Rival.io Runs DeepSeek V4 Flash on Eight NVIDIA L4 GPUs

Oct 30, 2026
Rival.io launched GraphX after running the 284 billion parameter DeepSeek V4 Flash model on eight NVIDIA L4 GPUs.

Rival.io launched GraphX after running the 284 billion parameter DeepSeek V4 Flash model on eight NVIDIA L4 GPUs. The company says the configuration served 32 concurrent users at about 10 generated tokens per second for each user without additional quantization, according to its technical write-up.

The eight GPUs provided 192 GB of combined memory, with the model occupying about 83% of each card's memory. The deployment also supported the model's context window and KV cache.

GraphX manages model placement, memory and computation across GPUs. Rival.io says it also deployed the DeepSeek model on eight NVIDIA RTX PRO 6000 GPUs, and that GraphX supports compatible hardware from AMD, Intel and NVIDIA. All figures are the company's own. Rival.io claims GraphX can cut open-weight model hosting costs by up to 95% and that its RTX PRO 6000 setup runs at roughly half the cost per output token of an eight-B200 reference.

We hope you enjoyed this article

Consider subscribing to one of our newsletters like Silicon Brief or Daily AI Brief.

Also, consider following us on social media:

AI Chips & Datacenters AI Brief AI Brief (X)
Free newsletter

Silicon Brief

Weekly coverage of AI hardware developments including chips, GPUs, cloud platforms, and data center technology.

Guide
Private AI chip companies, by latest reported valuation

Private AI chip companies, by latest reported valuation

Private companies building AI compute silicon (the chips that run AI workloads), shown with each company's most recent reported or disclosed valuation. The figures are measured on different dates and bases, so this is a snapshot, not a live ranking or a like-for-like comparison.

Whitepaper

Tensordyne Napier: What If One Rack Could Do the Work of Nine?

Tensordyne

This Tensordyne whitepaper presents Napier, an inference-focused AI processor and rack-scale system based on the company’s TDN Math logarithmic number system. It examines infrastructure requirements for large mixture-of-experts and agentic models, compares major inference architecture approaches, and details the TDN AIP processor, TDN72 pod, TDN Link fabric, and Napier Ultra configuration. The paper reports simulation-based performance, cost, and accuracy-validation results, including Tensordyne’s projected comparison of one Napier rack with a nine-rack Nvidia Rubin plus Groq deployment; the chip is reported as taped out and in fabrication.

Read more

You may also like

Free, six days a week

Daily AI Brief: the AI news that matters, in your inbox.