Mellanox Spins EZchip/Tilera IP Into BlueField Networking Silicon

By Tiffany Trader

June 1, 2016

Mellanox today introduced the BlueField family of programmable processors, the first product technology based on its $811 million acquisition of fellow Iraeli high-tech company EZchip. The fact that the product announcement is taking place just three months after the completion of the purchase speaks to the strong synergies between EZchip and Mellanox, said Bob Doud, senior director of marketing at Mellanox.

Doud came to Mellanox via EZchip and before that Tilera, which EZchip purchased for $130 million in July 2014. Mellanox is counting on the EZchip/Tilera intellectual property – the energy-efficient Tile multicore ARM CPU and the SkyMesh coherent on-chip interconnect scheme that enables scaling to more than one hundred cores on a single chip – to broaden its product portfolio and open up new markets.

The BlueField multicore SoC staples together Mellanox’s latest-generation ConnectX network acceleration technology with an array of Tile-based ARM A72 cores. Mellanox is targeting the fully-programmable Ethernet networking silicon at storage and networking applications.

“The multicore processor piece that originated with Tilera brings in very advanced IP to Mellanox,” said Doud. “Arguably Tilera was seen in the industry as the leader in high scale-out multicore processing. We had a 72-core processor in 2011, ahead of Cavium and certainly Intel. We really were the technology and thought leader. Revenue wise we were still in start-up mode, but when we talked to customers, they were eager to try the technology.

“Mellanox has over 100 patents in how to build this mesh of cores,” he continued. “We can scale up well above 100 cores on a chip, and we’ve figured out how to do cache coherency. These are big problems that folks like Intel are still working on.”

EZchip had been developing a 100-core ARM chip, but when the agreement with Mellanox was announced in September 2015, the development team was retasked to begin integrating the mesh architecture into the Mellanox portfolio. While Tilera and EZchip were focused more on big iron, Mellanox has pivoted the IP, scaling it down to play nicely with its adapters and give it a storage/networking play for the cloud, datacenter and security space. Mellanox didn’t specify the exact number of ARM cores that would be employed on the SoC, but a graphic supplied by the company depicts 16 cores in a 4×4 array. The associated text denotes the possibility of arrays with a smaller or larger number of cores.

Mellanox BlueField graphic

As Mellanox enters into the embedded processor space, one important target market is flash-based storage and NVMe. “We saw a need for an ideal storage controller chip,” said Doud. “This play in storage is really a silicon play. It’s selling silicon to the large storage OEMs – the EMCs, Hitachis and IBMs of the world – who are building storage appliances with flash and who need a storage controller that has all the right attributes and those attributes include really high-performance networking because the new storage architecture is scale-out. It’s a bunch of 1U, 2U appliances connected via fabric, 100GE or InfiniBand. That kind of thins out the market as to who can make such a controller. You need at least 20Gig just to service a single NVMe drive.”

Doud goes on to say that what Mellanox is doing with BlueField is designing a device that has a lot of PCI bandwidth on the south to talk to an array of drives (PCIe 3 and 4 are supported) and up to 100G fabric bandwidth on the north to talk to the fabric. Hardware acceleration is there to handle tasks like the NVMe over fabric protocol translation. ConnectX handles storage offloads like erasure coding for RAID.

“The addition of an array of ARM cores provides a highly programmable storage controller that can run not only the control plane functions to manage the array and even touch the data path, it can manipulate the data doing deduplication, data compression/decompression as well as disk sector encryption/decryption,” added Doud.

On the networking side, Mellanox is looking to sell adapters that integrate BlueField for applications in datacenters, cloud and security. The company is currently envisioning two models for this market. One, which Doud refers to as the sandbox, begins life as BlueField with just an operating system running on ARM cores and some essential plumbing enabling a hyperscaler like Google or Facebook to develop the software that it needs to run on those ARM cores. The datacenter operator develops the application code, and ConnectX and the hardware on the device helps to channel the flows to the right place and get them in and out of the cores. That second model is for Mellanox to develop the code for BlueField and sell it as a pure turnkey play much like they would ConnectX. For this angle, Mellanox would likely work with software partners to offer versions aimed at IPsec, SSL or intrusion protection — all turn-key and ready to go.

