watt.”

But the key advantage between them remains that applications optimized for Xeon Phi use the same core, thread, and SIMD elements – meaning applications optimized for one can be moved to the other – a critical advantage for developers.

Take a look at the chart below, which shows the 20 ways you can slice the E7 v2 for 2, 4 and 8 socket variants across different workload types via the basic, standard and advanced configurations, all of which offer some notable tradeoffs (or none at all for the big spenders) in core counts, cache and clock speeds to fit a variety of both memory, compute and energy demands. The segment optimized for HPC (bottom right with my red arrow) the E7 8857 v2, is available in an 8-socket version only with 12 cores running at 3.0 GHz in 30 megabytes of cache in 130 watt envelope for $3838 (versus the other 8-socket optimized variants that offer 3.2 ghz clockspeed, full cache and of course, more cores. The fact that specific optimized part is only available in 8-socket is definitely noteworthy; Intel has always offered 2 and 4-socket versions of previous offerings.

FullSlide66

Intel’s Curley told us that while there is a segment optimized offering in 8-node configuration, this by no means limits the options for HPC shops and OEMs. “If you wanted to try to build big a lot of memory capacity or bandwidth on a two-socket node you could use smaller E7 line products. OEMs will configure this in a number of ways. There are different feature sets across the line; the line is built for the greatest memory density per core, it has a lot of bandwidth per core, it’s built for even greater levels of resiliency to support mission-critical applications in ways that previous generations didn’t address to this degree.” In other words, as with everything in HPC–it all depends on where your priorities lie.

In this E7 world of multiple possibilities, there are a number of configurations that serve different needs, as highlighted below. As Intel demonstrates via the 120-core cluster in a box-themed slide below, if you have four 2-socket servers of the previous generation, you can do the same amount of work by jamming it into two four-socket servers, but run at around 2.5x faster (thanks to the QuickPath Interconnect, which is way faster than Ethernet).

Slide74The beauty of the model is that for HPC workloads in particular (keep in mind that the sweet spot for average job sizes in HPC hovers around the 128 core mark) it’s possible to speed up the work using the same number of processors and farm it out to larger, more memory-stuffed nodes that are designed to run faster.

To put this in some real-world performance context, consider the (somewhat wonky) chart below, which uses the baseline of the Ivy Bridge they rolled out last September. Intel wasn’t crossing out their own numbers there; rather, the crossed-out number represents the average performance boost for a broad range of similar workloads—in essence, an average figure across multiple applications in a domain. The higher number with the application name is a top performer, hitting the app high-water mark they reference on the top left.

In other words, using the life sciences example, while most life science applications they benchmarked came out at around 2x the performance, MILC hit 3x. Similarly, when it came to financial services, the average performance increase across a range of common applications was 2.4 while Black Scholes in particular topped out at 2.6x. Even without data on how many applications were benchmarked to come up with the average, you get the idea…

Slide73On another note, for those who are interested in bandwidth, Intel has had single device data correct, or SDDC, in the memory controllers used with its server processors since the 1990s. What it does, in essence anyway, is allow for a faulty DRAM memory chip to be taken out of the memory pool and isolated. You can heal around one bad chip and correct errors. With the Westmere E7s, given the large memory in the system, Intel added a new feature called Double Device Data Correct.

With this feature, Intel could correct errors after two chips failed on a memory module. With the new performance mode, Intel is allowing customers who want to speed up the memory buffer chips and therefore get more memory bandwidth out of the system to step back to single chip recovery mode. In the normal DDDC mode, which is called lockstep mode now, both the SMI2 memory buffers and the memory chips run at the same speed: 1600 MT/sec. With the new performance mode, the SMI2 memory buffers run at 2667 MT/sec, and the memory runs at half that speed. The net effect is that in performance mode, the system bandwidth is increased by 54 percent compared to lockstep mode. Some error correction capability is sacrificed for the sake of bandwidth.

slide40

All of this theoretical at this point, mind you–we can write the real story when some early users lend us their experiences and insights. As it stands, this story just broke and again, the real emphasis is on databases (the SAP HANA tale is certainly compelling–follow that at EnterpriseTech this week).

