diff --git a/Makefile b/Makefile
index 543be9c..81a0718 100644
--- a/Makefile
+++ b/Makefile
@@ -3,7 +3,7 @@ OPT = -Og
 LIBGCC = $(shell $(CC) -print-file-name=libgcc.a)
 LIB = ./saturn
 
-all: 
+all:
 
 include $(LIB)/common.mk
 
@@ -51,6 +51,9 @@ res/dejavusansmono.font.bin: tools/ttf-convert
 res/ipapgothic.font.bin: tools/ttf-convert
 	./tools/ttf-convert 3000 30ff 28 $(shell fc-match -f '%{file}' 'IPAPGothic') $@
 
+res/sperrypc.font.bin: tools/ttf-convert
+	./tools/ttf-convert 20 7f 8 res/Bm437_SperryPC_CGA.otb $@
+
 common/keyboard.hpp: common/keyboard.py
 	python common/keyboard.py header > $@
 
@@ -86,7 +89,7 @@ m68k/%.bin: m68k
 
 scsp/sound_cpu__slot.elf: scsp/sound_cpu__slot.o m68k/slot.bin.o
 
-scsp/sound_cpu__interrupt.elf: scsp/sound_cpu__interrupt.o m68k/interrupt.bin.o
+scsp/sound_cpu__interrupt.elf: scsp/sound_cpu__interrupt.o m68k/interrupt.bin.o sh/lib1funcs.o res/sperrypc.font.bin.o common/draw_font.o common/palette.o
 
 # clean
 clean: clean-sh
@@ -101,6 +104,7 @@ clean-sh:
 		common/keyboard.cpp \
 		common/keyboard.hpp \
 		wordle/word_list.hpp
-
+	make -C tools clean
+	make -C m68k clean
 
 PHONY: m68k tools
diff --git a/common/draw_font.hpp b/common/draw_font.hpp
index d30dabf..919a041 100644
--- a/common/draw_font.hpp
+++ b/common/draw_font.hpp
@@ -14,12 +14,13 @@ struct state {
 uint32_t font_data(void * buf, uint32_t top, state& state);
 
 template <typename T>
-uint32_t horizontal_string(state const& s,
-                           uint32_t& cmd_ix,
-                           const T * string,
-                           const uint32_t length,
-                           const int32_t x,
-                           const int32_t y)
+inline
+int32_t horizontal_string(state const& s,
+			  uint32_t& cmd_ix,
+			  const T * string,
+			  const uint32_t length,
+			  const int32_t x,
+			  const int32_t y)
 {
   int32_t total_advance = 0;
 
@@ -50,6 +51,7 @@ uint32_t horizontal_string(state const& s,
 }
 
 template <typename T>
+inline
 uint32_t single_character_centered(state const& s,
                                    uint32_t cmd_ix,
                                    const T c,
diff --git a/m68k/Makefile b/m68k/Makefile
index 68ee351..fbbd87c 100644
--- a/m68k/Makefile
+++ b/m68k/Makefile
@@ -2,7 +2,7 @@ CFLAGS = -I../saturn
 OPT = -Og
 LIB = ../saturn
 
-all: 
+all:
 
 include $(LIB)/m68k/common.mk
 
diff --git a/m68k/interrupt.cpp b/m68k/interrupt.cpp
index c3911ed..577f739 100644
--- a/m68k/interrupt.cpp
+++ b/m68k/interrupt.cpp
@@ -4,13 +4,15 @@
 
 extern void * _jojo_start __asm("_binary_jojo_11025_s16be_1ch_pcm_start");
 
-static volatile int32_t frame = 0;
+static volatile int32_t frame = 0x0;
 
 constexpr int32_t tactl = 7;
 constexpr int32_t frame_size = ((1 << tactl) * 256) / (44100 / 11025);
 constexpr int32_t frame_count = 507150 / (frame_size * 2);
 constexpr int32_t tima = 0;
 
