diff --git a/test/testextra/test_cfg_viz.py b/test/testextra/test_cfg_viz.py
index c6306482e6..9911fb5e0a 100644
--- a/test/testextra/test_cfg_viz.py
+++ b/test/testextra/test_cfg_viz.py
@@ -69,6 +69,8 @@ class TestCfg(unittest.TestCase):
     self.assertEqual(len(references["r0"]), 2)
     insts = [cfg["pc_tokens"][pc][0]["st"] for pc in references["r0"]]
     self.assertEqual(insts, ['s_mov_b32', 's_cmp_eq_u64'])
+    end_block_content = "\n".join(" ".join(t["st"] for t in cfg["pc_tokens"][pc]) for pc in list(cfg["blocks"].values())[-1])
+    self.assertEqual(end_block_content, "s_endpgm\ns_code_end (217x)")
 
   def test_loop(self):
     k = Kernel(arch=Device["AMD"].arch)
diff --git a/tinygrad/viz/serve.py b/tinygrad/viz/serve.py
index 5a3c4060d2..0e3cc801bb 100755
--- a/tinygrad/viz/serve.py
+++ b/tinygrad/viz/serve.py
@@ -533,6 +533,19 @@ def amdgpu_cfg(lib:bytes, target:str) -> dict:
       if isinstance(val:=getattr(inst, name), Reg): tokens.append({"st":val.fmt(), "keys":[f"r{val.offset+i}" for i in range(val.sz)], "kind":1})
       elif name in {"op","opx","opy"}: tokens.append({"st":(op_name:=val.name.lower()), "keys":[op_name], "kind":0})
       elif name != "encoding" and val != field.default: tokens.append({"st":(s:=repr(val)), "keys":[s], "kind":1})
+  # show a smaller view for repeated instructions in the graph
+  for pcs in blocks.values():
+    new_pcs:list[int] = []
+    i, n = 0, len(pcs)
+    while i < n:
+      j = i+1
+      while j<n and pc_table[pcs[j]] == pc_table[pcs[i]]: j += 1
+      new_pcs.append(pcs[i])
+      if j-i>1:
+        pc_tokens[pcs[i]].append({"st":f"({j-i}x)", "keys":[], "kind":0})
+        for k in range(i+1, j): del pc_tokens[pcs[k]]
+      i = j
+    pcs[:] = new_pcs
   from tinygrad.runtime.autogen import amdgpu_kd
   kd = amdgpu_kd.llvm_amdhsa_kernel_descriptor_t.from_buffer_copy(bytearray(get_elf_section(lib, ".rodata").content))
   vgpr_gran = kd.compute_pgm_rsrc1 & amdgpu_kd.COMPUTE_PGM_RSRC1_GRANULATED_WORKITEM_VGPR_COUNT