Donato Capitella retweeted
The first Lucebox Zero 495 batch is almost sold out. Only a few units left at shop.lucebox.com
Huge thank you to every customer who's come through so far. Once batch 1 sells out, new orders move to the next batch at a higher price.
I also updated the image to make pricing clearer, so it now shows the price for each configuration:
- 128GB + 2TB: $5,999
- 192GB + 2TB: $7,999
- 4TB SSD upgrade : +$250
- Cluster kit: +$799 per machine
Every configuration comes with the 32GB Radeon AI PRO R9700, and we've further optimized it for V4.1 Flash with our new quants. Full details in our latest blog post.
Donato Capitella retweeted
The end of software secrecy is almost upon us. The death of closed source, and the final triumph of open source.
I should have realized this a few months ago when I successfully decompiled the game of Firefighter from an ancient DOS shareware binary. My robot friend was able to tell it had been written in Borland Pascal from looking at the data layout; it gave me back very readable Pascal code with sensibly chosen function and variable names. I transpiled it to Rust and now ship it as part of my heritage games collection.
The thing is, I thought of this as a fun stunt but didn't have any confidence that the technique would scale up to large, real programs. I have since learned that people are now doing this sort of thing with entire AAA games, which are among the most complex software artifacts ever to be shipped as a binary. If they can be decompiled, anything can be.
This has implications. Massive implications. We need to think about what the world is like when, in general, there is no longer a secrecy moat around almost any software at all.
1/2
Donato Capitella retweeted
👇 @AMD… priced to impress
wccftech.com/super-fast-ai-w…
Donato Capitella retweeted
I spent all the last few days testing GLM 5.3 Flash on the @FrameworkPuter Gorgon and modified a quant that will work in ROCmFPX. Instructions are on the model card but I wanted to get this one up and get some feedback and opinions. Fingers crossed it works well for everyone like it does for me.
huggingface.co/cafonez/GLM-5…
@dcapitella @tekwendell
Donato Capitella retweeted
We're opening pre-orders for the 192GB Framework Desktop this Wednesday morning Pacific Time. This has AMD's new Ryzen AI Max+ Pro 495, an open-end x4 PCIe slot, a pre-installed Noctua fan, and an optional pre-built config with Fedora pre-loaded.
Donato Capitella retweeted
I’m not exaggerating: Codex usage feels completely out of line right now.
I burned through my entire ChatGPT Plus quota in roughly an hour of real work.
That is NOT acceptable.
Something here feels wrong.
Donato Capitella retweeted
Recently, we worked on something special: pushing ROCm to beat Vulkan on Strix Halo.
Lucebox engine (ROCm) and Llama.cpp (Vulkan) ran on the same Strix Halo on the same day with the same prompt. With the DSpark drafter Lucebox ROCm decodes at 41.9 tok/s at 8K and 38.0 tok/s at 123K prompt tokens.
That is 28% to 47% faster than llama.cpp and Vulkan v0.7.5 running the same drafter and its prefill is 17% to 47% faster 🧵
Donato Capitella retweeted
github.com/Nathanw1014/strix…
v0.7.6 is out, built in parallel with halo-box (PR coming)
- Prompt processing +16% to +31% on standard models tested
- Flash-Next PP flat with depth now, +34% at 32k
- MTP speculative decoding ~9% faster on 27B with slightly better acceptance
DwarfStar running DeepSeek v4.1 Flash on a 128GB M5 Max. I didn't expect with SSD streaming it could be so fast. Recent SSD streaming changes to retain the right experts surely helped, but also maybe DS4.1 uses the same experts more. Will push online when ready QA > ASAP.
Donato Capitella retweeted
We have a positive update today on LPCAMM2 pricing! We were able to source a limited quantity at lower cost, and are retroactively reducing prices for orders that shipped at higher price, as well as offering the lower price to orders through Batch 10.
