Editor's note: This post is part of the AI Decoded series, which demystifies AI by making the technology more accessible, and showcases new hardware, software, tools and accelerations for GeForce RTX PC and NVIDIA RTX workstation users.
Large language models (LLMs) are reshaping productivity. They're capable of drafting documents, summarizing web pages and, having been trained on vast quantities of data, accurately answering questions about nearly any topic.
LLMs are at the core of many emerging use cases in generative AI, including digital assistants, conversational avatars and customer service agents.
Many of the latest LLMs can run locally on PCs or workstations. This is useful for a variety of reasons: users can keep conversations and content private on-device, use AI without the internet, or simply take advantage of the powerful NVIDIA GeForce RTX GPUs in their system. Other models, because of their size and complexity, do no't fit into the local GPU's video memory (VRAM) and require hardware in large data centers.
However, Iit i's possible to accelerate part of a prompt on a data-center-class model locally on RTX-powered PCs using a technique called GPU offloading. This allows users to benefit from GPU acceleration without being as limited by GPU memory constraints.
Size and Quality vs. Performance There's a tradeoff between the model size and the quality of responses and the performance. In general, larger models deliver higher-quality responses, but run more slowly. With smaller models, performance goes up while quality goes down.
This tradeoff isn't always straightforward. There are cases where performance might be more important than quality. Some users may prioritize accuracy for use cases like content generation, since it can run in the background. A conversational assistant, meanwhile, needs to be fast while also providing accurate responses.
The most accurate LLMs, designed to run in the data center, are tens of gigabytes in size, and may not fit in a GPU's memory. This would traditionally prevent the application from taking advantage of GPU acceleration.
However, GPU offloading uses part of the LLM on the GPU and part on the CPU. This allows users to take maximum advantage of GPU acceleration regardless of model size.
Optimize AI Acceleration With GPU Offloading and LM Studio LM Studio is an application that lets users download and host LLMs on their desktop or laptop computer, with an easy-to-use interface that allows for extensive customization in how those models operate. LM Studio is built on top of llama.cpp, so it's fully optimized for use with GeForce RTX and NVIDIA RTX GPUs.
LM Studio and GPU offloading takes advantage of GPU acceleration to boost the performance of a locally hosted LLM, even if the model can't be fully loaded into VRAM.
With GPU offloading, LM Studio divides the model into smaller chunks, or subgraphs, which represent layers of the model architecture. Subgraphs aren't permanently fixed on the GPU, but loaded and unloaded as needed. With LM Studio's GPU offloading slider, users can decide how many of these layers are processed by the GPU.
LM Studio's interface makes it easy to decide how much of an LLM should be loaded to the GPU. For example, imagine using this GPU offloading technique with a large model like Gemma 2 27B. 27B refers to the number of parameters in the model, informing an estimate as to how much memory is required to run the model.
According to 4-bit quantization, a technique for reducing the size of an LLM without significantly reducing accuracy, each parameter takes up a half byte of memory. This means that the model should require about 13.5 billion bytes, or 13.5GB - plus some overhead, which generally ranges from 1-5GB.
Accelerating this model entirely on the GPU requires 19GB of VRAM, available on the GeForce RTX 4090 desktop GPU. With GPU offloading, the model can run on a system with a lower-end GPU and still benefit from acceleration.
The table above shows how to run several popular models of increasing size across a range of GeForce RTX and NVIDIA RTX GPUs. The maximum level of GPU offload is indicated for each combination. Note that even with GPU offloading, users still need enough system RAM to fit the whole model. In LM Studio, it's possible to assess the performance impact of different levels of GPU offloading, compared with CPU only. The below table shows the results of running the same query across different offloading levels on a GeForce RTX 4090 desktop GPU.
Depending on the percent of the model offloaded to GPU, users see increasing throughput performance compared with running on CPUs alone. For the Gemma 2 27B model, performance goes from an anemic 2.1 tokens per second to increasingly usable speeds the more the GPU is used. This enables users to benefit from the performance of larger models that they otherwise would've been unable to run. On this particular model, even users with an 8GB GPU can enjoy a meaningful speedup versus running only on CPUs. Of course, an 8GB GPU can always run a smaller model that fits entirely in GPU memory and get full GPU acceleration.
Achieving Optimal Balance LM Studio's GPU offloading feature is a powerful tool for unlocking the full potential of LLMs designed for the data center, like Gemma 2 27B, locally on RTX AI PCs. It makes larger, more complex models accessible across the entire lineup of PCs powered by GeForce RTX and NVIDIA RTX GPUs.
Download LM Studio to try GPU offloading on larger models, or experiment with a variety of RTX-accelerated LLMs running locally on RTX AI PCs and workstations.
Generative AI is transforming gaming, videoconferencing and interactive experiences of all kinds. Make sense of what's new and what's next by subscribing to the AI Decoded newsletter.
More from Nvidia
17/09/2025
Today's creators are equal parts entertainer, producer and gamer, juggling game commentary, scene changes, replay clips, chat moderation and technical troub...
16/09/2025
The U.K. is driving investments in sovereign AI, using the technology to advance...
13/09/2025
Celtic languages - including Cornish, Irish, Scottish Gaelic and Welsh - are the U.K.'s oldest living languages. To empower their speakers, the UK-LLM sover...
10/09/2025
GeForce NOW Blackwell RTX 5080-class SuperPODs are now rolling out, unlocking a new level of ultra high-performance, cinematic cloud gaming.
GeForce NOW Ultima...
