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 ...
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...
02/05/2026
Dalet, a leading technology and service provider for media-rich organizations, t...
01/05/2026
January 5 2026, 18:30 (PST) NBCUniversal's Peacock to Be First Streamer to ...
01/04/2026
January 4 2026, 18:00 (PST) DOLBY AND DOUYIN EMPOWER THE NEXT GENERATON OF CREATORS WITH DOLBY VISION
Douyin Users Can Now Create And Share Videos With Stun...
26/03/2026
Extron and Audio-Technica have announced that their Design Solutions devices, Ce...
26/03/2026
SPORTEL Asia took place March 24-25 at the Orchard Hotel Singapore, drawing 370 participants representing more than 220 companies from 35 countries. Approximate...
26/03/2026
LTN will demonstrate satellite-to-IP migration solutions for broadcasters, cable networks, MVPDs, and station groups at NAB Show 2026 (Booth W2100).
The C-Ban...
26/03/2026
The Seattle Mariners have partnered with Daktronics to install an upgraded outfi...
26/03/2026
The Esports Foundation (EF) has awarded National Team Partner status to organizations and individuals across more than 100 countries and territories for the Esp...
26/03/2026
Telestream has announced expanded integration between its Vantage workflow platf...
26/03/2026
FOR-A Corporation has announced a partnership with Nippon TV and NVIDIA to distr...
26/03/2026
CBS Sports and the WNBA have expanded their partnership through a new long-term, multiplatform rights agreement. CBS Sports will feature up to 20 regular-season...
26/03/2026
Zac Brown Band will star in the Sunday Night Baseball show open on NBC and Peacock, NBC Sports announced. The open was filmed at American Family Field, home of ...
26/03/2026
Major League Baseball opens the 2026 season with nationally televised games across multiple media partners. Fans can visit MLB.com/Watch for a complete schedule...
26/03/2026
DIRECTV has announced an agreement to carry nine new MLB-operated team channels, covering the Cincinnati Reds, Detroit Tigers (Detroit SportsNet), Kansas City R...
26/03/2026
Nationals.TV, the Washington Nationals' local broadcast channel, is now avai...
26/03/2026
Starting Thursday, March 26, MLB.TV will be available on the ESPN App, giving fans access to more than 2,000 out-of-market regular season games. Fans can also r...
26/03/2026
Spotify's partnership with NAVER, Korea's leading digital platform, is built around one simple goal: making it easier for listeners in Korea to discover...
26/03/2026
Exclusive to ComposerCloud for first month
Sample library and virtual instrument gurus EastWest have announced the upcoming release of a new virtual drum i...
26/03/2026
Friday 27 March
Friday 27 March will mark Dynamic Range Day 2026, the annual day of online activity that was started in 2010 by mastering engineer Ian Sheph...
26/03/2026
Includes four separate instruments
Max Richter's SRM Sounds venture have just announced their fifth release, ASMR Choir. Captured at the main live room ...
26/03/2026
The National Film and Video Foundation (NFVF), an agency of the Department of Sport, Arts and Culture (DSAC), has been tasked with administering the Sixth Presi...
26/03/2026
By Matt Klein
Hello to everyone reading this - whether you're a current YEP, HPA member, or just YEP-curious. I'm Matt Klein, one of the co-chairs of t...
26/03/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
26/03/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
26/03/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
26/03/2026
28 participating companies, from start-ups to blue-chips, lead the UK Group as part of the GREAT Britain and Northern Ireland presence across all the Halls at N...
26/03/2026
Arkona Unveils BLADE//planner and Major Usability Enhancements at NAB 2026
Brie Clayton March 26, 2026
0 Comments
New graphical configuration tool and...
26/03/2026
Bitfocus showcases complete control at NAB Show 2026
Brie Clayton March 26, 2026
0 Comments
Continuing development drives advances in security, availa...
26/03/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
26/03/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
26/03/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
26/03/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
26/03/2026
Nevion introduces powerful new Panel Builder to enhance VideoIPath broadcast con...
26/03/2026
2026 Oscar Nominated Films Powered by Blackmagic Design
Brie Clayton March 25, 2026
0 Comments
DaVinci Resolve Studio used on 27 of this year's no...
26/03/2026
Leader to present full suite of advanced Test & Measurement solutions at NAB Sho...
26/03/2026
Boston Conservatory to Present New England and Collegiate Premiere of Groundbrea...
26/03/2026
Thursday 26 March 2026
HBO Max included for Sky and NOW customers from 26 March...
26/03/2026
Thursday 26 March 2026
Sky Sets Iconic Costumes Dancing Together Like Never Before' in Brand CampaignTurn on cookies to view this content. Go to Privacy o...
26/03/2026
Wuppertal March 26, 2026
CS live Builds Future-Ready OB Van With Riedel MediorNet and hi Control SystemRiedel Communications today announced that Czech-based ...
26/03/2026
Back to All News
Have a Wish Worth Dying For? Netflix's First Korean YA Hor...
26/03/2026
Back to All News
The Truthers, Starring Jose Coronado and St phanie Magnin, Pre...
26/03/2026
Enhancements to Harmonic's Sports Streaming Solution Introduce Server-Side Multiview, Contextual Metadata-Driven Advertising, Dynamic In-Stream Ads and Robu...
26/03/2026
Open the report
Our 2025 report brings together results from 79 business magazine titles across 22 market sectors, published by 49 media owners. Over the year,...
26/03/2026
Editor's note: This post is part of Into the Omniverse, a series focused on ...
26/03/2026
RT Commercial today announced Fiat as sponsor of The Louise Duffy Show on RT Radio 1, Weekdays 3pm 4pm.
The Louise Duffy Show is the home of daytime music...
26/03/2026
That gaming backlog won't clear itself - GeForce NOW is here to help. Stream the latest titles straight from the cloud across a variety of devices.
This we...
26/03/2026
Wayne, N.J., March 26, 2026 Phantom High-Speed announces the latest product li...
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.
