AI COMPUTE•NVIDIA•GPU CLOUD•DATA CENTERS•SEMICONDUCTORS•ENERGY•FUNDING•M&A•AI COMPUTE•NVIDIA•GPU CLOUD•DATA CENTERS•SEMICONDUCTORS•ENERGY•FUNDING•M&A•
GPU·DATA HUB
Newsletter
GPUs

Rival.io Launches GraphX, Running DeepSeek V4 Flash on Just 8 NVIDIA L4 GPUs

Rival.io unveils GraphX kernel-optimisation platform to minimise AI inference hardware footprints

By GPU Data Hub Desk · GPU Data Hub·Editorial standards·
Rival.io Launches GraphX, Running DeepSeek V4 Flash on Just 8 NVIDIA L4 GPUs

Rival.io has launched GraphX, a proprietary framework focused on kernel optimisation and silicon placement to curtail the expenses associated with private artificial intelligence. To highlight the efficiency of the software, the developer successfully hosted DeepSeek V4 Flash on an array of just eight Nvidia L4 graphics processors.

Original source

This summary was written by the GPU Data Hub desk from reporting published by HPCwire on 29 Oct 2026, 18:28. Read the full article at the original publisher.

Read Original Source →

By subscribing you consent to receive the daily briefing. Unsubscribe at any time. See our privacy policy.

LinkedInX

More from GPU Data Hub

Latest →

Nvidia GeForce RTX 5080 prices hit $2,000, costing more than the RTX 5090 did at launch and with limited stock available — original $999 MSRP long forgotten, with prices now 2-3x higher

Street prices for Nvidia's GeForce RTX 5080 have reached $2,000, representing between two and three times the component's original recommended retail price of $999. The surge has pushed the graphics card above the debut cost of the flagship RTX 5090 amid scarce retailer inventory. Persistent inventory shortages continue to destabilise retail hardware markets for high-performance graphics processors.

GPUs··Tom's Hardware

HPC Data Infrastructure Wasn’t Built for HPC + AI

Established scientific research architectures are facing operational challenges as data centres incorporate GPU acceleration into legacy supercomputing environments. Traditional parallel file networks and long-term storage repositories, designed originally for classical modeling, must now adapt to handle the contrasting throughput requirements of modern artificial intelligence pipelines.

GPUs··HPCwire

Vertical DRAM Stacking Expands Thermal Design Space Beyond HBM-on-GPU (imec)

Research institute imec has analysed a novel three-dimensional memory layout that positions vertically oriented DRAM dies directly over a graphics processor. Thermal simulations indicate that integrating cooling channels within the structure provides a viable alternative to standard high-bandwidth memory packaging. The work evaluates how stack dimensions and memory density influence overall heat dissipation.

GPUs··Semiconductor Engineering

Nvidia discusses $2,000-per-GPU annual fee for software that 'runs agents on AI factories'

Nvidia is weighing the introduction of an annual software subscription priced at approximately $2,000 per processor to run autonomous workflows within enterprise clusters. Should the firm secure five million enterprise licences, the recurring billing structure could generate an estimated $10bn in revenue each year. While specific launch schedules and commercial terms remain under development, the chipmaker continues to bolster its corporate software and operational security suite.

GPUs··Tom's Hardware