diff --git a/fuzzers/037-iob18-pips/Makefile b/fuzzers/037-iob18-pips/Makefile
index cef61988..cc1b9e3a 100644
--- a/fuzzers/037-iob18-pips/Makefile
+++ b/fuzzers/037-iob18-pips/Makefile
@@ -13,7 +13,9 @@ TODO_RE=".*"
 EXCLUDE_RE=".*((PHASER)|(CLKDIVF)|(CLKDIVP)|(CLKDIVB)|(IOI_ILOGIC[01]_O)|(IOI_OLOGIC[01]_CLKB?\.)|(IOI_IMUX_RC)|(IOI_OLOGIC[01]_[OT]FB)|(OCLKM.*IMUX31)|IOI_ODELAY[01]_CLKIN.IOI_OCLK_[01]|RIOI_O[01].RIOI_ODELAY[01]_DATAOUT|RIOI_O[01].RIOI_OLOGIC[01]_OQ).*"
 
 MAKETODO_FLAGS=--pip-type ${PIP_TYPE} --seg-type $(PIP_TYPE) --re $(TODO_RE) --sides "xr" --exclude-re $(EXCLUDE_RE)
-N = 120
+# N=120 is excessive on a large device like xc7vx485t; 20 (matches other pip
+# fuzzers) is plenty for segmatch convergence and roughly 6x faster per iter.
+N = 20
 
 SPECIMENS_DEPS=build/cmt_regions.csv
 
diff --git a/fuzzers/037-iob18-pips/generate.tcl b/fuzzers/037-iob18-pips/generate.tcl
index 65d692d7..0c918824 100644
--- a/fuzzers/037-iob18-pips/generate.tcl
+++ b/fuzzers/037-iob18-pips/generate.tcl
@@ -7,28 +7,14 @@
 # SPDX-License-Identifier: ISC
 source "$::env(XRAY_DIR)/utils/utils.tcl"
 
-proc write_pip_txtdata {filename} {
-    puts "FUZ([pwd]): Writing $filename."
-    set fp [open $filename w]
-    set nets [get_nets -hierarchical]
-    set nnets [llength $nets]
-    set neti 0
-    foreach net $nets {
-        incr neti
-        if {($neti % 100) == 0 } {
-            puts "FUZ([pwd]): Dumping pips from net $net ($neti / $nnets)"
-        }
-        foreach pip [get_pips -of_objects $net] {
-            set tile [get_tiles -of_objects $pip]
-            set src_wire [get_wires -uphill -of_objects $pip]
-            set dst_wire [get_wires -downhill -of_objects $pip]
-            set num_pips [llength [get_nodes -uphill -of_objects [get_nodes -of_objects $dst_wire]]]
-            set dir_prop [get_property IS_DIRECTIONAL $pip]
-            puts $fp "$tile $pip $src_wire $dst_wire $num_pips $dir_prop"
-        }
-    }
-    close $fp
-}
+# NOTE: this fuzzer previously defined its own slow Tcl write_pip_txtdata proc
+# that iterated foreach net foreach pip and queried the routing graph per-pip
+# (set num_pips [llength [get_nodes -uphill ...]] ...). On a large device like
+# xc7vx485t this took hours per specimen (1296 nets * O(nodes) Tcl queries).
+# Vivado's built-in `write_pip_txtdata $filename` produces the same 6-column
+# output (tile pip src_wire dst_wire pnum pdir) that the prjxray segmakers
+# parse, but in seconds. Drop the override; the call at the end of this file
+# now resolves to the built-in.
 
 proc make_manual_routes {filename} {
     puts "MANROUTE: Loading routes from $filename"