Rival.io launched GraphX on October 29, 2026, its proprietary kernel-optimization and silicon-placement technology, and announced DeepSeek V4 Flash running on just eight NVIDIA L4 GPUs. The launch and the demonstration disclosed by the company are classified here as CONFIRMED; the announcement was distributed by ACCESS Newswire and republished by HPCwire.
According to Rival.io, GraphX optimizes how models are placed across GPUs and how they use memory and computing power. In the announced configuration, the eight L4 cards combine for 192 GB of GPU memory and support the model's context window and its associated KV cache — a setup the company says previously required more specialized, expensive hardware.
The company also says it has run DeepSeek V4 Flash on eight NVIDIA RTX 6000 GPUs with what it describes as full enterprise-grade performance and high concurrency. On costs, Rival.io states that, depending on the model and workload, GraphX can cut open-weight model hosting costs by up to 95% compared with equivalent internally managed deployments or hyperscale cloud providers. That percentage is a company STATEMENT, not an independently verified result in this article.
GraphX is presented as compatible with AMD, Intel and NVIDIA hardware and with DeepSeek, GLM and Kimi models. Supported models can be deployed in a customer's own cloud or data center, or on dedicated Rival.io infrastructure, in isolated environments with their own secure endpoints and access credentials.
For Geekiki readers, the practical point is that software optimization — model placement, memory management and computation execution — can expand what already-available hardware can deliver, without relying only on more expensive premium GPUs. Independent benchmarks and public technical details that would allow an apples-to-apples comparison with other configurations are still missing, however.
Image credit: NVIDIA, official L4 GPU page.
Sources: ACCESS Newswire — Rival.io Launches GraphX and HPCwire — Rival.io Launches GraphX. GraphX is available on demand at Rival.io/graph-x, according to the announcement.
Transparency: this article was produced with the assistance of artificial intelligence, based on cross-checking public posts and the internet sources cited below. Readers are advised to consult the original sources for full details.
Muse
