What's the ROI? Getting the Most Out of LLM Inference

09/10/2024

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

LINK:	https://blogs.nvidia.com/blog/llm-inference-roi/...
	See more stories from nvidia

Most recent headlines

05/01/2027

Worlds first 802.15.4ab-UWB chip verified by Calterah and Rohde & Schwarz to be demoed at CES 2026

Worlds first 802.15.4ab-UWB chip verified by Calterah and Rohde & Schwarz to be ...

01/06/2026

Dolby Sets the New Standard for Premium Entertainment at CES 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 Flex LTS Delivers Smarter Search, Faster Editing, and an AI-Ready Foundation for Modern Media

Dalet, a leading technology and service provider for media-rich organizations, t...

01/05/2026

NBCUniversal's Peacock to Be First Streamer to Integrate Dolby's Full Suite of Premium Picture and Sound Innovations

January 5 2026, 18:30 (PST) NBCUniversal's Peacock to Be First Streamer to ...

01/04/2026

DOLBY AND DOUYIN EMPOWER THE NEXT GENERATON OF CREATORS WITH DOLBY VISION

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...

18/03/2026

Net Insight Names Larissa GrnerMeeus CPO

Newly named chief product officer (CPO), Larissa G rner Meeus will return to Net Insight on May 4. Larissa G rner Meeus will become CPO of Net Insight on May 4...

18/03/2026

SVG Europe's The Football Summit 2026: Select Sessions Now Available to Watch on SVG PLAY

SVG Europe's The Football Summit 2026 explored how the sports broadcasting i...

18/03/2026

SVG's SportsTech@NABShow Blog Goes Live as Countdown to Vegas Heats Up

The 2026 NAB Show kicks off one month from today and SVG is once again set to cover the show from every angle. SVG's SportsTech@NAB Show Blog is now live - ...

18/03/2026

The Premiere of Dead Lover Showcases Grave Robbing and Sex With a Giant Finger

(L-R) The cast and crew of Dead Lover at The Ray Theater for its premiere at the 2025 Sundance Film Festival. (Photo by Robin Marshall/Shutterstock for Sundan...

18/03/2026

Music Row Piano from Wiltone Productions

Yamaha C7 captured in Nashville Wiltone Productions have announced the release of Music Row Piano, a deeply sampled Yamaha C7 piano library that's been ...

18/03/2026

Techivation introduce T-Warmer Mk2

Bass-enhancement plug-in upgraded Along with a steady stream of new releases, Techivation have recently been revisiting some of the older plug-ins in their ...

18/03/2026

Tonal Balance Control 3 from iZotope

Mix-reference plug-in overhauled iZotope's powerful mix-referencing plug-in has just reached its third major version, and now boasts a new capture proce...

18/03/2026

Toontrack Transistor Organ EKX

Latest EZKeys 2 expansion arrives Toontrack's staggering collection of EZKeys 2 expansions has grown once again, and the latest instalment delivers a on...

18/03/2026

Jess Ho explores the politics of food in new SBS Audio podcast For The Culture

Jess Ho explores the politics of food in new SBS Audio podcast For The Culture 18 March, 2026 Media releases New SBS Audio podcast For The Culture, hosted ...

18/03/2026

Future-ready broadcasts with ATSC 3.0 and enhanced services from Rohde & Schwarz at NAB 2026

Future-ready broadcasts with ATSC 3.0 and enhanced services from Rohde & Schwarz...

18/03/2026

LCTWS 2026

London Calling - When The Industry Convened to Help Streaming Find its MoJo In this blog, Laura Rognoni reflects on key discussions from the Connected TV World...

18/03/2026

SMPTE Unveils 2026 NAB Show Educational Presentations

SMPTE Unveils 2026 NAB Show Educational Presentations Brie Clayton March 18, 2026 0 Comments SMPTE , the home of media professionals, technologists, a...

18/03/2026

Auditel Ad Campaign Shot on Blackmagic PYXIS 12K

Auditel Ad Campaign Shot on Blackmagic PYXIS 12K Brie Clayton March 18, 2026 0 Comments LED wall virtual production blends 12K open gate acquisition w...

18/03/2026