diff --git a/src/include/86box/vid_8514a.h b/src/include/86box/vid_8514a.h
index fab504bbe..ac4085089 100644
--- a/src/include/86box/vid_8514a.h
+++ b/src/include/86box/vid_8514a.h
@@ -32,6 +32,13 @@ typedef struct hwcursor8514_t {
     uint32_t pitch;
 } hwcursor8514_t;
 
+typedef union {
+    uint64_t q;
+    uint32_t d[2];
+    uint16_t w[4];
+    uint8_t  b[8];
+} latch8514_t;
+
 typedef struct ibm8514_t {
     rom_t bios_rom;
     rom_t bios_rom2;
@@ -217,6 +224,8 @@ typedef struct ibm8514_t {
     int      ext_pitch;
     int      ext_crt_pitch;
     int      extensions;
+
+    latch8514_t latch;
 } ibm8514_t;
 
 #endif /*VIDEO_8514A_H*/
diff --git a/src/video/vid_ati_mach64.c b/src/video/vid_ati_mach64.c
index 9aa396383..8403e718d 100644
--- a/src/video/vid_ati_mach64.c
+++ b/src/video/vid_ati_mach64.c
@@ -621,11 +621,11 @@ mach64_updatemapping(mach64_t *mach64)
             break;
     }
 
-    mach64_log("Mach64 linear aperture = %08x.\n", mach64->linear_base);
     if (mach64->linear_base) {
         if (mach64->type == MACH64_GX) {
             if ((mach64->config_cntl & 3) == 2) {
                 /*8 MB aperture*/
+                mach64_log("Mach64 linear aperture=%08x, cfgcntl=%x, mapping=%x, VGAAP=%x.\n", mach64->linear_base + ((8 << 20) - 0x4000), mach64->config_cntl & 3, svga->gdcreg[6] & 0xc, mach64->config_cntl & 4);
                 mem_mapping_set_addr(&mach64->linear_mapping, mach64->linear_base, (8 << 20) - 0x4000);
                 mem_mapping_set_addr(&mach64->mmio_linear_mapping, mach64->linear_base + ((8 << 20) - 0x4000), 0x4000);
             } else {
@@ -2344,7 +2344,7 @@ mach64_ext_readb(uint32_t addr, void *priv)
                 ret = 0xff;
                 break;
         }
-    } else
+    } else {
         switch (addr & 0x3ff) {
             case 0x00:
             case 0x01:
@@ -2872,6 +2872,7 @@ mach64_ext_readb(uint32_t addr, void *priv)
                 ret = 0;
                 break;
         }
+    }
     if ((addr & 0x3fc) != 0x018)
         mach64_log("mach64_ext_readb : addr %08X ret %02X\n", addr, ret);
     return ret;
@@ -3050,7 +3051,7 @@ mach64_ext_writeb(uint32_t addr, uint8_t val, void *priv)
         mach64_log("nmach64_ext_writeb: addr=%04x val=%02x\n", addr, val);
     } else if (addr & 0x300) {
         mach64_accel_write_fifo(mach64, addr & 0x3ff, val);
-    } else
+    } else {
         switch (addr & 0x3ff) {
             case 0x00:
             case 0x01:
@@ -3232,17 +3233,20 @@ mach64_ext_writeb(uint32_t addr, uint8_t val, void *priv)
             case 0xc2:
             case 0xc3:
                 if (mach64->type == MACH64_GX)
-                    ati68860_ramdac_out((addr & 3) | ((mach64->dac_cntl & 3) << 2), val, mach64->svga.ramdac, &mach64->svga);
+                    ati68860_ramdac_out((addr & 3) | ((mach64->dac_cntl & 3) << 2), val, svga->ramdac, svga);
                 else
-                    ati68860_ramdac_out(addr & 3, val, mach64->svga.ramdac, &mach64->svga);
+                    ati68860_ramdac_out(addr & 3, val, svga->ramdac, svga);
                 break;
             case 0xc4:
             case 0xc5:
             case 0xc6:
             case 0xc7:
                 WRITE8(addr, mach64->dac_cntl, val);
-                svga_set_ramdac_type(svga, (mach64->dac_cntl & 0x100) ? RAMDAC_8BIT : RAMDAC_6BIT);
-                ati68860_set_ramdac_type(mach64->svga.ramdac, (mach64->dac_cntl & 0x100) ? RAMDAC_8BIT : RAMDAC_6BIT);
+                mach64_log("Ext RAMDAC TYPE write=%x, bit set=%03x.\n", addr & 0x3ff, mach64->dac_cntl & 0x100);
+                if ((addr & 3) >= 1) {
+                    svga_set_ramdac_type(svga, !!(mach64->dac_cntl & 0x100));
+                    ati68860_set_ramdac_type(svga->ramdac, !!(mach64->dac_cntl & 0x100));
+                }
                 i2c_gpio_set(mach64->i2c, !(mach64->dac_cntl & 0x20000000) || (mach64->dac_cntl & 0x04000000), !(mach64->dac_cntl & 0x10000000) || (mach64->dac_cntl & 0x02000000));
                 break;
 