Doud emphasized the security features of BlueField, which are prompting government interest. “There’s a certain degree of assuredness when you can lock it down and say this card’s job is to do network acceleration and security and it’s a contained environment with known code,” he said. “The new DOD model is to scale out with virtualized servers and it worries them to think what kind of exposure they may be getting that they can’t necessarily control because they don’t control VMware or KVM or other virtualized environments. They hope the community has done a good job of isolation.”

“Synergy-wise, what Mellanox sees is that compute is no longer relegated to just being in a server,” said Doud. “We think that the new model, the new hyperscale datacenter, is that things are blurring. You need compute in the storage appliance. It needs compute because you can add a lot of value functions right there close to the storage drive, things like deduplication and even other value add, e.g., some big data functions. The network itself is becoming more intelligent. Ethernet switches were really dumb 10 years ago; now they do L2 and L3 and they are starting to do other interesting functions. More and more we’re seeing they’re not just relegated to different encapsulation or decapsulation. Mellanox InfiniBand switches actually can do atomic compute operations on InfiniBand packets as they move through their switches. And in the adapter space, Mellanox has been preaching for years the need and the value of offloading certain functions from Intel and having a programmable array of multicore right there on the adapter takes that to the next level.

“On the NPU side, the network processors, you can see the synergy with our switching story. We have Spectrum, a pretty advanced L2/L3 switch, market consistent functionality going up against Broadcom and their line of switches, but what we now can bring to the table with this technology for the network processor is we can do 400G to 600G and our next-gen will be a terabit of network processor performance. That’s fully programmable in C code. You’re bringing the ability to not just mess with some headers and send the packet out a different port, but you can do security operations on packets as they move through or collective operations on the payload of the packet. So programmability is really the new differentiating capability. Some of this is capability is seen in the Dune products [from Broadcom], but what we do in our NPUs is vastly more sophisticated. We are not announcing this today, but it is certainly a vision.”

Mellanox is building BlueField on a 28nm process, which Linley Group analyst Tom Halfhill, who was briefed ahead of the announcement, characterized as a necessary compromise, one they couldn’t really avoid given the circumstances. “Putting production at late 2017 or early 2018, the 28nm technology will be getting dated. The reason for sticking with 28nm was that EZchip was already pretty far along in the project when the Mellanox acquisition happened. They didn’t want to do a complete switch where they would have to transition to a smaller process technology and change the whole design of the processor at the same time. Even Intel follows this Tick-Tock cycle, alternating process shrinks and process design to keep the design complexity down and stay on schedule.”

As for the potential low-power server play that once grabbed the imagination of HPCers, Doud explained that the cores that Tilera was building never really had the floating point performance required by HPC applications. “They were integer cores,” he said. “They were much more attuned to doing packet processing than they would be doing weather modeling. That was really a market choice. We chose not to go after HPC.”

He added that Tilera was, however, used in some HPC applications, but at the front end to do some of the more network-heavy stuff like sending traffic to the right nodes of a cluster. “It was more of a datacenter kind of use case,” he added. “But things are blurring now. You used to be able to put nice boxes around what’s HPC.”

While it might be fun to speculate whether Mellanox would ever direct the Tilera IP back at the server market, Doud was very clear that this is not part of Mellanox’s near-term plan. “The decisions that went into building BlueField were really to make this a combination of a great storage processor and a great networking front-end to Intel. It was not to displace Intel in a server. Companies like Cavium and Qualcomm — those guys are going after Intel. We’re taking a pass on that for now – in part because you have to pick your battles – and Mellanox is very much a networking-centric company, but secondly it’s an unproven proposition. There’s been a lot of talk about ARM killing Intel for a number of years and I think people are now beginning to wonder if that’s really going to happen or if it was a play to get Intel’s pricing back in line.”

Mellanox says it will begin sampling BlueField silicon in Q1 of 2017, with the goal of hitting production by mid-year 2017 and ramping revenue in 2018. Additional specifics about the first chip, a more precise timeline and pricing information should be ready by this winter, said the company.

