
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 ...
20/11/2026
The sports broadcaster's chairman sees the upcoming game as an opportunity t...
20/11/2026
In-venue and creative video staffers at the professional and collegiate level have one major thing in common: the intensity and attention to detail ramps up dur...
20/11/2026
A variety of LED displays, integrated broadcast infrastructure covers every angl...
20/11/2026
Real American Freestyle (RAF) has announced a distribution partnership with Migu Video, a Chinese digital sports platform, to carry the RAF15 main event between...
20/11/2026
Apple and F1 Academy have announced that the all-female racing series will stream on Apple TV in the U.S. beginning in 2027. Fans will be able to watch every qu...
20/11/2026
Regional broadcaster m nchen.tv has modernized its production infrastructure at a new location in northern Munich. Riedel MediorNet provides decentralized signa...
20/11/2026
Eutelsat has announced three appointments to its Executive Committee, effective November 2026, under CEO Jean-Fran ois Fallacher.
Jean-Hubert Lenotte is appoin...
20/11/2026
After more than two decades at Harbour Esplanade in Docklands, the Seven Network has relocated its Melbourne headquarters to a new, custom-built ST 2110 facilit...
20/11/2026
Brazil is among the first markets deploying Versatile Video Coding (VVC) as part...
20/11/2026
The Soccer Tournament (TST), the $1 million winner-take-all 7v7 world championsh...
20/11/2026
Enhancements include a larger compound, an elevated camera count, and a whole lo...
20/11/2026
Leaders from ESPN, iHeartMedia, and Just Women's Sports explore creating new...
20/11/2026
New EQ plug-in aimed at mastering duties
Lemur Audio's latest plug-in has just arrived, delivering a characterful EQ that's aimed primarily at maste...
20/11/2026
Discounts of up to 75% now available
Soundtoys' annual Black Friday sale is now live, with discounts of up to 75% applied across the company's range...
20/11/2026
New creative looping tool revealed
Described as a cycling audio fragmenter , the latest addition to the SoundGhost range draws its inspiration from tape ph...
20/11/2026
Fairchild recreation gets plug-in counterpart
Heritage Audio's Herchild 670 offers a meticulous hardware recreations of the much-loved Fairchild 670 and...
20/11/2026
BONDI: 7 Minutes That Changed Australia
16 November, 2026
Media releases
A landmark documentary chronicling the worst terror attack in modern Australian hi...
20/11/2026
Statement regarding SBS non-Executive Director Aaron Fa'Aoso
20 November, 2026
Media releases
Aaron Fa'Aoso has advised that he is stepping down a...
20/11/2026
INVITATION FOR PUBLIC COMMENTS AND OBJECTIONS ON NOMINEES FOR APPOINTMENT TO THE COUNCIL OF THE NATIONAL FILM AND VIDEO FOUNDATION (NFVF) The Department of Spor...
20/11/2026
UEFA Nations League football and autumn premieres propel TVP to number one as YouTube grows its TV screen share
Warsaw, Poland, November 19, 2026 - Nielsen, a...
20/11/2026
October's autumn weather clearly encouraged Poles to spend more time in front of their screens, resulting in both longer viewing times and a higher overall ...
20/11/2026
Share
{alert(Link copied)}).catch(err => {console.error(Failed to copy: , err)}) data-platform=copy_link data-action= class=font-ui-heading py-2 text-sm tex...
20/11/2026
DigitalGlue releases creative.space Pulse, a free Mac app that shows creatives w...
20/11/2026
How NBCUniversal Telemundo Used Virtual Production for its FIFA World Cup 2026 C...
20/11/2026
What Film School Pays and What it Costs: The Federal Numbers
Mark Santos November 20, 2026
0 Comments
Every editor has had the argument about film sch...
20/11/2026
Share
{alert(Link copied)}).catch(err => {console.error(Failed to copy: , err)}) data-platform=copy_link data-action= class=font-ui-heading py-2 text-sm tex...
20/11/2026
Share
{alert(Link copied)}).catch(err => {console.error(Failed to copy: , err)}) data-platform=copy_link data-action= class=font-ui-heading py-2 text-sm tex...
20/11/2026
Share
{alert(Link copied)}).catch(err => {console.error(Failed to copy: , err)}) data-platform=copy_link data-action= class=font-ui-heading py-2 text-sm tex...
20/11/2026
Share
{alert(Link copied)}).catch(err => {console.error(Failed to copy: , err)}) data-platform=copy_link data-action= class=font-ui-heading py-2 text-sm tex...
20/11/2026
Share
{alert(Link copied)}).catch(err => {console.error(Failed to copy: , err)}) data-platform=copy_link data-action= class=font-ui-heading py-2 text-sm tex...
20/11/2026
Share
{alert(Link copied)}).catch(err => {console.error(Failed to copy: , err)}) data-platform=copy_link data-action= class=font-ui-heading py-2 text-sm tex...
20/11/2026
Share
{alert(Link copied)}).catch(err => {console.error(Failed to copy: , err)}) data-platform=copy_link data-action= class=font-ui-heading py-2 text-sm tex...
20/11/2026
Beam Dynamics has introduced new Model Context Protocol (MCP) support, expanding how customers can work with the live operational data held in Beam. The new cap...
20/11/2026
Re:Vision Effects Cyber Week Sale Nov 24 to Nov 30
Pierre Jasmin November 19, 2026
0 Comments
Long time friends of the COW are up with a big sale!
Us...
20/11/2026
Avid is offering up to 33% off annual subscriptions!
Brie Clayton November 19, 2026
0 Comments
Hero image courtesy Deposit Photos
Pro Tools
Pro To...
19/11/2026
With a sports-bar vibe and all the teams onsite in a huge atrium in Ballpark Vil...
19/11/2026
The senior from Phoenix and England is balancing life as a Division I tennis player with a growing career behind the microphone
The future of live sports produ...
19/11/2026
RED Digital Cinema and Colorfront have partnered to integrate native Colorfront color management into RED Connect, letting productions apply Colorfront's co...
19/11/2026
SMPTE has published an engineering report summarizing an industry survey by its Study Group on Content Provenance and Authenticity in Media. The findings point ...
19/11/2026
CTech and Skyline Communications have announced a partnership to deliver integra...
19/11/2026
Euskal Irrati Telebista (EITB), the Basque Country's public media group, has completed Phase 1 of a five-studio renewal with the commissioning of a Lawo IP ...
19/11/2026
The Esports Foundation (EF) has announced an expanded Road to ENC 2027 for the E...
19/11/2026
All Women's Sports Network (AWSN) held its first awards show, VEON: Women W...
19/11/2026
Gracenote, the content intelligence business unit of Nielsen, has extended its relationship with DIRECTV. Gracenote will continue to supply visual imagery and p...
19/11/2026
The Audio Engineering Society (AES) Technical Committee on Broadcast and Online ...
19/11/2026
NDI has released NDI 6.4, which adds Recording and Playback APIs, support for HEVC 4:4:4 and a configurable receiver buffer for Wi-Fi. The update also includes ...
19/11/2026
Lemonlight, the video production company behind more than 50,000 brand videos, has announced Hero, an AI production operating system, along with a new AI agent ...
19/11/2026
In a first-of-its-kind deployment at XFINITY Center, the digital signage/display...
19/11/2026
For the first time on a US streaming service, real-time player identification, contextual stats, and visual insights appear right in the live stream, available ...