Asking an Encyclopedia-Sized Question: How To Make the World Smarter with Multi-Million Token Real-Time Inference

Modern AI applications increasingly rely on models that combine huge parameter counts with multi-million-token context windows. Whether it is AI agents following months of conversation, legal assistants reasoning through gigabytes of case law as big as an entire encyclopedia set, or coding copilots navigating sprawling repositories, preserving long-range context is essential for relevance and coherence. On top of that, users expect fast, interactive responses.

The growing demand to decode such massive amounts of data-and let multiple GPUs quickly scale and communicate with each other-underscores the importance of FP4 compute and the high-bandwidth large NVLink domain provided by NVIDIA Blackwell systems. Helix Parallelism, introduced in this blog, is co-designed with Blackwell. It enables up to a 32x increase in the number of concurrent users at a given latency, compared to the best known prior parallelism methods for real-time decoding with ultra-long context.

In other words, it lets AI agents and virtual assistants serve more people, faster than ever before.

(Note: Context in this blog refers to the sequence of previously generated tokens, whose intermediate key and value representations are stored as KV cache and accessed at every decoding step.)

Decoding bottlenecks: KV cache and FFN weight reads To support real-time decoding at scale, a system must overcome two major bottlenecks during the decoding (aka generation) phase:

Key-Value (KV) cache streaming: When handling multi-million-token contexts, each GPU must read a massive history of past tokens (KV cache) from DRAM per sample. This constant streaming can, in turn, saturate DRAM bandwidth, increase token-to-token latency (TTL), and quickly become a major bottleneck as context length grows.

Feed-Forward Network (FFN) weight loading: During autoregressive decoding, generating every new token requires loading large Feed-Forward Network (FFN) weights from DRAM. In low latency scenarios with small batch sizes, this memory access cost is not well amortized, making FFN weight reads a dominant source of latency.

These two bottlenecks, KV cache streaming and FFN weight loading, are difficult to optimize simultaneously using traditional parallelism strategies.

Let's take Tensor Parallelism (TP) as an example: Increasing TP can help reduce FFN stalls by distributing weight loading across multiple GPUs and improving TTL, but only up to a point. In attention schemes like Grouped Query Attention (GQA)-used in Llama models-or Multi-Latent Attention (MLA)-found in DeepSeek models-multiple query heads share a limited number of KV heads. As illustrated in Figure 2(c), when TP exceeds the number of KV heads, the system ends up duplicating the multi-million-token KV cache per sample across GPUs for self-attention. As a result, KV read volume stays high even with increased TP, once again saturating DRAM bandwidth and limiting scalability. In the case of MLA, the upper limit for TP is just one to avoid duplication of KV cache.

So how can developers scale both model size and context length without sacrificing real-time interactivity? Helix Parallelism offers a path forward.

Helix execution flow Helix is a hybrid sharding strategy that disaggregates the parallelism strategies of attention and FFNs in a temporal pipeline, effectively addressing both KV cache and FFN weight-read bottlenecks during multi-million-token decoding.

Figure 1 (below) shows how Helix orchestrates the execution of attention and FFN within a single transformer layer. Inspired by the structure of a DNA helix, Helix interweaves multiple dimensions of parallelism-KV, tensor, and expert-into a unified execution loop. By decoupling the parallelism strategy used for attention and FFN, Helix allows each stage to operate in a configuration tuned to its own bottleneck, all while reusing the same pool of GPUs. Helix reuse approach keeps GPUs efficiently utilized across stages, eliminating idle time as computation flows through the model.

data-src=https://developer-blogs.nvidia.com/wp-content/uploads/2025/07/image2-2-png.webp alt=A diagram showing the execution flow of Helix Parallelism. Helix reuses the same pool of N GPUs per layer by switching between N=KVPxTPA during attention and N=TPFxEP during FFN. class=lazyload wp-image-102939 data-srcset=https://developer-blogs.nvidia.com/wp-content/uploads/2025/07/image2-2-png.webp 1522w, https://developer-blogs.nvidia.com/wp-content/uploads/2025/07/image2-2-282x300-png.webp 282w, https://developer-blogs.nvidia.com/wp-content/uploads/2025/07/image2-2-625x665-png.webp 625w, https://developer-blogs.nvidia.com/wp-content/uploads/2025/07/image2-2-108x115-png.webp 108w, https://developer-blogs.nvidia.com/wp-content/uploads/2025/07/image2-2-768x817-png.webp 768w, https://developer-blogs.nvidia.com/wp-content/uploads/2025/07/image2-2-1443x1536-png.webp 1443w, https://developer-blogs.nvidia.com/wp-content/uploads/2025/07/image2-2-645x687-png.webp 645w, https://developer-blogs.nvidia.com/wp-content/uploads/2025/07/image2-2-85x90-png.webp 85w, https://developer-blogs.nvidia.com/wp-content/uploads/2025/07/image2-2-362x385-png.webp 362w, https://developer-blogs.nvidia.com/wp-content/uploads/2025/07/image2-2-103x110-png.webp 103w, https://developer-blogs.nvidia.com/wp-content/uploads/2025/07/image2-2-1024x1090-png.webp 1024w, https://developer-blogs.nvidia.com/wp-content/uploads/2025/07/image2-2-507x540-png.webp 507w data-sizes=(max-width: 1522px) 100vw, 1522px />Figure 1. Execution flow of Helix Parallelism. Helix reuses the same pool of N GPUs per layer by switching between N=KVPxTPA during attention and N=TPFxEP during FFN.

Attention phase Helix applies KV Parallelism (KVP) by sharding the multi-million-token KV cache along the sequence dimension across KVP GPUs, while applying Tensor Parallelism across attention heads (TPA), where TP

LINK:	https://developer.nvidia.com/blog/asking-an-encyclopedia-sized-questio...
	See more stories from nvidia

North America Stories