Subscribe to HPCwire's Weekly Update!

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

Slide Shows Samsung May be Developing a RISC-V CPU for In-memory AI Chip

June 19, 2024

Samsung may have unintentionally revealed its intent to develop a RISC-V CPU, which a presentation slide showed may be used in an AI chip. The company plans to release an AI accelerator with heavy in-memory processing Read more…

ASC24 Student Cluster Competition: Who Won and Why?

June 18, 2024

As is our tradition, we’re going to take a detailed look back at the recently concluded the ASC24 Student Cluster Competition (Asia Supercomputer Community) to see not only who won the various awards, but to figure out Read more…

Qubits 2024: D-Wave’s Steady March to Quantum Success

June 18, 2024

In his opening keynote at D-Wave’s annual Qubits 2024 user meeting, being held in Boston, yesterday and today, CEO Alan Baratz again made the compelling pitch that D-Wave’s brand of analog quantum computing (quantum Read more…

Apple Using Google Cloud Infrastructure to Train and Serve AI

June 18, 2024

Apple has built a new AI infrastructure to deliver AI features introduced in its devices and is utilizing resources available in Google's cloud infrastructure.  Apple's new AI backend includes: A homegrown foun Read more…

Argonne’s Rick Stevens on Energy, AI, and a New Kind of Science

June 17, 2024

The world is currently experiencing two of the largest societal upheavals since the beginning of the Industrial Revolution. One is the rapid improvement and implementation of artificial intelligence (AI) tools, while the Read more…

Under The Wire: Nearly HPC News (June 13, 2024)

June 13, 2024

As managing editor of the major global HPC news source, the term "news fire hose" is often mentioned. The analogy is quite correct. In any given week, there are many interesting stories, and only a few ever become headli Read more…

Slide Shows Samsung May be Developing a RISC-V CPU for In-memory AI Chip

June 19, 2024

Samsung may have unintentionally revealed its intent to develop a RISC-V CPU, which a presentation slide showed may be used in an AI chip. The company plans Read more…

Qubits 2024: D-Wave’s Steady March to Quantum Success

June 18, 2024

In his opening keynote at D-Wave’s annual Qubits 2024 user meeting, being held in Boston, yesterday and today, CEO Alan Baratz again made the compelling pitch Read more…

Shutterstock_666139696

Argonne’s Rick Stevens on Energy, AI, and a New Kind of Science

June 17, 2024

The world is currently experiencing two of the largest societal upheavals since the beginning of the Industrial Revolution. One is the rapid improvement and imp Read more…

Under The Wire: Nearly HPC News (June 13, 2024)

June 13, 2024

As managing editor of the major global HPC news source, the term "news fire hose" is often mentioned. The analogy is quite correct. In any given week, there are Read more…

Labs Keep Supercomputers Alive for Ten Years as Vendors Pull Support Early

June 12, 2024

Laboratories are running supercomputers for much longer, beyond the typical lifespan, as vendors prematurely deprecate the hardware and stop providing support. Read more…

MLPerf Training 4.0 – Nvidia Still King; Power and LLM Fine Tuning Added

June 12, 2024

There are really two stories packaged in the most recent MLPerf  Training 4.0 results, released today. The first, of course, is the results. Nvidia (currently Read more…

Highlights from GlobusWorld 2024: The Conference for Reimagining Research IT

June 11, 2024

The Globus user conference, now in its 22nd year, brought together over 180 researchers, system administrators, developers, and IT leaders from 55 top research Read more…

Nvidia Shipped 3.76 Million Data-center GPUs in 2023, According to Study

June 10, 2024

Nvidia had an explosive 2023 in data-center GPU shipments, which totaled roughly 3.76 million units, according to a study conducted by semiconductor analyst fir Read more…

Atos Outlines Plans to Get Acquired, and a Path Forward

May 21, 2024

Atos – via its subsidiary Eviden – is the second major supercomputer maker outside of HPE, while others have largely dropped out. The lack of integrators and Atos' financial turmoil have the HPC market worried. If Atos goes under, HPE will be the only major option for building large-scale systems. Read more…

Comparing NVIDIA A100 and NVIDIA L40S: Which GPU is Ideal for AI and Graphics-Intensive Workloads?

