diff --git a/src/include/86box/vid_8514a.h b/src/include/86box/vid_8514a.h
index 57e98cc44..bfde22d5e 100644
--- a/src/include/86box/vid_8514a.h
+++ b/src/include/86box/vid_8514a.h
@@ -213,7 +213,7 @@ typedef struct ibm8514_t {
     int     vdisp2;
     int     disp_cntl;
     int     interlace;
-    uint8_t subsys_cntl;
+    uint16_t subsys_cntl;
     uint8_t subsys_stat;
 
     atomic_int fifo_idx;
diff --git a/src/video/vid_8514a.c b/src/video/vid_8514a.c
index 379772820..765258237 100644
--- a/src/video/vid_8514a.c
+++ b/src/video/vid_8514a.c
@@ -715,27 +715,13 @@ ibm8514_accel_out(uint16_t port, uint32_t val, svga_t *svga, int len)
                 break;
 
             case 0x42e8:
-                if (val & 0x01)
-                    dev->subsys_stat &= ~0x01;
-                if (val & 0x02)
-                    dev->subsys_stat &= ~0x02;
-                if (val & 0x04)
-                    dev->subsys_stat &= ~0x04;
-                if (val & 0x08)
-                    dev->subsys_stat &= ~0x08;
+                ibm8514_log("VBLANK stat=%02x, val=%02x.\n", dev->subsys_stat, val);
+                dev->subsys_cntl = (dev->subsys_cntl & 0xff00) | val;
+                dev->subsys_stat &= ~val;
                 break;
             case 0x42e9:
-                dev->subsys_cntl = val;
-                if (val & 0x01)
-                    dev->subsys_stat |= 0x01;
-                if (val & 0x02)
-                    dev->subsys_stat |= 0x02;
-                if (val & 0x04)
-                    dev->subsys_stat |= 0x04;
-                if (val & 0x08)
-                    dev->subsys_stat |= 0x08;
-
-                if ((val & 0xc0) == 0xc0) {
+                dev->subsys_cntl = (dev->subsys_cntl & 0xff) | (val << 8);
+                if ((val & 0xc0) == 0x80) {
                     dev->fifo_idx = 0;
                     dev->force_busy = 0;
                     dev->force_busy2 = 0;
@@ -882,10 +868,10 @@ ibm8514_accel_in(uint16_t port, svga_t *svga)
 
     switch (port) {
         case 0x2e8:
-            if (dev->vc == dev->v_syncstart)
+            if (dev->vc == dev->dispend)
                 temp |= 0x02;
 
-            ibm8514_log("0x2E8 read: Display Status=%02x.\n", temp);
+            ibm8514_log("Read: Display Status1=%02x.\n", temp);
             break;
 
         case 0x6e8:
@@ -910,21 +896,25 @@ ibm8514_accel_in(uint16_t port, svga_t *svga)
 
         case 0x42e8:
         case 0x42e9:
-            if (dev->vc == dev->v_syncstart)
-                dev->subsys_stat |= 0x01;
+            if ((dev->subsys_cntl & 0x01) && !(dev->subsys_stat & 0x01) && (dev->vc == dev->dispend))
+                temp |= 0x01;
 
             if (cmd == 6) {
-                if ((dev->accel.dx >= clip_l) &&
+                if ((dev->subsys_cntl & 0x02) &&
+                    !(dev->subsys_stat & 0x02) &&
+                    (dev->accel.dx >= clip_l) &&
                     (dev->accel.dx <= clip_r_ibm) &&
                     (dev->accel.dy >= clip_t) &&
                     (dev->accel.dy <= clip_b_ibm))
-                    dev->subsys_stat |= 0x02;
+                    temp |= 0x02;
             } else {
-                if ((dev->accel.cx >= clip_l) &&
+                if ((dev->subsys_cntl & 0x02) &&
+                    !(dev->subsys_stat & 0x02) &&
+                    (dev->accel.cx >= clip_l) &&
                     (dev->accel.cx <= clip_r_ibm) &&
                     (dev->accel.cy >= clip_t) &&
                     (dev->accel.cy <= clip_b_ibm))
-                    dev->subsys_stat |= 0x02;
+                    temp |= 0x02;
             }
 
             if (!dev->fifo_idx) {
@@ -932,9 +922,10 @@ ibm8514_accel_in(uint16_t port, svga_t *svga)
                     temp |= 0x08;
             }
 
