diff --git a/src/include/86box/vid_8514a.h b/src/include/86box/vid_8514a.h
index ec8c5e168..0e6e0965a 100644
--- a/src/include/86box/vid_8514a.h
+++ b/src/include/86box/vid_8514a.h
@@ -42,6 +42,9 @@ typedef union {
 typedef struct ibm8514_t {
     rom_t bios_rom;
     rom_t bios_rom2;
+    mem_mapping_t bios_mapping;
+    uint8_t *rom1;
+    uint8_t *rom2;
     hwcursor8514_t hwcursor;
     hwcursor8514_t hwcursor_latch;
     uint8_t        pos_regs[8];
@@ -212,6 +215,8 @@ typedef struct ibm8514_t {
     uint8_t subsys_cntl;
     uint8_t subsys_stat;
 
+    atomic_int fifo_idx;
+    atomic_int ext_fifo_idx;
     atomic_int force_busy;
     atomic_int force_busy2;
 
@@ -223,11 +228,10 @@ typedef struct ibm8514_t {
     int      ext_crt_pitch;
     int      extensions;
     int      linear;
-    int      _4bpp;
     uint32_t vram_amount;
     int      vram_512k_8514;
-    PALETTE  _8514pal;
     int      vendor_mode;
+    PALETTE  _8514pal;
 
     latch8514_t latch;
 } ibm8514_t;
diff --git a/src/include/86box/vid_ati_eeprom.h b/src/include/86box/vid_ati_eeprom.h
index 99af36eda..c52b5bd4b 100644
--- a/src/include/86box/vid_ati_eeprom.h
+++ b/src/include/86box/vid_ati_eeprom.h
@@ -47,7 +47,8 @@ typedef struct ati_eeprom_t {
 } ati_eeprom_t;
 
 void ati_eeprom_load(ati_eeprom_t *eeprom, char *fn, int type);
-void ati_eeprom_load_mach8(ati_eeprom_t *eeprom, char *fn);
+void ati_eeprom_load_mach8(ati_eeprom_t *eeprom, char *fn, int mca);
+void ati_eeprom_load_mach8_vga(ati_eeprom_t *eeprom, char *fn);
 void ati_eeprom_write(ati_eeprom_t *eeprom, int ena, int clk, int dat);
 int  ati_eeprom_read(ati_eeprom_t *eeprom);
 
diff --git a/src/include/86box/vid_svga.h b/src/include/86box/vid_svga.h
index 8a9f1fdbe..d38b7614b 100644
--- a/src/include/86box/vid_svga.h
+++ b/src/include/86box/vid_svga.h
@@ -312,7 +312,6 @@ extern void     ibm8514_accel_out_pixtrans(svga_t *svga, uint16_t port, uint32_t
 extern void     ibm8514_short_stroke_start(int count, int cpu_input, uint32_t mix_dat, uint32_t cpu_dat, svga_t *svga, uint8_t ssv, int len);
 extern void     ibm8514_accel_start(int count, int cpu_input, uint32_t mix_dat, uint32_t cpu_dat, svga_t *svga, int len);
 
-#ifdef ATI_8514_ULTRA
 extern void     ati8514_out(uint16_t addr, uint8_t val, void *priv);
 extern uint8_t  ati8514_in(uint16_t addr, void *priv);
 extern void     ati8514_recalctimings(svga_t *svga);
@@ -320,7 +319,6 @@ extern uint8_t  ati8514_mca_read(int port, void *priv);
 extern void     ati8514_mca_write(int port, uint8_t val, void *priv);
 extern void     ati8514_pos_write(uint16_t port, uint8_t val, void *priv);
 extern void     ati8514_init(svga_t *svga, void *ext8514, void *dev8514);
-#endif
 
 extern void xga_write_test(uint32_t addr, uint8_t val, void *priv);
 extern uint8_t xga_read_test(uint32_t addr, void *priv);
diff --git a/src/video/vid_8514a.c b/src/video/vid_8514a.c
index 4d301e2f2..ceeb835e3 100644
--- a/src/video/vid_8514a.c
+++ b/src/video/vid_8514a.c
@@ -43,9 +43,7 @@
 #include <86box/vid_ati_mach8.h>
 #include "cpu.h"
 
-#ifdef ATI_8514_ULTRA
-#define BIOS_MACH8_ROM_PATH  "roms/video/mach8/11301113140.BIN"
-#endif
+#define BIOS_MACH8_ROM_PATH  "roms/video/mach8/11301113140_4k.BIN"
 
 static void     ibm8514_accel_outb(uint16_t port, uint8_t val, void *priv);
 static void     ibm8514_accel_outw(uint16_t port, uint16_t val, void *priv);
@@ -337,18 +335,21 @@ ibm8514_accel_out_fifo(svga_t *svga, uint16_t port, uint32_t val, int len)
     switch (port) {
         case 0x82e8:
         case 0xc2e8:
+            dev->fifo_idx++;
             if (len == 2)
                 dev->accel.cur_y = val & 0x7ff;
             break;
 
         case 0x86e8:
         case 0xc6e8:
+            dev->fifo_idx++;
             if (len == 2)
                 dev->accel.cur_x = val & 0x7ff;
             break;
 
         case 0x8ae8:
         case 0xcae8:
+            dev->fifo_idx++;
             if (len == 2) {
                 dev->accel.desty       = val & 0x7ff;
                 dev->accel.desty_axstp = val & 0x3fff;
@@ -359,6 +360,7 @@ ibm8514_accel_out_fifo(svga_t *svga, uint16_t port, uint32_t val, int len)
 
         case 0x8ee8:
         case 0xcee8:
+            dev->fifo_idx++;
             if (len == 2) {
                 dev->accel.destx       = val & 0x7ff;
                 dev->accel.destx_distp = val & 0x3fff;
@@ -368,6 +370,7 @@ ibm8514_accel_out_fifo(svga_t *svga, uint16_t port, uint32_t val, int len)
             break;
 
         case 0x92e8:
+            dev->fifo_idx++;
             if (len == 2)
                 dev->test = val;
             fallthrough;
@@ -382,6 +385,7 @@ ibm8514_accel_out_fifo(svga_t *svga, uint16_t port, uint32_t val, int len)
 
         case 0x96e8:
         case 0xd6e8:
+            dev->fifo_idx++;
             if (len == 2) {
                 dev->accel.maj_axis_pcnt = val & 0x7ff;
                 dev->accel.maj_axis_pcnt_no_limit = val;
@@ -390,6 +394,7 @@ ibm8514_accel_out_fifo(svga_t *svga, uint16_t port, uint32_t val, int len)
 
         case 0x9ae8:
         case 0xdae8:
+            dev->fifo_idx++;
             dev->accel.ssv_state = 0;
             if (len == 2) {
                 dev->data_available  = 0;
@@ -406,6 +411,7 @@ ibm8514_accel_out_fifo(svga_t *svga, uint16_t port, uint32_t val, int len)
 
         case 0x9ee8:
         case 0xdee8:
+            dev->fifo_idx++;
             dev->accel.ssv_state = 1;
             if (len == 2) {
                 dev->accel.short_stroke = val;
@@ -430,6 +436,7 @@ ibm8514_accel_out_fifo(svga_t *svga, uint16_t port, uint32_t val, int len)
 
         case 0xa2e8:
         case 0xe2e8:
+            dev->fifo_idx++;
             if (port == 0xe2e8) {
                 if (len == 2) {
                     if (dev->accel.cmd_back)
@@ -448,6 +455,7 @@ ibm8514_accel_out_fifo(svga_t *svga, uint16_t port, uint32_t val, int len)
 
         case 0xa6e8:
         case 0xe6e8:
+            dev->fifo_idx++;
             if (port == 0xe6e8) {
                 if (len == 2) {
                     if (dev->accel.cmd_back)
@@ -466,34 +474,40 @@ ibm8514_accel_out_fifo(svga_t *svga, uint16_t port, uint32_t val, int len)
 
         case 0xaae8:
         case 0xeae8:
+            dev->fifo_idx++;
             if (len == 2)
                 dev->accel.wrt_mask = val;
             break;
 
         case 0xaee8:
         case 0xeee8:
+            dev->fifo_idx++;
             if (len == 2)
                 dev->accel.rd_mask = val;
             break;
 