09/09/2025
Inference has emerged as the new frontier of complexity in AI. Modern models are...
09/09/2025
As large language models (LLMs) grow larger, they get smarter, with open models from leading developers now featuring hundreds of billions of parameters. At the...
09/09/2025
At this week's AI Infrastructure Summit in Silicon Valley, NVIDIA's VP o...
09/09/2025
Inference performance is critical, as it directly influences the economics of an AI factory. The higher the throughput of AI factory infrastructure, the more to...
09/09/2025
At this week's IAA Mobility conference in Munich, NVIDIA Vice President of A...
09/09/2025
ComfyUI - an open-source, node-based graphical interface for running and buildin...
04/09/2025
NVIDIA today announced new AI education support for K-12 programs at a White House event to celebrate public-private partnerships that advance artificial intell...
04/09/2025
Editor's note: This post is part of the AI On blog series, which explores the latest techniques and real-world applications of agentic AI, chatbots and copi...
04/09/2025
NVIDIA Blackwell RTX is coming to the cloud on Wednesday, Sept. 10 - an upgrade ...
03/09/2025
3D artists are constantly prototyping.
In traditional workflows, modelers must build placeholder, low-fidelity assets to populate 3D scenes, tinkering and adju...
02/09/2025
For more than a century, meteorologists have chased storms with chalkboards, equ...
28/08/2025
Brace yourself, COGs - the Locusts aren't the only thing rising up. The Coal...
28/08/2025
Last week at Gamescom, NVIDIA announced the winners of the NVIDIA and ModDB RTX ...
27/08/2025
AI models are advancing at a rapid rate and scale.
But what might they lack that (most) humans don't? Common sense: an understanding, developed through rea...
25/08/2025
Robots around the world are about to get a lot smarter as physical AI developers...
25/08/2025
As autonomous vehicle systems rapidly grow in complexity, equipped with reasonin...
22/08/2025
As the latest member of the NVIDIA Blackwell architecture family, the NVIDIA Blackwell Ultra GPU builds on core innovations to accelerate training and AI reason...
22/08/2025
AI reasoning, inference and networking will be top of mind for attendees of next...
21/08/2025
Japan is once again building a landmark high-performance computing system - not ...
21/08/2025
From AI assistants doing deep research to autonomous vehicles making split-second navigation decisions, AI adoption is exploding across industries.
Behind ever...
21/08/2025
Across the globe, AI factories are rising - massive new data centers built not to serve up web pages or email, but to train and deploy intelligence itself. Inte...
21/08/2025
Get a glimpse into the future of gaming.
The NVIDIA Blackwell RTX architecture is coming to GeForce NOW in September, marking the service's biggest upgrade...
20/08/2025
Editor's note: This blog is a part of Into the Omniverse, a series focused o...
18/08/2025
With over 175 games now supporting NVIDIA DLSS 4 - a suite of advanced, AI-power...
18/08/2025
At Gamescom, NVIDIA is releasing its first major update to Project G Assist - an...
15/08/2025
Of around 7,000 languages in the world, a tiny fraction are supported by AI lang...
14/08/2025
NVIDIA is partnering with the U.S. National Science Foundation (NSF) to create a...
14/08/2025
Warhammer 40,000: Dawn of War - Definitive Edition is marching onto GeForce NOW,...
13/08/2025
Black Forest Labs' FLUX.1 Kontext [dev] image editing model is now available as an NVIDIA NIM microservice.
FLUX.1 models allow users to edit existing imag...
11/08/2025
Using NVIDIA digital twin technologies, Amazon Devices & Services is powering bi...
11/08/2025
Packing the power of the NVIDIA Blackwell architecture in compact, energy-effici...
11/08/2025
Physical AI is becoming the foundation of smart cities, facilities and industria...
07/08/2025
This GFN Thursday brings an offer members can't refuse - 2K's highly ant...
05/08/2025
Two new open-weight AI reasoning models from OpenAI released today bring cutting...
05/08/2025
In collaboration with OpenAI, NVIDIA has optimized the company's new open-so...
05/08/2025
NVIDIA and OpenAI began pushing the boundaries of AI with the launch of NVIDIA D...
05/08/2025
NVIDIA GPUs are at the heart of modern computing. They're used across industries - from healthcare and finance to scientific research, autonomous systems an...
31/07/2025
August brings new levels of gaming excitement on GeForce NOW, with 2,300 titles now available to stream in the cloud.
Grab a controller and get ready for epic ...
31/07/2025
Interest in generative AI is continuing to grow, as new models include more capabilities. With the latest advancements, even enthusiasts without a developer bac...
29/07/2025
FourCastNet3 (FCN3) is the latest AI global weather forecasting system from NVID...
28/07/2025
The electrical grid is designed to support loads that are relatively steady, such as lighting, household appliances, and industrial machines that operate at con...
24/07/2025
For media company Black Mixture, AI isn't just a tool - it's an entire p...
24/07/2025
Sharpen the blade and brace for a journey steeped in myth and mystery. WUCHANG: Fallen Feathers has launched in the cloud.
Ride in style with skateboarding leg...
23/07/2025
In today's fast-evolving digital landscape, marketing teams face increasing ...
22/07/2025
Editor's note: This post is part of the AI On blog series, which explores th...
Editor's note: This post is part of the AI Decoded series, which demystifies AI by making the technology more accessible, and showcases new hardware, software, tools and accelerations for GeForce RTX PC and NVIDIA RTX workstation users.
