XM-Sparse pushes sparse attention throughput boundaries using X-Mem
11/04/2026 — 11/06, 12:16·1 sources (all)·1 reports (all)
Key takeaways
- XM-Sparse integrates Netpreme X-Mem with SGLang HiSparse, using high-bandwidth offloaded KV cache to reduce decoding latency while expanding…
Background & analysis
On November 4, 2026 at 08:00, LMSYS / SGLang announced XM-Sparse, a project that brings hardware memory into the serving stack for sparse attention workloads.
XM-Sparse integrates Netpreme X-Mem™ with SGLang HiSparse. According to the announcement material, the pairing targets the throughput limits of sparse attention inference. High-bandwidth access to offloaded KV caches reduces decode latency, while the added capacity from X-Mem supports larger batches and higher concurrency. In effect, the hardware component handles the memory side of KV cache management, while SGLang HiSparse provides the sparse attention serving logic.
That is the full extent of what the available source material covers. The material does not describe benchmark results, performance figures, deployment guidance, availability or pricing, and no follow-up announcements have been published since.
AI-generated from 1 reports · updated 1 hour ago
- Heat index
- 0
- All sources
- 1
- All reports
- 1
- First seen
- 2 days ago
Reports on this story headlines open the original
XM-Sparse integrates Netpreme X-Mem with SGLang HiSparse. High-bandwidth access to offloaded KV caches reduces decode latency, while the added capacity supports larger batches and higher concurrency.
LMSYS / SGLangFirst-partyAI score 56
Other stories people are talking about
- Heat index 1182RisingOpenAI rolls out GPT-6 Cyber to Daybreak Red users with API, Codex access15 sources
- Heat index 906SurgeGemini Omni 1.2 Flash enters Gemini API with video clip extension to 30 seconds11 sources
- Heat index 685RisingAnthropic launches Claude Personal Finance with read-only Plaid bank links9 sources
- Heat index 612OpenAI ships Images 2.6 with GIF and SVG output14 sources
- Heat index 586Tencent unveils Hy4 multimodal model and announces full open-sourcing13 sources
How is heat calculated?About the methodHide
Heat counts how many independent sources covered a story in the last 48 hours: one source counts once no matter how many posts it published, decaying with a 24-hour half-life. What ranks first is what many people are talking about.
This page aggregates public feeds. Headlines and summaries are machine-organized and remain the property of the original authors; verify important facts at the source. If you believe a headline or summary infringes your rights, email the contact address in the footer with the page URL and basis of your claim, and we will remove or replace it after review.
- Surge
- Discussion rising fast
- New
- First report within 6 hours
- Rising
- Still gathering discussion