         case 0xb2e8:
         case 0xf2e8:
+            dev->fifo_idx++;
             if (len == 2)
                 dev->accel.color_cmp = val;
             break;
 
         case 0xb6e8:
         case 0xf6e8:
+            dev->fifo_idx++;
             dev->accel.bkgd_mix = val & 0xff;
             break;
 
         case 0xbae8:
         case 0xfae8:
+            dev->fifo_idx++;
             dev->accel.frgd_mix = val & 0xff;
             break;
 
         case 0xbee8:
         case 0xfee8:
+            dev->fifo_idx++;
             if (len == 2) {
                 dev->accel.multifunc_cntl                             = val;
                 dev->accel.multifunc[dev->accel.multifunc_cntl >> 12] = dev->accel.multifunc_cntl & 0xfff;
@@ -601,7 +615,6 @@ void
 ibm8514_accel_out(uint16_t port, uint32_t val, svga_t *svga, int len)
 {
     ibm8514_t *dev = (ibm8514_t *) svga->dev8514;
-    uint8_t    old = 0;
 
     if (port & 0x8000)
         ibm8514_accel_out_fifo(svga, port, val, len);
@@ -687,26 +700,31 @@ ibm8514_accel_out(uint16_t port, uint32_t val, svga_t *svga, int len)
                 break;
 
             case 0x42e8:
-                if (val & 1)
-                    dev->subsys_stat &= ~1;
-                if (val & 2)
-                    dev->subsys_stat &= ~2;
-                if (val & 4)
-                    dev->subsys_stat &= ~4;
-                if (val & 8)
-                    dev->subsys_stat &= ~8;
+                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;
                 break;
             case 0x42e9:
-                old = dev->subsys_cntl;
                 dev->subsys_cntl = val;
-                if ((old ^ val) & 1)
-                    dev->subsys_stat |= 1;
-                if ((old ^ val) & 2)
-                    dev->subsys_stat |= 2;
-                if ((old ^ val) & 4)
-                    dev->subsys_stat |= 4;
-                if ((old ^ val) & 8)
-                    dev->subsys_stat |= 8;
+                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->fifo_idx = 0;
+                    dev->force_busy = 0;
+                    dev->force_busy2 = 0;
+                }
                 break;
 
             case 0x4ae8:
@@ -771,6 +789,22 @@ ibm8514_accel_in_fifo(svga_t *svga, uint16_t port, int len)
 
         case 0x9ae8:
         case 0xdae8:
+            if ((dev->fifo_idx >= 1) && (dev->fifo_idx <= 8)) {
+                temp |= (1 << (dev->fifo_idx - 1));
+                switch (dev->accel.cmd >> 13) {
+                    case 2:
+                    case 3:
+                    case 4:
+                    case 6:
+                        if (dev->accel.sy < 0)
+                            dev->fifo_idx = 0;
+                        break;
+                    default:
+                        if (!dev->accel.sy)
+                            dev->fifo_idx = 0;
+                        break;
+                }
+            }
             if (len == 2) {
                 if (dev->force_busy)
                     temp |= 0x200; /*Hardware busy*/
@@ -878,10 +912,15 @@ ibm8514_accel_in(uint16_t port, svga_t *svga)
                     dev->subsys_stat |= 0x02;
             }
 
+            if (!dev->fifo_idx) {
+                if (!dev->force_busy && !dev->force_busy2)
+                    temp |= 0x08;
+            }
+
             if (port & 1)
                 temp = dev->vram_512k_8514 ? 0x00 : 0x80;
             else {
-                temp |= (dev->subsys_stat | 0x80);
+                temp |= (dev->subsys_stat | (dev->vram_512k_8514 ? 0x00 : 0x80));
                 temp |= 0x20;
             }
             break;
@@ -932,6 +971,8 @@ ibm8514_short_stroke_start(int count, int cpu_input, uint32_t mix_dat, uint32_t
         dev->accel.ssv_draw = ssv & 0x10;
 
         if (ibm8514_cpu_src(svga)) {
+            dev->data_available  = 0;
+            dev->data_available2 = 0;
             return; /*Wait for data from CPU*/
         }
     }
@@ -1242,70 +1283,34 @@ ibm8514_accel_start(int count, int cpu_input, uint32_t mix_dat, uint32_t cpu_dat
                         break;
                     }
 
-                    if (dev->accel.err_term >= dev->accel.maj_axis_pcnt) {
-                        dev->accel.err_term += dev->accel.destx_distp;
-                        /*Step minor axis*/
-                        switch (dev->accel.cmd & 0xe0) {
-                            case 0x00:
-                                dev->accel.cy--;
-                                break;
-                            case 0x20:
-                                dev->accel.cy--;
-                                break;
-                            case 0x40:
-                                dev->accel.cx--;
-                                break;
-                            case 0x60:
-                                dev->accel.cx++;
-                                break;
-                            case 0x80:
-                                dev->accel.cy++;
-                                break;
-                            case 0xa0:
-                                dev->accel.cy++;
-                                break;
-                            case 0xc0:
-                                dev->accel.cx--;
-                                break;
-                            case 0xe0:
-                                dev->accel.cx++;
-                                break;
-
-                            default:
-                                break;
-                        }
-                    } else
-                        dev->accel.err_term += dev->accel.desty_axstp;
-
-                    /*Step major axis*/
-                    switch (dev->accel.cmd & 0xe0) {
-                        case 0x00:
-                            dev->accel.cx--;
-                            break;
-                        case 0x20:
-                            dev->accel.cx++;
-                            break;
-                        case 0x40:
-                            dev->accel.cy--;
-                            break;
-                        case 0x60:
-                            dev->accel.cy--;
-                            break;
-                        case 0x80:
-                            dev->accel.cx--;
-                            break;
-                        case 0xa0:
-                            dev->accel.cx++;
-                            break;
-                        case 0xc0:
+                    if (dev->accel.cmd & 0x40) {
+                        if (dev->accel.cmd & 0x80)
                             dev->accel.cy++;
-                            break;
-                        case 0xe0:
-                            dev->accel.cy++;
-                            break;
+                        else
+                            dev->accel.cy--;
 
-                        default:
-                            break;
+                        if (dev->accel.err_term >= 0) {
+                            dev->accel.err_term += dev->accel.destx_distp;
+                            if (dev->accel.cmd & 0x20)
+                                dev->accel.cx++;
+                            else
+                                dev->accel.cx--;
+                        } else
+                            dev->accel.err_term += dev->accel.desty_axstp;
+                    } else {
+                        if (dev->accel.cmd & 0x20)
+                            dev->accel.cx++;
+                        else
+                            dev->accel.cx--;
+
+                        if (dev->accel.err_term >= 0) {
+                            dev->accel.err_term += dev->accel.destx_distp;
+                            if (dev->accel.cmd & 0x80)
+                                dev->accel.cy++;
+                            else
+                                dev->accel.cy--;
+                        } else
+                            dev->accel.err_term += dev->accel.desty_axstp;
                     }
 
                     dev->accel.ssv_len--;
@@ -2356,8 +2361,8 @@ skip_nibble_rect_write:
                             }
                         }
                     } else {
-                        ibm8514_log("Polygon Draw Type=%02x, CL=%d, CR=%d.\n", dev->accel.multifunc[0x0a] & 0x06, clip_l, clip_r);
-                        while (count-- && dev->accel.sy >= 0) {
+                        ibm8514_log("Polygon Draw Type=%02x, CX=%d, CY=%d, SY=%d, CL=%d, CR=%d.\n", dev->accel.multifunc[0x0a] & 0x06, dev->accel.cx, dev->accel.cy, dev->accel.sy, clip_l, clip_r);
+                        while (count-- && (dev->accel.sy >= 0)) {
                             if ((dev->accel.cx >= clip_l) &&
                                 (dev->accel.cx <= clip_r) &&
                                 (dev->accel.cy >= clip_t) &&
@@ -3291,16 +3296,16 @@ ibm8514_render_8bpp(svga_t *svga)
 
