
Pinterest has engineered a way to serve its photo-sharing community more of the images they love.
The social-image service, with more than 400 million monthly active users, has trained bigger recommender models for improved accuracy at predicting people's interests.
Pinterest handles hundreds of millions of user requests an hour on any given day. And it must also narrow down relevant images from roughly 300 billion images on the site to roughly 50 for each person.
The last step - ranking the most relevant and engaging content for everyone using Pinterest - required a leap in acceleration to run heftier models, with minimal latency, for better predictions.
Pinterest has improved the accuracy of its recommender models powering people's home feeds and other areas, increasing engagement by as much as 16%.
The leap was enabled by switching from CPUs to NVIDIA GPUs, which could easily be applied next to other areas, including advertising images, according to Pinterest.
Normally we would be happy with a 2% increase, and 16% is just a beginning for home feeds. We see additional gains - it opens a lot of doors for opportunities, said Pong Eksombatchai, a software engineer at Pinterest.
Transformer models capable of better predictions are shaking up industries from retail to entertainment and advertising. But their leaps in performance gains of the past few years have come with a need to serve models that are some 100x bigger as their number of model parameters and computations skyrockets.
Huge Inference Gains, Same Infrastructure Cost Like many, Pinterest engineers wanted to tap into state-of-the-art recommender models to increase engagement. But serving these massive models on CPUs presented a 100x increase in cost and latency. That wasn't going to maintain its magical user experience - fresh and more appealing images - occurring within a fraction of a second.
If that latency happened, then obviously our users wouldn't like that very much because they would have to wait forever, said Eksombatchai. We are pretty close to the limit of what we can do on CPU basically.
The challenge was to serve these hundredfold larger recommender models within the same cost and latency constraints.
Working with NVIDIA, Pinterest engineers began architectural changes to optimize their inference pipeline and recommender models to enable the transition from CPU to GPU cloud instances. The technology transition began late last year and required major changes to how the company manages workloads. The result is a 100x gain in inference efficiency on the same IT budget, meeting their goals.
We are starting to use really, really big models now. And that is where the GPU comes in - to help make these models possible, Eksombatchai said.
Tapping Into cuCollections Switching from CPUs to GPUs required rethinking its inference systems architecture. Among other issues, engineers had to change how they send workloads to their inference servers. Fortunately, there are tools to assist in making the transition easier.
The Pinterest inference server built for CPUs had to be altered because it was set up to send smaller batch sizes to its servers. GPUs can handle much larger workloads, so it's necessary to set up larger batch requests to increase efficiency.
One area where this comes into play is with its embedding table lookup module. Embedding tables are used to track interactions between various context-specific features and interests of user profiles. They can track where you navigate, and what people Pin on Pinterest, share or numerous other actions, helping refine predictions on what users might like to click on next.
They are used to incrementally learn user preference based on context in order to make better content recommendations to those using Pinterest. Its embedding table lookup module required two computation steps repeated hundreds of times because of the number of features tracked.
Pinterest engineers greatly reduced this number of operations using a GPU-accelerated concurrent hash table from NVIDIA cuCollections. And they set up a custom consolidated embedding lookup module so they could merge requests into a single lookup. Better results were seen immediately.
Using cuCollections helped us to remove bottlenecks, said Eksombatchai.
Enlisting CUDA Graphs Pinterest relied on CUDA Graphs to eliminate what was remaining of the small batch operations, further optimizing its inference models.
CUDA Graphs helps reduce the CPU interactions when launching on GPUs. They're designed to enable workloads to be defined as graphs rather than single operations. They provide a mechanism to launch multiple GPU operations through a single CPU operation, reducing CPU overheads.
Pinterest enlisted CUDA Graphs to represent the model inference process as a static graph of operation instead of as those individually scheduled. This enabled the computation to be handled as a single unit without any kernel launching overhead.
The company now supports CUDA Graph as a new backend of its model server. When a model is first loaded, the model server runs the model inference once to build the graph instance. This graph can then be run repeatedly in inference to show content on its app or site.
Implementing CUDA Graphs helped Pinterest to significantly reduce inference latency of its recommender models, according to its engineers.
GPUs have enabled Pinterest to do something that was impossible with CPUs on the same budget, and by doing this they can make changes that have a direct impact on various business metrics.
Learn about Pinterest's GPU-driven inference and optimizations at its GTC session, Serving 100x Bigger Recommender Models, and in the Pinterest Engineering blog.
Register for GTC, running Sept. 19-22, for free to attend sessions with NVIDIA and dozens of industry leaders.
Most recent headlines
05/01/2027
Worlds first 802.15.4ab-UWB chip verified by Calterah and Rohde & Schwarz to be ...
06/09/2026
June 9 2026, 23:00 (PDT) Dolby and MagentaTV Bring Fans Closer to the FIFA Worl...
04/08/2026
Dalet, a leading technology and service provider for media-rich organizations, t...
04/07/2026
April 7 2026, 19:00 (PDT) Detective Conan: Fallen Angel of the Highway Opens in...
16/06/2026
Thomson's highly regarded expert-led online learning courses are now easier to access on the go via our new App.
Available now on Google Play Store, the J...
16/06/2026
Neumann.Berlin has released firmware version 2.0 for the MT 48 audio interface, adding plugin compatibility, expanded Dante networking options, broadcast encode...
