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 ...
07/10/2026
Dalet, a leading technology and service provider for media-rich organizations, today announced the latest Long-Term Supported (LTS) release of Dalet Flex. Build...
08/09/2026
Comcast Technology Solutions (CTS) has announced a multi-year agreement with A E...
08/09/2026
Spectra Logic today announced the availability of Media Insight, a new capabilit...
08/09/2026
Rogers Communications and Quebecor have announced a 12-year sublicensing agreement for French-language national NHL games beginning with the 2026-27 season. TVA...
08/09/2026
Skyline Communications and Bridge Technologies will demonstrate two integrations...
08/09/2026
Boland Communications will make the European debut of its 4K55HDR7SNB 55-inch super-narrow-bezel monitor at IBC2026 (Stand 12.C57, RAI Amsterdam, September 11-1...
08/09/2026
Ateme and Comcast Technology Solutions (CTS) have announced a collaboration comb...
08/09/2026
Haivision has joined the IBC Accelerator Media Innovation Programme as part of t...
08/09/2026
Blackmagic Design has announced price reductions on its PYXIS 6K and Cinema Camera 6K models, effective immediately through Blackmagic Design resellers worldwid...
08/09/2026
AI-Media will demonstrate its LEXI Suite at IBC2026 (Booth 5.C34, September 11-1...
08/09/2026
Bitmovin and Ad Insertion Platform (AIP) have announced the first live deploymen...
08/09/2026
Grass Valley will demonstrate story-centric media workflows at IBC2026 (Booth 9.A01, RAI Amsterdam, September 11-14), connecting newsroom production, content ma...
08/09/2026
M2A Media and Ateme have announced a partnership integrating Ateme's video processing solutions into M2A Media's MediaToAnywhere platform for live sport...
08/09/2026
Ikegami will exhibit at IBC2026 (Stand 12.A31, RAI Amsterdam, September 11-14), including a Day 1 introduction of new cameras in the UNICAM XE series. Additiona...
08/09/2026
Broadcast industry analysts will be gathering plenty of insights at the 2026 IBC Show this weekend in Amsterdam, but Caretta Research has offered up some though...
08/09/2026
SENSHI and DAZN have announced a global broadcast partnership that will bring SENSHI's live events free to viewers in more than 200 markets. The partnership...
08/09/2026
New architecture built with Evertz will bring 4K HDR to Monday Night Football, N...
08/09/2026
Built for rap producers, by rap producers
Sauceware Audio, the independent plug-in company founded by Grammy-nominated producer Jacob (Travis Scott, 21 Sava...
08/09/2026
Classic Sound Techniques design revived
Warm Audio's latest release delivers a meticulous recreation of the preamp circuitry employed by the Sound Techn...
08/09/2026
Latest soft synth inspired by the Memorymoog
XILS Lab have just launched a software instrument that draws its inspiration from Moog's iconic Memorymoog,...
08/09/2026
The Airport Chaplain takes off as SBS's biggest launch of 2026
8 September, 2026
Media releases
1.67 million viewers on National Total TV to date
Best...
08/09/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
08/09/2026
Mo-Sys, the world-leading developer of virtual production, camera robotics and image tracking technology, has completed a business restructuring, which has incl...
08/09/2026
New York, September 8, 2026 The TSAmediaHUB audiovisual contribution and distribution infrastructure developed by Telef nica's Audiovisual Services area re...
08/09/2026
Expanded Riyadh operation brings Asharq and Thmanyah services together on a resilient on-premises Pebble platform.
Pebble, the leading automation, content mana...
08/09/2026
Techex, a pioneering leader in live video workflows, will showcase its broadest software feature set to date at IBC 2026 (Stand 2.B45), extending its lead in so...
08/09/2026
Visitors to Grass Valley's booth at IBC2026 will be introduced to new Framelight X capabilities that connect content acquisition, management, editing and pu...
08/09/2026
BCNEXXT will showcase the latest evolution of Vipe at IBC2026 (Stand 2.B09), introducing a series of innovations designed to deliver higher-quality audience exp...
08/09/2026
Bitmovin, a provider of video streaming solutions, has partnered with Ad Insertion Platform (AIP), a leading innovator in advanced ad insertion technologies, to...
08/09/2026
Brainstorm, a leading manufacturer of real-time graphics and augmented and virtual production, is announcing a major package of enhancements to its Suite 7 soft...
08/09/2026
Big Blue Marble is supporting TDF, French infrastructure and network operator, broadcasting Digital Terrestrial Television (DTT) and radio services, as a techno...
08/09/2026
Agents work inside a single production environment, from incoming wire copy to a live studio show...
08/09/2026
Building on their strategic partnership, Ceeblue and Layercake make the unthinkable possible: scalable sub-second streaming over standard DRM, SSAI, and CDN wor...
08/09/2026
Brainstorm, a leading manufacturer of real-time graphics, augmented and virtual production, is completing the environment for fast-paced broadcast production wi...
08/09/2026
A SaaS portfolio built to work together, helping operators grow audiences, monetise smarter and run leaner, on their own terms
At IBC 2026 (stand 1.B49) Synam...
08/09/2026
Accedo, global provider of video streaming software and services, has been selected by PlayOn Sports, a national leader in high school sports technology, to exp...
08/09/2026
LiveU, the leader in live IP-video solutions, is enabling The Associated Press to take global newsgathering to the next stage with intelligent connectivity and ...
08/09/2026
Mediagenix, a global leader in smart content solutions to profitably connect the right content to the right audience, today announced the latest evolution of it...
08/09/2026
Compact, power-efficient X Platform performance brings broadcast-grade reliability and ultra-low latency to space-constrained remote production (REMI) workflows...
08/09/2026
Zixi, the leader in live video delivery and workflow orchestration, and Kiloview, a leading provider of professional IP video encoding and decoding solutions, t...
08/09/2026
Collaboration gives content owners scalable CDN capacity and broadcasters enhanced flexibility, including for C-band transition planning
Quortex today announce...
08/09/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
08/09/2026
Telos Alliance and Pluxbox Announce Dynamic Media Facility Integration Partnersh...
08/09/2026
J nger Audio Showcases Industry-First S-ADM Cloud Workflow at IBC 2026
Search
Dolby, MainConcept, Techex and J nger Audio will showcase the industry's ...
08/09/2026
Tuesday 8 September 2026
Sky Go gets a major upgrade for Sky Glass and Sky Stream customers
Today, Sky launches a new version of the Sky Go app for Sky Glass ...
08/09/2026
Documentary thriller simulating a Russian attack on UK soil will culminate in a live final episode reuniting the playersTuesday 8 September 2026
Trailer releas...
08/09/2026
Tuesday 8 September 2026
Jamie Demetriou and Amelia Dimoldenberg to host next t...
08/09/2026
Wuppertal September 8, 2026
Riedel Introduces New OnAir Headset PlatformRiedel Communications today announced the launch of its new OnAir headset platform, a...
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.
