The Habana Labs team is excited to join The Linley Group at its
upcoming fall conference, October 23 and 24 in Santa Clara. There we will show
our GOYA Inference and GAUDI Training Processors in operation—with both demonstrating
industry-leading, benchmarked performance.
In addition, Eitan Medina will provide a talk at 2pm on October 23, “Scaling AI Training Systems with the GAUDI Processor,” in which he will present GAUDI’s one-of-a-kind architecture and features, purpose-designed for scaling of AI Training systems.
If you’re a member of the media and wish to conduct a one-on-one briefing with Eitan on the company and our products, please feel free to contact us at PR@habana.ai to get on his schedule.
This week marked Habana’s second year of showing working hardware and demonstrating leading AI performance at the AI Hardware Summit. At last year’s summit, Habana stepped out of stealth to introduce the company and our Goya Inference Processor. This year at the Summit marked more progress for the company and its products—inference AND training.
This week Goya demonstrated that it continues to lead the inference processor market with performance on ResNet-50 for image classification, now with 15,453 images-per-second and latency of 1.0ms vs the T4, which is reported to have roughly one-third the inference throughput speed of Goya—with throughput of 5,013 and 26ms latency. Goya also demonstrated its BERT language processing performance; on BERT, Goya demonstrated 1,527 sentences-per-second and 16.3 ms latency vs. T4’s 738 sentences-per-second and 32.4ms—both for 24-Batch. In addition, Habana demonstrated Goya’s SSD (Single Shot MultiBox Detector) functionality live at the summit.
Also featured at this year’s summit was the Gaudi processor, introduced in June, demonstrating its ResNet-50 image classification training performance of 1,650 images-per-second on the industry’s only training processor to feature on-chip integrated RoCE RDMA. Habana’s innovative use of Ethernet enables unprecedented scaling of AI training systems—the topic of Habana’s “sold out” invitation-only workshop at the summit:A Standards-Based Approach to Scaling AI Datacenter Capacity and Performance.
As a strong proponent of standardization of AI datacenter hardware, Habana also participated in one of the summit’s key industry panels, “Developments and Impacts of Emerging Industry Standards on the AI Datacenter,”alongside industry leaders from Facebook, Google, Alibaba and Uber. The consensus of the panelist: for the AI industry to address datacenter customer needs for greater performance and increased flexibility, industry standards such as Ethernet and the OCP-OAM for hardware will be useful.
The Habana team was pleased to contribute
to the Summit with our live hardware demonstrations, presentation and panel
participation, and we look forward to many successful collaborations with
customers and partners we engaged at the conference.
The Habana team was thrilled with the turn-out of so many AI business leaders and academic gurus at this year’s Hot Chips. Thank you to those of you who attended and provided comment on Eitan Medina’s talk on Habana’s approach to scaling AI Training systems and update on our Goya Inference processor’s leading performance on BERT. For more info on that achievement, please see our Goya info here.
Dr. Ian Cutress of AnandTech live blogged the talk, which you can check out here. Thanks, Ian, for such a comprehensive blog.
Thank you, too, to all of you who came by our booth to see Habana’s live demos of Goya on BERT and Gaudi on ResNet, as well as our HLS-1 Training System featuring Habana’s HL-205 Mezzanine Card.
Thank you, David Schor for the photos you posted at the conference.
And finally, a big thank you, too, to the awesome volunteer team that made the Hot Chips 31 the super professional and smooth running, high-value conference that it was. Your customer and attendee focus really paid off in an extraordinarily pleasant 3 days of useful information, training and networking.
So, what’s next for Habana? We look forward to the upcoming AI Hardware Summit on September 17 & 18 where we’ll celebrate our one-year anniversary of coming out of stealth mode to introduce our company and Goya. Please come see Habana’s talk and product demonstrations. Here’s more info on the conference.
Habana features GAUDI Training and GOYA Inference Processors at this year’s conference
At Habana, we have fond memories of AI Hardware Summit 2018—it was where our company came out of stealth to surprise attendees with not only the emergence of our company, but also the launch of our GOYA AI Inference Processor with a live demonstration of industry-leading inference performance.