@@ -3275,7 +3279,9 @@ mach64_ext_writeb(uint32_t addr, uint8_t val, void *priv)
             default:
                 break;
         }
+    }
 }
+
 void
 mach64_ext_writew(uint32_t addr, uint16_t val, void *priv)
 {
@@ -3555,6 +3561,7 @@ void
 mach64_ext_outb(uint16_t port, uint8_t val, void *priv)
 {
     mach64_t *mach64 = (mach64_t *) priv;
+    svga_t *svga = &mach64->svga;
 
     mach64_log("mach64_ext_outb : port %04X val %02X\n", port, val);
     switch (port) {
@@ -3695,9 +3702,9 @@ mach64_ext_outb(uint16_t port, uint8_t val, void *priv)
         case 0x5eee:
         case 0x5eef:
             if (mach64->type == MACH64_GX)
-                ati68860_ramdac_out((port & 3) | ((mach64->dac_cntl & 3) << 2), val, mach64->svga.ramdac, &mach64->svga);
+                ati68860_ramdac_out((port & 3) | ((mach64->dac_cntl & 3) << 2), val, svga->ramdac, svga);
             else
-                ati68860_ramdac_out(port & 3, val, mach64->svga.ramdac, &mach64->svga);
+                ati68860_ramdac_out(port & 3, val, svga->ramdac, svga);
             break;
 
         case 0x62ec:
@@ -3818,8 +3825,7 @@ mach64_writew(uint32_t addr, uint16_t val, void *priv)
 {
     mach64_t *mach64 = (mach64_t *) priv;
     svga_t   *svga   = &mach64->svga;
-
-    addr = (addr & 0x7fff) + mach64->bank_w[(addr >> 15) & 1];
+    addr             = (addr & 0x7fff) + mach64->bank_w[(addr >> 15) & 1];
     svga_writew_linear(addr, val, svga);
 }
 void
@@ -3827,8 +3833,7 @@ mach64_writel(uint32_t addr, uint32_t val, void *priv)
 {
     mach64_t *mach64 = (mach64_t *) priv;
     svga_t   *svga   = &mach64->svga;
-
-    addr = (addr & 0x7fff) + mach64->bank_w[(addr >> 15) & 1];
+    addr             = (addr & 0x7fff) + mach64->bank_w[(addr >> 15) & 1];
     svga_writel_linear(addr, val, svga);
 }
 
@@ -3847,18 +3852,20 @@ mach64_readw(uint32_t addr, void *priv)
 {
     mach64_t *mach64 = (mach64_t *) priv;
     svga_t   *svga   = &mach64->svga;
-
+    uint16_t  ret;
     addr = (addr & 0x7fff) + mach64->bank_r[(addr >> 15) & 1];
-    return svga_readw_linear(addr, svga);
+    ret  = svga_readw_linear(addr, svga);
+    return ret;
 }
 uint32_t
 mach64_readl(uint32_t addr, void *priv)
 {
     mach64_t *mach64 = (mach64_t *) priv;
     svga_t   *svga   = &mach64->svga;
-
+    uint32_t  ret;
     addr = (addr & 0x7fff) + mach64->bank_r[(addr >> 15) & 1];
-    return svga_readl_linear(addr, svga);
+    ret  = svga_readl_linear(addr, svga);
+    return ret;
 }
 
 #define CLAMP(x)                      \
@@ -4202,16 +4209,34 @@ mach64_io_remove(mach64_t *mach64)
 static void
 mach64_io_set(mach64_t *mach64)
 {
+    uint16_t io_base = 0x02ec;
+
     mach64_io_remove(mach64);
 