+static uint16_t slot_ix = 0;
+
 extern "C"
 void auto_vector_1(void) __attribute__ ((interrupt_handler));
 void auto_vector_1(void)
@@ -20,15 +22,16 @@ void auto_vector_1(void)
   scsp.reg.ctrl.TIMA = TIMA__TACTL(tactl) | TIMA__TIMA(tima);
 
   if (frame > frame_count) frame = 0;
-  
+
   const uint16_t * jojo_start = reinterpret_cast<uint16_t *>(&_jojo_start);
   const uint32_t frame_addr = reinterpret_cast<uint32_t>(&jojo_start[frame * frame_size]);
-  
-  scsp_slot& slotp = scsp.reg.slot[(frame - 1) % 32];
+
+  scsp_slot& slotp = scsp.reg.slot[(slot_ix - 1) & 31];
   slotp.LOOP = 0;
   slotp.LOOP |= LOOP__KYONEX;
 
-  scsp_slot& slot = scsp.reg.slot[(frame) % 32];
+  scsp.reg.ctrl.STATUS = STATUS__MSLC(slot_ix & 31);
+  scsp_slot& slot = scsp.reg.slot[slot_ix & 31];
   slot.LOOP = LOOP__KYONB | LOOP__SA(frame_addr); // kx kb sbctl[1:0] ssctl[1:0] lpctl[1:0] 8b sa[19:16]
   slot.SA = SA__SA(frame_addr); // start address (bytes)
   slot.LSA = 0; // loop start address (samples)
@@ -42,54 +45,45 @@ void auto_vector_1(void)
 
   slot.LOOP |= LOOP__KYONEX;
 
+  scsp.ram.u32[0] = (0xdead << 16) | (slot_ix);
+  scsp.ram.u32[1] = frame;
   frame++;
-  
+  slot_ix++;
+
   return;
 }
 
 void main()
 {
   for (long i = 0; i < 807; i++) { asm volatile ("nop"); }   // wait for (way) more than 30µs
- 
+
   for (int i = 0; i < 32; i++) {
     scsp_slot& slot = scsp.reg.slot[i];
     slot.LOOP = 0;
+    slot.SA = 0;
+    slot.LSA = 0;
+    slot.LEA = 0;
+    slot.EG = 0;
+    slot.VOLUME = 0;
+    slot.FM = 0;
+    slot.PITCH = 0;
+    slot.LFO = 0;
+    slot.MIXER = 0;
     slot.LOOP |= LOOP__KYONEX;
   }
 
   scsp.reg.ctrl.MIXER = MIXER__MEM4MB | MIXER__MVOL(0xf);
-  
+
   // timer A is vector 1 (0b001)
   scsp.reg.ctrl.SCILV2 = 0;
   scsp.reg.ctrl.SCILV1 = 0;
   scsp.reg.ctrl.SCILV0 = SCILV__TIMER_A;
-  
+
   // enable TIMER_A
   scsp.reg.ctrl.SCIRE = INT__TIMER_A;
   scsp.reg.ctrl.SCIEB = INT__TIMER_A;
 
-  scsp.reg.ctrl.TIMA = TIMA__TACTL(tactl) | TIMA__TIMA(tima);  
-  
+  scsp.reg.ctrl.TIMA = TIMA__TACTL(tactl) | TIMA__TIMA(tima);
+
   asm volatile ("move.w #8192,%sr");
-
-  /*
-  const uint16_t * jojo_start = reinterpret_cast<uint16_t *>(&_jojo_start);
-  const uint32_t frame_addr = reinterpret_cast<uint32_t>(&jojo_start[frame * 128]);
-  scsp_slot& slot = scsp.reg.slot[0];
-  slot.LOOP = 0;
-  slot.LOOP |= LOOP__KYONEX;
-  
-  slot.LOOP = LOOP__KYONB | LOOP__SA(frame_addr); // kx kb sbctl[1:0] ssctl[1:0] lpctl[1:0] 8b sa[19:16]
-  slot.SA = SA__SA(frame_addr); // start address (bytes)
-  slot.LSA = 0; // loop start address (samples)
-  slot.LEA = 128; // loop end address (samples)
-  slot.EG = EG__AR(0x1f) | EG__EGHOLD; // d2r d1r ho ar krs dl rr
-  slot.VOLUME = 0; // stwinh sdir tl
-  slot.FM = 0; // mdl mdxsl mdysl
-  slot.PITCH = PITCH__OCT(-2) | PITCH__FNS(0); // oct fns
-  slot.LFO = 0; // lfof plfows
-  slot.MIXER = MIXER__DISDL(0b101); // disdl dipan efsdl efpan
-
-  slot.LOOP |= LOOP__KYONEX;
-  */
 }
diff --git a/m68k/slot.cpp b/m68k/slot.cpp
index 86bfeb5..0f57ee2 100644
--- a/m68k/slot.cpp
+++ b/m68k/slot.cpp
@@ -15,7 +15,7 @@ void main()
   scsp_slot& slot = scsp.reg.slot[0];
   slot.LOOP = 0;
   slot.LOOP |= LOOP__KYONEX;
-  
+
   slot.LOOP = LOOP__KYONB | LOOP__LPCTL__NORMAL | LOOP__SA(sine_start); // kx kb sbctl[1:0] ssctl[1:0] lpctl[1:0] 8b sa[19:16]
   slot.SA = SA__SA(sine_start); // start address (bytes)
   slot.LSA = 0; // loop start address (samples)
diff --git a/res/Bm437_SperryPC_CGA.otb b/res/Bm437_SperryPC_CGA.otb
new file mode 100644
index 0000000..10d9d0a
Binary files /dev/null and b/res/Bm437_SperryPC_CGA.otb differ
diff --git a/scsp/sound_cpu__interrupt.cpp b/scsp/sound_cpu__interrupt.cpp
index bccae66..993360b 100644
--- a/scsp/sound_cpu__interrupt.cpp
+++ b/scsp/sound_cpu__interrupt.cpp
@@ -2,13 +2,80 @@
 