         for (int x = 0; x <= dev->h_disp; x += 8) {
             dat  = *(uint32_t *) (&dev->vram[dev->ma & dev->vram_mask]);
-            p[0] = dev->pallook[dat & dev->dac_mask];
-            p[1] = dev->pallook[(dat >> 8) & dev->dac_mask];
-            p[2] = dev->pallook[(dat >> 16) & dev->dac_mask];
-            p[3] = dev->pallook[(dat >> 24) & dev->dac_mask];
+            p[0] = dev->pallook[dat & dev->dac_mask & 0xff];
+            p[1] = dev->pallook[(dat >> 8) & dev->dac_mask & 0xff];
+            p[2] = dev->pallook[(dat >> 16) & dev->dac_mask & 0xff];
+            p[3] = dev->pallook[(dat >> 24) & dev->dac_mask & 0xff];
 
             dat  = *(uint32_t *) (&dev->vram[(dev->ma + 4) & dev->vram_mask]);
-            p[4] = dev->pallook[dat & dev->dac_mask];
-            p[5] = dev->pallook[(dat >> 8) & dev->dac_mask];
-            p[6] = dev->pallook[(dat >> 16) & dev->dac_mask];
-            p[7] = dev->pallook[(dat >> 24) & dev->dac_mask];
+            p[4] = dev->pallook[dat & dev->dac_mask & 0xff];
+            p[5] = dev->pallook[(dat >> 8) & dev->dac_mask & 0xff];
+            p[6] = dev->pallook[(dat >> 16) & dev->dac_mask & 0xff];
+            p[7] = dev->pallook[(dat >> 24) & dev->dac_mask & 0xff];
 
             dev->ma += 8;
             p += 8;
@@ -3702,13 +3707,10 @@ ibm8514_recalctimings(svga_t *svga)
     ibm8514_t *dev = (ibm8514_t *) svga->dev8514;
 
     svga->render8514 = ibm8514_render_blank;
-#ifdef ATI_8514_ULTRA
     if (dev->extensions) {
         if (svga->ext8514 != NULL)
             ati8514_recalctimings(svga);
-    } else
-#endif
-    {
+    } else {
         if (dev->on) {
             dev->h_total = dev->htotal + 1;
             dev->rowcount = !!(dev->disp_cntl & 0x08);
@@ -3742,7 +3744,6 @@ ibm8514_recalctimings(svga_t *svga)
 
             dev->pitch = 1024;
             dev->rowoffset = 0x80;
-            svga->map8 = dev->pallook;
             if (dev->vram_512k_8514) {
                 if (dev->h_disp == 640)
                     dev->pitch = 640;
@@ -3795,11 +3796,9 @@ ibm8514_mca_reset(void *priv)
 
     ibm8514_log("MCA reset.\n");
     dev->on = 0;
-#ifdef ATI_8514_ULTRA
     if (dev->extensions)
         ati8514_mca_write(0x102, 0, svga);
     else
-#endif
         ibm8514_mca_write(0x102, 0, svga);
 
     timer_set_callback(&svga->timer, svga_poll);
@@ -3808,9 +3807,8 @@ ibm8514_mca_reset(void *priv)
 static void *
 ibm8514_init(const device_t *info)
 {
-#ifdef ATI_8514_ULTRA
     uint32_t bios_addr = 0;
-#endif
+    uint16_t bios_rom_eeprom = 0x0000;
 
     if (svga_get_pri() == NULL)
         return NULL;
@@ -3834,9 +3832,10 @@ ibm8514_init(const device_t *info)
     dev->type     = info->flags;
     dev->bpp      = 0;
 
-#ifdef ATI_8514_ULTRA
     dev->extensions = device_get_config_int("extensions");
     bios_addr = device_get_config_hex20("bios_addr");
+    if (dev->type & DEVICE_MCA)
+        bios_addr = 0xc6000;
 
     switch (dev->extensions) {
         case 1:
@@ -3844,29 +3843,28 @@ ibm8514_init(const device_t *info)
                 mach_t * mach = (mach_t *) calloc(1, sizeof(mach_t));
                 svga->ext8514 = mach;
 
+                rom_init(&dev->bios_rom,
+                         BIOS_MACH8_ROM_PATH,
+                         bios_addr, 0x1000, 0xfff,
+                         0, MEM_MAPPING_EXTERNAL);
+
+                ati8514_init(svga, svga->ext8514, svga->dev8514);
+                mach->accel.scratch0 = ((((bios_addr >> 7) - 0x1000) >> 4));
+                bios_rom_eeprom = mach->accel.scratch0;
                 if (dev->type & DEVICE_MCA) {
-                    rom_init(&dev->bios_rom,
-                             BIOS_MACH8_ROM_PATH,
-                             0xc6000, 0x2000, 0x1fff,
-                             0, MEM_MAPPING_EXTERNAL);
                     dev->pos_regs[0] = 0x88;
                     dev->pos_regs[1] = 0x80;
+                    mach->eeprom.data[0] = 0x0000;
+                    mach->eeprom.data[1] = bios_rom_eeprom | ((bios_rom_eeprom | 0x01) << 8);
                     mca_add(ati8514_mca_read, ati8514_mca_write, ibm8514_mca_feedb, ibm8514_mca_reset, svga);
-                    ati_eeprom_load_mach8(&mach->eeprom, "ati8514_mca.nvr");
+                    ati_eeprom_load_mach8(&mach->eeprom, "ati8514_mca.nvr", 1);
                     mem_mapping_disable(&dev->bios_rom.mapping);
-                } else {
-                    rom_init(&dev->bios_rom,
-                             BIOS_MACH8_ROM_PATH,
-                             bios_addr, 0x2000, 0x1fff,
-                             0, MEM_MAPPING_EXTERNAL);
-                    mach->accel.scratch0 = (((bios_addr >> 7) - 0x1000) >> 4);
-                    ati_eeprom_load_mach8(&mach->eeprom, "ati8514.nvr");
-                }
-                ati8514_init(svga, svga->ext8514, svga->dev8514);
+                } else
+                    ati_eeprom_load_mach8(&mach->eeprom, "ati8514.nvr", 0);
                 break;
             }
-            fallthrough;
 
+            fallthrough;
         default:
             ibm8514_io_set(svga);
 
@@ -3877,16 +3875,6 @@ ibm8514_init(const device_t *info)
             }
             break;
     }
-#else
-    ibm8514_io_set(svga);
-
-    if (dev->type & DEVICE_MCA) {
-        dev->pos_regs[0] = 0x7f;
-        dev->pos_regs[1] = 0xef;
-        mca_add(ibm8514_mca_read, ibm8514_mca_write, ibm8514_mca_feedb, ibm8514_mca_reset, svga);
-    }
-#endif
-
     return svga;
 }
 
@@ -3895,13 +3883,10 @@ ibm8514_close(void *priv)
 {
     svga_t    *svga = (svga_t *) priv;
     ibm8514_t *dev  = (ibm8514_t *) svga->dev8514;
-
-#ifdef ATI_8514_ULTRA
     mach_t    *mach = (mach_t *) svga->ext8514;
 
     if (mach)
         free(mach);
-#endif
 
     if (dev) {
         free(dev->vram);
@@ -3927,9 +3912,8 @@ ibm8514_force_redraw(void *priv)
     svga->fullchange = changeframecount;
 }
 
-#ifdef ATI_8514_ULTRA
 // clang-format off
-static const device_config_t ext8514_config[] = {
+static const device_config_t isa_ext8514_config[] = {
     {
         .name = "memory",
         .description = "Memory size",
@@ -3996,9 +3980,9 @@ static const device_config_t ext8514_config[] = {
         .type = CONFIG_END
     }
 };
-#else
+
 // clang-format off
-static const device_config_t ext8514_config[] = {
+static const device_config_t mca_ext8514_config[] = {
     {
         .name = "memory",
         .description = "Memory size",
@@ -4018,11 +4002,29 @@ static const device_config_t ext8514_config[] = {
             }
         }
     },
+    {
+        .name = "extensions",
+        .description = "Vendor",
+        .type = CONFIG_SELECTION,
+        .default_int = 0,
+        .selection = {
+            {
+                .description = "IBM",
+                .value = 0
+            },
+            {
+                .description = "ATI",
+                .value = 1
+            },
+            {
+                .description = ""
+            }
+        }
+    },
     {
         .type = CONFIG_END
     }
 };
-#endif
 