16/06/2026
TVNewsCheck has announced that nominations are now open for its 2027 Women in Technology Awards, to be presented at NAB Show 2027 on Tuesday, April 6 in the Med...
16/06/2026
Clear-Com has announced Avalon, a 1RU IP intercom platform for broadcast, live events, and production environments. Designed for IP-only workflows, Avalon suppo...
16/06/2026
SNS has published a guide to remote video editing workflows using its EVO shared storage platform and companion tools, covering use cases ranging from home edit...
16/06/2026
Grass Valley has announced that the Richmond Flying Squirrels, a Minor League Baseball affiliate of the San Francisco Giants, have deployed five Grass Valley LD...
16/06/2026
The Alliance for IP Media Solutions (AIMS) has announced the launch of the Official IPMX Training Series, a free online program covering the design, configurati...
16/06/2026
Swerve TV has announced distribution agreements with Fubo, Plex, Amazon Fire TV,...
16/06/2026
ATP and TikTok have announced an expansion of their global content partnership, extending the ATP's TikTok hub powered by TikTok GamePlan to cover all nine ...
16/06/2026
Network's LA facility serves as the heart of a sprawling operation built to ...
16/06/2026
At Pico, the network's media-management team is supporting a flood of HBS fe...
16/06/2026
The NHL will no longer air on CBC after the pulic broadcasters and national rights-holder Rogers Sportsnet were unable to come to agreement.
After a successfu...
16/06/2026
As live sports broadcasters continue to seek new ways to make complex action mor...
16/06/2026
Built with the 2026 FIFA World Cup in mind, these small but mighty IP-based tran...
16/06/2026
Boasts individual synths for each band
UVI's latest synth takes an interesting approach to synthesis, offering a trio of synth engines that each operate...
16/06/2026
New intelligent auto-fader plug-in unveiled
PSPaudioware's latest release offers automatic level adjustment and provides more detailed control than many...
16/06/2026
New performance-focused library announced
Crystal Pads is the latest addition to The Crow Hill Company's ever-growing product range, and according to th...
16/06/2026
Developed in partnership with Sequential
In recent years, GForce Software have branched into official emulations of classic hardware synths, delivering a ha...
16/06/2026
Designed specifically for live performance monitoring
beyerdynamic's latest announcement sees the company introduce an affordable in-ear monitoring syst...
16/06/2026
Official emulation celebrates iconic synth's 40th anniversary
Cherry Audio have just introduced Ensoniq ESQ-1, an official recreation of the 1986 polyph...
16/06/2026
Australians place growing trust in SBS News
16 June, 2026
Media releases
SBS has been recognised as one of Australia's most trusted news providers, ran...
16/06/2026
Rohde & Schwarz achieves highest number of GCF validated 3GPP NR NTN test cases ...
16/06/2026
Bydgoszcz to Become a Local Centre of Excellence for Advanced Rail Technologies....
16/06/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
16/06/2026
87 Percent of Creators Who Use Creative AI Say It Is Growing Their Business and ...
16/06/2026
Historic Zhuque-3 Reusable Rocket Test Mission Captured with URSA Cine Immersive
Brie Clayton June 16, 2026
0 Comments
Apple Immersive Video puts view...
16/06/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
16/06/2026
Rise WIB, the award-winning advocacy group championing gender diversity and career progression across the broadcast and media technology industry, today announc...
16/06/2026
Limecraft today announced the availability of Limecraft 2026.4, the fourth of eight planned platform releases this year. The update introduces Team-Based Access...
16/06/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
16/06/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
16/06/2026
Free Program Supports IPMX Education from Foundational Concepts Through System and Network Design
The Alliance for IP Media Solutions (AIMS) today announced t...
16/06/2026
Share your views on Screen Australia and the future of the industry 15 June 2026
Your feedback matters.
Following the instrumental insights provided in 2025,...
16/06/2026
Enterprises are moving agentic AI from proof of concept to production - and the next generation of AI factories are built for the era of agents.
At HPE Discove...
16/06/2026
AI runs at the speed of light. More and more, that light is made in Texas.
Cohe...
16/06/2026
Techtel Supports T-Motion RCCP-2A Controller Upgrade for Major Australian Broadc...
16/06/2026
Tuesday 16 June 2026
Record audiences tune in for opening weekend of ICC Women&...
16/06/2026
Every breakthrough AI model starts the same way: with a training run. The infrastructure running those training jobs shapes everything: how fast teams can itera...
15/06/2026
One of the more exciting internal video production divisions within a college at...
15/06/2026
The deal valued at $22 Billion is expected to close in the first half of 2027...
15/06/2026
Golf Channel and the Arnold Palmer Cup have announced a partnership to livestream the 2026 Arnold Palmer Cup on Golf Channel Mobile and GolfChannel.com. The tou...
15/06/2026
TikTok and Panini have announced a partnership to bring a digital collectible ca...
15/06/2026
Cosm and Monster Energy have announced the debut of the first full-dome immersiv...
15/06/2026
Real American Freestyle (RAF) and Fox Nation have announced an exclusive streaming agreement for three RAF international events, beginning with RAF Georgia on J...
15/06/2026
FanConnect has announced a partnership with Extreme Networks integrating FanConn...
15/06/2026
Ten Emerging Filmmakers Ages 18 to 25 Will Start Fellowship Year at Ignite Lab from June 14-19
LOS ANGELES, CA, June 15, 2026 - The nonprofit Sundance Institut...