-            if (port & 1)
+            if (port & 1) {
                 temp = dev->vram_512k_8514 ? 0x00 : 0x80;
-            else {
+                temp |= (dev->subsys_cntl >> 8);
+            } else {
                 temp |= (dev->subsys_stat | (dev->vram_512k_8514 ? 0x00 : 0x80));
                 temp |= 0x20;
             }
@@ -1155,6 +1146,7 @@ ibm8514_accel_start(int count, int cpu_input, uint32_t mix_dat, uint32_t cpu_dat
                         (dev->accel.cx <= clip_r) &&
                         (dev->accel.cy >= clip_t) &&
                         (dev->accel.cy <= clip_b)) {
+                        dev->subsys_stat |= 0x02;
                         switch ((mix_dat & mix_mask) ? frgd_mix : bkgd_mix) {
                             case 0:
                                 src_dat = bkgd_color;
@@ -1250,6 +1242,7 @@ ibm8514_accel_start(int count, int cpu_input, uint32_t mix_dat, uint32_t cpu_dat
                         (dev->accel.cx <= clip_r) &&
                         (dev->accel.cy >= clip_t) &&
                         (dev->accel.cy <= clip_b)) {
+                        dev->subsys_stat |= 0x02;
                         switch ((mix_dat & mix_mask) ? frgd_mix : bkgd_mix) {
                             case 0:
                                 src_dat = bkgd_color;
@@ -1385,6 +1378,7 @@ ibm8514_accel_start(int count, int cpu_input, uint32_t mix_dat, uint32_t cpu_dat
                         (dev->accel.cx <= clip_r) &&
                         (dev->accel.cy >= clip_t) &&
                         (dev->accel.cy <= clip_b)) {
+                        dev->subsys_stat |= 0x02;
                         if (ibm8514_cpu_dest(svga) && (pixcntl == 0)) {
                             mix_dat = mix_mask; /* Mix data = forced to foreground register. */
                         } else if (ibm8514_cpu_dest(svga) && (pixcntl == 3)) {
@@ -1543,6 +1537,7 @@ ibm8514_accel_start(int count, int cpu_input, uint32_t mix_dat, uint32_t cpu_dat
                             (dev->accel.cx <= clip_r) &&
                             (dev->accel.cy >= clip_t) &&
                             (dev->accel.cy <= clip_b)) {
+                            dev->subsys_stat |= 0x02;
                             if (ibm8514_cpu_dest(svga)) {
                                 READ((dev->accel.cy * dev->pitch) + dev->accel.cx, src_dat);
                             } else
@@ -1634,6 +1629,7 @@ ibm8514_accel_start(int count, int cpu_input, uint32_t mix_dat, uint32_t cpu_dat
                             (dev->accel.cx <= clip_r) &&
                             (dev->accel.cy >= clip_t) &&
                             (dev->accel.cy <= clip_b)) {
+                            dev->subsys_stat |= 0x02;
                             if (ibm8514_cpu_dest(svga) && (pixcntl == 0)) {
                                 mix_dat = mix_mask; /* Mix data = forced to foreground register. */
                             } else if (ibm8514_cpu_dest(svga) && (pixcntl == 3)) {
@@ -1832,6 +1828,7 @@ ibm8514_accel_start(int count, int cpu_input, uint32_t mix_dat, uint32_t cpu_dat
                                 (dev->accel.cx <= clip_r) &&
                                 (dev->accel.cy >= clip_t) &&
                                 (dev->accel.cy <= clip_b)) {
+                                dev->subsys_stat |= 0x02;
                                 switch ((mix_dat & mix_mask) ? frgd_mix : bkgd_mix) {
                                     case 0:
                                         src_dat = bkgd_color;
@@ -1989,6 +1986,7 @@ skip_vector_rect_write:
                             (dev->accel.cx <= clip_r) &&
                             (dev->accel.cy >= clip_t) &&
                             (dev->accel.cy <= clip_b)) {
+                            dev->subsys_stat |= 0x02;
                             if (ibm8514_cpu_dest(svga) && (pixcntl == 0)) {
                                 mix_dat = mix_mask; /* Mix data = forced to foreground register. */
                             } else if (ibm8514_cpu_dest(svga) && (pixcntl == 3)) {
@@ -2150,6 +2148,7 @@ skip_nibble_rect_write:
                                     (dev->accel.cx <= clip_r) &&
                                     (dev->accel.cy >= clip_t) &&
                                     (dev->accel.cy <= clip_b)) {
+                                    dev->subsys_stat |= 0x02;
                                     switch ((mix_dat & mix_mask) ? frgd_mix : bkgd_mix) {
                                         case 0:
                                             src_dat = bkgd_color;
@@ -2232,6 +2231,7 @@ skip_nibble_rect_write:
                                     (dev->accel.cx <= clip_r) &&
                                     (dev->accel.cy >= clip_t) &&
                                     (dev->accel.cy <= clip_b)) {
+                                    dev->subsys_stat |= 0x02;
                                     switch ((mix_dat & 0x01) ? frgd_mix : bkgd_mix) {
                                         case 0:
                                             src_dat = bkgd_color;
@@ -2313,6 +2313,7 @@ skip_nibble_rect_write:
                                 (dev->accel.cx <= clip_r) &&
                                 (dev->accel.cy >= clip_t) &&
                                 (dev->accel.cy <= clip_b)) {
+                                dev->subsys_stat |= 0x02;
                                 switch ((mix_dat & mix_mask) ? frgd_mix : bkgd_mix) {
                                     case 0:
                                         src_dat = bkgd_color;
@@ -2424,6 +2425,7 @@ skip_nibble_rect_write:
                                 (dev->accel.cx <= clip_r) &&
                                 (dev->accel.cy >= clip_t) &&
                                 (dev->accel.cy <= clip_b)) {
+                                dev->subsys_stat |= 0x02;
                                 switch ((mix_dat & mix_mask) ? frgd_mix : bkgd_mix) {
                                     case 0:
                                         src_dat = bkgd_color;
@@ -2546,6 +2548,7 @@ skip_nibble_rect_write:
                         (dev->accel.cx <= clip_r) &&
                         (dev->accel.cy >= clip_t) &&
                         (dev->accel.cy <= clip_b)) {
+                        dev->subsys_stat |= 0x02;
                         switch ((mix_dat & mix_mask) ? frgd_mix : bkgd_mix) {
                             case 0:
                                 src_dat = bkgd_color;
@@ -2650,9 +2653,10 @@ skip_nibble_rect_write:
                     dev->accel.cx = CLAMP(dev->accel.cx, clip_l, clip_r);
 