+    switch (mach64->io_base) {
+        default:
+        case 0:
+            io_base = 0x02ec;
+            break;
+        case 1:
+            io_base = 0x01cc;
+            break;
+        case 2:
+            io_base = 0x01c8;
+            break;
+        case 3:
+            fatal("Attempting to use the reserved value for I/O Base\n");
+            return;
+    }
+
     io_sethandler(0x03c0, 0x0020, mach64_in, NULL, NULL, mach64_out, NULL, NULL, mach64);
 
     if (!mach64->use_block_decoded_io) {
         for (uint8_t c = 0; c < 8; c++) {
-            io_sethandler((c * 0x1000) + 0x2ec, 0x0004, mach64_ext_inb, mach64_ext_inw, mach64_ext_inl, mach64_ext_outb, mach64_ext_outw, mach64_ext_outl, mach64);
-            io_sethandler((c * 0x1000) + 0x6ec, 0x0004, mach64_ext_inb, mach64_ext_inw, mach64_ext_inl, mach64_ext_outb, mach64_ext_outw, mach64_ext_outl, mach64);
-            io_sethandler((c * 0x1000) + 0xaec, 0x0004, mach64_ext_inb, mach64_ext_inw, mach64_ext_inl, mach64_ext_outb, mach64_ext_outw, mach64_ext_outl, mach64);
-            io_sethandler((c * 0x1000) + 0xeec, 0x0004, mach64_ext_inb, mach64_ext_inw, mach64_ext_inl, mach64_ext_outb, mach64_ext_outw, mach64_ext_outl, mach64);
+            io_sethandler((c * 0x1000) + 0x0000 + io_base, 0x0004, mach64_ext_inb, mach64_ext_inw, mach64_ext_inl, mach64_ext_outb, mach64_ext_outw, mach64_ext_outl, mach64);
+            io_sethandler((c * 0x1000) + 0x0400 + io_base, 0x0004, mach64_ext_inb, mach64_ext_inw, mach64_ext_inl, mach64_ext_outb, mach64_ext_outw, mach64_ext_outl, mach64);
+            io_sethandler((c * 0x1000) + 0x0800 + io_base, 0x0004, mach64_ext_inb, mach64_ext_inw, mach64_ext_inl, mach64_ext_outb, mach64_ext_outw, mach64_ext_outl, mach64);
+            io_sethandler((c * 0x1000) + 0x0c00 + io_base, 0x0004, mach64_ext_inb, mach64_ext_inw, mach64_ext_inl, mach64_ext_outb, mach64_ext_outw, mach64_ext_outl, mach64);
         }
     }
 
@@ -4221,6 +4246,107 @@ mach64_io_set(mach64_t *mach64)
         io_sethandler(mach64->block_decoded_io, 0x0400, mach64_block_inb, mach64_block_inw, mach64_block_inl, mach64_block_outb, mach64_block_outw, mach64_block_outl, mach64);
 }
 