We've published the latest set of updates around pricing. We have a major and unfortunate update around LPCAMM2 pricing and available, alongside a few smaller increases and decreases on other memory and storage: frame.work/blog/updates-on-m…
Donato Capitella retweeted
Big open source news for Strix Halo owners. 🚀
For the last two weeks we have been working on a community fork of llama.cpp to make it easier for the community to use a stable version and avoid having to switch forks on a weekly basis.
Try it out here:
github.com/halo-box/strix-ll…
Well done @HiGaetan, @NathanW1014 and others, let's continue this work to push the performance of local inference, fun working with you guys!
For users with other hardware we have a more mainline compatible version: github.com/halo-box/llama.cp…
Donato Capitella retweeted
Threadripper Halo station: the world's most powerful workstation @AMD
Donato Capitella retweeted
Excited to announce that Lucebox engine now serves Qwen3.8-27B on a single AMD Radeon AI PRO R9700 (32 GB, RDNA4) with the DFlash2 block-diffusion drafter from z-lab:
- up to 227 tok/s on code
- 208 tok/s HumanEval average
- math at 133 tok/s
on an off-the-shelf quant that matches an 8-bit reference on HumanEval and GSM8K.
And it is lossless: greedy verification commits only tokens the model itself would produce, so the drafter changes how fast you get the answer, never what the answer is.
Full breakdown in the article below👇
Donato Capitella retweeted
It's an excellent day for running open weights models locally! Qwen3.8-Flash-Next fits in its entirety at 4-bit on a 128GB Framework Desktop, and at 8-bit on the upcoming 192GB Framework Desktop.
Qwen3.8-Flash can now be run locally! 🔥
The 125B MoE model outperforms Claude-Opus-4.6 (Max).
Run on 75GB RAM via Unsloth GGUFs.
Qwen3.8-Flash-Next enables CPU RAM / unified mem setups to deliver near VRAM speeds.
Guide: unsloth.ai/docs/models/qwen3…
GGUF: huggingface.co/unsloth/Qwen3…
Donato Capitella retweeted
We reached sota speed for coding with Qwen3.8 27B on the AMD R9700 GPU, directly integrated with DFlash2 🏎️
Soon we will release custom quants to make it even faster.
Excited to announce that Lucebox engine now serves Qwen3.8-27B on a single AMD Radeon AI PRO R9700 (32 GB, RDNA4) with the DFlash2 block-diffusion drafter from z-lab:
- up to 227 tok/s on code
- 208 tok/s HumanEval average
- math at 133 tok/s
on an off-the-shelf quant that matches an 8-bit reference on HumanEval and GSM8K.
And it is lossless: greedy verification commits only tokens the model itself would produce, so the drafter changes how fast you get the answer, never what the answer is.
Full breakdown in the article below👇
Donato Capitella retweeted
Wrote up a blog for the PipeWire sandbox escape via malicious library loading
CVE-2026-5674
Details ⬇️
embracethered.com/blog/posts…
Donato Capitella retweeted
Today, we're really excited to announce that @luceboxai is partnering with @AMD to bring heterogenous consumer hardware to life.
We worked really hard on this, and were able to have Lucebox (AMD Radeon AI PRO R9700 + Strix Halo) to beat one NVIDIA DGX Spark by 3.63x on DeepSeek V4 Flash Decode Speed.
R9700 takes the dense path, the hot experts, the cache and the draft model. Strix Halo holds the other experts in 128gb and computes them at the same time, not after.
51.1 tok/s on the full 284B, 3.63x one DGX Spark, and $2,899 less than having two of them.
More details on the machine at lucebox.com and full technical breakdown in the article below ⬇️
Open has always been at the heart of how we build at AMD.
That’s why we’re joining @Microsoft and others in signing an open letter supporting open-weight AI models. A strong AI ecosystem depends on open standards, interoperability and choice across software, systems and hardware: aka.ms/OpenLetter.



