                     if ((dev->accel.cx >= clip_l) &&
-                        (dev->accel.cx < clip_r) &&
+                        (dev->accel.cx <= clip_r) &&
                         (dev->accel.cy >= clip_t) &&
                         (dev->accel.cy <= clip_b)) {
+                        dev->subsys_stat |= 0x02;
                         switch ((mix_dat & mix_mask) ? frgd_mix : bkgd_mix) {
                             case 0:
                                 src_dat = bkgd_color;
@@ -2811,6 +2815,7 @@ skip_nibble_rect_write:
                         (dev->accel.dx <= clip_r) &&
                         (dev->accel.dy >= clip_t) &&
                         (dev->accel.dy <= clip_b)) {
+                        dev->subsys_stat |= 0x02;
                         if (pixcntl == 3) {
                             if (!(dev->accel.cmd & 0x10) && ((frgd_mix != 3) || (bkgd_mix != 3))) {
                                 READ(dev->accel.src + dev->accel.cx, mix_dat);
@@ -2978,6 +2983,7 @@ skip_nibble_bitblt_write:
                                 (dev->accel.dx <= clip_r) &&
                                 (dev->accel.dy >= clip_t) &&
                                 (dev->accel.dy <= clip_b)) {
+                                dev->subsys_stat |= 0x02;
                                 switch ((mix_dat & mix_mask) ? frgd_mix : bkgd_mix) {
                                     case 0:
                                         src_dat = bkgd_color;
@@ -3075,6 +3081,7 @@ skip_nibble_bitblt_write:
                                 (dev->accel.dx <= clip_r) &&
                                 (dev->accel.dy >= clip_t) &&
                                 (dev->accel.dy <= clip_b)) {
+                                dev->subsys_stat |= 0x02;
                                 switch ((mix_dat & 0x01) ? frgd_mix : bkgd_mix) {
                                     case 0:
                                         src_dat = bkgd_color;
@@ -3174,6 +3181,7 @@ skip_nibble_bitblt_write:
                                 (dx <= (((uint64_t)clip_r) * 3)) &&
                                 (dev->accel.dy >= (clip_t << 1)) &&
                                 (dev->accel.dy <= (clip_b << 1))) {
+                                dev->subsys_stat |= 0x02;
 
