4 comments

  • syntaxing0 minutes ago
    Highly recommend lemonade server if you have a strix halo desktop. Been using Qwen3.6-35B @ Q_8 as my main driver and it’s been great. I occasionally use the 27B @ q6 but only get 20-25 TPS for generation with MTP.
  • chmod7758 minutes ago
    This is more of an ad, not a review, and reads like the author has hardly any experience with the things he's trying out. That Z Image Turbo diffusion model would've also run on many consumer GPUs and with way higher performance for a fraction of the price. Misleading.
  • netinstructions33 minutes ago
    Unfortunately, the table of models and tokens per second (TPS) and time to first token (TTFT) is not helpful without specifying the quantization of the model.
  • cmrdporcupine35 minutes ago
    Until RAM prices drop and can economically get machines with 256GB, 512GB and higher bandwidth... I frankly think the local AI story is going to be still fairly muted for most people.<p>My Spark can do Qwen3.6 MoE A3B at 60 to 70-ish token&#x2F;second and that&#x27;s really good, but there&#x27;s limits the usefulness of that model. It&#x27;s not useful for coding, in any case.<p>Once people can run something like GLM 5.2 at lower quants (512GB could do a passable job), then I think the story changes.<p>Whether we ever see DRAM as cheap as it was ever again, I don&#x27;t know.
    • science4sail18 minutes ago
      Agree - the 128GB Strix Halo is capable if you use LLMs as <i>assistants</i>, but it&#x27;s not so good if you use LLMs as <i>agents</i> (or worse, agent teams&#x2F;swarms) since all of the models that can fit on it are pretty dumb compared to frontier or near-frontier models. You can at best hope for Sonnet-level capabilities.<p>That doesn&#x27;t mean that local models are useless though! If Mythos&#x2F;Sol is an ASI that threatens to take your job and turn you into paperclips, then Qwen&#x2F;Gemma is an old-fashioned office secretary that loyally helps you with tasks but doesn&#x27;t have a good grasp of details. Every white-collar worker 50 years ago would have killed to have a hard-working personal secretary.
    • whaleofatw202211 minutes ago
      Part of me wonders, would 3d xpoint (if still around) be a viable option?<p>Yeah it is slower than real RAM by a good amount for latency, but you can get similar bandwidth and the cost was history about half of the same size DDR.
      • science4sail5 minutes ago
        [delayed]
      • cmrdporcupine4 minutes ago
        I was thought experimenting the other day... ~10 nVME drives striped and running parallel could approximate the memory bandwidth of DDR5 DRAM in a box like this. Like you say, latency wouldn&#x27;t compete but on raw throughput would be comparable.<p>Not anymore cost effective, I guess, but gets you the ability to work over very large model sizes maybe. But the problem is that tensor matmul etc hardware wouldn&#x27;t work effectively with it.<p>Useful for KVCache though.