
AI in Public 🛠️ | ex @SentinelOne 1st SRE Go Team & @Comcast VIPER Startup | Building : Strix Halo / R9700 K8S Cluster for Compound AI Systems
🏔 DENVER 🏔
Joined September 2016
- Tweets1.9K
- Following306
- Followers1.2K
- Likes34.4K
Ryan Wright retweeted
Hearing the creator of Mario, Shigeru Miyamoto, say that the Typerwriter to Word Processor is the equilivent of Generative AI and is part of technological human evolution is probably one of the best statements of 2026.
It will not make human creativity obsolete
Human intention and individual visions will become even more important
Ryan Wright retweeted
DIY tower 3d printed ✅
Awaiting fans from Amazon now
Dual Asus GX10 setup is getting nicer every day between new recipes from @MiaAI_lab and creators in the 3d printing space haha
Ryan Wright retweeted
Replying to @IQReactorAI
This was done on a phone. 8B getting 20-30 tok/s
xitter.cf/trevorwood222/status/2…
Ryan Wright retweeted
New @AMDembedded Ryzen AI Embedded X100 CPU family, COM-HPC module, and developer platform.
cnx-software.com/2026/07/24/…
The 55W Ryzen AI Embedded X100 family offers up to 16 Zen 5 cores, an RDNA 3.5 iGPU, and up to 50 TOPS NPU. The company also provides the Kria AI SOM COM-HPC module featuring the Ryzen AI Embedded X199 CPU and up to 128LPDDR5, and the Kria AI Robotics developer platform with 10GbE networking, FAKRA camera connector, an Oculink connector, and more.
The new family supports Linux, the AMD ROCm software stack, Xen Hypervisor, and standard AI frameworks like PyTorch, ONNX, and TensorFlow. It's designed to compete against Intel Core Ultra Series 3 and NVIDIA Jetson Thor T5000 solutions.
Ryan Wright retweeted
AMD Gorgon Halo Cluster preview. 2x 192GB unified memory, same Strix Halo memory bandwidth.
Ryan Wright retweeted
Ported our 8x GB10 cluster to the v16 unified base + b12x sparse indexer. All measured, TP8.
prefill pp2048: 880 -> 1,101 tok/s @4k depth (+28% vs our old stack)
depth prefill: 1052 / 940 / 838 at 77K / 231K / 395K (was 653/441/319)
single-stream decode: 42.1 tok/s median
860K max_len serving, KV pool 903K tokens. needle recovered at 845K, still decoding 10.9 tok/s at that depth.
Two traps if you're on this base: cold inductor cache makes first-boot lazy compiles (~53s each) look exactly like a shm-broadcast hang. warm restart -> 3s, gone. and available KV drifts 47->41 GiB over repeated boots from GPU memory fragmentation. full node reboot resets it.
Thanks @anuntrapid_auto your repo + patch pointer were the starting point. GLM-5.2 Int4-Int8Mix, fp8 sparse-MLA KV, MTP.
xitter.cf/anuntrapid_auto/status…
GLM-5.2-Int4-Int8 on 8× GB10: ~1,200 t/s prefill, 33–54 t/s avg decode (generic - coding/structured) and memory remaining to run also a Mimo 2.5 in parallel for image/audio input, both tp 8. forums.developer.nvidia.com/…
github.com/ciprianveg/gb10-g…
Ryan Wright retweeted
Claude code source code has been leaked via a map file in their npm registry!
Code: pub-aea8527898604c1bbb12468b…
Ryan Wright retweeted
Google Nano Banana just broke the Internet yesterday
People are getting crazy creative with it.
10 wild examples:
Ryan Wright retweeted
365 days of UNLIMITED FREE Nano Banana.
Only for Higgsfield Pro, Ultimate & Creator users.
Limited-time offer until the end of this week.
You read that right. Unlimited & Free for one WHOLE year.
You've got 4 days to secure the offer.
A HUGE Banana, indeed. Go wild.
Retweet to get the offer link in DMs 📨
Ryan Wright retweeted
Spend 80% of your time with people who help you generate 80% of your desired results.
Ryan Wright retweeted
Dear developers, good night! 🌚🦸♂️
People deserve to know the truth about programmers
Ryan Wright retweeted
✅ How to output QR codes in your app? Its simple.
👇Attached are 2 code examples from Python and JavaScript (Photos)
🧠 An interesting fact about the data they hold :
Numeric. 7,089 chars
Alphanumeric 4,296 chars
That's actually alot the data !
This quoted post is unavailable.
Ryan Wright retweeted
=> Free hosting for static sites <=
✅ GitHub Pages
✅ Netifly
✅ Firebase
✅ Amazon S3
✅ Zeit
✅ Forge





































