La boutique est ouverte

Nous construisons les machines Workstation AI pour les sites web, les agents IA et l'entraînement de LLM — configurez-les et achetez-les en ligne.

Obtenir les prix pro
Produits IA
Labs IAAgents OpenAIAgents ClaudeGrok BotCRM natif Workstation AIMarketingTous les produits
Solutions IA
IA appliquéeAgents IAIA générativeAutomatisation par IAMontée en compétence IAOffres IA pour PMELaboratoire IA d'entrepriseStations de travail IAMatériel et logiciels NVIDIACalcul accéléré NVIDIAIA privéeCloud privé GPUClusters GPUIA en périphérieIA par secteur
Services IA
Modernisation des plateformesIngénierie numériqueFondations data et IAOpérations autonomesConseil IAAutomatisation DevOpsCybersécuritéDéveloppement logicielCréation d'agentsMise en place MLOpsLocation de GPUCloud privé GPU
À propos
EntrepriseNotre histoirePartenairesTémoignages clients
Centres de données IA
Centres de données IACloud privé GPUClusters GPULocation de GPU
Analyses IASérie JevLabo de benchmarksChoix de modèles
Notes techniques
Documentation
OS LinuxUbuntuRed Hat Enterprise LinuxSUSE LinuxAlpine LinuxDebian, Rocky & AlmaLinux
Plateformes cloud d'IAAWS & Amazon BedrockGoogle Cloud & Vertex AIMicrosoft Azure & AI FoundryClouds et places de marché GPU
Centres de données IA privésPour les cabinets d'avocatsPour les cabinets comptablesPlateformes sur siteCloud entièrement sécurisé
OutilsRing PromoterJobshout
Actualités
Blog
Nous contacterSe connecter

Stations de travail IA, logiciels multi-agents IA, infrastructure GPU et solutions d'agents intelligents pour les entreprises modernes.

Nous contacter

Solutions IA

IA appliquéeAgents IAIA générativeAutomatisation par IAMontée en compétence IAOffres IA pour PMELaboratoire IA d'entrepriseIA par secteur

Produits

Tous les produitsCRM & ERP natifs Workstation AIMarketingAgents OpenAIRing PromoterJobshout

Entreprise

À proposEntrepriseCentres de données IAPourquoi WorkstationPartenairesTémoignages clientsTarifsContactTenthMatrix

Ressources

Analyses IANotes techniquesDocumentationOS LinuxPlateformes cloud d'IACentres de données IA privésBlogRechercherPlan du site
Bureau de Belgique
Workstation SRL, Rue Vanderkindere 34, 1180 Uccle, BrusselsBE 0751.518.683Lun - Ven : 9h00 - 18h00 CET
+32 492 45 67 46
Siège social au Royaume-Uni
77-79 Marlowes, Hemel Hempstead HP1 1LFItinéraire : prenez la sortie 20 de la M25, Outer LondonN° d'entreprise: 11641870Lun - Ven : 9h00 - 18h00 GMT
+44 7515 356 146
Bureau de l'Inde
#159 Sector 9, Pocket 1, DDA Flats, 110077 Dwarka, New Delhi
+91 98881 98841
Bureau de Taïwan
3F, No. 2-4, Aisi Rd., Ren'ai Dist., Keelung City 200, Taiwan基隆市仁愛區愛四路2之4號3樓Lun - Ven : 9 h 00 - 18 h 00, heure de Taïwan (GMT+8)
+886 926 892 701+32 492 45 67 46
Bureau en Estonie
Aleksander Puškini tn 30-25, 20306 Narva, EstoniaLun - Ven : 9 h 00 - 18 h 00, heure d’Estonie (EET)
+32 492 45 67 46

© 2026 Workstation AI. Tous droits réservés.

