James T

Chronara AI Announces Infrastructure Milestone with Ascent GX10 Integration. London, UK — Chronara AI, the leader in decentralized quantum networking and airgapped wallet technology, today announced a major upgrade to its internal processing capabilities with the deployment of the ASUS Ascent GX10 personal supercomputer for inferencing, faster pre-loading and IP protection. This move marks a shift toward localized, high-performance AI operations, allowing Chronara to refine its proprietary models with unprecedented speed and security, getting ready for its claims on global intellectual property submissions. The "Desktop Supercomputer" Advantage: By moving away from cloud-dependency for core development, Chronara AI is leveraging the GX10’s NVIDIA GB10 Grace Blackwell Superchip to achieve: * 2 Petaflop of AI Performance: Massive local compute for real-time model fine-tuning. In addition to our Mac Mini and Studio cluster of 160GB Unified memory. * 256GB Unified Memory: Seamless handling of large language models (LLMs) up to 200B parameters. * Enhanced Data Privacy: By processing sensitive data on-premise, Chronara strengthens its commitment to user security and decentralized integrity. "Our goal has always been to redefine safety and performance in the Web3 space," said James, Founder of Chronara. "Integrating the Ascent GX10 allows our AI agents to learn and adapt faster than ever before, ensuring our community stays ahead of market shifts with local-speed intelligence." This upgrade paves the way for the next phase of the Chronara roadmap, including quantum-encrypted personal portals and real-time AI behavioral monitoring to run just in time flash bots and finally to maximise Nvidia inception backing and credits. About Chronara AI Chronara is a provider of decentralized pico level timing & entanglement based quantum networking and airgapped wallet technology. Founded in 2025, the company focuses on building a secure, AI-driven future for decentralized finance and digital identity. Media Contact: Info@chronara.io More info below: chronara.io/news/ascent-gx… # #

@KislayParashar1 @__tinygrad__ Agreed I'm keen to see where we get too

@JamesTervit @__tinygrad__ That’s a fun mix. Curious how Tinygrad behaves across Metal vs CUDA. eGPU always sounds great until bandwidth kicks in, would be interesting to see real gap vs native GPU.

I will get stuck into tuning and testing Tinygrad Mac M3 and RTX Pro 6000 Workstation 300w edition and a Razer eGPU TB5 edition. @__tinygrad__ Lets see how we go.

@DegenApeDev @__tinygrad__ its wild how AI generates crazy images like thta now aam enthused to make my desk look as good hahahaha

@JamesTervit @__tinygrad__ That looks legendary!

@squareroundcurl @__tinygrad__ travelling and havent got it unboxed yet.

@JamesTervit @__tinygrad__ why on earth would you ai generate this image?

@MemonZain1 @__tinygrad__ Thanks for your comment with Mac being great on Zero Copy and RTX on raw compute I see them as a complete complimentary setup. I will post the iontegration soon and results, exciting times.

@JamesTervit @__tinygrad__ that's a solid setup for testing across different architectures. curious how the m3 stacks up against the rtx pro for tinygrad workloads

@no_stp_on_snek thanks

@JamesTervit sent!

Thank you everyone for sending me data, info, and support throughout the past few days. Been a tremendous lift from the community to help with making TurboQuant a reality!

@zarathustra_3 IRGC nonesense

40 years spent in a country, granted a Golden Visa and he has been asked to pack up 40 years of his life in two days. To those who say UAE is their home, is this what home looks like? Today it’s Iranians next time it can be you.

@no_stp_on_snek @justbyte_ sterling work.

@justbyte_ shipped asymmetric K/V cache compression for llama.cpp. you can now mix cache types independently for keys and values. also shipped turbo2 yesterday (2-bit KV cache, 6.4x compression). the full turbo family is done: 2-bit, 3-bit, 4-bit. sleep is overrated

Developers, what are you building this weekend?

@alexocheema @kDasme @nvidia @Apple @GroqInc godamm teaser

@kDasme @nvidia @Apple Here’s a clue. @nvidia x @GroqInc

Big moment coming for local AI. @nvidia x @apple

@alexocheema @nvidia @Apple looking forward to this Alex.

@amjadt25 It needs GCC to fix this once and for all and let everyone move forward in modernsing a more peaceful world

The UAE will show the world what strength is made of. The UAE will show the EU and NATO how to be brave. The UAE will make sure the Islamic regime in Iran never, ever targets us again. Never again. We lead. We protect. We win. And the Middle East moves forward without Islamist terrorists.

@Jokee_web3 @__tinygrad__ I am travelling in singapore for now talking to my engineers who are building my atomic clock holdover system to give me independent entanglement time for my quantum project. @no_stp_on_snek got me fired up with the turboquant thats awesome solid work.

@JamesTervit @__tinygrad__ Excited to see the results of your hard work!