 #include "smpc.h"
 #include "scsp.h"
+#include "vdp1.h"
+#include "vdp2.h"
+#include "scu.h"
+#include "sh2.h"
 
 #include "../common/copy.hpp"
+#include "../common/font.hpp"
+#include "../common/draw_font.hpp"
+#include "../common/palette.hpp"
+#include "../common/vdp2_func.hpp"
 
 extern void * _m68k_start __asm("_binary_m68k_interrupt_bin_start");
 extern void * _m68k_size __asm("_binary_m68k_interrupt_bin_size");
+extern void * _sperrypc_start __asm("_binary_res_sperrypc_font_bin_start");
 
-void main()
+struct draw_state {
+  uint32_t cmd_ix;
+  struct draw_font::state font;
+};
+
+static struct draw_state draw_state;
+
+uint32_t print_hex(uint8_t * c, uint32_t len, uint32_t n)
+{
+  uint32_t ret = 0;
+
+  while (len > 0) {
+    uint32_t nib = n & 0xf;
+    n = n >> 4;
+
+    if (nib > 9) {
+      nib += (97 - 10);
+    } else {
+      nib += (48 - 0);
+    }
+
+    c[--len] = nib;
+
+    ret++;
+  }
+  return ret;
+}
+
+int32_t draw_string(const uint8_t * string,
+		    const uint32_t length,
+		    const int32_t x,
+		    const int32_t y)
+{
+  return draw_font::horizontal_string(draw_state.font,
+				      draw_state.cmd_ix,
+				      string,
+				      length,
+				       (x    ),
+				      ((y + 1) * 8) << 6);
+}
+
+static inline uint32_t init_font(uint32_t top)
+{
+  // 256 is the number of colors in the color palette, not the number of grays
+  // that are used by the font.
+  constexpr uint32_t colors_per_palette = 256;
+  constexpr uint32_t color_bank_index = 0; // completely random and arbitrary value
+
+  palette::vdp2_cram_32grays(colors_per_palette, color_bank_index);
+  // For color bank color, COLR is concatenated bitwise with pixel data. See
+  // Figure 6.17 in the VDP1 manual.
+  draw_state.font.color_address = color_bank_index << 8;
+
+  top = font_data(&_sperrypc_start, top, draw_state.font);
+
+  return top;
+}
+
+void init_sound()
 {
   /* SEGA SATURN TECHNICAL BULLETIN # 51
 
@@ -23,6 +90,12 @@ void main()
 
   scsp.reg.ctrl.MIXER = MIXER__MEM4MB;
 
+  /*
+    The Saturn BIOS does not (un)initialize the DSP. Without zeroizing the DSP
+    program, the SCSP DSP appears to have a program that continuously writes to
+    0x30000 through 0x3ffff in sound RAM, which has the effect of destroying any
+    samples stored there.
+  */  
   reg32 * dsp_steps = reinterpret_cast<reg32*>(&(scsp.reg.dsp.STEP[0].MPRO[0]));
   fill<reg32>(dsp_steps, 0, (sizeof (scsp.reg.dsp.STEP)));
 