 // clang-format off
 const device_t gen8514_isa_device = {
@@ -4036,7 +4038,7 @@ const device_t gen8514_isa_device = {
     { .available = NULL },
     .speed_changed = ibm8514_speed_changed,
     .force_redraw = ibm8514_force_redraw,
-    .config = ext8514_config
+    .config = isa_ext8514_config
 };
 
 const device_t ibm8514_mca_device = {
@@ -4050,7 +4052,7 @@ const device_t ibm8514_mca_device = {
     { .available = NULL },
     .speed_changed = ibm8514_speed_changed,
     .force_redraw = ibm8514_force_redraw,
-    .config = ext8514_config
+    .config = mca_ext8514_config
 };
 
 
diff --git a/src/video/vid_ati_eeprom.c b/src/video/vid_ati_eeprom.c
index 054d83d36..a057f1f39 100644
--- a/src/video/vid_ati_eeprom.c
+++ b/src/video/vid_ati_eeprom.c
@@ -43,11 +43,12 @@ ati_eeprom_load(ati_eeprom_t *eeprom, char *fn, int type)
     }
     if (fread(eeprom->data, 1, size, fp) != size)
         memset(eeprom->data, 0, size);
+
     fclose(fp);
 }
 
 void
-ati_eeprom_load_mach8(ati_eeprom_t *eeprom, char *fn)
+ati_eeprom_load_mach8(ati_eeprom_t *eeprom, char *fn, int mca)
 {
     FILE *fp;
     int   size;
@@ -55,14 +56,43 @@ ati_eeprom_load_mach8(ati_eeprom_t *eeprom, char *fn)
     strncpy(eeprom->fn, fn, sizeof(eeprom->fn) - 1);
     fp   = nvr_fopen(eeprom->fn, "rb");
     size = 128;
-    if (!fp) { /*The ATI Graphics Ultra bios expects an immediate write to nvram if none is present at boot time otherwise
+    if (!fp) {
+        if (mca) {
+            (void) fseek(fp, 2L, SEEK_SET);
+            memset(eeprom->data + 2, 0xff, size - 2);
+            fp = nvr_fopen(eeprom->fn, "wb");
+            fwrite(eeprom->data, 1, size, fp);
+            fclose(fp);
+        } else
+            memset(eeprom->data, 0xff, size);
+        return;
+    }
+    if (fread(eeprom->data, 1, size, fp) != size)
+        memset(eeprom->data, 0, size);
+
+    fclose(fp);
+}
+
+void
+ati_eeprom_load_mach8_vga(ati_eeprom_t *eeprom, char *fn)
+{
+    FILE *fp;
+    int   size;
+    eeprom->type = 0;
+    strncpy(eeprom->fn, fn, sizeof(eeprom->fn) - 1);
+    fp   = nvr_fopen(eeprom->fn, "rb");
+    size = 128;
+    if (!fp) { /*The ATI Graphics Ultra bios expects a fresh nvram zero'ed at boot time otherwise
             it would hang the machine.*/
         memset(eeprom->data, 0, size);
         fp = nvr_fopen(eeprom->fn, "wb");
         fwrite(eeprom->data, 1, size, fp);
+        fclose(fp);
+        return;
     }
     if (fread(eeprom->data, 1, size, fp) != size)
         memset(eeprom->data, 0, size);
+
     fclose(fp);
 }
 
@@ -79,9 +109,9 @@ ati_eeprom_save(ati_eeprom_t *eeprom)
 void
 ati_eeprom_write(ati_eeprom_t *eeprom, int ena, int clk, int dat)
 {
-    if (!ena) {
+    if (!ena)
         eeprom->out = 1;
-    }
+
     if (clk && !eeprom->oldclk) {
         if (ena && !eeprom->oldena) {
             eeprom->state  = EEPROM_WAIT;
diff --git a/src/video/vid_ati_mach8.c b/src/video/vid_ati_mach8.c
index b426b8876..f488edace 100644
--- a/src/video/vid_ati_mach8.c
+++ b/src/video/vid_ati_mach8.c
@@ -56,19 +56,17 @@ static video_timings_t timing_mach32_pci   = { .type = VIDEO_PCI, .write_b = 2,
 
 static void     mach_accel_outb(uint16_t port, uint8_t val, void *priv);
 static void     mach_accel_outw(uint16_t port, uint16_t val, void *priv);
+static void     mach_accel_outl(uint16_t port, uint32_t val, void *priv);
 static uint8_t  mach_accel_inb(uint16_t port, void *priv);
 static uint16_t mach_accel_inw(uint16_t port, void *priv);
-static uint8_t  mach_in(uint16_t addr, void *priv);
+static uint32_t mach_accel_inl(uint16_t port, void *priv);
 
-#ifdef ATI_8514_ULTRA
 static void     ati8514_accel_outb(uint16_t port, uint8_t val, void *priv);
 static void     ati8514_accel_outw(uint16_t port, uint16_t val, void *priv);
 static void     ati8514_accel_outl(uint16_t port, uint32_t val, void *priv);
 static uint8_t  ati8514_accel_inb(uint16_t port, void *priv);
 static uint16_t ati8514_accel_inw(uint16_t port, void *priv);
 static uint32_t ati8514_accel_inl(uint16_t port, void *priv);
-#endif
-
 
 static void mach32_updatemapping(mach_t *mach, svga_t *svga);
 
@@ -1254,10 +1252,12 @@ mach_accel_start(int cmd_type, int cpu_input, int count, uint32_t mix_dat, uint3
 
                 if ((mono_src == 2) || (bkgd_sel == 2) || (frgd_sel == 2) || mach_pixel_read(mach)) {
                     if (mach_pixel_write(mach)) {
+                        mach_log("Write PIXTRANS.\n");
                         dev->data_available  = 0;
                         dev->data_available2 = 0;
                         return;
                     } else if (mach_pixel_read(mach)) {
+                        mach_log("Read PIXTRANS.\n");
                         dev->data_available  = 1;
                         dev->data_available2 = 1;
                         return;
@@ -2484,14 +2484,9 @@ mach_in(uint16_t addr, void *priv)
     return temp;
 }
 
-
-#ifdef ATI_8514_ULTRA
 void
 ati8514_out(uint16_t addr, uint8_t val, void *priv)
 {
-    svga_t *svga = (svga_t *)priv;
-    ibm8514_t *dev  = (ibm8514_t *) svga->dev8514;
-
     mach_log("[%04X:%08X]: ADDON OUT addr=%03x, val=%02x.\n", CS, cpu_state.pc, addr, val);
 
     svga_out(addr, val, priv);
@@ -2500,8 +2495,6 @@ ati8514_out(uint16_t addr, uint8_t val, void *priv)
 uint8_t
 ati8514_in(uint16_t addr, void *priv)
 {
-    svga_t *svga = (svga_t *)priv;
-    ibm8514_t *dev  = (ibm8514_t *) svga->dev8514;
     uint8_t temp = 0xff;
 
     temp = svga_in(addr, priv);
@@ -2526,15 +2519,20 @@ ati8514_recalctimings(svga_t *svga)
         dev->accel.ge_offset            = (mach->accel.ge_offset_lo | (mach->accel.ge_offset_hi << 16));
         mach->accel.ge_offset           = dev->accel.ge_offset;
 
