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...
04/06/2026
Sony Electronics is introducing the SRG-AS10, a 4K 60p-compatible PTZ auto-frami...
04/06/2026
This recent grad from Spring, TX, led creative-video output for the Aggies' men's basketball team last season and has been producing video and creating ...
04/06/2026
For the first time at a women's golf major, every player in the field will r...
04/06/2026
Three Panasonic PT-RQ45 40,000-lumen 3-Chip DLP projectors made their first live...
04/06/2026
Bitmovin and Akamai have announced a collaboration with NRJ Group, a French mult...
04/06/2026
Telestream will exhibit at InfoComm 2026 (Booth N7952), demonstrating media work...
04/06/2026
Sony has announced the development of RIALTO 65, a 65mm format image sensor block for the VENICE 2 digital cinema camera, targeting release in the first half of...
04/06/2026
KOKUSAI DENKI Electric America will exhibit at InfoComm 2026 (Booth N8025, June ...
04/06/2026
Bell Media's TSN and RDS are the exclusive Canadian broadcasters of FIFA Wor...
04/06/2026
The Challenge: Receiving Heavy Media Files From Road Games Quickly and ReliablyT...
04/06/2026
MASV, a managed file transfer platform used in broadcast and live sports product...
04/06/2026
NESN has announced the appointment of Fahad Haider as Vice President of Operations and Engineering. Haider returns to NESN, where he previously served as Vice P...
04/06/2026
David J. Halberstam, who spent almost 50 years in sports as a broadcaster and an executive, died June 2 after a years-long battle with brain cancer.
Over his l...
04/06/2026
Although collegiate production programs are tasked with delivering high-quality ...
04/06/2026
California studio, two production trucks, global distribution system are combine...
04/06/2026
New global program empowers and supports storytellers through scriptwriting course and access to industry experts
TikTok and Sundance Institute today announce...
04/06/2026
Steinberg DAWs now boast in-depth Tonalic integration
Celemony's innovative virtual session musician plug-in has just received an update that brings ARA...
04/06/2026
Get Hands-On With Over 20 Mic Brands
GearExpo UK is fast approaching, and if you've been looking for a chance to check out some new mics, then you'r...
04/06/2026
Combos feature new Amplifier Intelligence engine
Positive Grid's latest release sees the company introduce two new combo amplifiers that promise to offe...
04/06/2026
Is Your Job Making You Work this June?
4 June, 2026
Media releases
SBS Launches the World Cup Watchers' Rights Association to Stand Up For Australians&...
04/06/2026
Statement regarding unauthorised use of SBS logos on third party social content
4 June, 2026
Media releases
SBS has become aware of social media posts in c...
04/06/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
04/06/2026
Bitmovin, a leading provider of video streaming infrastructure, and Akamai, the cybersecurity and cloud computing company that powers and protects business onli...
04/06/2026
American Underground (AU), the Startup Hub of the South and a community of mor...
04/06/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
04/06/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
04/06/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
04/06/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
04/06/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
04/06/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
04/06/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
04/06/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
04/06/2026
NVIDIA and Microsoft Reinvent Windows PCs for the Age of Personal AI
Brie Clayton June 3, 2026
0 Comments
RTX Spark - a 1-Petaflop Superchip, the Full...
04/06/2026
Inside the Peaky Blinders: The Immortal Man Grade
Brie Clayton June 3, 2026
0 Comments
Simone Grattarola discusses shaping the look in DaVinci Resolve...
04/06/2026
Cine Gear Expo Announces 2026 Awards of Excellence Recipients
Brie Clayton June 3, 2026
0 Comments
Ed Lachman ASC, Caleb Deschanel ASC, and M. David M...
04/06/2026
St. Vincents Live with Orchestra Tour to Feature Berklee Alum Ruby Plume Berklee alumni St. Vincent and Ruby Plume will appear on the same bill across seven d...
04/06/2026
Back to All News
All the Truth in My Lies Coming to Netflix on August 28
Entertainment
04 June 2026
GlobalSpain
Link copied to clipboard
Download the imag...
04/06/2026
Back to All News
Larry Tanz, VP of Content, EMEA, Delivers a Keynote Speech at ...
04/06/2026
Back to All News
Turn Your Living Room into a Stadium With the New FIFA World ...
04/06/2026
Keeping conversations real on LinkedIn Published on Jun 4, 2026 Categories: Product News
LinkedIn Corporate Communications
Share
LinkedIn
Facebook ...
04/06/2026
BackStory follows four Irish young people as they travel back to their parents' homelands
Modern Irish identity is enriched by cultures and influences from...
04/06/2026
June's forecast with GeForce NOW: 100% chance of gaming.
GeForce NOW is lining up new adventures for the month, from big-name blockbusters to quirky indies...
03/06/2026
SES and Viva, Mexico's ultra low-cost airline, have launched multi-orbit satellite inflight connectivity on Viva's Airbus aircraft. A total of 60 A320s ...
03/06/2026
The College Football Playoff, ESPN, and TNT Sports have announced kick times and...
03/06/2026
RED Digital Cinema will exhibit at Cine Gear Expo 2026 (Booth 33, June 5-6, Universal Studios Lot), hosting three panels and hands-on product demonstrations.
P...
03/06/2026
Roku has announced the Soccer Zone, a dedicated hub for FIFA World Cup 2026 content available across the United States, Canada, Mexico, Brazil, Colombia, Argent...
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.
