diff --git a/bitsandbytes/backends/cpu_xpu_common.py b/bitsandbytes/backends/cpu_xpu_common.py
index 87ffc7360..2a29604ba 100644
--- a/bitsandbytes/backends/cpu_xpu_common.py
+++ b/bitsandbytes/backends/cpu_xpu_common.py
@@ -369,8 +369,9 @@ def quantize_4bit_impl(
             out_uint8[abs_scaled_A > key] = val
         out_uint8 += sign.to(torch.uint8) * 8
     elif quant_type == "int8":
-        for i in range(len(INT8_QUANT_TABLE)):
-            out_uint8[scaled_A > INT8_QUANT_TABLE[i]] = i
+        map = torch.tensor(INT8_QUANT_TABLE, device=scaled_A.device)
+        diff = torch.abs(scaled_A.unsqueeze(-1) - map)
+        out_uint8 = torch.argmin(diff, dim=-1).to(torch.uint8).to(scaled_A.device)
 
     if quant_type == "int8":
         out = out_uint8