-        mach_log("HDISP=%d, VDISP=%d, shadowset=%x, 8514/A mode=%x, clocksel=%02x.\n", dev->hdisp, dev->vdisp, mach->shadow_set & 0x03, dev->accel.advfunc_cntl & 0x04, mach->accel.clock_sel & 0xfe);
+        mach_log("HDISP=%d, VDISP=%d, shadowset=%x, 8514/A mode=%x, clocksel=%02x.\n", dev->hdisp, dev->vdisp, mach->shadow_set & 0x03, dev->accel.advfunc_cntl & 0x05, mach->accel.clock_sel & 0x01);
 
         if (mach->accel.clock_sel & 0x01) {
             dev->h_disp = dev->hdisp;
             dev->dispend = dev->vdisp;
         } else {
             if (dev->accel.advfunc_cntl & 0x04) {
-                dev->h_disp = dev->hdisp;
-                dev->dispend = dev->vdisp;
+                if (dev->hdisp == 640) {
+                    dev->h_disp = 1024;
+                    dev->dispend = 768;
+                } else {
+                    dev->h_disp = dev->hdisp;
+                    dev->dispend = dev->vdisp;
+                }
             } else {
                 dev->h_disp = 640;
                 dev->dispend = 480;
@@ -2563,7 +2561,6 @@ ati8514_recalctimings(svga_t *svga)
         svga->render8514 = ibm8514_render_8bpp;
     }
 }
-#endif
 
 static void
 mach_recalctimings(svga_t *svga)
@@ -2803,14 +2800,18 @@ mach_accel_out_fifo(mach_t *mach, svga_t *svga, ibm8514_t *dev, uint16_t port, u
         case 0x86e8:
         case 0xc2e8:
         case 0xc6e8:
+            dev->ext_fifo_idx++;
             ibm8514_accel_out_fifo(svga, port, val, len);
             break;
         case 0xf6ee:
-            ibm8514_accel_out_fifo(svga, 0x82e8, val, len);
+            dev->ext_fifo_idx++;
+            if (len == 2)
+                dev->accel.cur_y = val & 0x7ff;
             break;
 
         case 0x8ae8:
         case 0xcae8:
+            dev->ext_fifo_idx++;
             ibm8514_accel_out_fifo(svga, port, val, len);
             if (len == 2) {
                 mach_log("SRCY=%d.\n", val & 0x07ff);
@@ -2820,6 +2821,7 @@ mach_accel_out_fifo(mach_t *mach, svga_t *svga, ibm8514_t *dev, uint16_t port, u
 
         case 0x8ee8:
         case 0xcee8:
+            dev->ext_fifo_idx++;
             ibm8514_accel_out_fifo(svga, port, val, len);
             if (len == 2) {
                 mach_log("SRCX=%d.\n", val & 0x07ff);
@@ -2829,11 +2831,13 @@ mach_accel_out_fifo(mach_t *mach, svga_t *svga, ibm8514_t *dev, uint16_t port, u
 
         case 0x92e8:
         case 0xd2e8:
+            dev->ext_fifo_idx++;
             ibm8514_accel_out_fifo(svga, port, val, len);
             break;
 
         case 0x96e8:
         case 0xd6e8:
+            dev->ext_fifo_idx++;
             ibm8514_accel_out_fifo(svga, port, val, len);
             if (len == 2)
                 mach->accel.test = val & 0x1fff;
@@ -2841,17 +2845,21 @@ mach_accel_out_fifo(mach_t *mach, svga_t *svga, ibm8514_t *dev, uint16_t port, u
 
         case 0x9ae8:
         case 0xdae8:
+            dev->ext_fifo_idx++;
             mach->accel.cmd_type = -1;
             ibm8514_accel_out_fifo(svga, port, val, len);
             break;
 
         case 0x9ee8:
         case 0xdee8:
+            dev->ext_fifo_idx++;
             ibm8514_accel_out_fifo(svga, port, val, len);
             break;
 
         case 0xa2e8:
         case 0xe2e8:
+            dev->ext_fifo_idx++;
+            dev->fifo_idx++;
             if (port == 0xe2e8) {
                 if (len == 2) {
                     if (dev->accel.cmd_back) {
@@ -2883,6 +2891,8 @@ mach_accel_out_fifo(mach_t *mach, svga_t *svga, ibm8514_t *dev, uint16_t port, u
 
         case 0xa6e8:
         case 0xe6e8:
+            dev->ext_fifo_idx++;
+            dev->fifo_idx++;
             if (port == 0xe6e8) {
                 if (len == 2) {
                     if (dev->accel.cmd_back)
@@ -2913,6 +2923,8 @@ mach_accel_out_fifo(mach_t *mach, svga_t *svga, ibm8514_t *dev, uint16_t port, u
 
         case 0xe2e9:
         case 0xe6e9:
+            dev->ext_fifo_idx++;
+            dev->fifo_idx++;
             mach_log("Write PORT=%04x, 8514/A=%x, val=%04x, len=%d.\n", port, dev->accel.cmd_back, val, len);
             if (len == 1) {
                 if (mach->accel.cmd_type >= 0) {
@@ -2964,11 +2976,13 @@ mach_accel_out_fifo(mach_t *mach, svga_t *svga, ibm8514_t *dev, uint16_t port, u
         case 0xf2e8:
         case 0xf6e8:
         case 0xfae8:
+            dev->ext_fifo_idx++;
             ibm8514_accel_out_fifo(svga, port, val, len);
             break;
 
         case 0xbee8:
         case 0xfee8:
+            dev->ext_fifo_idx++;
             ibm8514_accel_out_fifo(svga, port, val, len);
             if (len == 2) {
                 if ((dev->accel.multifunc_cntl >> 12) == 5) {
@@ -2980,6 +2994,7 @@ mach_accel_out_fifo(mach_t *mach, svga_t *svga, ibm8514_t *dev, uint16_t port, u
 
             /*ATI Mach8/32 specific registers*/
         case 0x82ee:
+            dev->ext_fifo_idx++;
             mach->accel.patt_data_idx_reg = val & 0x1f;
             mach->accel.patt_data_idx = mach->accel.patt_data_idx_reg;
 
@@ -2992,6 +3007,7 @@ mach_accel_out_fifo(mach_t *mach, svga_t *svga, ibm8514_t *dev, uint16_t port, u
             break;
 
         case 0x8eee:
+            dev->ext_fifo_idx++;
             if (len == 2) {
                 if (mach->accel.patt_data_idx_reg < 0x10) {
                     mach->accel.color_pattern[mach->accel.patt_data_idx] = val & 0xff;
@@ -3007,6 +3023,7 @@ mach_accel_out_fifo(mach_t *mach, svga_t *svga, ibm8514_t *dev, uint16_t port, u
             break;
 
         case 0x96ee:
+            dev->ext_fifo_idx++;
             if (len == 2) {
                 mach->accel.bres_count = val & 0x7ff;
                 mach_log("BresenhamDraw=%04x.\n", mach->accel.dp_config);
@@ -3026,10 +3043,12 @@ mach_accel_out_fifo(mach_t *mach, svga_t *svga, ibm8514_t *dev, uint16_t port, u
             break;
 
         case 0x9aee:
+            dev->ext_fifo_idx++;
             mach->accel.line_idx = val & 0x07;
             break;
 
         case 0xa2ee:
+            dev->ext_fifo_idx++;
             mach_log("Line OPT=%04x.\n", val);
             if (len == 2) {
                 mach->accel.linedraw_opt = val;
@@ -3047,16 +3066,19 @@ mach_accel_out_fifo(mach_t *mach, svga_t *svga, ibm8514_t *dev, uint16_t port, u
             break;
 
         case 0xa6ee:
+            dev->ext_fifo_idx++;
             if (len == 2)
                 mach->accel.dest_x_start = val & 0x7ff;
             break;
 
         case 0xaaee:
+            dev->ext_fifo_idx++;
             if (len == 2)
                 mach->accel.dest_x_end = val & 0x7ff;
             break;
 
         case 0xaeee:
+            dev->ext_fifo_idx++;
             if (len == 2) {
                 mach->accel.dest_y_end = val & 0x7ff;
                 if ((val + 1) == 0x10000) {
@@ -3082,28 +3104,34 @@ mach_accel_out_fifo(mach_t *mach, svga_t *svga, ibm8514_t *dev, uint16_t port, u
             break;
 
         case 0xb2ee:
+            dev->ext_fifo_idx++;
             if (len == 2)
                 mach->accel.src_x_start = val & 0x7ff;
             break;
 
         case 0xb6ee:
-            ibm8514_accel_out_fifo(svga, 0xb6e8, val, len);
+            dev->ext_fifo_idx++;
+            dev->accel.bkgd_mix = val & 0xff;
             break;
 
         case 0xbaee:
-            ibm8514_accel_out_fifo(svga, 0xbae8, val, len);
+            dev->ext_fifo_idx++;
+            dev->accel.frgd_mix = val & 0xff;
             break;
 
         case 0xbeee:
+            dev->ext_fifo_idx++;
             if (len == 2)
                 mach->accel.src_x_end = val & 0x7ff;
             break;
 
         case 0xc2ee:
+            dev->ext_fifo_idx++;
             mach->accel.src_y_dir = val & 1;
             break;
 