ConfidentialitéSuppression des donnéesCookiesConditions d'utilisationPlan du site web

NVIDIA, le logo NVIDIA, RTX, RTX PRO, Blackwell, CUDA, NVLink, MGX, Omniverse, Spectrum-X, Quantum, BlueField, DGX et HGX sont des marques commerciales et/ou des marques déposées de NVIDIA Corporation aux États-Unis et dans d'autres pays. AMD, le logo AMD Arrow, EPYC, Instinct, Ryzen et Threadripper sont des marques commerciales d'Advanced Micro Devices, Inc. Dell et PowerEdge sont des marques commerciales de Dell Inc. ou de ses filiales. Intel et Xeon sont des marques commerciales d'Intel Corporation ou de ses filiales. AMAX, AceleMax et LiquidMax sont des marques commerciales d'AMAX Engineering Corporation. Tous les autres noms de produits, logos et marques appartiennent à leurs propriétaires respectifs. Nous les utilisons uniquement pour identifier les produits que nous décrivons, fournissons ou prenons en charge ; leur utilisation ne signifie pas que leurs propriétaires nous cautionnent.

Loading blog...

Accueil / Blogs
Engineering

LLM Inference Boosts 4-5x with NVIDIA Grace

JobShout AI Blog Writer Agent (opens in a new window)19 novembre 20266 min read
03529290 2598 4fa0 8c37 8311c61c219f

Desktop-scale clusters are moving from prototype curiosity to operational reality for manufacturing, supermarkets, banking, education, healthtech, fintech and tax returns.

The NVIDIA Grace Blackwell GB10 Superchip is positioned as the desktop anchor for this shift. The GB10 Superchip delivers up to 1 petaflop of AI performance at FP4 precision [1]. Each Project DIGITS features 128GB of unified, coherent memory and up to 4TB of NVMe storage [1]. With the supercomputer, developers can run up to 200-billion-parameter large language models to supercharge AI innovation [1].

Challenges and Opportunities

For regulated and operationally constrained industries, the challenge is not just model size. It is memory, energy and consistency at the edge of the plant, branch or campus.

Large language models exceed single-GPU memory, so inference requires either model parallelism, offloading, or compression. FlexGen can be flexibly configured under various hardware resource constraints by aggregating memory and computation from the GPU, CPU, and disk [2]. FlexGen further compresses the weights and the attention cache to 4 bits with negligible accuracy loss [2].

Methods collectively enable running models up to twice the size of the available DRAM, with a 4-5x and 20-25x increase in inference speed compared to naive loading approaches in CPU and GPU, respectively [3]. For a supermarket chain running private product description generation or a bank drafting compliant explanations, that speedup turns a batch overnight job into an interactive tool.

Energy and latency trade-offs remain. One Spark uses less GPU energy per output token at matched concurrency [4]. That matters in manufacturing where floor systems run 24/7 and in healthtech where power budgets are tied to clinical facilities.

Benchmarking Process and Results

Benchmarking desktop inference must mirror production request patterns, not just peak tokens per second.

A typical evaluation flow can be modeled as:

Diagram

TP=2 delivers a median 1.59x peak-throughput gain over one Spark, versus 1.39x for PP=2 [4]. TP=2 also reduces unloaded inter-token latency to 0.52x to 0.73x of one Spark, whereas PP=2 provides little reduction [4]. TP=2 also achieves the highest goodput under a strict 100 ms inter-token latency objective wherever both strategies are supported and the objective is met [4].

These results are relevant for sectors with strict service-level expectations. A fintech fraud review assistant or an education tutoring bot needs consistent inter-token latency, not just average throughput. In contrast, a nightly tax document summarization batch may prioritize throughput and energy.

Strategy Peak Throughput Gain vs 1 Spark Unloaded Inter-Token Latency Goodput at 100 ms latency objective Energy per token at matched concurrency
TP=2 median 1.59x [4] 0.52x to 0.73x [4] highest where supported [4] higher than one Spark [4]
PP=2 median 1.39x [4] little reduction [4] lower than TP=2 [4] —
One Spark baseline baseline baseline less GPU energy per token [4]

FlexGen’s own benchmarking demonstrates practical limits. When running OPT-175B on a single 16GB GPU, FlexGen achieves significantly higher throughput compared to state-of-the-art offloading systems, reaching a generation throughput of 1 token/s for the first time with an effective batch size of 144 [2]. On the HELM benchmark, FlexGen can benchmark a 30B model with a 16GB GPU on 7 representative sub-scenarios in 21 hours [2].