@no_stp_on_snek @__tinygrad__ Thanks tom for highlighting this and as you have done a lot of hard work on this can I push the RFC to your repo? I should be done with doing a full agile research soon and hopefully we have a way forward. This is a truly exciting development to get deeper in the architecture

that is a genuinely exciting problem. the turbo KV format is identical bytes on both backends (same struct in ggml-common.h, same WHT rotation seed) so cross-device deserialization is architecturally clean. and the compression works in your favor twice: less data to serialize AND less data moving over TB5 during inference. up to 4.6x less vs fp16, about 2.4x vs q8_0. using compressed KV as the wire format for cross-backend streaming is a new angle as far as I know. tinygrad and others have done heterogeneous multi-device inference but not with compression as the bandwidth strategy. that could make eGPU setups actually practical for long-context workloads where the interconnect is normally the wall. I am very interested in collaborating on this. let me know when you start the RFC and I can help with the KV format details and Metal-side integration.

@no_stp_on_snek @__tinygrad__ @no_stp_on_snek on the bridge, I will work on an RFC, lets see if we can build deserialisation between cuda and metal with a double buffer approach. Very interesting problem that could change entry into this costly game. Also extend Mac M3 ultra into M5 territory.

@DCBK2LA @genspark_ai @ericjing_ai How it felt for me when I connected my 8bit micro with my first modem and hit dial to another 8 bit with file exchanging.

🚨 Genspark AI + Curious Teenage Minds + Me = ❓ Last Thursday proved something to me: kids don’t need to be protected from AI, they need to be taught how to use it wisely, creatively, and fearlessly because once they are, the future gets very bright very fast. So I had the opportunity to speak to a small AI-focused program at my nephew’s high school last week, and honestly, it left me insanely inspired. Standing in that classroom, looking at these students as they leaned in, asked incredibly thoughtful questions, and explored what’s possible with AI, I felt something much bigger than just excitement about technology… I felt hope! But what made it so powerful wasn’t just that they were learning new tools, it was watching them realize in real time that the future is not something that just happens to them. It’s something they can help shape. You could see and feel it in the room: the curiosity, the confidence starting to build, the spark that happens when young people begin to understand that they have access to tools that can expand how they think, create, dream and ultimately build wealth for themselves and their families! As for me, it hit even deeper because this was my nephew’s high school. There was something incredibly meaningful about being in that space, giving back in a way that felt both personal and purposeful. It reminded me that sometimes the most important rooms we walk into are the ones where we can help young people see themselves differently… not just as students, but as builders, creators, and future leaders. A small classroom, students engaged, laptops open, real learning happening in real time. To some people, it may just look like a class. To me, it looked like the beginning of something. It looked and felt like possibility. I’ll likely be teaching in this program over the next few months, and that makes me even more excited because if Thursday showed me anything, it’s that the future is incredibly bright for kids who learn how to use AI with intention, creativity, and guidance. Not as a shortcut like most teenagers, but as a superpower. And these kids get it! None of this would’ve been possible without the kindness and generosity of @ericjing_ai and the entire @genspark_ai team who are the official AI sponsor of this program and helping make this experience possible. Access changes everything. When students are given the chance to engage with technology early, in the right environment, it can open their minds to futures they may not have imagined for themselves before. That matters more than people realize. Have a listen below and there is so much more to come, stay tuned!

@Ascend_Domains @no_stp_on_snek @__tinygrad__ Thanks one of my GX10 had a pcie failure, RMA'd and now back working. They are awesome machines.

@JamesTervit @no_stp_on_snek @__tinygrad__ Very nice setup I have 2 @ GX10s and 4090 RN ... waiting for M5 Ultra studios. Hope you can get this going nicely.

@Ascend_Domains @no_stp_on_snek @__tinygrad__ Thanks once I see what the MacM3 ultra can do, I will certainly look at the helios

@JamesTervit @no_stp_on_snek @__tinygrad__ If 10GBE is an issue get Helios 5S & PCIE to MicroTik switch and bump that pipeline to 6GB/s to the Mac

@Mk__0168 @__tinygrad__ Initially 24 x 7 qlora training it should take me from 3 days to under a day. I will report back a list of info like thermals and performance.

@JamesTervit @__tinygrad__ Getting ready to push Tinygrad to the limit! What specific tests do you plan to run first on your Mac M3 and RTX Pro 6000 setup?

@asian42419 @__tinygrad__ Yes I agree and puts the Mac M3 Ultra up there in terms of compute power.

@JamesTervit @__tinygrad__ Can wait for results , it’s best possible set up Mac Studio + RTX PRO 6000

@madebyollin @BderKhan @__tinygrad__ Yes

@BderKhan @__tinygrad__ @JamesTervit Also it has the nano banana watermark on the bottom right?