diff --git a/.github/actions/setup-tinygrad/action.yml b/.github/actions/setup-tinygrad/action.yml
index 050c51fab8..3d061fcb92 100644
--- a/.github/actions/setup-tinygrad/action.yml
+++ b/.github/actions/setup-tinygrad/action.yml
@@ -228,6 +228,11 @@ runs:
 
         sudo chown -R $USER:$USER /var/cache/apt/archives/
 
+    - name: Add clang to PATH (Linux)
+      if: inputs.llvm == 'true' && runner.os == 'Linux'
+      shell: bash
+      run: echo "/usr/lib/llvm-20/bin" >> "$GITHUB_PATH"
+
     # **** AMD ****
     - name: Setup AMD (Linux)
       if: inputs.amd == 'true' && runner.os == 'Linux'
diff --git a/.github/workflows/test.yml b/.github/workflows/test.yml
index c5e9f5af99..a00ad97d8f 100644
--- a/.github/workflows/test.yml
+++ b/.github/workflows/test.yml
@@ -17,7 +17,7 @@ on:
 jobs:
   llvmspeed:
     name: LLVM Speed
-    runs-on: ubuntu-24.04
+    runs-on: &linux ${{ github.repository == 'tinygrad/tinygrad' && github.event_name == 'pull_request' && github.event.pull_request.author_association == 'COLLABORATOR' && 'namespace-profile-tinygrad' || 'ubuntu-24.04' }}
     timeout-minutes: 20
     steps:
     - name: Checkout Code
@@ -35,7 +35,7 @@ jobs:
 
   docs:
     name: Docs
-    runs-on: ubuntu-24.04
+    runs-on: *linux
     timeout-minutes: 10
     env:
       CHECK_OOB: 0
@@ -89,7 +89,7 @@ jobs:
 
   torchbackend:
     name: Torch Backend Tests
-    runs-on: ubuntu-24.04
+    runs-on: *linux
     timeout-minutes: 15
     steps:
     - name: Checkout Code
@@ -125,7 +125,7 @@ jobs:
 
   torchbackendmore:
     name: Torch Backend Tests More
-    runs-on: ubuntu-24.04
+    runs-on: *linux
     timeout-minutes: 15
     steps:
     - name: Checkout Code
@@ -147,7 +147,7 @@ jobs:
 
   bepython:
     name: Python Backend
-    runs-on: ubuntu-24.04
+    runs-on: *linux
     timeout-minutes: 15
     steps:
     - name: Checkout Code
@@ -215,7 +215,7 @@ jobs:
 
   linter:
     name: Linters
-    runs-on: ubuntu-24.04
+    runs-on: *linux
     timeout-minutes: 10
 
     steps:
@@ -246,7 +246,7 @@ jobs:
 
   nulltest:
     name: Null Tests
-    runs-on: ubuntu-24.04
+    runs-on: *linux
     timeout-minutes: 15
 
     steps:
@@ -277,7 +277,7 @@ jobs:
 
   unittest:
     name: Unit Tests
-    runs-on: ubuntu-24.04
+    runs-on: *linux
     timeout-minutes: 15
 
     steps:
@@ -320,7 +320,7 @@ jobs:
       matrix:
         group: [1, 2]
     name: SPEC=2 (${{ matrix.group }})
-    runs-on: ubuntu-24.04
+    runs-on: *linux
     timeout-minutes: 15
     steps:
     - name: Checkout Code
@@ -331,12 +331,13 @@ jobs:
         key: spec-unit
         deps: testing_unit
         python-version: '3.14'
+        llvm: 'true'
     - name: Test SPEC=2
       run: SPEC=2 pytest --maxfail=10 -n auto --durations=30 test/unit test/backend test/opt --ignore test/backend/test_custom_kernel.py --ignore test/unit/test_hashing.py --timeout 60 -k "not test_setitem_big" -k "not test_conv2d_ceildiv_edge_case" --splits 2 --group ${{ matrix.group }}
 
   fuzzing:
     name: Fuzzing
-    runs-on: ubuntu-24.04
+    runs-on: *linux
     timeout-minutes: 10
     steps:
     - name: Checkout Code
@@ -357,7 +358,7 @@ jobs:
 
   testopenclimage:
     name: CL IMAGE Tests
-    runs-on: ubuntu-24.04
+    runs-on: *linux
     timeout-minutes: 15
     steps:
       - name: Checkout Code
@@ -377,7 +378,7 @@ jobs:
 
   testgpumisc:
     name: CL Misc tests
-    runs-on: ubuntu-24.04
+    runs-on: *linux
     timeout-minutes: 10
     steps:
       - name: Checkout Code
@@ -402,7 +403,7 @@ jobs:
 
   testopenpilot:
     name: openpilot Compile Tests
-    runs-on: ubuntu-24.04
+    runs-on: *linux
     timeout-minutes: 15
     steps:
       - name: Checkout Code
@@ -432,7 +433,7 @@ jobs:
 
   testonnxcpu:
     name: ONNX (CPU) Tests
-    runs-on: ubuntu-24.04
+    runs-on: *linux
     timeout-minutes: 20
 
     steps:
@@ -460,7 +461,7 @@ jobs:
 
   testopencl:
     name: ONNX (CL)+Optimization Tests
-    runs-on: ubuntu-24.04
+    runs-on: *linux
     timeout-minutes: 20
     steps:
       - name: Checkout Code
@@ -494,7 +495,7 @@ jobs:
 
