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...
02/06/2026
blueCORE standalone processors headline solutions designed to simplify the trans...
02/06/2026
If you've kept up with this article series, you know by now where to start w...
02/06/2026
With new DSP configurations and flexible licensing options, Calrec is removing the barriers to virtualised audio, giving broadcasters the freedom to scale produ...
02/06/2026
The TV agency was one of the earliest adopters of Nielsen's local television...
02/06/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
02/06/2026
Grass Valley today announced that Australian News Channel (ANC), operator of Sky News Australia, has deployed Grass Valley AMPP to transform its newsroom produc...
02/06/2026
Studio Technologies, a leading manufacturer of high-quality audio, video, and fiber-optic solutions, announces its new Model 385 Mic/Intercom Beltpack. The Mode...
02/06/2026
The Riedel Group today announced the appointment of Gudrun Scharler as CEO of Riedel Networks. She succeeds Michael Martens, who has led Riedel Networks since 2...
02/06/2026
More signals, higher quality, and outstanding ingest and streaming flexibility deliver professional results in a small, all-in-one footprint...
02/06/2026
farmerswife will be featured on the Modena Media & Entertainment stand at this year's Mediatech Africa 2026, giving visitors an opportunity to explore the l...
02/06/2026
PTZOptics will showcase a new generation of intelligent video workflows at InfoComm 2026, June 17 19, Las Vegas. Visitors to booth N8227 will see how PTZOptics ...
02/06/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
02/06/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
02/06/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
02/06/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
02/06/2026
Trust has become a commercial issue With global advertising spend forecast to exceed US$1 trillion this year*, the commercial consequences of weak governance co...
02/06/2026
June sees Ireland's cultural calendar in full bloom, as RT Supporting the Arts showcases a vibrant and wide-ranging programme spanning music, theatre, visu...
02/06/2026
After The Traitors Ireland launched in 2025, Irish audiences proved to have a taste for the global hit reality show. This Bank Holiday Monday fans can indulge e...
01/06/2026
In its sixth year, the broadcaster's coverage has become a global brand and ...
01/06/2026
Designed to solve a common problem in broadcasting, the automated workflow detects, identifies, removes, and documents copyrighted music
AudioShake has introdu...
01/06/2026
The sports-analytics company combines its data with proprietary AI to help leagu...
01/06/2026
Category line-up & sponsors announced
Photo: Paul Clarke
The Production Music Awards (PMA) have announced that submissions are now officially open ahead of...
01/06/2026
New hybrid sample/synthesis instrument revealed
Excite Audio have just released the latest instalment in their Evolve series, which has been developed in co...
01/06/2026
Latest TONEX expansion captures three rare vintage amps
The newest addition to IK Multimedia's ever-growing TONEX line-up introduces a set of three incr...
01/06/2026
Musically intelligent soft synth gets upgraded
Scaler Music will be probably be best known to many for their music theory tools, but their product range al...
01/06/2026
SBS confirms its broadcast sponsors for FIFA World Cup 2026
1 June, 2026
Media releases
SBS has secured Hyundai, Hisense, Macca's, Rexona, bet365, Com...
01/06/2026
Rohde & Schwarz Satellite Industry Days 2026 guided by the motto From Earth to ...
01/06/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
01/06/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
01/06/2026
Marshall Electronics is showcasing a comprehensive lineup of next-generation POV cameras, purpose-built to power today's connected AV environments, at InfoC...
01/06/2026
Adobe Announces Concept to Vector
Deepa Subramaniam June 1, 2026
0 Comments
One of the biggest frustrations we hear from designers is how difficult it...
01/06/2026
Vampire Feature Night Patrol Graded with DaVinci Resolve Studio
Brie Clayton June 1, 2026
0 Comments
Colorist shapes dark, gritty tone for horror thri...
01/06/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
01/06/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
01/06/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
01/06/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
01/06/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
01/06/2026
Luxembourg, June 1, 2026 - SES, a leading space solutions company, and Viva, Mexico's ultra low-cost airline, launched fast and reliable multi-orbit satelli...
01/06/2026
Agentic AI is getting physical.
At COMPUTEX on Tuesday, NVIDIA announced NVIDIA JetPack 7.2 and NVIDIA NemoClaw support on NVIDIA Jetson.
JetPack 7.2 brings a...
01/06/2026
Financial institutions have spent years building AI: fraud models, credit models...
01/06/2026
Mardi 2 juin 14h00
FilmLight (ARRI), 10 rue Ren Boulanger, 75010 Paris
Rejoignez-nous pour d couvrir comment FLAPI (l'API FilmLight) peut transformer e...
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...
31/05/2026
Olivia P rez-Collellmir to Premiere Original Work at Gaud Centennial in Barcelona The Berklee graduate and faculty member will debut her choral symphony with...
31/05/2026
Back to All News
Netflix Wins 15 Awards at the Canadian Screen Awards - See Pho...
31/05/2026
Taiwan is home to more than 500 NVIDIA ecosystem partners. More than 1 million N...
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.
