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...
04/03/2026
Lega Basket Serie A (LBA), the governing body for Italy's premier basketball...
04/03/2026
Wrexham AFC co-chairmen Rob Mac and Ryan Reynolds will host a first-of-its-kind ...
04/03/2026
The countdown is underway and with just 100 days to go until the world's greatest sporting event begins on Thurs., June 11, FOX Sports, America's Englis...
04/03/2026
No matter where they are in the world, service members and veterans can stream N...
04/03/2026
Telemundo officially releases Somos M s, the anthem for the network's cove...
04/03/2026
The Hollywood Professional Association (HPA) concluded the 2026 HPA Tech Retreat, convening more than 800 industry leaders, technologists, creatives, and execut...
04/03/2026
Smith Entertainment Group transformed how local sports are consumed by creating ...
04/03/2026
The production method, which spans a distance of 36.2 miles, was designed and im...
04/03/2026
Haivision, a global provider of mission-critical, real-time video networking and...
04/03/2026
When Hugues Meyrath came out of retirement to take the helm as CEO of Quantum, i...
04/03/2026
Last week's launch of Banana Ball Championship League has spurred a significant upgrade of production facilities
The Savannah Bananas, arguably the hottest...
04/03/2026
sldkfjsdlfkjsldkfjsldkjfslkdjfslkdjfsl
Lorem ipsum dolor sit amet, consectetur adipiscing elit. Ut elit tellus, luctus nec ullamcorper mattis, pulvinar dapibus...
04/03/2026
For many fans, a song's backstory can be just as compelling as the final pro...
04/03/2026
In 2006, Spotify was founded on the belief that technology could bring artists a...
04/03/2026
Spotify is back on the ground for the Houston Livestock Show and Rodeo, and we&#...
04/03/2026
Spotify had an energizing week in Sydney, Australia, filled with powerful conver...
04/03/2026
Earlier this year, we launched Directed By, a documentary-style series that pull...
04/03/2026
100 Days to Go: SBS Unveils World Class Team for The Greatest Show on Earth -The...
04/03/2026
KT and Rohde & Schwarz to showcase AI-enhanced radio transmission performance In a joint 6G AI proof-of-concept demonstration, the CMX500 one-box tester from ...
04/03/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
04/03/2026
Luxembourg, 3 March 2026 - SES S.A. has today published its 2025 Annual Report, following the announcement of the company's full year financial results for ...
04/03/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
04/03/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
04/03/2026
Luxembourg, March 3, 2026 - SES, a leading space solutions company, along with I...
04/03/2026
04 Mar 2026
VEON Partners with GSMA Innovation Fund to Accelerate Digital Innov...
04/03/2026
04 Mar 2026
VEON's Beeline Uzbekistan and Rakuten Symphony Partner for Open...
04/03/2026
Wednesday 4 March 2026
Sky Sports unveils plans for 2026 Formula 1 coverage
Sky Sports is preparing for one of the most highly anticipated F1 seasons in recen...
04/03/2026
Back to All News
Bloodhounds' Season 2 Gears Up for April 3 Premiere with ...
04/03/2026
Back to All News
Netflix Ads Suite Expands Capabilities
Business
04 March 2026
GlobalUnited States
Link copied to clipboard
After launching the Netflix Ad...
04/03/2026
Scripps Research welcomes healthcare innovator Joe Kiani to the Board of Directors Kiani brings decades of experience in patient safety and public service.
Mar...
04/03/2026
Nanoparticle vaccine approach takes on a new target: Hepatitis C virus Scripps Research scientists reengineer critical proteins on the surface of HCV, paving th...
03/03/2026
Beyond Sports, a Sony group company, and LIV Golf, the world's golf league, ...
03/03/2026
Ilitch Sports + Entertainment announces the launch of Detroit SportsNet (DSN), a year-round broadcast home for two of Detroit's franchises. With flexible op...
03/03/2026
Advanced Systems Group, LLC (ASG), a technology and services provider for media ...
03/03/2026
The PGA of America, NBC Sports and USA Sports extend their media rights agreemen...
03/03/2026
AI device ecosystem company HONOR enters into a strategic technical collaboratio...
03/03/2026
Cleveland's Telos Alliance, pioneers in broadcast technology for 30 years, l...
03/03/2026
The MD 9235 microphone head for wireless handhelds has been a firm favorite with many engineers and artists for its ability to cut through high on-stage levels ...
03/03/2026
Haivision Systems Inc. (Haivision), a global provider of mission-critical, real-...
03/03/2026
Broadcast Management Group (BMG) announces the expansion of its 62,000-square-fo...
03/03/2026
Maroon 5's musical tour in 2025 marked a leap forward in live audio as Monitor Engineer Dave Rupsch utilized Sennheiser's all-digital Spectera wireless ...
03/03/2026
Designed specifically for pro basketball, the renovated space at Gainbridge Fiel...
03/03/2026
As part of the move, former CEO Phillipp Lawo joins the broadcast-tech provider's Supervisory Board
Lawo has announced appointment of Jamie Dunn as chief e...
03/03/2026
A team of legendary announcers and analysts and a classic graphics look will bring the past to life
NBC Sports and Peacock will return to yesteryear for tonigh...
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.
