Large language models and the applications they power enable unprecedented opportunities for organizations to get deeper insights from their data reservoirs and to build entirely new classes of applications.
But with opportunities often come challenges.
Both on premises and in the cloud, applications that are expected to run in real time place significant demands on data center infrastructure to simultaneously deliver high throughput and low latency with one platform investment.
To drive continuous performance improvements and improve the return on infrastructure investments, NVIDIA regularly optimizes the state-of-the-art community models, including Meta's Llama, Google's Gemma, Microsoft's Phi and our own NVLM-D-72B, released just a few weeks ago.
Relentless Improvements Performance improvements let our customers and partners serve more complex models and reduce the needed infrastructure to host them. NVIDIA optimizes performance at every layer of the technology stack, including TensorRT-LLM, a purpose-built library to deliver state-of-the-art performance on the latest LLMs. With improvements to the open-source Llama 70B model, which delivers very high accuracy, we've already improved minimum latency performance by 3.5x in less than a year.
We're constantly improving our platform performance and regularly publish performance updates. Each week, improvements to NVIDIA software libraries are published, allowing customers to get more from the very same GPUs. For example, in just a few months' time, we've improved our low-latency Llama 70B performance by 3.5x.
NVIDIA has increased performance on the Llama 70B model by 3.5x. In the most recent round of MLPerf Inference 4.1, we made our first-ever submission with the Blackwell platform. It delivered 4x more performance than the previous generation.
This submission was also the first-ever MLPerf submission to use FP4 precision. Narrower precision formats, like FP4, reduces memory footprint and memory traffic, and also boost computational throughput. The process takes advantage of Blackwell's second-generation Transformer Engine, and with advanced quantization techniques that are part of TensorRT Model Optimizer, the Blackwell submission met the strict accuracy targets of the MLPerf benchmark.
Blackwell B200 delivers up to 4x more performance versus previous generation on MLPerf Inference v4.1's Llama 2 70B workload. Improvements in Blackwell haven't stopped the continued acceleration of Hopper. In the last year, Hopper performance has increased 3.4x in MLPerf on H100 thanks to regular software advancements. This means that NVIDIA's peak performance today, on Blackwell, is 10x faster than it was just one year ago on Hopper.
These results track progress on the MLPerf Inference Llama 2 70B Offline scenario over the past year. Our ongoing work is incorporated into TensorRT-LLM, a purpose-built library to accelerate LLMs that contain state-of-the-art optimizations to perform inference efficiently on NVIDIA GPUs. TensorRT-LLM is built on top of the TensorRT Deep Learning Inference library and leverages much of TensorRT's deep learning optimizations with additional LLM-specific improvements.
Improving Llama in Leaps and Bounds More recently, we've continued optimizing variants of Meta's Llama models, including versions 3.1 and 3.2 as well as model sizes 70B and the biggest model, 405B. These optimizations include custom quantization recipes, as well as efficient use of parallelization techniques to more efficiently split the model across multiple GPUs, leveraging NVIDIA NVLink and NVSwitch interconnect technologies. Cutting-edge LLMs like Llama 3.1 405B are very demanding and require the combined performance of multiple state-of-the-art GPUs for fast responses.
Parallelism techniques require a hardware platform with a robust GPU-to-GPU interconnect fabric to get maximum performance and avoid communication bottlenecks. Each NVIDIA H200 Tensor Core GPU features fourth-generation NVLink, which provides a whopping 900GB/s of GPU-to-GPU bandwidth. Every eight-GPU HGX H200 platform also ships with four NVLink Switches, enabling every H200 GPU to communicate with any other H200 GPU at 900GB/s, simultaneously.
Many LLM deployments use parallelism over choosing to keep the workload on a single GPU, which can have compute bottlenecks. LLMs seek to balance low latency and high throughput, with the optimal parallelization technique depending on application requirements.
For instance, if lowest latency is the priority, tensor parallelism is critical, as the combined compute performance of multiple GPUs can be used to serve tokens to users more quickly. However, for use cases where peak throughput across all users is prioritized, pipeline parallelism can efficiently boost overall server throughput.
The table below shows that tensor parallelism can deliver over 5x more throughput in minimum latency scenarios, whereas pipeline parallelism brings 50% more performance for maximum throughput use cases.
For production deployments that seek to maximize throughput within a given latency budget, a platform needs to provide the ability to effectively combine both techniques like in TensorRT-LLM.
Read the technical blog on boosting Llama 3.1 405B throughput to learn more about these techniques.
Different scenarios have different requirements, and parallelism techniques bring optimal performance for each of these scenarios. The Virtuous Cycle Over the lifecycle of our architectures, we deliver significant performance gains from ongoing software tuning and optimization. These improvements translate into additional value for customers who train and deploy on our platforms. They're able to create more capable models and applications and deploy their existing models using less infrastructure, enhancing th
Most recent headlines
05/01/2027
Worlds first 802.15.4ab-UWB chip verified by Calterah and Rohde & Schwarz to be ...
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...
01/06/2026
January 6 2026, 05:30 (PST) Dolby Sets the New Standard for Premium Entertainment at CES 2026
Throughout the week, Dolby brings to life the latest innovatio...
