diff --git a/emul/hw/sms/.gitignore b/emul/hw/sms/.gitignore
new file mode 100644
index 0000000..2ad1042
--- /dev/null
+++ b/emul/hw/sms/.gitignore
@@ -0,0 +1 @@
+/sms
diff --git a/emul/hw/sms/Makefile b/emul/hw/sms/Makefile
new file mode 100644
index 0000000..3e802de
--- /dev/null
+++ b/emul/hw/sms/Makefile
@@ -0,0 +1,15 @@
+EXTOBJS = ../../emul.o ../../libz80/libz80.o
+OBJS = sms.o vdp.o
+TARGET = sms
+CFLAGS += `pkg-config --cflags xcb`
+
+.PHONY: all
+all: $(TARGET)
+
+$(TARGET): $(OBJS) $(EXTOBJS)
+	$(CC) `pkg-config --libs xcb` $(OBJS) $(EXTOBJS) -o $@
+
+.PHONY: clean
+clean:
+	rm -f $(TARGET) $(OBJS)
+	
diff --git a/emul/hw/sms/sms.c b/emul/hw/sms/sms.c
new file mode 100644
index 0000000..89e4865
--- /dev/null
+++ b/emul/hw/sms/sms.c
@@ -0,0 +1,194 @@
+#include <stdlib.h>
+#include <stdio.h>
+#include <stdbool.h>
+
+#include <xcb/xcb.h>
+
+#include "../../emul.h"
+#include "vdp.h"
+
+#define RAMSTART 0xc000
+#define VDP_CMD_PORT 0xbf
+#define VDP_DATA_PORT 0xbe
+#define MAX_ROMSIZE 0x8000
+
+static xcb_connection_t    *conn;
+static xcb_screen_t        *screen;
+
+/* graphics contexts */
+static xcb_gcontext_t       fg;
+/* win */
+static xcb_drawable_t       win;
+
+// pixels to draw. We draw them in one shot.
+static xcb_rectangle_t rectangles[VDP_SCREENW*VDP_SCREENH];
+
+static Machine *m;
+static VDP vdp;
+static bool vdp_changed;
+
+static uint8_t iord_vdp_cmd()
+{
+    return vdp_cmd_rd(&vdp);
+}
+
+static uint8_t iord_vdp_data()
+{
+    return vdp_data_rd(&vdp);
+}
+
+static void iowr_vdp_cmd(uint8_t val)
+{
+    vdp_cmd_wr(&vdp, val);
+}
+
+static void iowr_vdp_data(uint8_t val)
+{
+    vdp_changed = true;
+    vdp_data_wr(&vdp, val);
+}
+
+void create_window()
+{
+    uint32_t mask;
+    uint32_t values[2];
+
+    /* Create the window */
+    win = xcb_generate_id(conn);
+    mask = XCB_CW_BACK_PIXEL | XCB_CW_EVENT_MASK;
+    values[0] = screen->white_pixel;
+    values[1] = XCB_EVENT_MASK_EXPOSURE | XCB_EVENT_MASK_KEY_PRESS |
+        XCB_EVENT_MASK_KEY_RELEASE;
+    xcb_create_window(
+        conn,
+        screen->root_depth,
+        win,
+        screen->root,
+        0, 0,
+        150, 150,
+        10,
+        XCB_WINDOW_CLASS_INPUT_OUTPUT,
+        screen->root_visual,
+        mask, values);
+
+    fg = xcb_generate_id(conn);
+    mask = XCB_GC_FOREGROUND | XCB_GC_GRAPHICS_EXPOSURES;
+    values[0] = screen->black_pixel;
+    values[1] = 0;
+    xcb_create_gc(conn, fg, screen->root, mask, values);
+
+    /* Map the window on the screen */
+    xcb_map_window(conn, win);
+}
+
+void draw_pixels()
+{
+    xcb_get_geometry_reply_t *geom;
+
+    geom = xcb_get_geometry_reply(conn, xcb_get_geometry(conn, win), NULL);
+
+    xcb_clear_area(
+        conn, 0, win, 0, 0, geom->width, geom->height);
+    // Figure out inner size to maximize our screen's aspect ratio
+    int psize = geom->height / VDP_SCREENH;