+static uint8_t
+mach64_read_linear(uint32_t addr, void *priv)
+{
+    const svga_t *svga = (svga_t *) priv;
+
+    cycles -= svga->monitor->mon_video_timing_read_b;
+
+    if (!svga->fast) {
+        if (svga->chain2_read) {
+            addr &= ~1;
+            addr <<= 2;
+        }
+    }
+
+    addr &= svga->decode_mask;
+    if (addr >= svga->vram_max)
+        return 0xff;
+
+    return svga->vram[addr & svga->vram_mask];
+}
+
+static uint16_t
+mach64_readw_linear(uint32_t addr, void *priv)
+{
+    svga_t *svga = (svga_t *) priv;
+
+    cycles -= svga->monitor->mon_video_timing_read_w;
+
+    addr &= svga->decode_mask;
+    if (addr >= svga->vram_max)
+        return 0xffff;
+
+    return *(uint16_t *) &svga->vram[addr & svga->vram_mask];
+}
+
+static uint32_t
+mach64_readl_linear(uint32_t addr, void *priv)
+{
+    svga_t *svga = (svga_t *) priv;
+
+    cycles -= svga->monitor->mon_video_timing_read_l;
+
+    addr &= svga->decode_mask;
+    if (addr >= svga->vram_max)
+        return 0xffffffff;
+
+    return *(uint32_t *) &svga->vram[addr & svga->vram_mask];
+}
+
+static void
+mach64_write_linear(uint32_t addr, uint8_t val, void *priv)
+{
+    svga_t *svga = (svga_t *) priv;
+
+    cycles -= svga->monitor->mon_video_timing_write_b;
+
+    if (!svga->fast) {
+        if (svga->chain2_write) {
+            addr &= ~1;
+            addr <<= 2;
+        }
+    }
+
+    addr &= svga->decode_mask;
+    if (addr >= svga->vram_max)
+        return;
+    addr &= svga->vram_mask;
+    svga->changedvram[addr >> 12] = svga->monitor->mon_changeframecount;
+    svga->vram[addr]              = val;
+}
+
+static void
+mach64_writew_linear(uint32_t addr, uint16_t val, void *priv)
+{
+    svga_t *svga = (svga_t *) priv;
+
+    cycles -= svga->monitor->mon_video_timing_write_w;
+
+    addr &= svga->decode_mask;
+    if (addr >= svga->vram_max)
+        return;
+    addr &= svga->vram_mask;
+    svga->changedvram[addr >> 12]   = svga->monitor->mon_changeframecount;
+    *(uint16_t *) &svga->vram[addr] = val;
+}
+
+static void
+mach64_writel_linear(uint32_t addr, uint32_t val, void *priv)
+{
+    svga_t *svga = (svga_t *) priv;
+
+    cycles -= svga->monitor->mon_video_timing_write_l;
+
+    addr &= svga->decode_mask;
+    if (addr >= svga->vram_max)
+        return;
+    addr &= svga->vram_mask;
+    svga->changedvram[addr >> 12]   = svga->monitor->mon_changeframecount;
+    *(uint32_t *) &svga->vram[addr] = val;
+}
+
 uint8_t
 mach64_pci_read(UNUSED(int func), int addr, void *priv)
 {
@@ -4364,9 +4490,9 @@ mach64_pci_write(UNUSED(int func), int addr, uint8_t val, void *priv)
         case 0x33:
             mach64->pci_regs[addr] = val;
             if (mach64->pci_regs[0x30] & 0x01) {
-                uint32_t addr = (mach64->pci_regs[0x32] << 16) | (mach64->pci_regs[0x33] << 24);
-                mach64_log("Mach64 bios_rom enabled at %08x\n", addr);
-                mem_mapping_set_addr(&mach64->bios_rom.mapping, addr, 0x8000);
+                uint32_t biosaddr = (mach64->pci_regs[0x32] << 16) | (mach64->pci_regs[0x33] << 24);
+                mach64_log("Mach64 bios_rom enabled at %08x\n", biosaddr);
+                mem_mapping_set_addr(&mach64->bios_rom.mapping, biosaddr, 0x8000);
             } else {
                 mach64_log("Mach64 bios_rom disabled\n");
                 mem_mapping_disable(&mach64->bios_rom.mapping);
@@ -4411,7 +4537,7 @@ mach64_common_init(const device_t *info)
               mach64_overlay_draw);
     svga->dac_hwcursor.cur_ysize = 64;
 
-    mem_mapping_add(&mach64->linear_mapping, 0, 0, svga_read_linear, svga_readw_linear, svga_readl_linear, svga_write_linear, svga_writew_linear, svga_writel_linear, NULL, MEM_MAPPING_EXTERNAL, svga);
+    mem_mapping_add(&mach64->linear_mapping, 0, 0, mach64_read_linear, mach64_readw_linear, mach64_readl_linear, mach64_write_linear, mach64_writew_linear, mach64_writel_linear, NULL, MEM_MAPPING_EXTERNAL, svga);
     mem_mapping_add(&mach64->mmio_linear_mapping, 0, 0, mach64_ext_readb, mach64_ext_readw, mach64_ext_readl, mach64_ext_writeb, mach64_ext_writew, mach64_ext_writel, NULL, MEM_MAPPING_EXTERNAL, mach64);
     mem_mapping_add(&mach64->mmio_linear_mapping_2, 0, 0, mach64_ext_readb, mach64_ext_readw, mach64_ext_readl, mach64_ext_writeb, mach64_ext_writew, mach64_ext_writel, NULL, MEM_MAPPING_EXTERNAL, mach64);
     mem_mapping_add(&mach64->mmio_mapping, 0xbc000, 0x04000, mach64_ext_readb, mach64_ext_readw, mach64_ext_readl, mach64_ext_writeb, mach64_ext_writew, mach64_ext_writel, NULL, MEM_MAPPING_EXTERNAL, mach64);
@@ -4458,24 +4584,20 @@ mach64gx_init(const device_t *info)
     mach64->config_chip_id = 0x000000d7;
     mach64->dac_cntl       = 5 << 16;             /*ATI 68860 RAMDAC*/
     mach64->config_stat0   = (5 << 9) | (3 << 3); /*ATI-68860, 256Kx16 DRAM*/
-    if (info->flags & DEVICE_PCI)
-        mach64->config_stat0 |= 0; /*PCI, 256Kx16 DRAM*/
-    else if (info->flags & DEVICE_VLB)
-        mach64->config_stat0 |= 1; /*VLB, 256Kx16 DRAM*/
-    else if (info->flags & DEVICE_ISA)
-        mach64->config_stat0 |= 7; /*ISA 16-bit, 256k16 DRAM*/
-
-    ati_eeprom_load(&mach64->eeprom, "mach64.nvr", 1);
-
-    if (info->flags & DEVICE_PCI)
+    if (info->flags & DEVICE_PCI) {
+        mach64->config_stat0 |= 7; /*PCI, 256Kx16 DRAM*/
+        ati_eeprom_load(&mach64->eeprom, "mach64_pci.nvr", 1);
         rom_init(&mach64->bios_rom, BIOS_ROM_PATH, 0xc0000, 0x8000, 0x7fff, 0, MEM_MAPPING_EXTERNAL);
-    else if (info->flags & DEVICE_VLB)
-        rom_init(&mach64->bios_rom, BIOS_VLB_ROM_PATH, 0xc0000, 0x8000, 0x7fff, 0, MEM_MAPPING_EXTERNAL);
-    else if (info->flags & DEVICE_ISA)
-        rom_init(&mach64->bios_rom, BIOS_ISA_ROM_PATH, 0xc0000, 0x8000, 0x7fff, 0, MEM_MAPPING_EXTERNAL);
-
-    if (info->flags & DEVICE_PCI)
         mem_mapping_disable(&mach64->bios_rom.mapping);
+    } else if (info->flags & DEVICE_VLB) {
+        mach64->config_stat0 |= 6; /*VLB, 256Kx16 DRAM*/
+        ati_eeprom_load(&mach64->eeprom, "mach64_vlb.nvr", 1);
+        rom_init(&mach64->bios_rom, BIOS_VLB_ROM_PATH, 0xc0000, 0x8000, 0x7fff, 0, MEM_MAPPING_EXTERNAL);
+    } else if (info->flags & DEVICE_ISA) {
+        mach64->config_stat0 |= 0; /*ISA 16-bit, 256k16 DRAM*/
+        ati_eeprom_load(&mach64->eeprom, "mach64.nvr", 1);
+        rom_init(&mach64->bios_rom, BIOS_ISA_ROM_PATH, 0xc0000, 0x8000, 0x7fff, 0, MEM_MAPPING_EXTERNAL);
+    }
 
