diff --git a/tinygrad/renderer/assembly.py b/tinygrad/renderer/assembly.py
index 86f576eead..c2c95eaf85 100644
--- a/tinygrad/renderer/assembly.py
+++ b/tinygrad/renderer/assembly.py
@@ -220,7 +220,7 @@ def uops_to_asm(lang:AssemblyLanguage, function_name:str, uops:UOpGraph) -> str:
   return lang.render_kernel(kernel, function_name, bufs, c.items())
 
 class PTXLanguage(AssemblyLanguage):
-  kernel_prefix = """.version 7.5
+  kernel_prefix = """.version VERSION
 .target TARGET
 .address_size 64
 .visible .entry"""
diff --git a/tinygrad/runtime/ops_cuda.py b/tinygrad/runtime/ops_cuda.py
index 3a1895a285..ba85c81311 100644
--- a/tinygrad/runtime/ops_cuda.py
+++ b/tinygrad/runtime/ops_cuda.py
@@ -56,9 +56,10 @@ class PTXCompiler(Compiler):
   linearizer_opts = LinearizerOptions("CUDA", suffix="PTX", global_max=[65535, 65535, 2147483647], local_max=[64, 1024, 1024], shared_max=49152)
   def __init__(self, arch:str):
     self.arch = arch
+    self.version = "7.8" if int(arch[3:]) >= 89 else "7.5"
     PTXCompiler.linearizer_opts = PTXCompiler.linearizer_opts._replace(has_tensor_cores=int(arch[3:]) >= 80)
     super().__init__(f"compile_ptx_{self.arch}")
-  def render(self, name:str, uops) -> str: return PTXRenderer(name, uops).replace("TARGET", self.arch)
+  def render(self, name:str, uops) -> str: return PTXRenderer(name, uops).replace("TARGET", self.arch).replace("VERSION", self.version)
   def compile(self, src:str) -> bytes: return src.encode()
 
 class CUDACompiler(Compiler):