We’re heading to NYC with our friends at @VAST_Data!
Our VP of Engineering, Nick Barcet, will join VAST’s team to talk KV cache at scale and the challenges behind production AI inference.
📆Dec 9, 5:30-8:30 PM
📍VAST Data NYC Office
Request to join: na2.hubs.ly/H0911xn0
Running LMCache across a fleet? Each KV cache server only knows about itself, so the same prompt gets prefilled again on every node.
The LMCache Coordinator adds a control plane on top: 𝐨𝐧𝐞 𝐊𝐕 𝐜𝐚𝐜𝐡𝐞 𝐟𝐨𝐫 𝐭𝐡𝐞 𝐰𝐡𝐨𝐥𝐞 𝐟𝐥𝐞𝐞𝐭.
It’s the cross-node layer in Tensormesh Platform, our self-hosted KV caching for vLLM on Kubernetes 👇
na2.hubs.ly/H08-ygm0
We're at #KubeCon NA! 👋
Come say hi at 𝐁𝐨𝐨𝐭𝐡 𝟑𝟒𝟑 in the Solutions Showcase and see how to reuse more of the GPU work you've already paid for.
Wed, Nov 11: 10:15 AM–5 PM
Thu, Nov 12: 10:15 AM–2 PM
Grab a hat while they last 🧢 #CloudNativeCon #AIInference
Prefix caching stops at the first change in your prompt. RAG and agents reorder the same docs on every call.
𝐂𝐚𝐜𝐡𝐞𝐁𝐥𝐞𝐧𝐝 reuses KV cache anywhere in the prompt:
- 2.6x more traffic, same GPUs
- Up to 2.9x faster TTFT
- ~$2.7M/yr saved on 256 GPUs
na2.hubs.ly/H08VHzK0 @CloudDude_
Today at #KubeCon ⏰ 1:30 PM · @SUSE booth #107
Sandro Mazziotta on sovereign AI inference at a predictable cost. 𝟒𝟓 𝐦𝐢𝐧𝐮𝐭𝐞𝐬, 𝐛𝐫𝐢𝐧𝐠 𝐲𝐨𝐮𝐫 𝐪𝐮𝐞𝐬𝐭𝐢𝐨𝐧𝐬.
#SUSE #SUSEAIFactorywithNVIDIA
1
73
Landed in Salt Lake City for #KubeCon 👋
Tomorrow at 1:30 PM: Sandro Mazziotta at the @SUSE booth #107 on one simple idea: stop recomputing what your GPUs already know.
+𝐂𝐨𝐦𝐞 𝐟𝐢𝐧𝐝 𝐮𝐬 𝐚𝐭 𝐛𝐨𝐨𝐭𝐡 #𝟑𝟒𝟑 𝐚𝐥𝐥 𝐰𝐞𝐞𝐤.
#SUSE #SUSEAIFactorywithNVIDIA
Next Tuesday at #KubeCon, Sandro Mazziotta, our CPO, will share:
→ Where production inference wastes GPU cycles → How sovereign AI teams get cost under control → Up to 4x faster time to first token
@SUSE booth #107 · Nov 10 · 1:30 PM
#SUSEAIFactorywithNVIDIA
Sovereign AI shouldn't mean unpredictable costs. @SUSE secures the AI platform. Tensormesh makes inference on it faster and cost-efficient.
See how at SUSE booth #107 at #KubeCon — Tue, Nov 10, 1:30 PM.
#SUSE #SUSEAIFactorywithNVIDIA
Running LLMs at scale? Your GPUs are probably doing the same work twice.
Same prompts, same docs, same context — 𝐫𝐞𝐜𝐨𝐦𝐩𝐮𝐭𝐞𝐝 𝐨𝐧 𝐞𝐯𝐞𝐫𝐲 𝐫𝐞𝐪𝐮𝐞𝐬𝐭.
Our CPO, Sandro Mazziotta, will show how to stop paying for it at @SUSE booth #107, Tue Nov 10, 1:30 PM. Can't make this time?
Stop by our dedicated Booth #343 and chat with our team!
#KubeCon #SUSE #SUSEAIFactorywithNVIDIA
62
We're joining @SUSE at #KubeCon in Salt Lake City! 🏔️ Our CPO Sandro Mazziotta takes the SUSE booth stage Tue, Nov 10 at 1:30 PM to answer one question: How do you scale LLM inference without scaling your GPU bill? 📍 SUSE Booth #107 #SUSEAIFactorywithNVIDIA
1
4
136
Tensormesh is heading to @KubeCon_ + CloudNativeCon NA 2026!
We’ll be in Salt Lake City Nov. 9–12 talking production AI inference, context caching, and how to manage reusable model state beyond GPU memory.
Booth #343
#KubeCon #CloudNativeCon
3
131
[𝗣𝗮𝗽𝗲𝗿 𝗔𝗹𝗲𝗿𝘁] 🧙 Gandalf: Flash Shall Not Die!
KV cache has (quite literally) created an existential problem for flash storage. When SSDs are used to offload KV cache, they can wear out in just a few WEEKS, because KV cache is huge and generated nonstop.
Imagine having to put a new SSD in your MacBook every few weeks. That's what is happening in LLM inference systems today.
Researchers and engineers at Tensormesh and @UChicago show that SSD lifetime can be extended 3–7x with software-only changes, no new hardware needed. The key insight: exploit the intrinsic structure of KV cache in modern LLMs.
Read the paper: na2.hubs.ly/H08zNcv0
CacheBlend reuses KV cache anywhere in the prompt, not just shared prefixes: 𝟮.𝟱𝘅 𝗳𝗮𝘀𝘁𝗲𝗿 𝘁𝗵𝗮𝗻 𝗽𝗿𝗲𝗳𝗶𝘅 𝗰𝗮𝗰𝗵𝗶𝗻𝗴 𝗼𝗻 𝗥𝗔𝗚.
So what happens when that cache moves off RAM onto IBM Storage Scale? @IBM @nvidia @Supermicro
🔗👇
na2.hubs.ly/H08zm740
Missed our VP of Engineering, Nick Barcet at #𝗡𝗩𝗜𝗗𝗜𝗔𝗚𝗧𝗖 𝗕𝗲𝗿𝗹𝗶𝗻? Here’s the short version.
Tensormesh Platform adds a persistent, shared KV cache to vLLM on @SUSE AI Factory with @NVIDIA, so your GPUs stop recomputing context.
#SUSEAIFactorywithNVIDIA
Full blog here 👇 na2.hubs.ly/H08xjSJ0
4
136
Live today at #GTCBerlin! 🎉 Nick Barcet presents 𝐓𝐞𝐧𝐬𝐨𝐫𝐦𝐞𝐬𝐡 𝐏𝐥𝐚𝐭𝐟𝐨𝐫𝐦 on SUSE AI Factory with Nvidia at 3:30pm CEST. Find us at @SUSE Booth #3053!
𝙎𝙚𝙚 𝙮𝙤𝙪 𝙩𝙤𝙢𝙤𝙧𝙧𝙤𝙬, 𝘽𝙚𝙧𝙡𝙞𝙣! 👋 Catch Nick Barcet presenting Tensormesh Platform on SUSE AI Factory with Nvidia at @SUSE Booth #3053, 3:30pm CEST. #GTCBerlin #KVCache
82