@@ -34,8 +107,157 @@ void main()
   smpc.reg.SF = 1;
   smpc.reg.COMREG = COMREG__SNDON;
   while (smpc.reg.oreg[31] != OREG31__SNDON);
+}
 
-  // do nothing while the sound CPU manipulates the SCSP
+static inline void init_vdp()
+{
+  // wait for the beginning of a V blank
+  v_blank_in();
+
+  // DISP: Please make sure to change this bit from 0 to 1 during V blank.
+  vdp2.reg.TVMD = ( TVMD__DISP | TVMD__LSMD__NON_INTERLACE
+                  | TVMD__VRESO__240 | TVMD__HRESO__NORMAL_320);
+
+  // disable all VDP2 backgrounds (e.g: the Sega bios logo)
+  vdp2.reg.BGON = 0;
+
+  // zeroize BACK color
+  vdp2.reg.BKTAU = 0;
+  vdp2.reg.BKTAL = 0;
+  vdp2.vram.u16[0] = 0;
+
+  vdp2.reg.PRISA = PRISA__S0PRIN(1); // Sprite register 0 PRIority Number
+
+  /* TVM settings must be performed from the second H-blank IN interrupt after the
+  V-blank IN interrupt to the H-blank IN interrupt immediately after the V-blank
+  OUT interrupt. */
+  // "normal" display resolution, 16 bits per pixel, 512x256 framebuffer
+  vdp1.reg.TVMR = TVMR__TVM__NORMAL;
+
+  // swap framebuffers every 1 cycle; non-interlace
+  vdp1.reg.FBCR = 0;
+
+  // during a framebuffer erase cycle, write the color "black" to each pixel
+  constexpr uint16_t black = 0x0000;
+  vdp1.reg.EWDR = black;
+
+  // erase upper-left coordinate
+  vdp1.reg.EWLR = EWLR__16BPP_X1(0) | EWLR__Y1(0);
+
+  // erase lower-right coordinate
+  vdp1.reg.EWRR = EWRR__16BPP_X3(319) | EWRR__Y3(239);
+
+  vdp1.vram.cmd[0].CTRL = CTRL__JP__JUMP_NEXT | CTRL__COMM__SYSTEM_CLIP_COORDINATES;
+  vdp1.vram.cmd[0].LINK = 0;
+  vdp1.vram.cmd[0].XC = 319;
+  vdp1.vram.cmd[0].YC = 239;
+
+  vdp1.vram.cmd[1].CTRL = CTRL__JP__JUMP_NEXT | CTRL__COMM__LOCAL_COORDINATE;
+  vdp1.vram.cmd[1].LINK = 0;
+  vdp1.vram.cmd[1].XA = 0;
+  vdp1.vram.cmd[1].YA = 0;
+
+  vdp1.vram.cmd[2].CTRL = CTRL__END;
+
+  draw_state.cmd_ix = 2;
+
+  // start drawing (execute the command list) on every frame
+  vdp1.reg.PTMR = PTMR__PTM__FRAME_CHANGE;
+}
+
+template <uint32_t n, uint32_t size>
+inline void draw_label(const uint8_t(&label)[size], int32_t advance, int32_t & row, uint32_t value)
+{
+  advance = (advance * 8) << 6;
+  advance += draw_string(label, (sizeof (label)) - 1, advance, row);
+  uint8_t v[n];
+  print_hex(v, (sizeof (v)), value);
+  draw_string(v, (sizeof (v)), advance, row);
+
+  row++;  
+};
+
+static inline void render()
+{
+  draw_state.cmd_ix = 2;
+
+  if ((scsp.ram.u32[0] & 0xffff0000) != static_cast<uint32_t>(0xdead << 16))
+    return;
+
+  uint16_t slot_ix = scsp.ram.u32[0] & 0xffff;
+  uint16_t slotp_ix = (slot_ix - 1) & 31;
+  slot_ix &= 31;
+  uint32_t frame = scsp.ram.u32[1];
+
+  int32_t row = 1;
+
+  const uint8_t frame_l[] = "frame: 0x";
+  draw_label<4>(frame_l, 1, row, frame);
+
+  row++;
+
+  // previous slot
+  const uint8_t slotp_l[] = "slot_previous: 0x";
+  draw_label<2>(slotp_l, 1, row, slotp_ix);
+
+  scsp_slot& slot_p = scsp.reg.slot[slotp_ix];
+  
+  const uint8_t kyonb_l[] = "KYONB: ";  
+  const uint32_t kyonb_p = (slot_p.LOOP >> 11) & 1;
+  draw_label<1>(kyonb_l, 3, row, kyonb_p);
+
+  row++;
+
+  // current slot
+  const uint8_t slot_l[] = "slot: 0x";
+  draw_label<2>(slot_l, 1, row, slot_ix);
+
+  scsp_slot& slot = scsp.reg.slot[slot_ix];
+
+  uint32_t kyonb = (slot.LOOP >> 11) & 1;
+  draw_label<1>(kyonb_l, 3, row, kyonb);
+
+  const uint8_t sa_l[] = "SA: 0x";
+  const uint32_t sa = ((slot.LOOP & 0b1111) << 16) | slot.SA;
+  draw_label<5>(sa_l, 3, row, sa);
+
+  const uint8_t lsa_l[] = "LSA: 0x";
+  draw_label<4>(lsa_l, 3, row, slot.LSA);
+
+  const uint8_t lea_l[] = "LEA: 0x";
+  draw_label<4>(lea_l, 3, row, slot.LEA);    
+  
+  row++;
+
+  // global
+  const uint8_t ca_l[] = "STATUS__CA: 0x";
+  draw_label<1>(ca_l, 1, row, STATUS__CA(scsp.reg.ctrl.STATUS));
+
+  vdp1.vram.cmd[draw_state.cmd_ix].CTRL = CTRL__END;
+}
+
+void v_blank_in_int(void) __attribute__ ((interrupt_handler));
+void v_blank_in_int()
+{
+  scu.reg.IST &= ~(IST__V_BLANK_IN);
+  scu.reg.IMS = ~(IMS__V_BLANK_IN);
+
+  render();
+}
+
+void main()
+{
+  uint32_t top = (sizeof (union vdp1_vram));
+  top = init_font(top);
+
+  init_vdp();
+
+  sh2_vec[SCU_VEC__V_BLANK_IN] = (u32)(&v_blank_in_int);
+
+  scu.reg.IST = 0;
+  scu.reg.IMS = ~(IMS__V_BLANK_IN);
+
+  init_sound();
 }
 