+    if (geom->width / VDP_SCREENW < psize) {
+        // width is the constraint
+        psize = geom->width / VDP_SCREENW;
+    }
+    int innerw = psize * VDP_SCREENW;
+    int innerh = psize * VDP_SCREENH;
+    int innerx = (geom->width - innerw) / 2;
+    int innery = (geom->height - innerh) / 2;
+    int drawcnt = 0;
+    for (int i=0; i<VDP_SCREENW; i++) {
+        for (int j=0; j<VDP_SCREENH; j++) {
+            if (vdp_pixel(&vdp, i, j)) {
+                int x = innerx + (i*psize);
+                int y = innery + (j*psize);
+                rectangles[drawcnt].x = x;
+                rectangles[drawcnt].y = y;
+                rectangles[drawcnt].height = psize;
+                rectangles[drawcnt].width = psize;
+                drawcnt++;
+            }
+        }
+    }
+    if (drawcnt) {
+        xcb_poly_fill_rectangle(
+            conn, win, fg, drawcnt, rectangles);
+    }
+    vdp_changed = false;
+    xcb_flush(conn);
+}
+
+void event_loop()
+{
+    while (1) {
+        emul_step();
+        if (vdp_changed) {
+            // To avoid overdrawing, we'll let the CPU run a bit to finish its
+            // drawing operation.
+            emul_steps(100);
+            draw_pixels();
+        }
+        xcb_generic_event_t *e = xcb_poll_for_event(conn);
+        if (!e) {
+            continue;
+        }
+        switch (e->response_type & ~0x80) {
+        /* ESC to exit */
+        case XCB_KEY_RELEASE:
+        case XCB_KEY_PRESS: {
+            xcb_key_press_event_t *ev = (xcb_key_press_event_t *)e;
+            if (ev->detail == 0x09) return;
+            break;
+        }
+        case XCB_EXPOSE: {
+            draw_pixels();
+            break;
+        }
+        default: {
+            break;
+        }
+        }
+        free(e);
+    }
+}
+
+int main(int argc, char *argv[])
+{
+    if (argc != 2) {
+        fprintf(stderr, "Usage: ./sms /path/to/rom\n");
+        return 1;
+    }
+    FILE *fp = fopen(argv[1], "r");
+    if (fp == NULL) {
+        fprintf(stderr, "Can't open %s\n", argv[1]);
+        return 1;
+    }
+    m = emul_init();
+    m->ramstart = RAMSTART;
+    int i = 0;
+    int c;
+    while ((c = fgetc(fp)) != EOF && i < MAX_ROMSIZE) {
+        m->mem[i++] = c & 0xff;
+    }
+    pclose(fp);
+    if (c != EOF) {
+        fprintf(stderr, "ROM image too large.\n");
+        return 1;
+    }
+    vdp_init(&vdp);
+    vdp_changed = false;
+    m->iord[VDP_CMD_PORT] = iord_vdp_cmd;
+    m->iord[VDP_DATA_PORT] = iord_vdp_data;
+    m->iowr[VDP_CMD_PORT] = iowr_vdp_cmd;
+    m->iowr[VDP_DATA_PORT] = iowr_vdp_data;
+    conn = xcb_connect(NULL, NULL);
+    screen = xcb_setup_roots_iterator(xcb_get_setup(conn)).data;
+    create_window();
+    draw_pixels();
+    event_loop();
+    emul_printdebug();
+    return 0;
+}
diff --git a/emul/hw/sms/vdp.c b/emul/hw/sms/vdp.c