         case 0xc6ee:
+            dev->ext_fifo_idx++;
             if (len == 2) {
                 mach->accel.cmd_type = 0;
                 mach_log("TODO: Short Stroke.\n");
@@ -3118,6 +3146,7 @@ mach_accel_out_fifo(mach_t *mach, svga_t *svga, ibm8514_t *dev, uint16_t port, u
             break;
 
         case 0xcaee:
+            dev->ext_fifo_idx++;
             if (len == 2) {
                 mach->accel.scan_to_x = (val & 0x7ff);
                 if ((val + 1) == 0x10000) {
@@ -3143,6 +3172,7 @@ mach_accel_out_fifo(mach_t *mach, svga_t *svga, ibm8514_t *dev, uint16_t port, u
             break;
 
         case 0xceee:
+            dev->ext_fifo_idx++;
             mach_log("CEEE write val = %04x.\n", val);
             if (len == 2) {
                 dev->data_available  = 0;
@@ -3152,6 +3182,7 @@ mach_accel_out_fifo(mach_t *mach, svga_t *svga, ibm8514_t *dev, uint16_t port, u
             break;
 
         case 0xd2ee:
+            dev->ext_fifo_idx++;
             mach->accel.patt_len = val & 0x1f;
             mach_log("Write Port d2ee: Pattern Length=%d, val=%04x.\n", val & 0x1f, val);
             mach->accel.mono_pattern_enable = !!(val & 0x80);
@@ -3162,11 +3193,13 @@ mach_accel_out_fifo(mach_t *mach, svga_t *svga, ibm8514_t *dev, uint16_t port, u
             break;
 
         case 0xd6ee:
+            dev->ext_fifo_idx++;
             mach->accel.patt_idx = val & 0x1f;
             mach_log("Write Port d6ee: Pattern Index=%d.\n", val & 0x1f);
             break;
 
         case 0xdaee:
+            dev->ext_fifo_idx++;
             if (len == 2) {
                 dev->accel.multifunc[2] = val & 0x7ff;
                 dev->accel.clip_left = dev->accel.multifunc[2];
@@ -3177,6 +3210,7 @@ mach_accel_out_fifo(mach_t *mach, svga_t *svga, ibm8514_t *dev, uint16_t port, u
             break;
 
         case 0xdeee:
+            dev->ext_fifo_idx++;
             if (len == 2) {
                 dev->accel.multifunc[1] = val & 0x7ff;
                 dev->accel.clip_top = dev->accel.multifunc[1];
@@ -3188,6 +3222,7 @@ mach_accel_out_fifo(mach_t *mach, svga_t *svga, ibm8514_t *dev, uint16_t port, u
             break;
 
         case 0xe2ee:
+            dev->ext_fifo_idx++;
             if (len == 2) {
                 dev->accel.multifunc[4] = val & 0x7ff;
                 dev->accel.clip_right = dev->accel.multifunc[4];
@@ -3198,6 +3233,7 @@ mach_accel_out_fifo(mach_t *mach, svga_t *svga, ibm8514_t *dev, uint16_t port, u
             break;
 
         case 0xe6ee:
+            dev->ext_fifo_idx++;
             if (len == 2) {
                 dev->accel.multifunc[3] = val & 0x7ff;
                 dev->accel.clip_bottom = dev->accel.multifunc[3];
@@ -3208,17 +3244,20 @@ mach_accel_out_fifo(mach_t *mach, svga_t *svga, ibm8514_t *dev, uint16_t port, u
             break;
 
         case 0xeeee:
+            dev->ext_fifo_idx++;
             if (len == 2)
                 mach->accel.dest_cmp_fn = val;
             break;
 
         case 0xf2ee:
+            dev->ext_fifo_idx++;
             mach_log("F2EE.\n");
             if (len == 2)
                 mach->accel.dst_clr_cmp_mask = val;
             break;
 
         case 0xfeee:
+            dev->ext_fifo_idx++;
             mach_log("LineDraw=%04x.\n", mach->accel.dp_config);
             if (len == 2) {
                 mach->accel.line_array[mach->accel.line_idx] = val;
@@ -3263,8 +3302,14 @@ mach_accel_out_call(uint16_t port, uint8_t val, mach_t *mach, svga_t *svga, ibm8
         case 0x1ee8:
         case 0x1ee9:
         case 0x42e8:
+            ibm8514_accel_out(port, val, svga, 2);
+            break;
         case 0x42e9:
             ibm8514_accel_out(port, val, svga, 2);
+            if ((val & 0xc0) == 0xc0) {
+                dev->ext_fifo_idx = 0;
+                mach->force_busy = 0;
+            }
             break;
 
         case 0x6e8:
@@ -3443,11 +3488,13 @@ mach_accel_out_call(uint16_t port, uint8_t val, mach_t *mach, svga_t *svga, ibm8
                 else
                     dev->ext_crt_pitch <<= 1;
             }
-            dev->on |= 0x01;
-            dev->vendor_mode = 1;
+            if ((dev->local & 0xff) >= 0x02) {
+                dev->on |= 0x01;
+                dev->vendor_mode = 1;
+            }
             svga_recalctimings(svga);
             mach32_updatemapping(mach, svga);
-            mach_log("ATI 8514/A: (0x%04x) val=0x%02x.\n", port, val);
+            mach_log("ATI 8514/A: (0x%04x) val=0x%02x, extended 8514/A mode=%02x.\n", port, val, mach->regs[0xb0] & 0x20);
             break;
 
         case 0x32ee:
@@ -3459,8 +3506,10 @@ mach_accel_out_call(uint16_t port, uint8_t val, mach_t *mach, svga_t *svga, ibm8
         case 0x36ee:
         case 0x36ef:
             mach_log("ATI 8514/A: (0x%04x) val = %04x.\n", port, val);
-            WRITE8(port, mach->misc, val);
-            mach->misc &= 0xfff0;
+            if ((dev->local & 0xff) >= 0x02) {
+                WRITE8(port, mach->misc, val);
+                mach->misc &= 0xfff0;
+            }
             break;
 
         case 0x3aee:
@@ -3599,17 +3648,8 @@ mach_accel_out_call(uint16_t port, uint8_t val, mach_t *mach, svga_t *svga, ibm8
                 svga_set_ramdac_type(svga, !!(mach->accel.ext_ge_config & 0x4000));
                 mach_log("ATI 8514/A: (0x%04x) val=%02x.\n", port, val);
                 svga_recalctimings(svga);
-            } else {
-                mach_log("ATI 8514/A: (0x%04x) val=%02x.\n", port, val & 0x30);
+            } else
                 ati_eeprom_write(&mach->eeprom, !!(mach->accel.ext_ge_config & 0x4000), !!(mach->accel.ext_ge_config & 0x2000), !!(mach->accel.ext_ge_config & 0x1000));
-            }
-            break;
-
-        case 0x7eee:
-        case 0x7eef:
-            WRITE8(port, mach->accel.eeprom_control, val);
-            ati_eeprom_write(&mach->eeprom, !!(mach->accel.eeprom_control & 8), !!(mach->accel.eeprom_control & 2), !!(mach->accel.eeprom_control & 1));
-            mach_log("ATI 8514/A: (0x%04x) val=%02x.\n", port, val);
             break;
 
         default:
@@ -3631,7 +3671,7 @@ static uint16_t
 mach_accel_in_fifo(mach_t *mach, svga_t *svga, ibm8514_t *dev, uint16_t port, int len)
 {
     const uint16_t *vram_w = (uint16_t *) dev->vram;
-    uint16_t        temp = 0;
+    uint16_t        temp = 0x0000;
     int             cmd;
     int             frgd_sel;
     int             bkgd_sel;
@@ -3649,6 +3689,11 @@ mach_accel_in_fifo(mach_t *mach, svga_t *svga, ibm8514_t *dev, uint16_t port, in
 
         case 0x9ae8:
         case 0xdae8:
+            if ((dev->fifo_idx >= 1) && (dev->fifo_idx <= 8)) {
+                temp |= (1 << (dev->fifo_idx - 1));
+                dev->fifo_idx = 0;
+            }
+
             if (len == 2) {
                 if (dev->force_busy)
                     temp |= 0x200; /*Hardware busy*/
@@ -3672,8 +3717,19 @@ mach_accel_in_fifo(mach_t *mach, svga_t *svga, ibm8514_t *dev, uint16_t port, in
                                 break;
                         }
                     } else {
-                        if (dev->accel.sy < 0)
-                            dev->data_available = 0;
+                        switch (dev->accel.cmd >> 13) {
+                            case 2:
+                            case 3:
+                            case 4:
+                            case 6:
+                                if (dev->accel.sy < 0)
+                                    dev->data_available = 0;
+                                break;
+                            default:
+                                if (!dev->accel.sy)
+                                    dev->data_available = 0;
+                                break;
+                        }
                     }
                 }
             }
@@ -3846,6 +3902,15 @@ mach_accel_in_fifo(mach_t *mach, svga_t *svga, ibm8514_t *dev, uint16_t port, in
                 temp = mach->accel.test >> 8;
             break;
 