                                 READ(dev->accel.src + cx, src_dat);
                                 READ(dev->accel.dest + dx, dest_dat);
@@ -3196,6 +3204,7 @@ skip_nibble_bitblt_write:
                                 (dev->accel.dx <= clip_r) &&
                                 (dev->accel.dy >= clip_t) &&
                                 (dev->accel.dy <= clip_b)) {
+                                dev->subsys_stat |= 0x02;
                                 if (pixcntl == 3) {
                                     if (!(dev->accel.cmd & 0x10) && ((frgd_mix != 3) || (bkgd_mix != 3))) {
                                         READ(dev->accel.src + dev->accel.cx, mix_dat);
@@ -3704,6 +3713,8 @@ ibm8514_poll(void *priv)
             dev->vc &= 0xfff;
 
             if (dev->vc == dev->dispend) {
+                dev->subsys_stat |= 0x01;
+                ibm8514_log("VBLANK irq.\n");
                 dev->dispon = 0;
 
                 for (x = 0; x < ((dev->vram_mask + 1) >> 12); x++) {
diff --git a/src/video/vid_ati68860_ramdac.c b/src/video/vid_ati68860_ramdac.c
index bb5ccbf31..17b2a6d22 100644
--- a/src/video/vid_ati68860_ramdac.c
+++ b/src/video/vid_ati68860_ramdac.c
@@ -70,20 +70,19 @@ void
 ati68860_ramdac_out(uint16_t addr, uint8_t val, void *priv, svga_t *svga)
 {
     ati68860_ramdac_t *ramdac = (ati68860_ramdac_t *) priv;
-    const ibm8514_t   *dev    = (ibm8514_t *) svga->dev8514;
 
     switch (addr) {
         case 0:
-            svga_out((dev && dev->on) ? 0x2ec : 0x3c8, val, svga);
+            svga_out(0x3c8, val, svga);
             break;
         case 1:
-            svga_out((dev && dev->on) ? 0x2ed : 0x3c9, val, svga);
+            svga_out(0x3c9, val, svga);
             break;
         case 2:
-            svga_out((dev && dev->on) ? 0x2ea : 0x3c6, val, svga);
+            svga_out(0x3c6, val, svga);
             break;
         case 3:
-            svga_out((dev && dev->on) ? 0x2eb : 0x3c7, val, svga);
+            svga_out(0x3c7, val, svga);
             break;
         default:
             ramdac->regs[addr & 0xf] = val;
@@ -176,21 +175,20 @@ uint8_t
 ati68860_ramdac_in(uint16_t addr, void *priv, svga_t *svga)
 {
     const ati68860_ramdac_t *ramdac = (ati68860_ramdac_t *) priv;
-    const ibm8514_t         *dev    = (ibm8514_t *) svga->dev8514;
     uint8_t                  temp   = 0;
 