 extern "C"
diff --git a/scsp/sound_cpu__slot.cpp b/scsp/sound_cpu__slot.cpp
index a5b6b14..4559011 100644
--- a/scsp/sound_cpu__slot.cpp
+++ b/scsp/sound_cpu__slot.cpp
@@ -15,23 +15,32 @@ void main()
      The document suggests that Sound RAM is (somewhat) preserved
      during SNDOFF.
    */
-  
+
   while ((smpc.reg.SF & 1) != 0);
   smpc.reg.SF = 1;
   smpc.reg.COMREG = COMREG__SNDOFF;
   while (smpc.reg.oreg[31] != OREG31__SNDOFF);
 
   scsp.reg.ctrl.MIXER = MIXER__MEM4MB;
+
+  /*
+    The Saturn BIOS does not (un)initialize the DSP. Without zeroizing the DSP
+    program, the SCSP DSP appears to have a program that continuously writes to
+    0x30000 through 0x3ffff in sound RAM, which has the effect of destroying any
+    samples stored there.
+  */
+  reg32 * dsp_steps = reinterpret_cast<reg32*>(&(scsp.reg.dsp.STEP[0].MPRO[0]));
+  fill<reg32>(dsp_steps, 0, (sizeof (scsp.reg.dsp.STEP)));
   
   uint32_t * m68k_main_start = reinterpret_cast<uint32_t*>(&_m68k_start);
   uint32_t m68k_main_size = reinterpret_cast<uint32_t>(&_m68k_size);
   copy<uint32_t>(&scsp.ram.u32[0], m68k_main_start, m68k_main_size);
-  
+
   while ((smpc.reg.SF & 1) != 0);
   smpc.reg.SF = 1;
   smpc.reg.COMREG = COMREG__SNDON;
   while (smpc.reg.oreg[31] != OREG31__SNDON);
-  
+
   // do nothing while the sound CPU manipulates the SCSP
 }
 