     return mach64;
 }
diff --git a/src/video/vid_ati_mach8.c b/src/video/vid_ati_mach8.c
index 29a687f24..be2c5d547 100644
--- a/src/video/vid_ati_mach8.c
+++ b/src/video/vid_ati_mach8.c
@@ -332,8 +332,9 @@ mach_accel_start(int cmd_type, int cpu_input, int count, uint32_t mix_dat, uint3
     }
 
     if ((dev->accel_bpp == 8) || (dev->accel_bpp == 15) || (dev->accel_bpp == 16) || (dev->accel_bpp == 24)) {
-        if (cpu_input && (cmd_type == 2))
+        if ((cmd_type == 2) && cpu_input) {
             mach_log("RdMask=%04x, DPCONFIG=%04x, Clipping: l=%d, r=%d, t=%d, b=%d, LineDrawOpt=%04x, BPP=%d, CMDType = %d, offs=%08x, cnt = %d, input = %d, mono_src = %d, frgdsel = %d, d(%d,%d), dstxend = %d, pitch = %d, extcrt = %d, rw = %x, monpattern = %x.\n", rd_mask, mach->accel.dp_config, clip_l, clip_r, clip_t, clip_b, mach->accel.linedraw_opt, dev->accel_bpp, cmd_type, mach->accel.ge_offset, count, cpu_input, mono_src, frgd_sel, dev->accel.cur_x, dev->accel.cur_y, mach->accel.dest_x_end, dev->ext_pitch, dev->ext_crt_pitch, mach->accel.dp_config & 1, mach->accel.mono_pattern_enable);
+        }
     }
 
     switch (cmd_type) {
@@ -816,7 +817,7 @@ mach_accel_start(int cmd_type, int cpu_input, int count, uint32_t mix_dat, uint3
                 if (dev->accel.cur_y >= 0x600)
                     dev->accel.dy |= ~0x5ff;
 