     switch (addr) {
         case 0:
-            temp = svga_in((dev && dev->on) ? 0x2ec : 0x3c8, svga);
+            temp = svga_in(0x3c8, svga);
             break;
         case 1:
-            temp = svga_in((dev && dev->on) ? 0x2ed : 0x3c9, svga);
+            temp = svga_in(0x3c9, svga);
             break;
         case 2:
-            temp = svga_in((dev && dev->on) ? 0x2ea : 0x3c6, svga);
+            temp = svga_in(0x3c6, svga);
             break;
         case 3:
-            temp = svga_in((dev && dev->on) ? 0x2eb : 0x3c7, svga);
+            temp = svga_in(0x3c7, svga);
             break;
         case 4:
         case 8:
diff --git a/src/video/vid_ati_mach8.c b/src/video/vid_ati_mach8.c
index 2038f1b1d..6e2798295 100644
--- a/src/video/vid_ati_mach8.c
+++ b/src/video/vid_ati_mach8.c
@@ -438,6 +438,7 @@ mach_accel_start(int cmd_type, int cpu_input, int count, uint32_t mix_dat, uint3
                         (dev->accel.dx <= clip_r) &&
                         (dev->accel.dy >= clip_t) &&
                         (dev->accel.dy <= clip_b)) {
+                        dev->subsys_stat |= 0x02;
                         switch (mix ? frgd_sel : bkgd_sel) {
                             case 0:
                                 src_dat = dev->accel.bkgd_color;
@@ -661,6 +662,7 @@ mach_accel_start(int cmd_type, int cpu_input, int count, uint32_t mix_dat, uint3
                         (dev->accel.dx <= clip_r) &&
                         (dev->accel.dy >= clip_t) &&
                         (dev->accel.dy <= clip_b)) {
+                        dev->subsys_stat |= 0x02;
                         switch (mix ? frgd_sel : bkgd_sel) {
                             case 0:
                                 src_dat = dev->accel.bkgd_color;
@@ -1067,6 +1069,7 @@ mach_accel_start(int cmd_type, int cpu_input, int count, uint32_t mix_dat, uint3
                     (dev->accel.dx <= clip_r) &&
                     (dev->accel.dy >= clip_t) &&
                     (dev->accel.dy <= clip_b)) {
+                    dev->subsys_stat |= 0x02;
                     if (mach->accel.dp_config & 0x02) {
                         READ(dev->accel.src + dev->accel.cx, poly_src);
                         poly_src = ((poly_src & rd_mask) == rd_mask);
@@ -1308,6 +1311,7 @@ mach_accel_start(int cmd_type, int cpu_input, int count, uint32_t mix_dat, uint3
                             (dev->accel.cx <= clip_r) &&
                             (dev->accel.cy >= clip_t) &&
                             (dev->accel.cy <= clip_b)) {
+                            dev->subsys_stat |= 0x02;
                             mach->accel.clip_overrun = 0;
                             switch (mix ? frgd_sel : bkgd_sel) {
                                 case 0:
@@ -1441,6 +1445,7 @@ mach_accel_start(int cmd_type, int cpu_input, int count, uint32_t mix_dat, uint3
                             (dev->accel.cx <= clip_r) &&
                             (dev->accel.cy >= clip_t) &&
                             (dev->accel.cy <= clip_b)) {
+                            dev->subsys_stat |= 0x02;
                             mach->accel.clip_overrun = 0;
                             if (mach->accel.linedraw_opt & 0x02) {
                                 if (dev->bpp) {
@@ -1599,6 +1604,7 @@ mach_accel_start(int cmd_type, int cpu_input, int count, uint32_t mix_dat, uint3
                             (dev->accel.cx <= clip_r) &&
                             (dev->accel.cy >= clip_t) &&
                             (dev->accel.cy <= clip_b)) {
+                            dev->subsys_stat |= 0x02;
                             mach->accel.clip_overrun = 0;
                             switch (mix ? frgd_sel : bkgd_sel) {
                                 case 0:
@@ -1733,6 +1739,7 @@ mach_accel_start(int cmd_type, int cpu_input, int count, uint32_t mix_dat, uint3
                             (dev->accel.cx <= clip_r) &&
                             (dev->accel.cy >= clip_t) &&
                             (dev->accel.cy <= clip_b)) {
+                            dev->subsys_stat |= 0x02;
                             mach->accel.clip_overrun = 0;
                             switch (mix ? frgd_sel : bkgd_sel) {
                                 case 0:
@@ -2028,6 +2035,7 @@ mach_accel_start(int cmd_type, int cpu_input, int count, uint32_t mix_dat, uint3
                     (dev->accel.dx <= clip_r) &&
                     (dev->accel.dy >= clip_t) &&