13/05/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
13/05/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
13/05/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
13/05/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
13/05/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
13/05/2026
Creative software developer Foundry today announced the latest developments on Nuke Stage. A purpose-built application for end-to-end virtual production and in-...
12/05/2026
Beyond the Hype: A Strategic Post-Hoc Analysis of NAB 2026 If NAB Show 2026 had an underlying theme, it was a quiet, industry-wide pivot from the high-energy sp...
12/05/2026
Guntermann and Drunck (G&D), a Panoptec Technologies Group company, and CT Square, led by Chandresh Shah, have announced a joint venture to distribute G&D and V...
12/05/2026
With 30 days until the start of the FIFA World Cup 2026, Telemundo, the exclusive Spanish-language home of the tournament in the United States, has announced th...
12/05/2026
The NHL has announced the return of Stanley Pup for its third consecutive year, a 90-minute special featuring adoptable rescue dogs competing on a miniature rin...
12/05/2026
NBCUniversal presented its 2026 Upfront to advertisers at Radio City Music Hall, detailing upcoming programming across NBC, Peacock, Bravo, and Versant properti...
12/05/2026
FOX Sports has announced funding for the Fandom and Social Connection Initiative at Harvard Kennedy School's Shorenstein Center on Media, Politics, and Publ...
12/05/2026
TNDV and Live Media, both divisions of Live Media Group, supported live broadcast coverage around NCAA Final Four weekend in Indianapolis, including the March M...
12/05/2026
The European Football Alliance (EFA) has announced a content distribution agreement with Fubo Sports Network, the free ad-supported streaming TV (FAST) channel ...
12/05/2026
For the first time, Spanish-speaking fans in the U.S. will have two separate tel...
12/05/2026
CP Communications led a comprehensive spectrum management initiative on behalf of Churchill Downs during Kentucky Derby week, coordinating RF assets across the ...
12/05/2026
LiveU has announced a strategic partnership with DRONERESPONDERS, a 501(c)3 non-...
12/05/2026
Open Broadcast Systems has announced that BMC TV, a specialist in IP transport of broadcast content, has selected the Open Broadcast Systems 5G Flyaway solution...
12/05/2026
NEP Europe, part of NEP Group, has announced it will deliver broadcast solutions...
12/05/2026
Grass Valley has announced continued collaboration with Ravensbourne University ...
12/05/2026
Stats Perform has announced the launch of Opta Pulse, an AI-assisted video creation and distribution platform for leagues, rights holders, and broadcasters. The...
12/05/2026
FOX Sports has announced a collaboration with Sesame Workshop to integrate Sesame Street characters into FOX Sports' FIFA World Cup 2026 programming. Conten...
12/05/2026
To date, NHL Productions has produced 19 broadcasts with commentary in American Sign Language
NHL in ASL (American Sign Language) may be just one show, but the...
12/05/2026
Google's Brian Albert: creators, athletes, highlights, nostalgia, second-scr...
12/05/2026
A still from Past Lives by Celine Song, an official selection of the Premieres program at the 2023 Sundance Film Festival. (Courtesy of Sundance Institute | p...
12/05/2026
Spotify is where fans and artists come together, turning discovery into somethin...
12/05/2026
Features patented Marco-MMC clocking technology
Black Lion Audio's latest release combines the company's expertise in clocking with their renowned p...
12/05/2026
New Track Panel, sequencer upgrades & more
Following their recent public beta release, Reason Studios have announced the full release of Reason 14. With the...
12/05/2026
One month to go! SBS reveals expansive FIFA World Cup 2026 lineup beyond the pi...
12/05/2026
Rohde & Schwarz presents its advanced solutions for power electronics testing at...
12/05/2026
aconnic AG (ISIN: DE000A0LBKW6), Munich, has developed a modified fund raising p...
12/05/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
12/05/2026
Tyrell Corporation, specialists in high-end live sports and entertainment broadcasts, was tasked with delivering compelling broadcast coverage of premier equest...
12/05/2026
Registration is now open for IBC2026 as the global media, entertainment and technology community prepares to converge on the RAI Amsterdam from 11 14 September ...
12/05/2026
Ross Video, a global leader in live video production technology, will present its latest innovations and integrated production workflows at BroadcastAsia 2026, ...
12/05/2026
500 selected leaders from around the world across start-ups, corporates, and venture capital. Over 50bn in assets under management among attending investors, a...
12/05/2026
CVP, one of Europe's leading suppliers of professional video and broadcast solutions, has appointed Randy Koeman as Sales Manager, Netherlands, marking anot...
12/05/2026
Taiwan Television has expanded its HD content ingest workflow capabilities by investing in the PlayBox Neo Suite - a leading multi-channel multi-server UHD/HD/S...
12/05/2026
Jigsaw24 is strengthening its media team with the appointment of Tom Laker as Client Director and Ashish Chauhan as Project Manager. These strategic hires bring...
12/05/2026
Broadcast playout leader showcases hybrid IP infrastructure and workflow automation with Videoland deployment as regional benchmark...
12/05/2026
dzjinius, the all-in-one production management platform for the media and entertainment industry, will showcase its centralised approach to production workflows...
12/05/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
12/05/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
12/05/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
12/05/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
Large language models and the applications they power enable unprecedented opportunities for organizations to get deeper insights from their data reservoirs and to build entirely new classes of applications.