new file mode 100644
index 0000000..2a7f789
--- /dev/null
+++ b/emul/hw/sms/vdp.c
@@ -0,0 +1,77 @@
+#include <string.h>
+#include "vdp.h"
+
+void vdp_init(VDP *vdp)
+{
+    memset(vdp->vram, 0, VDP_VRAM_SIZE);
+    memset(vdp->regs, 0, 0x10);
+    vdp->has_cmdlsb = false;
+    vdp->curaddr = 0;
+}
+
+uint8_t vdp_cmd_rd(VDP *vdp)
+{
+    return 0;
+}
+
+void vdp_cmd_wr(VDP *vdp, uint8_t val)
+{
+    if (!vdp->has_cmdlsb) {
+        vdp->cmdlsb = val;
+        vdp->has_cmdlsb = true;
+        return;
+    }
+    vdp->has_cmdlsb = false;
+    if ((val & 0xc0) == 0x80) {
+        // set register
+        vdp->regs[val&0xf] = vdp->cmdlsb;
+    } else if ((val & 0xc0) == 0xc0) {
+        // palette RAM
+        vdp->curaddr = 0x4000 + (vdp->cmdlsb&0x1f);
+    } else {
+        // VRAM
+        vdp->curaddr = ((val&0x3f) << 8) + vdp->cmdlsb;
+    }
+}
+
+uint8_t vdp_data_rd(VDP *vdp)
+{
+    uint8_t res = vdp->vram[vdp->curaddr];
+    if (vdp->curaddr < VDP_VRAM_SIZE) {
+        vdp->curaddr++;
+    }
+    return res;
+}
+
+void vdp_data_wr(VDP *vdp, uint8_t val)
+{
+    vdp->vram[vdp->curaddr] = val;
+    if (vdp->curaddr < VDP_VRAM_SIZE) {
+        vdp->curaddr++;
+    }
+}
+
+uint8_t vdp_pixel(VDP *vdp, uint16_t x, uint16_t y)
+{
+    if (x >= VDP_SCREENW) {
+        return 0;
+    }
+    if (y >= VDP_SCREENH) {
+        return 0;
+    }
+    // name table offset
+    uint16_t offset = 0x3800 + ((y/8) << 6) + ((x/8) << 1);
+    uint16_t tableval = vdp->vram[offset] + (vdp->vram[offset+1] << 8);
+    uint16_t tilenum = tableval & 0x1ff;
+    // tile offset this time. Each tile is 0x20 bytes long.
+    offset = tilenum * 0x20;
+    // Each 4 byte is a row. Find row first.
+    offset += ((y%8) * 4);
+    uint8_t bitnum = 7 - (x%8);
+    // Now, let's compose the result by pushing the right bit of our 4 bytes
+    // into our result.
+    return ((vdp->vram[offset] >> bitnum) & 1) + \
+           (((vdp->vram[offset+1] >> bitnum) & 1) << 1) + \
+           (((vdp->vram[offset+2] >> bitnum) & 1) << 2) + \
+           (((vdp->vram[offset+3] >> bitnum) & 1) << 3);
+}
diff --git a/emul/hw/sms/vdp.h b/emul/hw/sms/vdp.h
new file mode 100644
index 0000000..e36e014
--- /dev/null
+++ b/emul/hw/sms/vdp.h
@@ -0,0 +1,26 @@
+#include <stdint.h>
+#include <stdbool.h>
+
+#define VDP_VRAM_SIZE 0x4020
+#define VDP_SCREENW (32*8)
+#define VDP_SCREENH (24*8)
+// Offset of the name table
+#define VDP_NTABLE_OFFSET 0x3800
+
+
+typedef struct {
+    // the last 0x20 is palette RAM
+    uint8_t vram[VDP_VRAM_SIZE];
+    uint8_t regs[0x10];
+    uint8_t cmdlsb;
+    bool has_cmdlsb;
+    uint16_t curaddr;
+} VDP;
+
+void vdp_init(VDP *vdp);
+uint8_t vdp_cmd_rd(VDP *vdp);
+void vdp_cmd_wr(VDP *vdp, uint8_t val);
+uint8_t vdp_data_rd(VDP *vdp);
+void vdp_data_wr(VDP *vdp, uint8_t val);
+// result is a RGB value
+uint8_t vdp_pixel(VDP *vdp, uint16_t x, uint16_t y);