                     (dev->accel.dy <= clip_b)) {
+                    dev->subsys_stat |= 0x02;
                     switch (mix ? frgd_sel : bkgd_sel) {
                         case 0:
                             src_dat = dev->accel.bkgd_color;
@@ -2333,6 +2341,7 @@ mach_out(uint16_t addr, uint8_t val, void *priv)
         case 0x2ed:
             rs2 = !!(mach->accel.ext_ge_config & 0x1000);
             rs3 = !!(mach->accel.ext_ge_config & 0x2000);
+            mach_log("8514/A RS2=%d, RS3=%d, addr=%03x.\n", rs2, rs3, addr);
             if ((dev->local & 0xff) >= 0x02) {
                 if (mach->regs[0xb0] & 0x20) { /*ATI extended 8514/A mode.*/
                     mach_log("Extended 8514/A mode.\n");
@@ -2341,10 +2350,14 @@ mach_out(uint16_t addr, uint8_t val, void *priv)
                     svga_recalctimings(svga);
                     mach32_updatemapping(mach, svga);
                 }
-                if (mach->pci_bus && !mach->ramdac_type)
-                    ati68860_ramdac_out((addr & 0x03) | (rs2 << 2) | (rs3 << 3), val, svga->ramdac, svga);
-                else
-                    ati68875_ramdac_out(addr, rs2, rs3, val, svga->ramdac, svga);
+                if (dev->on)
+                    svga_out(addr, val, svga);
+                else {
+                    if (mach->pci_bus && !mach->ramdac_type)
+                        ati68860_ramdac_out((addr & 0x03) | (rs2 << 2) | (rs3 << 3), val, svga->ramdac, svga);
+                    else
+                        ati68875_ramdac_out(addr, rs2, rs3, val, svga->ramdac, svga);
+                }
             } else
                 svga_out(addr, val, svga);
             return;
@@ -2355,6 +2368,7 @@ mach_out(uint16_t addr, uint8_t val, void *priv)
         case 0x3C9:
             rs2 = !!(mach->regs[0xa0] & 0x20);
             rs3 = !!(mach->regs[0xa0] & 0x40);
+            mach_log("VGA RS2=%d, RS3=%d, addr=%03x.\n", rs2, rs3, addr);
             if ((dev->local & 0xff) >= 0x02) {
                 if (svga->attrregs[0x10] & 0x40) {
                     mach_log("VGA mode.\n");
@@ -2363,10 +2377,14 @@ mach_out(uint16_t addr, uint8_t val, void *priv)
                     svga_recalctimings(svga);
                     mach32_updatemapping(mach, svga);
                 }
-                if (mach->pci_bus && !mach->ramdac_type)
-                    ati68860_ramdac_out((addr & 0x03) | (rs2 << 2) | (rs3 << 3), val, svga->ramdac, svga);
-                else
-                    ati68875_ramdac_out(addr, rs2, rs3, val, svga->ramdac, svga);
+                if (dev->on)
+                    svga_out(addr, val, svga);
+                else {
+                    if (mach->pci_bus && !mach->ramdac_type)
+                        ati68860_ramdac_out((addr & 0x03) | (rs2 << 2) | (rs3 << 3), val, svga->ramdac, svga);
+                    else
+                        ati68875_ramdac_out(addr, rs2, rs3, val, svga->ramdac, svga);
+                }
             } else
                 svga_out(addr, val, svga);
             return;
@@ -2481,10 +2499,14 @@ mach_in(uint16_t addr, void *priv)
             rs2 = !!(mach->accel.ext_ge_config & 0x1000);
             rs3 = !!(mach->accel.ext_ge_config & 0x2000);
             if ((dev->local & 0xff) >= 0x02) {
-                if (mach->pci_bus && !mach->ramdac_type)
-                    temp = ati68860_ramdac_in((addr & 3) | (rs2 << 2) | (rs3 << 3), svga->ramdac, svga);
-                else
-                    temp = ati68875_ramdac_in(addr, rs2, rs3, svga->ramdac, svga);
+                if (dev->on)
+                    temp = svga_in(addr, svga);
+                else {
+                    if (mach->pci_bus && !mach->ramdac_type)
+                        temp = ati68860_ramdac_in((addr & 3) | (rs2 << 2) | (rs3 << 3), svga->ramdac, svga);
+                    else
+                        temp = ati68875_ramdac_in(addr, rs2, rs3, svga->ramdac, svga);
+                }
             } else
                 temp = svga_in(addr, svga);
             break;
@@ -3407,7 +3429,7 @@ mach_accel_out_fifo(mach_t *mach, svga_t *svga, ibm8514_t *dev, uint16_t port, u
 static void
 mach_accel_out_call(uint16_t port, uint8_t val, mach_t *mach, svga_t *svga, ibm8514_t *dev)
 {
-    if (port != 0x42e8 && port != 0x42e9)
+    if (port == 0x42e8 || port == 0x42e9)
         mach_log("[%04X:%08X]: Port CALL OUT=%04x, val=%02x.\n", CS, cpu_state.pc, port, val);
 