-                if (mach->accel.dp_config == 0x5211) {
+                if ((mach->accel.dp_config == 0x5211) || (mach->accel.dp_config == 0x3251)) {
                     if (mach->accel.dest_x_end == 1024) {
                         goto skip_dx;
                     }
@@ -2602,7 +2603,7 @@ mach_recalctimings(svga_t *svga)
         dev->rowcount                   = !!(dev->disp_cntl & 0x08);
 
         mach_log("HDISP=%d, VDISP=%d, shadowset=%x, 8514/A mode=%x.\n", dev->hdisp, dev->vdisp, mach->shadow_set & 0x03, dev->accel.advfunc_cntl & 0x04);
-        if ((dev->hdisp == 640) || (dev->hdisp == 800) || (dev->hdisp == 1280) || dev->bpp) {
+        if ((dev->hdisp != 1024) || dev->bpp) {
             /*For VESA/ATI modes in 8514/A mode.*/
             dev->h_disp = dev->hdisp;
             dev->dispend = dev->vdisp;
@@ -4865,14 +4866,22 @@ mach32_write_common(uint32_t addr, uint8_t val, int linear, mach_t *mach)
     ibm8514_t *dev = (ibm8514_t *) svga->dev8514;
     int     writemask2 = svga->writemask;
     int     reset_wm   = 0;
-    latch_t vall;
+    latch8514_t vall;
     uint8_t wm = svga->writemask;
     uint8_t count;
     uint8_t i;
 
     cycles -= svga->monitor->mon_video_timing_write_b;
 
-    if (!linear) {
+    if (linear) {
+        addr &= svga->decode_mask;
+        if (addr >= dev->vram_size)
+            return;
+        addr &= dev->vram_mask;
+        dev->changedvram[addr >> 12] = svga->monitor->mon_changeframecount;
+        dev->vram[addr]              = val;
+        return;
+    } else {
         addr = mach32_decode_addr(svga, addr, 1);
         if (addr == 0xffffffff)
             return;
@@ -4881,15 +4890,12 @@ mach32_write_common(uint32_t addr, uint8_t val, int linear, mach_t *mach)
     if (!(svga->gdcreg[6] & 1))
         svga->fullchange = 2;
 
-    mach_log("WriteCommon chain4 = %x.\n", svga->chain4);
     if (((svga->chain4 && (svga->packed_chain4 || svga->force_old_addr)) || svga->fb_only) && (svga->writemode < 4)) {
         writemask2 = 1 << (addr & 3);
         addr &= ~3;
     } else if (svga->chain4 && (svga->writemode < 4)) {
         writemask2 = 1 << (addr & 3);
-        if (!linear)
-            addr &= ~3;
-
+        addr &= ~3;
         addr = ((addr & 0xfffc) << 2) | ((addr & 0x30000) >> 14) | (addr & ~0x3ffff);
     } else if (svga->chain2_write) {
         writemask2 &= ~0xa;
@@ -4934,7 +4940,7 @@ mach32_write_common(uint32_t addr, uint8_t val, int linear, mach_t *mach)
         case 1:
             for (i = 0; i < count; i++) {
                 if (writemask2 & (1 << i))
-                    dev->vram[addr | i] = svga->latch.b[i];
+                    dev->vram[addr | i] = dev->latch.b[i];
             }
             return;
         case 2:
@@ -4944,7 +4950,7 @@ mach32_write_common(uint32_t addr, uint8_t val, int linear, mach_t *mach)
             if (!(svga->gdcreg[3] & 0x18) && (!svga->gdcreg[1] || svga->set_reset_disabled)) {
                 for (i = 0; i < count; i++) {
                     if (writemask2 & (1 << i))
-                        dev->vram[addr | i] = (vall.b[i] & svga->gdcreg[8]) | (svga->latch.b[i] & ~svga->gdcreg[8]);
+                        dev->vram[addr | i] = (vall.b[i] & svga->gdcreg[8]) | (dev->latch.b[i] & ~svga->gdcreg[8]);
                 }
                 return;
             }
@@ -4967,25 +4973,25 @@ mach32_write_common(uint32_t addr, uint8_t val, int linear, mach_t *mach)
         case 0x00: /* Set */
             for (i = 0; i < count; i++) {
                 if (writemask2 & (1 << i))
-                    dev->vram[addr | i] = (vall.b[i] & svga->gdcreg[8]) | (svga->latch.b[i] & ~svga->gdcreg[8]);
+                    dev->vram[addr | i] = (vall.b[i] & svga->gdcreg[8]) | (dev->latch.b[i] & ~svga->gdcreg[8]);
             }
             break;
         case 0x08: /* AND */
             for (i = 0; i < count; i++) {
                 if (writemask2 & (1 << i))
-                    dev->vram[addr | i] = (vall.b[i] | ~svga->gdcreg[8]) & svga->latch.b[i];
+                    dev->vram[addr | i] = (vall.b[i] | ~svga->gdcreg[8]) & dev->latch.b[i];
             }
             break;
         case 0x10: /* OR */
             for (i = 0; i < count; i++) {
                 if (writemask2 & (1 << i))
-                    dev->vram[addr | i] = (vall.b[i] & svga->gdcreg[8]) | svga->latch.b[i];
+                    dev->vram[addr | i] = (vall.b[i] & svga->gdcreg[8]) | dev->latch.b[i];
             }
             break;
         case 0x18: /* XOR */
             for (i = 0; i < count; i++) {
                 if (writemask2 & (1 << i))
-                    dev->vram[addr | i] = (vall.b[i] & svga->gdcreg[8]) ^ svga->latch.b[i];
+                    dev->vram[addr | i] = (vall.b[i] & svga->gdcreg[8]) ^ dev->latch.b[i];
             }
             break;
 