October 30, 2023

With long lead times for the NVIDIA H100 and A100 GPUs, many organizations are looking at the new NVIDIA L40S GPU, which it’s a new GPU optimized for AI and g Read more…

Nvidia H100: Are 550,000 GPUs Enough for This Year?

August 17, 2023

The GPU Squeeze continues to place a premium on Nvidia H100 GPUs. In a recent Financial Times article, Nvidia reports that it expects to ship 550,000 of its lat Read more…

Everyone Except Nvidia Forms Ultra Accelerator Link (UALink) Consortium

May 30, 2024

Consider the GPU. An island of SIMD greatness that makes light work of matrix math. Originally designed to rapidly paint dots on a computer monitor, it was then Read more…

Nvidia’s New Blackwell GPU Can Train AI Models with Trillions of Parameters

March 18, 2024

Nvidia's latest and fastest GPU, codenamed Blackwell, is here and will underpin the company's AI plans this year. The chip offers performance improvements from Read more…

Choosing the Right GPU for LLM Inference and Training

December 11, 2023

Accelerating the training and inference processes of deep learning models is crucial for unleashing their true potential and NVIDIA GPUs have emerged as a game- Read more…

Synopsys Eats Ansys: Does HPC Get Indigestion?

February 8, 2024

Recently, it was announced that Synopsys is buying HPC tool developer Ansys. Started in Pittsburgh, Pa., in 1970 as Swanson Analysis Systems, Inc. (SASI) by John Swanson (and eventually renamed), Ansys serves the CAE (Computer Aided Engineering)/multiphysics engineering simulation market. Read more…

Some Reasons Why Aurora Didn’t Take First Place in the Top500 List

May 15, 2024

The makers of the Aurora supercomputer, which is housed at the Argonne National Laboratory, gave some reasons why the system didn't make the top spot on the Top Read more…

Leading Solution Providers

Contributors

AMD MI3000A

How AMD May Get Across the CUDA Moat

October 5, 2023

When discussing GenAI, the term "GPU" almost always enters the conversation and the topic often moves toward performance and access. Interestingly, the word "GPU" is assumed to mean "Nvidia" products. (As an aside, the popular Nvidia hardware used in GenAI are not technically... Read more…

The NASA Black Hole Plunge

May 7, 2024

We have all thought about it. No one has done it, but now, thanks to HPC, we see what it looks like. Hold on to your feet because NASA has released videos of wh Read more…

Google Announces Sixth-generation AI Chip, a TPU Called Trillium

May 17, 2024

On Tuesday May 14th, Google announced its sixth-generation TPU (tensor processing unit) called Trillium.  The chip, essentially a TPU v6, is the company's l Read more…

Intel’s Next-gen Falcon Shores Coming Out in Late 2025 

April 30, 2024

It's a long wait for customers hanging on for Intel's next-generation GPU, Falcon Shores, which will be released in late 2025.  "Then we have a rich, a very Read more…

Nvidia Shipped 3.76 Million Data-center GPUs in 2023, According to Study

June 10, 2024

Nvidia had an explosive 2023 in data-center GPU shipments, which totaled roughly 3.76 million units, according to a study conducted by semiconductor analyst fir Read more…

Q&A with Nvidia’s Chief of DGX Systems on the DGX-GB200 Rack-scale System

March 27, 2024

Pictures of Nvidia's new flagship mega-server, the DGX GB200, on the GTC show floor got favorable reactions on social media for the sheer amount of computing po Read more…

GenAI Having Major Impact on Data Culture, Survey Says

February 21, 2024

While 2023 was the year of GenAI, the adoption rates for GenAI did not match expectations. Most organizations are continuing to invest in GenAI but are yet to Read more…

A Big Memory Nvidia GH200 Next to Your Desk: Closer Than You Think

February 22, 2024

Students of the microprocessor may recall that the original 8086/8088 processors did not have floating point units. The motherboard often had an extra socket fo Read more…

  • arrow
  • Click Here for More Headlines
  • arrow
HPCwire