     switch (port) {
@@ -3422,7 +3444,7 @@ mach_accel_out_call(uint16_t port, uint8_t val, mach_t *mach, svga_t *svga, ibm8
             break;
         case 0x42e9:
             ibm8514_accel_out(port, val, svga, 2);
-            if ((val & 0xc0) == 0xc0) {
+            if ((val & 0xc0) == 0x80) {
                 dev->ext_fifo_idx = 0;
                 mach->force_busy = 0;
             }
@@ -3513,6 +3535,11 @@ mach_accel_out_call(uint16_t port, uint8_t val, mach_t *mach, svga_t *svga, ibm8
             svga_recalctimings(svga);
             break;
 
+        case 0x46e8:
+        case 0x46e9:
+            mach_log("0x%04x write: VGA subsystem enable add-on=%02x.\n", port, val);
+            break;
+
         case 0x4ae8:
             dev->accel.advfunc_cntl = val;
             dev->on = dev->accel.advfunc_cntl & 0x01;
@@ -4201,6 +4228,7 @@ mach_accel_in_call(uint16_t port, mach_t *mach, svga_t *svga, ibm8514_t *dev)
 
     switch (port) {
         case 0x2e8:
+        case 0x2e9:
         case 0x6e8:
         case 0x22e8:
         case 0x26e8:
@@ -4212,18 +4240,22 @@ mach_accel_in_call(uint16_t port, mach_t *mach, svga_t *svga, ibm8514_t *dev)
 
         case 0x42e8:
         case 0x42e9:
-            if (dev->vc == dev->v_syncstart)
+            if ((dev->subsys_cntl & 0x01) && !(dev->subsys_stat & 0x01) && (dev->vc == dev->dispend))
                 temp |= 0x01;
 