@@ -5022,11 +5028,43 @@ mach32_writel(uint32_t addr, uint32_t val, void *priv)
     mach32_write_common(addr + 3, val >> 24, 0, mach);
 }
 
+static __inline void
+mach32_writew_linear(uint32_t addr, uint16_t val, mach_t *mach)
+{
+    svga_t *svga = &mach->svga;
+    ibm8514_t *dev = (ibm8514_t *) svga->dev8514;
+
+    cycles -= svga->monitor->mon_video_timing_write_w;
+
+    addr &= svga->decode_mask;
+    if (addr >= dev->vram_size)
+        return;
+    addr &= dev->vram_mask;
+    dev->changedvram[addr >> 12] = svga->monitor->mon_changeframecount;
+    *(uint16_t *) &dev->vram[addr] = val;
+}
+
+static __inline void
+mach32_writel_linear(uint32_t addr, uint32_t val, mach_t *mach)
+{
+    svga_t *svga = &mach->svga;
+    ibm8514_t *dev = (ibm8514_t *) svga->dev8514;
+
+    cycles -= svga->monitor->mon_video_timing_write_l;
+
+    addr &= svga->decode_mask;
+    if (addr >= dev->vram_size)
+        return;
+    addr &= dev->vram_mask;
+    dev->changedvram[addr >> 12] = svga->monitor->mon_changeframecount;
+    *(uint32_t *) &dev->vram[addr] = val;
+}
+
 static __inline uint8_t
 mach32_read_common(uint32_t addr, int linear, mach_t *mach)
 {
     svga_t          *svga       = &mach->svga;
-    const ibm8514_t *dev        = (ibm8514_t *) svga->dev8514;
+    ibm8514_t       *dev        = (ibm8514_t *) svga->dev8514;
     uint32_t         latch_addr = 0;
     int              readplane  = svga->readplane;
     uint8_t          count;
@@ -5035,7 +5073,13 @@ mach32_read_common(uint32_t addr, int linear, mach_t *mach)
 
     cycles -= svga->monitor->mon_video_timing_read_b;
 
-    if (!linear) {
+    if (linear) {
+        addr &= svga->decode_mask;
+        if (addr >= dev->vram_size)
+            return 0xff;
+
+        return dev->vram[addr & dev->vram_mask];
+    } else {
         addr = mach32_decode_addr(svga, addr, 0);
         if (addr == 0xffffffff)
             return 0xff;
@@ -5046,14 +5090,13 @@ mach32_read_common(uint32_t addr, int linear, mach_t *mach)
     latch_addr = (addr << count) & svga->decode_mask;
     count      = (1 << count);
 
-    mach_log("ReadCommon chain4 = %x.\n", svga->chain4);
     if ((svga->chain4 && (svga->packed_chain4 || svga->force_old_addr)) || svga->fb_only) {
         addr &= svga->decode_mask;
         if (addr >= dev->vram_size)
             return 0xff;
         latch_addr = (addr & dev->vram_mask) & ~3;
         for (uint8_t i = 0; i < count; i++)
-            svga->latch.b[i] = dev->vram[latch_addr | i];
+            dev->latch.b[i] = dev->vram[latch_addr | i];
         return dev->vram[addr & dev->vram_mask];
     } else if (svga->chain4 && !svga->force_old_addr) {
         readplane = addr & 3;
@@ -5068,7 +5111,7 @@ mach32_read_common(uint32_t addr, int linear, mach_t *mach)
             return 0xff;
         latch_addr = (addr & dev->vram_mask) & ~3;
         for (uint8_t i = 0; i < count; i++)
-            svga->latch.b[i] = dev->vram[latch_addr | i];
+            dev->latch.b[i] = dev->vram[latch_addr | i];
         return dev->vram[addr & dev->vram_mask];
     }
 