   testllm:
     name: Test LLM
-    runs-on: ubuntu-24.04
+    runs-on: *linux
     timeout-minutes: 15
     env:
       CHECK_OOB: 0
@@ -519,7 +520,7 @@ jobs:
 
   testmodels:
     name: Models (llvm+cpu+gpu)
-    runs-on: ubuntu-24.04
+    runs-on: *linux
     timeout-minutes: 15
     steps:
       - name: Checkout Code
@@ -562,7 +563,7 @@ jobs:
 
   testdsp:
     name: Linux (DSP)
-    runs-on: ubuntu-24.04
+    runs-on: *linux
     timeout-minutes: 15
     steps:
     - name: Checkout Code
@@ -580,13 +581,13 @@ jobs:
     - name: Run test_tiny on DSP
       run: DEBUG=2 DEV=DSP python test/test_tiny.py
     - name: Test transcendentals
-      run: CC=clang-20 DEBUG=2 DEV=DSP python test/backend/test_transcendental.py TestTranscendentalVectorized
+      run: DEBUG=2 DEV=DSP python test/backend/test_transcendental.py TestTranscendentalVectorized
     - name: Test quantize onnx
       run: DEBUG=2 DEV=DSP python3 test/backend/test_quantize_onnx.py
 
   testwebgpu:
     name: Linux (WebGPU)
-    runs-on: ubuntu-24.04
+    runs-on: *linux
     timeout-minutes: 20
     steps:
     - name: Checkout Code
@@ -610,7 +611,7 @@ jobs:
 
   testamdasm:
     name: AMD ASM IDE
-    runs-on: ubuntu-24.04
+    runs-on: *linux
     timeout-minutes: 20
     env:
       DEV: MOCKKFD+AMD
@@ -657,7 +658,7 @@ jobs:
 
   testmockam:
     name: Linux (am)
-    runs-on: ubuntu-24.04
+    runs-on: *linux
     timeout-minutes: 15
     env:
       DEV: MOCKPCI+AMD
@@ -693,7 +694,7 @@ jobs:
         arch: [gfx1100, gfx1201, gfx950]
 
     name: Linux (${{ matrix.backend }} ${{ matrix.arch }})
-    runs-on: ubuntu-24.04
+    runs-on: *linux
     timeout-minutes: 15
     env:
       DEV: MOCKKFD+AMD:${{ matrix.backend == 'amdllvm' && 'LLVM' || '' }}:${{ matrix.arch }}
@@ -728,7 +729,7 @@ jobs:
         backend: [ptx, nv]
 
     name: Linux (${{ matrix.backend }})
-    runs-on: ubuntu-24.04
+    runs-on: *linux
     timeout-minutes: 20
     env:
       FORWARD_ONLY: 1
@@ -763,7 +764,7 @@ jobs:
         backend: [llvm, cpu, opencl, lvp, x86]
 
     name: Linux (${{ matrix.backend }})
-    runs-on: ubuntu-24.04
+    runs-on: *linux
     timeout-minutes: 20
     steps:
       - name: Checkout Code
@@ -777,7 +778,7 @@ jobs:
           llvm: ${{ matrix.backend == 'llvm' || matrix.backend == 'cpu' || matrix.backend == 'lvp' }}
           mesa: ${{ matrix.backend == 'lvp' && 'cpu' }}
       - name: Set env
-        run: printf "${{ matrix.backend == 'llvm' && 'DEV=CPU:LLVM' || matrix.backend == 'cpu' && 'CC=clang-20\nDEV=CPU\nCPU_COUNT=2' || matrix.backend == 'opencl' && 'DEV=CL' || matrix.backend == 'lvp' && 'DEV=CPU:LVP' || matrix.backend == 'x86' && 'DEV=CPU:X86' }}" >> $GITHUB_ENV
+        run: printf "${{ matrix.backend == 'llvm' && 'DEV=CPU:LLVM' || matrix.backend == 'cpu' && 'DEV=CPU\nCPU_COUNT=2' || matrix.backend == 'opencl' && 'DEV=CL' || matrix.backend == 'lvp' && 'DEV=CPU:LVP' || matrix.backend == 'x86' && 'DEV=CPU:X86' }}" >> $GITHUB_ENV
       - name: Check Device.DEFAULT and print some source
         run: |
           python3 -c "from tinygrad import Device; assert Device.DEFAULT in ['CPU','CL'], Device.DEFAULT"
@@ -962,7 +963,7 @@ jobs:
       matrix:
         backend: [ir3, nak]
     name: Compile-only (${{ matrix.backend }})
-    runs-on: ubuntu-24.04
+    runs-on: *linux
     timeout-minutes: 15
     steps:
       - name: Checkout Code
@@ -994,7 +995,7 @@ jobs:
           python -m pytest -n=auto test/backend/test_ops.py --durations=20
   qcomclcompiletests:
     name: Compile-only (QCOM CL)
-    runs-on: ubuntu-24.04-arm
+    runs-on: ${{ github.repository == 'tinygrad/tinygrad' && github.event_name == 'pull_request' && github.event.pull_request.author_association == 'MEMBER' && 'namespace-profile-tinygrad-arm64' || 'ubuntu-24.04-arm' }}
     timeout-minutes: 15
     steps:
       - name: Checkout Code