+        case 0x9aee:
+            if (len == 2) {
+                if ((dev->ext_fifo_idx >= 1) && (dev->ext_fifo_idx <= 16)) {
+                    temp |= (1 << (dev->ext_fifo_idx - 1));
+                    dev->ext_fifo_idx = 0;
+                }
+            }
+            break;
+
         case 0xa2ee:
             if (len == 1)
                 temp = mach->accel.linedraw_opt & 0xff;
@@ -3991,8 +4056,7 @@ mach_accel_in_fifo(mach_t *mach, svga_t *svga, ibm8514_t *dev, uint16_t port, in
             break;
     }
 
-    if (port != 0x9aee && port != 0x62ee)
-        mach_log("[%04X:%08X]: Port FIFO IN=%04x, temp=%04x, len=%d.\n", CS, cpu_state.pc, port, temp, len);
+    mach_log("[%04X:%08X]: Port FIFO IN=%04x, temp=%04x, len=%d.\n", CS, cpu_state.pc, port, temp, len);
 
     return temp;
 }
@@ -4023,7 +4087,7 @@ 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)
-                dev->subsys_stat |= 0x01;
+                temp |= 0x01;
 
             if (mach->accel.cmd_type == -1) {
                 if (cmd == 6) {
@@ -4031,13 +4095,13 @@ mach_accel_in_call(uint16_t port, mach_t *mach, svga_t *svga, ibm8514_t *dev)
                         (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) &&
                         (dev->accel.cx <= clip_r_ibm) &&
                         (dev->accel.cy >= clip_t) &&
                         (dev->accel.cy <= clip_b_ibm))
-                        dev->subsys_stat |= 0x02;
+                        temp |= 0x02;
                 }
             } else {
                 switch (mach->accel.cmd_type) {
@@ -4048,25 +4112,29 @@ mach_accel_in_call(uint16_t port, mach_t *mach, svga_t *svga, ibm8514_t *dev)
                             (dev->accel.dx <= clip_r) &&
                             (dev->accel.dy >= clip_t) &&
                             (dev->accel.dy <= clip_b))
-                            dev->subsys_stat |= 0x02;
+                            temp |= 0x02;
                         break;
                     case 3:
                     case 4:
-                    if ((dev->accel.cx >= clip_l) &&
-                        (dev->accel.cx <= clip_r) &&
-                        (dev->accel.cy >= clip_t) &&
-                        (dev->accel.cy <= clip_b))
-                            dev->subsys_stat |= 0x02;
+                        if ((dev->accel.cx >= clip_l) &&
+                            (dev->accel.cx <= clip_r) &&
+                            (dev->accel.cy >= clip_t) &&
+                            (dev->accel.cy <= clip_b))
+                            temp |= 0x02;
                         break;
                     default:
                         break;
                 }
             }
 
+            if ((!dev->fifo_idx || !dev->ext_fifo_idx)) {
+                if ((!dev->force_busy && !dev->force_busy2) || !mach->force_busy)
+                    temp |= 0x08;
+            }
             if (port & 1)
                 temp = dev->vram_512k_8514 ? 0x00 : 0x80;
             else {
-                temp |= (dev->subsys_stat | 0x80);
+                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
@@ -4103,23 +4171,25 @@ mach_accel_in_call(uint16_t port, mach_t *mach, svga_t *svga, ibm8514_t *dev)
 
         case 0x36ee:
         case 0x36ef:
-            READ8(port, mach->misc);
+            if ((dev->local & 0xff) >= 0x02) {
+                READ8(port, mach->misc);
 
-            if (!(port & 1)) {
-                temp &= ~0x0c;
-                switch (dev->vram_amount) {
-                    case 1024:
-                        temp |= 0x04;
-                        break;
-                    case 2048:
-                        temp |= 0x08;
-                        break;
-                    case 4096:
-                        temp |= 0x0c;
-                        break;
+                if (!(port & 1)) {
+                    temp &= ~0x0c;
+                    switch (dev->vram_amount) {
+                        case 1024:
+                            temp |= 0x04;
+                            break;
+                        case 2048:
+                            temp |= 0x08;
+                            break;
+                        case 4096:
+                            temp |= 0x0c;
+                            break;
 
-                    default:
-                        break;
+                        default:
+                            break;
+                    }
                 }
             }
             break;
@@ -4142,16 +4212,13 @@ mach_accel_in_call(uint16_t port, mach_t *mach, svga_t *svga, ibm8514_t *dev)
         case 0x52ee:
         case 0x52ef:
             READ8(port, mach->accel.scratch0);
-#ifdef ATI_8514_ULTRA
-            if (dev->extensions && ((dev->local & 0xff) == 0x00)) {
-                if (mach->mca_bus) {
+            if (mach->mca_bus) {
+                if (svga->ext8514 != NULL) {
+                    temp = (((dev->bios_rom.mapping.base >> 7) - 0x1000) >> 4);
                     if (port & 1)
-                        temp = dev->pos_regs[5];
-                    else
-                        temp = dev->pos_regs[4];
+                        temp |= 0x01;
                 }
             }
-#endif
             break;
 
         case 0x56ee:
@@ -4210,13 +4277,11 @@ mach_accel_in_call(uint16_t port, mach_t *mach, svga_t *svga, ibm8514_t *dev)
         default:
             break;
     }
-    if (port != 0x42e8 && port != 0x42e9 && port != 0x62ee && port != 0x62ef && port != 0x02e8 && port != 0x02e9)
-        mach_log("[%04X:%08X]: Port NORMAL IN=%04x, temp=%04x.\n", CS, cpu_state.pc, port, temp);
+    mach_log("[%04X:%08X]: Port NORMAL IN=%04x, temp=%04x.\n", CS, cpu_state.pc, port, temp);
 
     return temp;
 }
 
-#ifdef ATI_8514_ULTRA
 static void
 ati8514_accel_out(uint16_t port, uint8_t val, svga_t *svga)
 {
@@ -4267,7 +4332,6 @@ ati8514_accel_outl(uint16_t port, uint32_t val, void *priv)
         ati8514_accel_out(port + 3, (val >> 24), svga);
     }
 }
-#endif
 
 static void
 mach_accel_outb(uint16_t port, uint8_t val, void *priv)
@@ -4312,7 +4376,6 @@ mach_accel_outl(uint16_t port, uint32_t val, void *priv)
     }
 }
 
-#ifdef ATI_8514_ULTRA
 static uint8_t
 ati8514_accel_in(uint16_t port, svga_t *svga)
 {
@@ -4368,7 +4431,6 @@ ati8514_accel_inl(uint16_t port, void *priv)
     }
     return temp;
 }
-#endif
 
 static uint8_t
 mach_accel_in(uint16_t port, mach_t *mach)
@@ -5316,10 +5378,10 @@ mach32_hwcursor_draw(svga_t *svga, int displine)
         dev->hwcursor_latch.addr += 16;
 }
 