It’s been a landmark year for Habana. GOYA has been shipping since Q4 last year and is the first back end for Facebook’s GLOW compiler and runtime project. We also fulfilled our promise—to launch the GAUDI AI Training processor during Q2 of 2019 and are pleased GAUDI has delivered two industry firsts: record-breaking performance in small batches as well as at large scale, and the industry’s only on-chip integration of RDMA over Converged Ethernet (RoCEv2), an innovation we expect to be transformative for AI systems scale-out and scale-up.
So, please come to check out our live demos of both GOYA and GAUDI in our booth. And, join us for the deep-dive workshop and panel that we’re sponsoring.
If you’re not yet registered for the Summit, be sure to use the Habana10 code to get a discount of 10% off the price. See details.
DAY 1 LUNCHTIME WORKSHOP: 1:45pm A Standards-Based Approach to Scaling AI Datacenter Capacity and Performance; a by invitation session on “how to” efficiently scale AI datacenter capacity and performance. Please let the event organizer know if you’re interested in attending this workshop. Presenter: Eitan Medina, Habana Labs Chief Business Officer
DAY 1 PANEL: 5:10pm Developments and Impacts of Emerging Industry Standards on the AI Datacenter; immediately precedes the Day 1 closing keynote, followed by cocktail reception. Panelists: Samar Dalal, Senior Manager Architecture & Design Uber; Cliff Young, Software Engineer Google Brain; Lingjie Xu, Director Applied AI Architecture Alibaba; Whitney Zhao, AI Hardware System Architect Facebook; Eitan Medina, Chief Business Officer, Habana Labs
Join Habana Labs on Monday, August 19 at Hot Chips 2019: A Symposium on High Performance Chips.
The Hot Chips conference, which is sponsored by the IEEE Technical Committee on Microprocessors, focuses on the latest advances in semiconductor chip design and implementation.
On Monday, August 19 at 6:15 at the conference, Eitan Medina, Habana’s CBO, will share the company’s recently announced, innovative approach to scaling AI Training systems with its GAUDI AI Processor. Eitan will show how AI Training in datacenter and cloud operations can reap previously unavailable advantages with GAUDI’s integrated on-chip RoCE RDMA, leveraging industry-standard Ethernet networking. These advantages include…
Increased sourcing options with standards-based technologies
Avoidance of proprietary lock-in that limits growth and customization
…all of which add up to increased productivity and
efficiency for AI datacenter customers.
Eitan will show how GAUDI eases implementation of AI systems
from a small-scale server to medium and large-scale systems. And, brings a new level
of performance to data parallel and model parallel computation.
When: June 18 – 21 Where: Beijing International Hotel Convention Center, Booth #206 What: Unmatched AI Performance and Versatility with GAUDI and GOYA
Habana Labs was pleased to demonstrate its newly debuted GAUDI AI Training processor and GOYA AI Inference processor, shipping since 2018, at the O’Reilly AI Conference in Beijing. Both processors, built from the ground up to address the specific demands of AI datacenter and cloud operations, offer record-breaking performance.
We introduced attendees to the advantages of Integrated On-chip RoCE
In addition to our focus on performance, Habana introduced conference attendees to GAUDI’s industry-only integration of on-chip Ethernet (RDMA over Converged Ethernet). Customers learned how integrated Ethernet will bring a radically new level of productivity and flexibility to scale up and scale out AI systems of all sizes—to address the networking needs of both data parallel and model parallel systems. GAUDI supports virtually unlimited scalability in speeds and port-count with Ethernet technology that’s already deployed and familiar in data centers and cloud operations.
Habana Keynote Increasing AI productivity and efficiency with purpose-built AI Processors, delivered by Eitan Medina. Watch this keynote.
Habana Deep-dive Session A Fresh Approach to Building High-Performance AI Systems, conducted by Eitan Medina. Watch the video.