@@ -5077,12 +5120,12 @@ mach32_read_common(uint32_t addr, int linear, mach_t *mach)
     /* standard VGA latched access */
     if (latch_addr >= dev->vram_size) {
         for (uint8_t i = 0; i < count; i++)
-            svga->latch.b[i] = 0xff;
+            dev->latch.b[i] = 0xff;
     } else {
         latch_addr &= dev->vram_mask;
 
         for (uint8_t i = 0; i < count; i++)
-            svga->latch.b[i] = dev->vram[latch_addr | i];
+            dev->latch.b[i] = dev->vram[latch_addr | i];
     }
 
     if (addr >= dev->vram_size)
@@ -5097,7 +5140,7 @@ mach32_read_common(uint32_t addr, int linear, mach_t *mach)
             for (uint8_t plane = 0; plane < count; plane++) {
                 if (svga->colournocare & (1 << plane)) {
                     /* If we care about a plane, and the pixel has a mismatch on it, clear its bit. */
-                    if (((svga->latch.b[plane] >> pixel) & 1) != ((svga->colourcompare >> plane) & 1))
+                    if (((dev->latch.b[plane] >> pixel) & 1) != ((svga->colourcompare >> plane) & 1))
                         temp &= ~(1 << pixel);
                 }
             }
@@ -5144,6 +5187,36 @@ mach32_readl(uint32_t addr, void *priv)
     return ret;
 }
 
+static __inline uint16_t
+mach32_readw_linear(uint32_t addr, mach_t *mach)
+{
+    svga_t          *svga       = &mach->svga;
+    ibm8514_t       *dev        = (ibm8514_t *) svga->dev8514;
+
+    cycles -= svga->monitor->mon_video_timing_read_w;
+
+    addr &= svga->decode_mask;
+    if (addr >= dev->vram_size)
+        return 0xffff;
+
+    return *(uint16_t *) &dev->vram[addr & dev->vram_mask];
+}
+
+static __inline uint32_t
+mach32_readl_linear(uint32_t addr, mach_t *mach)
+{
+    svga_t          *svga       = &mach->svga;
+    ibm8514_t       *dev        = (ibm8514_t *) svga->dev8514;
+
+    cycles -= svga->monitor->mon_video_timing_read_l;
+
+    addr &= svga->decode_mask;
+    if (addr >= dev->vram_size)
+        return 0xffffffff;
+
+    return *(uint32_t *) &dev->vram[addr & dev->vram_mask];
+}
+
 static void
 mach32_ap_writeb(uint32_t addr, uint8_t val, void *priv)
 {
@@ -5190,8 +5263,7 @@ mach32_ap_writew(uint32_t addr, uint16_t val, void *priv)
     } else {
         mach_log("Linear WORDW Write=%08x, val=%04x.\n", addr, val);
         if (dev->on[0] || dev->on[1]) {
-            mach32_write_common(addr, val & 0xff, 1, mach);
-            mach32_write_common(addr + 1, val >> 8, 1, mach);
+            mach32_writew_linear(addr, val, mach);
         } else
             svga_writew_linear(addr, val, svga);
     }
@@ -5219,10 +5291,7 @@ mach32_ap_writel(uint32_t addr, uint32_t val, void *priv)
     } else {
         mach_log("Linear WORDL Write=%08x, val=%08x.\n", addr, val);
         if (dev->on[0] || dev->on[1]) {
-            mach32_write_common(addr, val & 0xff, 1, mach);
-            mach32_write_common(addr + 1, val >> 8, 1, mach);
-            mach32_write_common(addr + 2, val >> 16, 1, mach);
-            mach32_write_common(addr + 3, val >> 24, 1, mach);
+            mach32_writel_linear(addr, val, mach);
         } else
             svga_writel_linear(addr, val, svga);
     }
@@ -5272,8 +5341,7 @@ mach32_ap_readw(uint32_t addr, void *priv)
             temp = mach_accel_inw(0x02e8 + (port_dword << 8), mach);
     } else {
         if (dev->on[0] || dev->on[1]) {
-            temp  = mach32_read_common(addr, 1, mach);
-            temp  |= (mach32_read_common(addr + 1, 1, mach) << 8);
+            temp = mach32_readw_linear(addr, mach);
         } else
             temp = svga_readw_linear(addr, svga);
 
@@ -5303,10 +5371,7 @@ mach32_ap_readl(uint32_t addr, void *priv)
         }
     } else {
         if (dev->on[0] || dev->on[1]) {
-            temp  = mach32_read_common(addr, 1, mach);
-            temp  |= (mach32_read_common(addr + 1, 1, mach) << 8);
-            temp  |= (mach32_read_common(addr + 2, 1, mach) << 16);
-            temp  |= (mach32_read_common(addr + 3, 1, mach) << 24);
+            temp = mach32_readl_linear(addr, mach);
         } else
             temp = svga_readl_linear(addr, svga);