Subscribe to HPCwire's Weekly Update!

Be the most informed person in the room! Stay ahead of the tech trends with industy updates delivered to you every week!

China’s Expanding Effort to Win in Microchips

July 27, 2017

The global battle for preeminence, or at least national independence, in semiconductor technology and manufacturing continues to heat up with Europe, China, Japan, and the U.S. all vying for sway. A fascinating article ( Read more…

By John Russell

Hyperion: Storage to Lead HPC Growth in 2016-2021

July 27, 2017

Global HPC external storage revenues will grow 7.8% over the 2016-2021 timeframe according to an updated forecast released by Hyperion Research this week. HPC server sales, by comparison, will grow a modest 5.8% to $14.8 Read more…

By John Russell

Exascale FY18 Budget – The Senate Provides Their Input

July 27, 2017

In the federal budgeting world, “regular order” is a meaningful term that is fondly remembered by members of both the Congress and the Executive Branch. Regular order is the established process whereby an Administrat Read more…

By Alex R. Larzelere

HPE Extreme Performance Solutions

HPE Servers Deliver High Performance Remote Visualization

Whether generating seismic simulations, locating new productive oil reservoirs, or constructing complex models of the earth’s subsurface, energy, oil, and gas (EO&G) is a highly data-driven industry. Read more…

India Plots Three-Phase Indigenous Supercomputing Strategy

July 26, 2017

Additional details on India's plans to stand up an indigenous supercomputer came to light earlier this week. As reported in the Indian press, the Rs 4,500-crore (~$675 million) supercomputing project, approved by the Ind Read more…

By Tiffany Trader

Exascale FY18 Budget – The Senate Provides Their Input

July 27, 2017

In the federal budgeting world, “regular order” is a meaningful term that is fondly remembered by members of both the Congress and the Executive Branch. Reg Read more…

By Alex R. Larzelere

India Plots Three-Phase Indigenous Supercomputing Strategy

July 26, 2017

Additional details on India's plans to stand up an indigenous supercomputer came to light earlier this week. As reported in the Indian press, the Rs 4,500-crore Read more…

By Tiffany Trader

Tuning InfiniBand Interconnects Using Congestion Control

July 26, 2017

InfiniBand is among the most common and well-known cluster interconnect technologies. However, the complexities of an InfiniBand (IB) network can frustrate the Read more…

By Adam Dorsey

NSF Project Sets Up First Machine Learning Cyberinfrastructure – CHASE-CI

July 25, 2017

Earlier this month, the National Science Foundation issued a $1 million grant to Larry Smarr, director of Calit2, and a group of his colleagues to create a comm Read more…

By John Russell

Graphcore Readies Launch of 16nm Colossus-IPU Chip

July 20, 2017

A second $30 million funding round for U.K. AI chip developer Graphcore sets up the company to go to market with its “intelligent processing unit” (IPU) in Read more…

By Tiffany Trader

Fujitsu Continues HPC, AI Push

July 19, 2017

Summer is well under way, but the so-called summertime slowdown, linked with hot temperatures and longer vacations, does not seem to have impacted Fujitsu's out Read more…

By Tiffany Trader

Researchers Use DNA to Store and Retrieve Digital Movie

July 18, 2017

From abacus to pencil and paper to semiconductor chips, the technology of computing has always been an ever-changing target. The human brain is probably the com Read more…

By John Russell

The Exascale FY18 Budget – The Next Step

July 17, 2017

On July 12, 2017, the U.S. federal budget for its Exascale Computing Initiative (ECI) took its next step forward. On that day, the full Appropriations Committee Read more…

By Alex R. Larzelere

Google Pulls Back the Covers on Its First Machine Learning Chip

April 6, 2017

This week Google released a report detailing the design and performance characteristics of the Tensor Processing Unit (TPU), its custom ASIC for the inference Read more…

By Tiffany Trader

Nvidia Responds to Google TPU Benchmarking

April 10, 2017

Nvidia highlights strengths of its newest GPU silicon in response to Google's report on the performance and energy advantages of its custom tensor processor. Read more…