             if (mach->accel.cmd_type == -1) {
                 if (cmd == 6) {
-                    if ((dev->accel.dx >= clip_l) &&
+                    if ((dev->subsys_cntl & 0x02) &&
+                        !(dev->subsys_stat & 0x02) &&
+                        (dev->accel.dx >= clip_l) &&
                         (dev->accel.dx <= clip_r_ibm) &&
                         (dev->accel.dy >= clip_t) &&
                         (dev->accel.dy <= clip_b_ibm))
                         temp |= 0x02;
                 } else {
-                    if ((dev->accel.cx >= clip_l) &&
+                    if ((dev->subsys_cntl & 0x02) &&
+                        !(dev->subsys_stat & 0x02) &&
+                        (dev->accel.cx >= clip_l) &&
                         (dev->accel.cx <= clip_r_ibm) &&
                         (dev->accel.cy >= clip_t) &&
                         (dev->accel.cy <= clip_b_ibm))
@@ -4234,7 +4266,9 @@ mach_accel_in_call(uint16_t port, mach_t *mach, svga_t *svga, ibm8514_t *dev)
                     case 1:
                     case 2:
                     case 5:
-                        if ((dev->accel.dx >= clip_l) &&
+                        if ((dev->subsys_cntl & 0x02) &&
+                            !(dev->subsys_stat & 0x02) &&
+                            (dev->accel.dx >= clip_l) &&
                             (dev->accel.dx <= clip_r) &&
                             (dev->accel.dy >= clip_t) &&
                             (dev->accel.dy <= clip_b))
@@ -4242,7 +4276,9 @@ mach_accel_in_call(uint16_t port, mach_t *mach, svga_t *svga, ibm8514_t *dev)
                         break;
                     case 3:
                     case 4:
-                        if ((dev->accel.cx >= clip_l) &&
+                        if ((dev->subsys_cntl & 0x02) &&
+                            !(dev->subsys_stat & 0x02) &&
+                            (dev->accel.cx >= clip_l) &&
                             (dev->accel.cx <= clip_r) &&
                             (dev->accel.cy >= clip_t) &&
                             (dev->accel.cy <= clip_b))
@@ -4257,16 +4293,17 @@ mach_accel_in_call(uint16_t port, mach_t *mach, svga_t *svga, ibm8514_t *dev)
                 if ((!dev->force_busy && !dev->force_busy2) || !mach->force_busy)
                     temp |= 0x08;
             }
-            if (port & 1)
+            if (port & 1) {
                 temp = dev->vram_512k_8514 ? 0x00 : 0x80;
-            else {
+                temp |= (dev->subsys_cntl >> 8);
+            } else {
                 temp |= (dev->subsys_stat | (dev->vram_512k_8514 ? 0x00 : 0x80));
                 if (mach->accel.ext_ge_config & 0x08)
                     temp |= ((mach->accel.ext_ge_config & 0x07) << 4);
                 else
                     temp |= 0x20;
             }
-            mach_log("0x%04x read: Subsystem Status=%02x.\n", port, temp);
+            mach_log("0x%04x read: Subsystem Status=%02x, monitoralias=%02x.\n", port, temp, mach->accel.ext_ge_config & 0x07);
             break;
 
             /*ATI Mach8/32 specific registers*/
@@ -4403,7 +4440,8 @@ mach_accel_in_call(uint16_t port, mach_t *mach, svga_t *svga, ibm8514_t *dev)
         default:
             break;
     }
-    mach_log("[%04X:%08X]: Port NORMAL IN=%04x, temp=%04x.\n", CS, cpu_state.pc, port, temp);
+    if (port == 0x2ee8 || port == 0x2ee9 || port == 0x42e8 || port == 0x42e9)
+        mach_log("[%04X:%08X]: Port NORMAL IN=%04x, temp=%04x.\n", CS, cpu_state.pc, port, temp);
 
     return temp;
 }
@@ -5419,6 +5457,10 @@ mach32_updatemapping(mach_t *mach, svga_t *svga)
             mem_mapping_set_handler(&svga->mapping, mach32_read, mach32_readw, mach32_readl, mach32_write, mach32_writew, mach32_writel);
             mem_mapping_set_p(&svga->mapping, mach);
         } else {
+            if (!dev->on) {
+                memset(dev->vram, 0, dev->vram_size);
+                memset(dev->changedvram, 0, (dev->vram_size >> 12) + 1);
+            }
             mach_log("IBM compatible banked mapping.\n");
             mem_mapping_set_handler(&svga->mapping, svga_read, svga_readw, svga_readl, svga_write, svga_writew, svga_writel);
             mem_mapping_set_p(&svga->mapping, svga);