diff --git a/sh/lib1funcs.S b/sh/lib1funcs.S
index 2805841..2b32b94 100644
--- a/sh/lib1funcs.S
+++ b/sh/lib1funcs.S
@@ -45,7 +45,7 @@ see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see
 #if defined __vxworks && defined __PIC__
 #define NO_FPSCR_VALUES
 #endif
-	
+
 #ifdef L_ashiftrt
 	.global	GLOBAL(ashiftrt_r4_0)
 	.global	GLOBAL(ashiftrt_r4_1)
@@ -552,7 +552,7 @@ LOCAL(ashlsi_29):
 LOCAL(ashlsi_30):
 	shlr2	r0
 	rts
-	shll16	r0	
+	shll16	r0
 
 	ENDFUNC(GLOBAL(ashlsi3))
 	ENDFUNC(GLOBAL(ashlsi3_r0))
@@ -728,7 +728,7 @@ LOCAL(lshrsi_29):
 LOCAL(lshrsi_30):
 	shll2	r0
 	rts
-	shlr16	r0	
+	shlr16	r0
 
 	ENDFUNC(GLOBAL(lshrsi3))
 	ENDFUNC(GLOBAL(lshrsi3_r0))
@@ -1063,7 +1063,7 @@ GLOBAL(sdivsi3_i4):
 	fdiv	dr2,dr0
 	ftrc	dr0,fpul
 	rts
-	fpchg	
+	fpchg
 
 #endif /* __SH4A__  */
 
@@ -1722,7 +1722,7 @@ LOCAL(div_ge64k_2):
 	and r1,r0
 	bra LOCAL(div_ge64k_end)
 	xor r4,r0
-	
+
 LOCAL(div_r8):
 	shll16 r4
 	bra LOCAL(div_r8_2)
@@ -2257,7 +2257,7 @@ GLOBAL(udiv_qrnnd_16):
 	cmp/hi r6,r0
 	bt .Lots
 	.rept 16
-	div1 r6,r0 
+	div1 r6,r0
 	.endr
 	extu.w r0,r1
 	bt 0f
diff --git a/tools/Makefile b/tools/Makefile
index e654c0d..cade8b0 100644
--- a/tools/Makefile
+++ b/tools/Makefile
@@ -10,6 +10,11 @@ all: ttf-convert
 %: %.o
 	$(CXX) $(LDFLAGS) $< -o $@
 
-.SUFFIXES:
+clean:
+	rm -f *.o ttf-convert
+
+.SUFFIXES:
+.INTERMEDIATE:
+.SECONDARY:
+.PHONY: all clean
 
-# /usr/share/fonts/OTF/ipagp.ttf
diff --git a/tools/ttf-convert.cpp b/tools/ttf-convert.cpp
index df9d57a..39d8446 100644
--- a/tools/ttf-convert.cpp
+++ b/tools/ttf-convert.cpp
@@ -68,7 +68,7 @@ load_outline_char(const FT_Face face,
     return -1;
   }
 
-  assert(face->glyph->format == FT_GLYPH_FORMAT_OUTLINE);
+  //assert(face->glyph->format == FT_GLYPH_FORMAT_OUTLINE);
 
   error = FT_Render_Glyph(face->glyph, FT_RENDER_MODE_NORMAL);
   if (error) {
@@ -79,6 +79,8 @@ load_outline_char(const FT_Face face,
   uint32_t pitch = (face->glyph->bitmap.pitch + 8 - 1) & -8;
   uint32_t bitmap_size = face->glyph->bitmap.rows * pitch;
 
+  //printf("num_grays %d\n", face->glyph->bitmap.num_grays);
+
   if (!(face->glyph->bitmap.pitch > 0)) {
     assert(pitch == 0);
     assert(bitmap_size == 0);
@@ -86,11 +88,41 @@ load_outline_char(const FT_Face face,
     assert(face->glyph->bitmap.rows == 0);
   }
 