By Tiffany Trader

Quantum Bits: D-Wave and VW; Google Quantum Lab; IBM Expands Access

March 21, 2017

For a technology that’s usually characterized as far off and in a distant galaxy, quantum computing has been steadily picking up steam. Just how close real-wo Read more…

By John Russell

HPC Compiler Company PathScale Seeks Life Raft

March 23, 2017

HPCwire has learned that HPC compiler company PathScale has fallen on difficult times and is asking the community for help or actively seeking a buyer for its a Read more…

By Tiffany Trader

Trump Budget Targets NIH, DOE, and EPA; No Mention of NSF

March 16, 2017

President Trump’s proposed U.S. fiscal 2018 budget issued today sharply cuts science spending while bolstering military spending as he promised during the cam Read more…

By John Russell

CPU-based Visualization Positions for Exascale Supercomputing

March 16, 2017

In this contributed perspective piece, Intel’s Jim Jeffers makes the case that CPU-based visualization is now widely adopted and as such is no longer a contrarian view, but is rather an exascale requirement. Read more…

By Jim Jeffers, Principal Engineer and Engineering Leader, Intel

Nvidia’s Mammoth Volta GPU Aims High for AI, HPC

May 10, 2017

At Nvidia's GPU Technology Conference (GTC17) in San Jose, Calif., this morning, CEO Jensen Huang announced the company's much-anticipated Volta architecture a Read more…

By Tiffany Trader

Facebook Open Sources Caffe2; Nvidia, Intel Rush to Optimize

April 18, 2017

From its F8 developer conference in San Jose, Calif., today, Facebook announced Caffe2, a new open-source, cross-platform framework for deep learning. Caffe2 is the successor to Caffe, the deep learning framework developed by Berkeley AI Research and community contributors. Read more…

By Tiffany Trader

Leading Solution Providers

How ‘Knights Mill’ Gets Its Deep Learning Flops

June 22, 2017

Intel, the subject of much speculation regarding the delayed, rewritten or potentially canceled “Aurora” contract (the Argonne Lab part of the CORAL “ Read more…

By Tiffany Trader

Reinders: “AVX-512 May Be a Hidden Gem” in Intel Xeon Scalable Processors

June 29, 2017

Imagine if we could use vector processing on something other than just floating point problems.  Today, GPUs and CPUs work tirelessly to accelerate algorithms Read more…

By James Reinders

Russian Researchers Claim First Quantum-Safe Blockchain

May 25, 2017

The Russian Quantum Center today announced it has overcome the threat of quantum cryptography by creating the first quantum-safe blockchain, securing cryptocurrencies like Bitcoin, along with classified government communications and other sensitive digital transfers. Read more…

By Doug Black

MIT Mathematician Spins Up 220,000-Core Google Compute Cluster

April 21, 2017

On Thursday, Google announced that MIT math professor and computational number theorist Andrew V. Sutherland had set a record for the largest Google Compute Engine (GCE) job. Sutherland ran the massive mathematics workload on 220,000 GCE cores using preemptible virtual machine instances. Read more…

By Tiffany Trader

Google Debuts TPU v2 and will Add to Google Cloud

May 25, 2017

Not long after stirring attention in the deep learning/AI community by revealing the details of its Tensor Processing Unit (TPU), Google last week announced the Read more…

By John Russell

Groq This: New AI Chips to Give GPUs a Run for Deep Learning Money

April 24, 2017

CPUs and GPUs, move over. Thanks to recent revelations surrounding Google’s new Tensor Processing Unit (TPU), the computing world appears to be on the cusp of Read more…

By Alex Woodie

Six Exascale PathForward Vendors Selected; DoE Providing $258M

June 15, 2017

The much-anticipated PathForward awards for hardware R&D in support of the Exascale Computing Project were announced today with six vendors selected – AMD Read more…

By John Russell

Top500 Results: Latest List Trends and What’s in Store

June 19, 2017

Greetings from Frankfurt and the 2017 International Supercomputing Conference where the latest Top500 list has just been revealed. Although there were no major Read more…

By Tiffany Trader

  • arrow
  • Click Here for More Headlines
  • arrow
Share This