-#ifdef ATI_8514_ULTRA
 static void
 ati8514_io_set(svga_t *svga)
 {
+    io_sethandler(0x2e8, 0x0002, ati8514_accel_inb, ati8514_accel_inw, ati8514_accel_inl, ati8514_accel_outb, ati8514_accel_outw, ati8514_accel_outl, svga);
     io_sethandler(0x6e8, 0x0002, ati8514_accel_inb, ati8514_accel_inw, ati8514_accel_inl, ati8514_accel_outb, ati8514_accel_outw, ati8514_accel_outl, svga);
     io_sethandler(0xae8, 0x0002, ati8514_accel_inb, ati8514_accel_inw, ati8514_accel_inl, ati8514_accel_outb, ati8514_accel_outw, ati8514_accel_outl, svga);
     io_sethandler(0xee8, 0x0002, ati8514_accel_inb, ati8514_accel_inw, ati8514_accel_inl, ati8514_accel_outb, ati8514_accel_outw, ati8514_accel_outl, svga);
@@ -5430,7 +5492,6 @@ ati8514_io_set(svga_t *svga)
     io_sethandler(0xf6ee, 0x0002, ati8514_accel_inb, ati8514_accel_inw, ati8514_accel_inl, ati8514_accel_outb, ati8514_accel_outw, ati8514_accel_outl, svga);
     io_sethandler(0xfeee, 0x0002, ati8514_accel_inb, ati8514_accel_inw, ati8514_accel_inl, ati8514_accel_outb, ati8514_accel_outw, ati8514_accel_outl, svga);
 }
-#endif
 
 static void
 mach_io_remove(mach_t *mach)
@@ -5447,7 +5508,6 @@ mach_io_remove(mach_t *mach)
     io_removehandler(0x26e8, 0x0002, mach_accel_inb, mach_accel_inw, mach_accel_inl, mach_accel_outb, mach_accel_outw, mach_accel_outl, mach);
     io_removehandler(0x2ee8, 0x0002, mach_accel_inb, mach_accel_inw, mach_accel_inl, mach_accel_outb, mach_accel_outw, mach_accel_outl, mach);
     io_removehandler(0x42e8, 0x0002, mach_accel_inb, mach_accel_inw, mach_accel_inl, mach_accel_outb, mach_accel_outw, mach_accel_outl, mach);
-    io_removehandler(0x46e8, 0x0002, mach_accel_inb, mach_accel_inw, mach_accel_inl, mach_accel_outb, mach_accel_outw, mach_accel_outl, mach);
     io_removehandler(0x4ae8, 0x0002, mach_accel_inb, mach_accel_inw, mach_accel_inl, mach_accel_outb, mach_accel_outw, mach_accel_outl, mach);
     io_removehandler(0x52e8, 0x0002, mach_accel_inb, mach_accel_inw, mach_accel_inl, mach_accel_outb, mach_accel_outw, mach_accel_outl, mach);
     io_removehandler(0x56e8, 0x0002, mach_accel_inb, mach_accel_inw, mach_accel_inl, mach_accel_outb, mach_accel_outw, mach_accel_outl, mach);
@@ -5564,7 +5624,6 @@ mach_io_set(mach_t *mach)
     io_sethandler(0x26e8, 0x0002, mach_accel_inb, mach_accel_inw, mach_accel_inl, mach_accel_outb, mach_accel_outw, mach_accel_outl, mach);
     io_sethandler(0x2ee8, 0x0002, mach_accel_inb, mach_accel_inw, mach_accel_inl, mach_accel_outb, mach_accel_outw, mach_accel_outl, mach);
     io_sethandler(0x42e8, 0x0002, mach_accel_inb, mach_accel_inw, mach_accel_inl, mach_accel_outb, mach_accel_outw, mach_accel_outl, mach);
-    io_sethandler(0x46e8, 0x0002, mach_accel_inb, mach_accel_inw, mach_accel_inl, mach_accel_outb, mach_accel_outw, mach_accel_outl, mach);
     io_sethandler(0x4ae8, 0x0002, mach_accel_inb, mach_accel_inw, mach_accel_inl, mach_accel_outb, mach_accel_outw, mach_accel_outl, mach);
     io_sethandler(0x52e8, 0x0002, mach_accel_inb, mach_accel_inw, mach_accel_inl, mach_accel_outb, mach_accel_outw, mach_accel_outl, mach);
     io_sethandler(0x56e8, 0x0002, mach_accel_inb, mach_accel_inw, mach_accel_inl, mach_accel_outb, mach_accel_outw, mach_accel_outl, mach);
@@ -5715,7 +5774,6 @@ mach_mca_reset(void *priv)
     timer_set_callback(&svga->timer, svga_poll);
 }
 
-#ifdef ATI_8514_ULTRA
 uint8_t
 ati8514_mca_read(int port, void *priv)
 {
@@ -5747,7 +5805,6 @@ ati8514_pos_write(uint16_t port, uint8_t val, void *priv)
 {
     ati8514_mca_write(port, val, priv);
 }
-#endif
 
 static uint8_t
 mach32_pci_read(UNUSED(int func), int addr, void *priv)
@@ -6031,10 +6088,10 @@ mach8_init(const device_t *info)
     io_sethandler(0x03c0, 32, mach_in, NULL, NULL, mach_out, NULL, NULL, mach);
     io_sethandler(0x02ea, 4,  mach_in, NULL, NULL, mach_out, NULL, NULL, mach);
     mach_io_set(mach);
+    mach->accel.cmd_type = -2;
 
     if ((dev->local & 0xff) >= 0x02) {
         svga->decode_mask     = (4 << 20) - 1;
-        mach->accel.cmd_type  = -1;
         mach->cursor_col_1    = 0xff;
         mach->ext_cur_col_1_r = 0xff;
         mach->ext_cur_col_1_g = 0xff;
@@ -6055,19 +6112,18 @@ mach8_init(const device_t *info)
             } else
                 pci_add_card(PCI_ADD_VIDEO, mach32_pci_read, mach32_pci_write, mach, &mach->pci_slot);
 
-            mach->pci_regs[PCI_REG_COMMAND] = 0x83;
+            mach->pci_regs[PCI_REG_COMMAND] = 0x87;
             mach->pci_regs[0x30]            = 0x00;
             mach->pci_regs[0x32]            = 0x0c;
             mach->pci_regs[0x33]            = 0x00;
         } else
             ati_eeprom_load(&mach->eeprom, "mach32.nvr", 1);
     } else
-        ati_eeprom_load_mach8(&mach->eeprom, "mach8.nvr");
+        ati_eeprom_load_mach8_vga(&mach->eeprom, "mach8.nvr");
 
     return mach;
 }
 
-#ifdef ATI_8514_ULTRA
 void
 ati8514_init(svga_t *svga, void *ext8514, void *dev8514)
 {
@@ -6086,7 +6142,7 @@ ati8514_init(svga_t *svga, void *ext8514, void *dev8514)
     ati8514_io_set(svga);
     mach->mca_bus = !!(dev->type & DEVICE_MCA);
 
-    mach->config1 = 0x01 | 0x02 | 0x08 | 0x80;
+    mach->config1 = 0x02 | 0x08 | 0x80;
 
     if (mach->mca_bus)
         mach->config1 |= 0x04;
@@ -6096,7 +6152,6 @@ ati8514_init(svga_t *svga, void *ext8514, void *dev8514)
 
     mach->config2 = 0x02;
 }
-#endif
 
 static int
 mach8_vga_available(void)
diff --git a/src/video/vid_svga.c b/src/video/vid_svga.c
index e192ecd49..5054bd8e7 100644
--- a/src/video/vid_svga.c
+++ b/src/video/vid_svga.c
@@ -458,11 +458,16 @@ svga_in(uint16_t addr, void *priv)
                 ret = 0x10;
             break;
         case 0x3c3:
-            ret = 0x00;
-            if (xga_active && xga)
-                ret |= !xga->on;
-            if (ibm8514_active && dev)
-                ret |= !dev->on;
+            ret = 0x01;
+            if (xga_active && xga) {
+                if (xga->on)
+                    ret = 0x00;
+            }
+            if (ibm8514_active && dev) {
+                if (dev->on)
+                    ret = 0x00;
+            }
+            svga_log("VGA read: (0x%04x) ret=%02x.\n", addr, ret);
             break;
         case 0x3c4:
             ret = svga->seqaddr;
@@ -560,6 +565,8 @@ svga_in(uint16_t addr, void *priv)
 
     if ((addr >= 0x3c6) && (addr <= 0x3c9))
         svga_log("VGA IN addr=%03x, temp=%02x.\n", addr, ret);
+    else if ((addr >= 0x2ea) && (addr <= 0x2ed))
+        svga_log("8514/A IN addr=%03x, temp=%02x.\n", addr, ret);
 
     return ret;
 }