Updating my vLLM with new Qwen3.5. It is down for less than an hour and OpenClaw is already complaining about not having tokens from my Gemini Pro subscription. Hungry bastard.
Updating my vLLM with new Qwen3.5. It is down for less than an hour and OpenClaw is already complaining about not having tokens from my Gemini Pro subscription. Hungry bastard.
Week 1 with my new AI Rig
blog.matyasprokop.com/ai/rig/-/wee...
This talk www.youtube.com/watch?v=_c9C... reminded me @bcantrill.bsky.social for two reasons:
1) @oxide.computer offices are across the street
2) Lawrence Levy wrote great book about his journey in Pixar which I'm not sure has been mentioned in any of his podcasts.
The eagle has landed. Little bit more info on my new AI rig.
www.matyasprokop.com/ai/linux/hom...
Finally was able to finish the base built for my AI rig this weekend. Installed Proxmox, installed couple of VMs and was able to migrate my web site from AWS to this beauty. My blog is now running from my living room. Time to shut down my AWS environment.
AI agents and more new benchmarks
blog.matyasprokop.com/llm/gaia/oss...
Met with Positron today. New AI inferencing company. Few notes on them. I will follow up with more detailed write up in few weeks.
blog.matyasprokop.com/ai/inferenci...
A look at the impressive capabilities of the Unitree G1 robot, and some thoughts on the exciting convergence of AI and hardware in the field of robotics.
Llm-d team released new post which is focusing on intelligent inference serving and how LLM is different from stateless web requests. Worth to read.
blog.matyasprokop.com/ai/linux/llm...
Does GPU passthrough has any impact on performance? Looks like it does but it is negligible.
My home AI rig AKA great data repatriation project is nicely starting to take some shape
blog.matyasprokop.com/articles/202...
Nokia becoming new challenger in DC space
#TFDx Biggest launch since Cisco ACI in Cisco Datacenter networking. Cisco HyperFabric presentation now. "There is no IPv4 traffic in fabric unless you configure IPv4 SVI" - This is probably the first IPv6 native fabric in history of Cisco. Brave.
Is it 2019 again?
#TFDx Cisco Silicon One is arriving into SP world. We will see simplification in the SP portfolio but we will see some trade offs in locking you in more into their ecosystem. Cisco is becoming vertically integrated company like Apple in phone and computer world.
It feels like Intel still don't have the focus. If they plan to beat AMD in CPU and Nvidia in GPU they need laser focus on what is their niche. Like every small company trying to beat big guys you have to understand what is your niche. Intel lost it in all those years of dominance.
2) GPUs - Falcon Shores cancelled (I assume they wouldn't be competitive) and focus is now on Jaguar. Those fast cycles in GPU generations are just too fast for Intel. And based on the parameters Intel assumes they will compete with H100 in 2-3 years? Nvidia will be at that time 2 generations ahead.
Yes there are some positive signs however - 1) CPUs - still very much behind in terms of amounts of cores per CPU. Diamond Rapids with 182 cores pushed out behind 2026(AMD has 192 cores in 2024). I know cores is not everything but for most of enterprises amount of cores is THE number they decide on.
I'm back at #CiscoLiveEMEA and spent first couple of hours with #intel product team.
I listen barely any podcasts but I try never miss your new episodes. Finally managed to listen this episode and it is one of the best ones. So much goodness there.
Just got off an hour internal call with engineering, where we have discussed tools like Microsoft Copilot, agents, Cursor, Perplexity, o3-mini, DeepSeek, and how each of us is leveraging them for different tasks. This transformation in our field is already happening.
It teaches also one important lesson which I very often remind to my team: "Necessity is the mother of innovation". That's important not just for your career but for life in general.
This is interesting. Perplexity Pro basically offers either 10x ChatGPT-o1 or unlimited DeepSeek R1 requests. If they are both comparable in terms of quality (and based on my brief testing they are) this is no brainer and I'm sure that's where this is heading. Race to the bottom starts this week.
Spinning off venture capital arm and also spinning off large innovation incubator. This very short sighted and will hit Intel in longterm.
I just published The Data Dash #2: AI New Frontiers, Zucky and bananas. Check it out here newsletter.matyasprokop.com/posts/the-da...
My returns to Prague always starts with proper lunch. Good to be back home.
I'm launching my newsletter: newsletter.matyasprokop.com
Each month, I'll dive into interesting developments in AI, semiconductors, datacenters, open source, and the cloud. A curated list of 10 short articles or topics with quick summaries, 2-3 longer articles for a deeper dive and short essay.
Since we are on Bluesky I would really recommend to read this book. Fantastic read not just about takeover of Twitter by Elon but also beginnings of Bluesky.
Who did this?!
βThis shouldnβt have work! I donβt understand!β π Max from Pure trying to break FlashArray with pulling 3 drives, NVRAM module and single power supply. Unsuccessfully. #puretechnical