
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...
06/09/2026
June 9 2026, 23:00 (PDT) Dolby and MagentaTV Bring Fans Closer to the FIFA Worl...
02/09/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
02/09/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
02/09/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
02/09/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
02/09/2026
KOKKEDAL, DENMARK, SEPTEMBER 1, 2026 DPA Microphones announces its new 6388 CORE Headset Microphone, a high-performance directional headset designed for dema...
02/09/2026
Pliant Technologies will highlight its CrewCom CXD 32CF 32x32 Dante /AES67 Digital Audio Network Interface, a powerful connectivity solution designed to expand ...
02/09/2026
iWedia, a leading provider of software solutions for connected TV devices, today announced two enhancements to Teatro Pulse, its end-to-end TV service operation...
02/09/2026
Signiant, trusted by the media industry for accelerated content movement for more than 20 years, will showcase the latest evolution of the Signiant Platform at ...
02/09/2026
With a contractual agreement to supply a leading UK sports broadcaster with outside broadcast services for 104 football matches per season over the next four ye...
02/09/2026
IBC2026 will put sport firmly at the centre of the global media and entertainment technology conversation with the strongest sports showing in the event's h...
02/09/2026
PTZOptics will bring a hands-on showcase of intelligent video production to IBC2026 (11-14 September, RAI Amsterdam), with demonstrations spanning new camera te...
02/09/2026
OOONA, a global provider of professional management and production tools for the media localization industry, today announced the launch of OOONA Go, a ready-to...
02/09/2026
EON Media will make its European debut at IBC2026 (September 11-14, RAI Amsterdam) in the show's Future Tech Zone (Hall 14.FT5). The company is launching it...
02/09/2026
Frequency, the engine behind many of the worlds leading streaming television channels, will demonstrate at IBC2026 how integrating intelligence throughout chann...
02/09/2026
On-device AI platform gives broadcasters and content creators purpose-built tools for real-time vertical video production and AI-enhanced sports coverage...
02/09/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
02/09/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
02/09/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
02/09/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
02/09/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
02/09/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
02/09/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
02/09/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
02/09/2026
Next-Generation Freeform Control Surfaces Lay the Foundation for AI-Enabled Media Operations
At IBC2026 (Sept. 11 14, RAI Amsterdam, Stand #1.B73), Imagine Com...
02/09/2026
Friedrichstadt-Palast in Berlin, the world's largest theater stage, has modernized its backstage production infrastructure with a comprehensive Riedel solut...
02/09/2026
Women in Media announces that it will host the two-day Reel/Forward in-person conference devoted to navigating today's production scene. Guests can enjoy pa...
02/09/2026
The expanded partnership brings Teradek's Live Production solutions to New Media AV's European dealer and integrator network, opening new opportunities ...
02/09/2026
QuickLink, a leading provider of award-winning video production and remote guest contribution solutions, announces a major expansion to its award winning Studio...
02/09/2026
BCNEXXT, the developers of the advanced playout platform Vipe, today announced that premium entertainment brand Foxtel has selected Vipe to underpin playout for...
02/09/2026
SubtitleNEXT helps HOT Telecom's subtitling team keep a cool head with reliable QC capabilities
Profuz Digital, global developer of digital media and local...
02/09/2026
Case Study: Mission Digital Strengthens Battery Safety with Safe-ion
As productions become increasingly dependent on battery-powered equipment, the safe storag...
02/09/2026
LA JOLLA, CA-Influenza viruses constantly shapeshift to evade recognition by the immune system. This shapeshifting occurs in critical proteins like hemagglutini...
02/09/2026
LA JOLLA, CA-Clinicians often use a medical test called a 12-lead electrocardiogram (ECG) to diagnose heart problems, which uses electrodes placed on the chest ...
02/09/2026
LA JOLLA, CA-Proteins may look like rigid molecular structures, but inside our c...
02/09/2026
RT has today announced a major media partnership with Lidl around Katie Taylor's historic Croke Park clash with Flora Pili on Saturday, September 5th.
The...
01/09/2026
Bridge Technologies will announce at IBC2026 (Booth 1.A71) that its VB440 produc...
01/09/2026
DAZN has announced an agreement to acquire EverPass Media, a commercial sports distribution platform serving bars, restaurants, hotels, and other venues in the ...
01/09/2026
The USTA and Kalshi have announced a multi-year partnership making Kalshi the Official Prediction Market Partner of the US Open, beginning with the 2026 US Open...
01/09/2026
MASV will exhibit at IBC2026 (Booth 5.B40, RAI Amsterdam, September 11-14). CTO Majed Alhajry will appear on a panel session, A Tale of Two Feeds: Automation o...
01/09/2026
FingerWorks Telestrators has announced a partnership with Broadcast Rental to ex...
01/09/2026
FOR-A has announced the expansion of viztrick AiDi into two dedicated products: Go Vertical! AiDi and AiDi Sports Hub. Developed by Nippon TV and delivered glob...
01/09/2026
Sony Pictures Television has launched Live TV on PS5, a free ad-supported streaming experience available to PS5 players in the United States. The service is acc...
01/09/2026
Friedrichstadt-Palast in Berlin has installed a Riedel production infrastructure spanning video, intercom, control, and I/O across the venue, implemented in col...
01/09/2026
Arkona technologies GmbH has announced an OEM partnership with BFE Studio und Medien Systeme GmbH to offer arkona-branded control panels for its SDI-IP routing ...
01/09/2026
Cricket Australia has launched Cricket TV, a direct-to-consumer streaming service built on Eluvio's Content Fabric platform. The service is currently in ear...
01/09/2026
Eluvio has announced it was selected to power Pasifika , a new direct-to-consume...
01/09/2026
Layercake, the company behind the intelligent media workflow platform Streamcake, has appointed Jon Marquard as Director and Chairperson. The appointment marks ...