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...
20/08/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
20/08/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
20/08/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
20/08/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
20/08/2026
IPC Expands Brazilian Presence with New Channel Partner CCATech
Industry veterans Carlos Abrah o and Bruno Lopes strengthen local support and expand IPCs prese...
20/08/2026
Zixi Enters Mauritius Market Through Strategic Deployment with MTC Jockey Club
Worlds Oldest Horse Racing Club in the Southern Hemisphere Selects Zixis Softwar...
20/08/2026
LA JOLLA, CA-Scripps Research assistant professor Tiantian Liu has received a five-year NIAID DP2 award from the National Institute of Allergy and Infectious Di...
20/08/2026
Audiences reveal appetite for multicultural Australia on-screen 20 August 2026
Dr Georgie McClean, Dr Rebecca Huntley (L-R)
By Dr Georgie McClean and Dr Rebec...
20/08/2026
How Mavis Studio Is Expanding Live Production Teaching at Sussex
Brie Clayton August 19, 2026
0 Comments
Hero image displays a student about to mix be...
20/08/2026
Vertical Unscripted Series Legal Queens Captured with Blackmagic Design
Brie Clayton August 19, 2026
0 Comments
Reality TV veteran Eric Gardner create...
19/08/2026
As the cost of IP nears price parity with traditional SDI, Reynolds argues that ...
19/08/2026
A private 5G bubble was chosen instead of traditional broadcast RF for the race ...
19/08/2026
(L-R) Rafael Manuel and Isabel Sicat attend the Filipi ana premiere during the 2026 Sundance Film Festival at The Ray Theatre on January 23, 2026, in Park Cit...
19/08/2026
Studio owners invited to have their say on rates cuts
The Music Producers Guild (MPG) are inviting UK-based studios to take part in an in-depth survey that ...
19/08/2026
Vintage-inspired desktop interface revealed
Harrison's latest hardware offering sees them head into the world of audio interfaces, delivering a 10-in/12...
19/08/2026
Granular effect remains locked to scale and tempo
Described as your dream granular effect , Minimal Audio's latest creation is capable of turning any s...
19/08/2026
First polyphonic SWAM instrument incoming
Audio Modeling's latest physical-modelling orchestral instrument is just around the corner, and is now availab...
19/08/2026
World Cup Games Generate 84 Billion Minutes Across FOX, Fox Sports 1 and NBCU...
19/08/2026
July brought typical peak-holiday drops in viewership, driven primarily by viewers stepping away from screens and a visible decline in reach across both traditi...
19/08/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
19/08/2026
Henry joins BCE as the company accelerates its commercial strategy and expands its international presence across the media, broadcast and technology markets. He...
19/08/2026
iWedia, a leading provider of software solutions for connected TV devices, today announced a major milestone in Android platform optimization with the successfu...
19/08/2026
NTP Technology announces that ST 2110-30 audio-over-IP connectivity with AES67 and RAVENNA capabilities is now available for the brand's world-class audio i...
19/08/2026
Net Insight has been selected to provide its Nimbra platform by the service provider Big Blue Marble, formerly ORS Group, as the core media transport solution f...
19/08/2026
LiveU has supported Coventry City FC and several other top-flight UK football clubs through one of their most demanding live production challenges of the year, ...
19/08/2026
Nature Seekers Documentary Shot with Blackmagic URSA Cine 12K LF
Brie Clayton August 18, 2026
0 Comments
Filming endangered turtles required camera...
19/08/2026
Neyrinck Brings D-Control Consoles Back to Life on Apple Silicon with D-Control ...
19/08/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
19/08/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
19/08/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
19/08/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
19/08/2026
Share
Copy link
Facebook
X
Linkedin
Bluesky
Email...
19/08/2026
Riedel Communications today announced that Edith Cowan University (ECU) in Perth, Australia, implemented a comprehensive communications infrastructure that brin...
19/08/2026
QuickLink, a leading provider of award-winning video production and remote guest contribution solutions, will spotlight a major leap forward in remote guest con...
19/08/2026
Integrated platform melds media asset management, playout automation and newsroom workflows
EMAM, PlayBox Neo and SNEWS have officially launched Nuvex, a conne...
19/08/2026
Screen Australia releases the Private Investment Toolkit to strengthen screen in...
19/08/2026
Do Aussies truly value local content? Unpacking Screen Currency 2026 with Deirdr...
19/08/2026
There's an early Christmas gift for Bookish fans, as UKTV confirms the re-commission of the hit U&alibi drama, created by and starring Mark Gatiss. Viewers ...
19/08/2026
Wednesday 19 August 2026
Sky gives sports fans the chance to watch multiple liv...
19/08/2026
On Saturday September 5 Katie Taylor will step into the ring for the fight of her life against Flora Pili in front of 80,000 fans in Croke Park. A legendary fig...
18/08/2026
SportStream 2026, a free three-day virtual event, will take place September 15-17. One registration provides access to all three days of live programming and on...
18/08/2026
BeckTV will introduce BeckFlow to the European market at IBC2026, demonstrated on the Providius stand (10.F53).
BeckFlow is a web-based schematic documentation...
18/08/2026
Boland Communications will introduce a SMPTE ST 2110 input option for its 4K monitors at IBC2026, along with a new 55-inch super-narrow bezel display for video ...
18/08/2026
Dolby Laboratories and the Seattle Seahawks have announced a partnership integrating Dolby OptiView into the Seahawks' live streaming coverage, making the S...
18/08/2026
For BMG Executive Producer of Sports Production, Graham Taylor, producing live s...
18/08/2026
TMRW Sports has appointed Rufus Hack as President of Golf, a newly created role in which he will lead the company's golf businesses including TGL presented ...
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.
