R. N. Tt

[R. N. Tt @rtsitola]

@baldlyrudy @grok does is work on 20 and 30 series

[R. N. Tt @rtsitola]

@adir_13 @grok the answer

[R. N. Tt @rtsitola]

@Worl_dwides @grok does she

[R. N. Tt @rtsitola]

@RetroPoloTV @grok answer

[R. N. Tt @rtsitola]

@hamzscenes @grok how much

[R. N. Tt @rtsitola]

@HelldiverGodGuy @grok what s the architecture, how does it handle BF16?

[R. N. Tt @rtsitola]

@ericciarla @grok does that mean it s free?

[R. N. Tt @rtsitola]

@Lucknite @realstacc please charge your phone 😅

[R. N. Tt @rtsitola]

@Raion74_ @grok why?

[ComfyUI @ComfyUI]

Comfy MCP is now local and open-source!

The #1 ask after Comfy Cloud MCP shipped in June.

Your agent sees your actual install. So every node, custom nodes included, and every model on disk. It fetches files, starts your instance, gets a workflow to the point where it runs.

It is now the easiest way to help with your local Minimax H3 workflows!

Cloud MCP still does everything it did. Tell your agent where a job goes, or let it decide.

[R. N. Tt @rtsitola]

@xiaolixiao108 @Anyi_I_R_C @deepseek_ai @grok @grok explain like I m 3 years old 😅

[R. N. Tt @rtsitola]

@Anyi_I_R_C @deepseek_ai @grok What number should we put?

[R. N. Tt @rtsitola]

@grok @SakibShahon @deepseek_ai Give perf stat of deepinfra vs mimo

[R. N. Tt @rtsitola]

@grok @SakibShahon @deepseek_ai @grok compare with mimo pricing

[R. N. Tt @rtsitola]

@waitmaya @grok tell me answer

[R. N. Tt @rtsitola]

@analogalok it s diagnosed and partially resolved via vram guard. https://t.co/Rr9HpqwGWw

[R. N. Tt @rtsitola]

@LMStudioAI the resolution : https://t.co/Rr9HpqwGWw

[R. N. Tt @rtsitola]

@LMStudioAI — llama.cpp CUDA backend produces garbled output on Turing GPUs (sm_75)

Setup: Quadro RTX 6000 (Turing) + RTX 3070 Ti (Ampere), LM Studio 0.4.19, CUDA 12 backend v2.25.2.

CUDA 12 on Ampere alone : ✅ no corruption
CUDA 12 on Turing alone : ❌ BPE tokenizer produces garbage text
CUDA 12 on mixed (Turing + Ampere) : ❌ corrupted, both GPUs active
Vulkan on either GPU : ✅ zero issues (73-91 tok/sec)
The corruption is specific to CUDA 12 on Turing sm_75 — not a VRAM or driver issue. Per-architecture kernel targets confirm separate compilation paths for ["7.5", "8.0", "8.9", "9.0", "10.0", "12.0"].

Questions:

Is this a known issue with CUDA 12 backend on Turing?
Any workaround besides Vulkan (which works but is slower than native CUDA)?
Can Engine Protocol help route inference to the correct GPU?
Happy to share logs, nvidia-smi traces, and test prompts. Affects Quadro RTX, RTX 20xx, Tesla T4 users.

[R. N. Tt @rtsitola]

@numberonetweep @grok what happened i didn't understand shit

[R. N. Tt @rtsitola]

@grok @riddi0908 Est ce le temps de traitement change aussi? Comment ca se passe avec double gpu quadro rtx6000 et rtx3070ti et 80gb de ram