FlexGen: Flexible Configuration Framework

FlexGen is presented as a high-throughput generation engine for running LLMs with limited GPU memory [2]. The key claim for desktop operators is flexibility: it can be flexibly configured under various hardware resource constraints by aggregating memory and computation from the GPU, CPU, and disk [2].

Compression is central to that flexibility. FlexGen compresses the weights and the attention cache to 4 bits with negligible accuracy loss [2]. For a healthtech deployment handling de-identified notes, or a bank running on-prem policy QA, the ability to fit larger models into 128GB unified memory without cloud egress is operationally decisive.

Throughput gains are measurable. FlexGen achieves a generation throughput of 1 token/s for the first time with an effective batch size of 144 when running OPT-175B on a single 16GB GPU [2]. Benchmarking speed also improves: FlexGen can benchmark a 30B model with a 16GB GPU on 7 representative sub-scenarios in 21 hours on the HELM benchmark [2].

TensorRT-LLM: Optimized Large Language Model Inference

TensorRT-LLM is the NVIDIA software stack for production inference. TensorRT-LLM includes fully optimized, ready-to-run versions of many LLMs widely used in production today [5].

In-flight batching enables improved GPU usage and minimally doubles the throughput on a benchmark of real-world LLM requests on NVIDIA H100 Tensor Core GPUs [5]. FP8 quantization retains higher accuracy compared to other data formats like INT8 or INT4 while achieving the fastest performance and offering the simplest implementation [5].

A minimal Python invocation pattern looks like:

import tensorrt_llm
llm = tensorrt_llm.LLM(model_name="meta-llama/Llama-3-8B")
output = llm.generate("Summarize this production log:", max_tokens=256)

The ready-to-run models and Python API lower the barrier for teams in manufacturing and education who lack dedicated ML infrastructure engineers.

Implications for Developers

For developers in manufacturing, supermarkets, banking, education, healthtech, fintech and tax returns, desktop-scale clusters change the deployment calculus.

First, privacy and compliance become easier. With 128GB unified coherent memory and up to 4TB NVMe per Project DIGITS unit [1], sensitive data can stay on-prem. That is critical for healthtech clinical documentation and tax return processing where data residency is non-negotiable.

Second, iteration speed improves. The 4-5x inference speed increase compared to naive loading approaches in CPU [3] shortens experiment cycles for prompt tuning and evaluation. A supermarket merchandising team can test dozens of product copy variants in an afternoon instead of overnight.

Third, trade-offs must be explicit. TP=2 improves latency and goodput under a 100 ms inter-token objective [4] but One Spark uses less GPU energy per output token at matched concurrency [4]. For a 24/7 manufacturing copilot, energy may dominate; for an interactive banking chatbot, latency dominates.

Developers should benchmark with their own request mix, measure goodput under latency SLOs, and evaluate energy per token. Desktop Grace Blackwell hardware gives them a realistic path to run up to 200-billion-parameter models locally [1] without rebuilding data pipelines for the cloud.

The net effect is practical: LLMs become a deployable component rather than a research experiment, with measurable throughput, latency and cost metrics that map directly to operational outcomes in regulated industries.

References

  1. NVIDIA Puts Grace Blackwell on Every Desk and at Every AI Developer’s Fingertips — nvidianews.nvidia.com (7 January 2025)
  2. [2303.06865] FlexGen: High-Throughput Generative Inference of Large Language Models with a Single GPU — arxiv.org (14 March 2023)
  3. LLM in a flash: Efficient Large Language Model Inference with Limited Memory — arxiv.org (24 January 2024)
  4. Benchmarking LLM Inference on Desktop-Scale NVIDIA Grace Blackwell Clusters — arxiv.org
  5. NVIDIA TensorRT-LLM Supercharges Large Language Model Inference on NVIDIA H100 GPUs — developer.nvidia.com (8 September 2023)