ci: route GPU verification by machine

[davide221]

Routes the RTX 3090 jobs exclusively to the dedicated lucebox-rtx3090 runner and shares its concurrency group with the speed profile.

Adds the DGX GB10 (ARM64, sm_121, CUDA 13) as an independent verifier, and updates the lucebox3 HIP lane and Docker PR target for its current R9700/gfx1201 hardware.

Both NVIDIA runners are installed, online, and registered with machine-specific labels. The stale RTX/Strix labels were removed from lucebox3.

[cubic-dev-ai]

1 issue found across 4 files

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name=".github/workflows/ci.yml">

<violation number="1" location=".github/workflows/ci.yml:260">
P2: On the GB10 path, this `121` value incorrectly activates the consumer-Blackwell workaround for every `12x` target. The DGX verifier therefore omits FP4 kernels and aborts valid FP4 dispatches; restrict the workaround to consumer SM 12.0 or add a distinct non-consumer GB10 configuration using the `121a` target.</violation>
</file>

<sub>Reply with feedback, questions, or to request a fix.

Re-trigger cubic</sub>

[cubic-dev-ai]

P2: On the GB10 path, this 121 value incorrectly activates the consumer-Blackwell workaround for every 12x target. The DGX verifier therefore omits FP4 kernels and aborts valid FP4 dispatches; restrict the workaround to consumer SM 12.0 or add a distinct non-consumer GB10 configuration using the 121a target.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At .github/workflows/ci.yml, line 260:

<comment>On the GB10 path, this `121` value incorrectly activates the consumer-Blackwell workaround for every `12x` target. The DGX verifier therefore omits FP4 kernels and aborts valid FP4 dispatches; restrict the workaround to consumer SM 12.0 or add a distinct non-consumer GB10 configuration using the `121a` target.</comment>

<file context>
@@ -219,33 +224,75 @@ jobs:
+          cmake -B build \
+            -DCMAKE_CUDA_COMPILER="$CUDACXX" \
+            -DCMAKE_CUDA_ARCHITECTURES="121" \
+            -DDFLASH27B_USER_CUDA_ARCHITECTURES="121" \
+            -DDFLASH27B_ENABLE_BSA=OFF \
+            -DDFLASH27B_FA_ALL_QUANTS=OFF \
</file context>

[cubic-dev-ai]

All reported issues were addressed across 2 files (changes from recent commits).

<sub>Tip: Review your code locally with the cubic CLI to iterate faster.

Re-trigger cubic</sub>

[cubic-dev-ai]

All reported issues were addressed across 3 files (changes from recent commits).

<sub>Tip: Review your code locally with the cubic CLI to iterate faster.

Re-trigger cubic</sub>