-  for (uint32_t y = 0; y < face->glyph->bitmap.rows; y++) {
-    for (uint32_t x = 0; x < face->glyph->bitmap.width; x++) {
-      uint8_t i = face->glyph->bitmap.buffer[y * face->glyph->bitmap.pitch + x];
-      glyph_bitmaps[bitmap_offset + (y * pitch + x)] = i >> 3;
+  switch (face->glyph->bitmap.num_grays) {
+  case 2:
+    {
+      for (unsigned int y = 0; y < face->glyph->bitmap.rows; y++) {
+	uint8_t * row = &face->glyph->bitmap.buffer[y * face->glyph->bitmap.pitch];
+	//uint8_t row_out = 0;
+	for (unsigned int x = 0; x < face->glyph->bitmap.width; x++) {
+	  int bit;
+	  bit = (row[x / 8] >> (7 - (x % 8))) & 1;
+	  //fprintf(stderr, bit ? "█" : " ");
+	  //row_out |= (bit << x);
+
+	  // FIXME: this is lazy: this bloats the storage format from
+	  // 1 bit per pixel to 8 bits per pixel, even though this
+	  // expansion could be done at runtime inside vdp1 vram
+	  // instead.
+	  glyph_bitmaps[bitmap_offset + (y * pitch + x)] = bit ? 31 : 0;
+	}
+	//fprintf(stderr, "\n");
+	//buf[y] = row_out;
+      }
     }
+    break;
+  case 256:
+    {
+      for (uint32_t y = 0; y < face->glyph->bitmap.rows; y++) {
+	for (uint32_t x = 0; x < face->glyph->bitmap.width; x++) {
+	  uint8_t i = face->glyph->bitmap.buffer[y * face->glyph->bitmap.pitch + x];
+	  glyph_bitmaps[bitmap_offset + (y * pitch + x)] = i >> 3;
+	}
+      }
+    }
+    break;
+  default:
+    assert(-1 == face->glyph->bitmap.num_grays);
   }
 
   //memcpy(&glyph_bitmaps[bitmap_offset], face->glyph->bitmap.buffer, bitmap_size);
@@ -130,7 +162,7 @@ int main(int argc, char *argv[])
   FT_Face face;
   FT_Error error;
 
-  if (argc < 5) {
+  if (argc != 6) {
     std::cerr << "usage: " << argv[0] << " [start-hex] [end-hex] [pixel-size] [font-file-path] [output-file-path]\n\n";
     std::cerr << "   ex: " << argv[0] << " 3000 30ff 30 ipagp.ttf font.bin\n";
     return -1;
@@ -178,9 +210,6 @@ int main(int argc, char *argv[])
   ss2 << std::hex << argv[2];
   ss2 >> end;
 
-  //uint32_t start = 0x20;
-  //uint32_t end = 0x7f;
-
   glyph glyphs[(end - start) + 1];
   uint8_t glyph_bitmaps[1024 * 1024];
   memset(glyph_bitmaps, 0x00, 1024 * 1024);
@@ -218,6 +247,10 @@ int main(int argc, char *argv[])
   std::cerr << "glyph_index: 0x" << std::hex << glyph_index << '\n';
 
   FILE * out = fopen(argv[5], "w");
+  if (out == NULL) {
+    perror("fopen(w)");
+    return -1;
+  }
 
   fwrite(reinterpret_cast<void*>(&font), (sizeof (font)), 1, out);
   fwrite(reinterpret_cast<void*>(&glyphs[0]), (sizeof (glyph)), glyph_index, out);
diff --git a/wordle/main_saturn.cpp b/wordle/main_saturn.cpp
index 20aed35..8afe441 100644
--- a/wordle/main_saturn.cpp
+++ b/wordle/main_saturn.cpp
@@ -203,7 +203,8 @@ void render()
 }
 
 void v_blank_in_int(void) __attribute__ ((interrupt_handler));
-void v_blank_in_int() {
+void v_blank_in_int()
+{
   scu.reg.IST &= ~(IST__V_BLANK_IN);
   scu.reg.IMS = ~(IMS__SMPC | IMS__V_BLANK_IN);
 
@@ -262,14 +263,14 @@ void main()
   uint32_t top = (sizeof (union vdp1_vram));
   top = init_font(top);
 
-  // wait for the beginning of a V blank
-  v_blank_in();
-
   // wordle init
   const uint8_t word_ix = 6;
   wordle::init_screen(wordle_state, word_ix);
   // end wordle init
 
+  // wait for the beginning of a V blank
+  v_blank_in();
+
   // DISP: Please make sure to change this bit from 0 to 1 during V blank.
   vdp2.reg.TVMD = ( TVMD__DISP | TVMD__LSMD__NON_INTERLACE
                   | TVMD__VRESO__240 | TVMD__HRESO__NORMAL_320);