diff --git a/wled00/FX_fcn.cpp b/wled00/FX_fcn.cpp
index 34fde40058..41106053db 100644
--- a/wled00/FX_fcn.cpp
+++ b/wled00/FX_fcn.cpp
@@ -150,11 +150,9 @@ bool Segment::allocateData(size_t len) {
   if (len == 0) return false;    // nothing to do
   if (data && _dataLen >= len) { // already allocated enough (reduce fragmentation)
     if (call == 0) {
-      if (_dataLen < FAIR_DATA_PER_SEG) { // segment data is small
         //DEBUG_PRINTF_P(PSTR("--   Clearing data (%d): %p\n"), len, this);
         memset(data, 0, len);  // erase buffer if called during effect initialisation
         return true; // no need to reallocate
-      }
     }
     else
       return true;
@@ -198,7 +196,7 @@ void Segment::deallocateData() {
     DEBUG_PRINTF_P(PSTR("---- Released data (%p): inconsistent UsedSegmentData (%d/%d), cowardly refusing to free nothing.\n"), this, _dataLen, Segment::getUsedSegmentData());
   }
   data = nullptr;
-  Segment::addUsedSegmentData(_dataLen <= Segment::getUsedSegmentData() ? -_dataLen : -Segment::getUsedSegmentData());
+  Segment::addUsedSegmentData(-_dataLen); // addUsedSegmentData guards agains negative and sets zero if underrun so this is safe
   _dataLen = 0;
 }
 
diff --git a/wled00/bus_manager.cpp b/wled00/bus_manager.cpp
index 1ba808d8b0..ad9c510be6 100644
--- a/wled00/bus_manager.cpp
+++ b/wled00/bus_manager.cpp
@@ -1003,7 +1003,7 @@ BusHub75Matrix::BusHub75Matrix(const BusConfig &bc) : Bus(bc.type, bc.start, bc.
   this->_len = (display->width() * display->height()); // note: this returns correct number of pixels but incorrect dimensions if using virtual display (updated below)
 
   DEBUGBUS_PRINTF("Length: %u\n", _len);
-  if (this->_len >= MAX_LEDS) {
+  if (this->_len > MAX_LEDS) {
     DEBUGBUS_PRINTLN("MatrixPanel_I2S_DMA Too many LEDS - playing safe");
     return;
   }
@@ -1033,7 +1033,6 @@ BusHub75Matrix::BusHub75Matrix(const BusConfig &bc) : Bus(bc.type, bc.start, bc.
     display->clearScreen();   // initially clear the screen buffer
     DEBUGBUS_PRINTLN("MatrixPanel_I2S_DMA clear ok");
 
-    if (_ledBuffer) d_free(_ledBuffer);                 // should not happen
     if (_ledsDirty) d_free(_ledsDirty);                 // should not happen
     DEBUGBUS_PRINTLN("MatrixPanel_I2S_DMA allocate memory");
     _ledsDirty = (byte*) d_malloc(getBitArrayBytes(_len));  // create LEDs dirty bits
@@ -1048,9 +1047,6 @@ BusHub75Matrix::BusHub75Matrix(const BusConfig &bc) : Bus(bc.type, bc.start, bc.
       return;  //  fail is we cannot get memory for the buffer
     }
     setBitArray(_ledsDirty, _len, false);             // reset dirty bits
-
-    // create LEDs buffer (initialized to BLACK), prefer DRAM if enough heap is available (faster in case global _pixels buffer is in PSRAM as not both will fit the cache)
-    _ledBuffer = static_cast<CRGB*>(allocate_buffer(_len * sizeof(CRGB), BFRALLOC_PREFER_DRAM | BFRALLOC_CLEAR));
   }
 
   PANEL_CHAIN_TYPE chainType = CHAIN_NONE; // default for quarter-scan panels that do not use chaining
@@ -1098,11 +1094,10 @@ BusHub75Matrix::BusHub75Matrix(const BusConfig &bc) : Bus(bc.type, bc.start, bc.
   DEBUGBUS_PRINT(F("MatrixPanel_I2S_DMA "));
   DEBUGBUS_PRINTF_P(PSTR("%sstarted, width=%u, %u pixels.\n"), _valid? "":"not ", _panelWidth, _len);
 
-  if (_ledBuffer != nullptr) DEBUGBUS_PRINTLN(F("MatrixPanel_I2S_DMA LEDS buffer enabled."));
-  if (_ledsDirty != nullptr) DEBUGBUS_PRINTLN(F("MatrixPanel_I2S_DMA LEDS dirty bit optimization enabled."));
-  if ((_ledBuffer != nullptr) || (_ledsDirty != nullptr)) {
-    DEBUGBUS_PRINT(F("MatrixPanel_I2S_DMA LEDS buffer uses "));
-    DEBUGBUS_PRINT((_ledBuffer? _len*sizeof(CRGB) :0) + (_ledsDirty? getBitArrayBytes(_len) :0));
+  if (_ledsDirty != nullptr) {
+    DEBUGBUS_PRINTLN(F("MatrixPanel_I2S_DMA LEDS dirty bit optimization enabled."));
+    DEBUGBUS_PRINT(F("MatrixPanel_I2S_DMA LEDS dirty buffer uses "));
+    DEBUGBUS_PRINT((_ledsDirty? getBitArrayBytes(_len) :0));
     DEBUGBUS_PRINTLN(F(" bytes."));
   }
 }
@@ -1110,40 +1105,27 @@ BusHub75Matrix::BusHub75Matrix(const BusConfig &bc) : Bus(bc.type, bc.start, bc.
 void IRAM_ATTR BusHub75Matrix::setPixelColor(unsigned pix, uint32_t c) {
   if (!_valid) return; // note: no need to check pix >= _len as that is checked in containsPixel()
   // if (_cct >= 1900) c = colorBalanceFromKelvin(_cct, c); //color correction from CCT
+  if ((c == IS_BLACK) && (getBitFromArray(_ledsDirty, pix) == false)) return; // ignore black if pixel is already black
+  setBitInArray(_ledsDirty, pix, c != IS_BLACK);                              // dirty = true means "color is not BLACK"
 
-  if (_ledBuffer) {
-    CRGB fastled_col = CRGB(c);
-    if (_ledBuffer[pix] != fastled_col) {
-      _ledBuffer[pix] = fastled_col;
-      setBitInArray(_ledsDirty, pix, true);  // flag pixel as "dirty"
-    }
-  }
-  else {
-    if ((c == IS_BLACK) && (getBitFromArray(_ledsDirty, pix) == false)) return; // ignore black if pixel is already black
-    setBitInArray(_ledsDirty, pix, c != IS_BLACK);                              // dirty = true means "color is not BLACK"
+  uint8_t r = R(c);
+  uint8_t g = G(c);
+  uint8_t b = B(c);
 
-    uint8_t r = R(c);
-    uint8_t g = G(c);
-    uint8_t b = B(c);
-
-    if (virtualDisp != nullptr) {
-      int x = pix % _panelWidth; // TODO: check if using & and shift would be faster here, it limits to power-of-2 widths though
-      int y = pix / _panelWidth;
-      virtualDisp->drawPixelRGB888(int16_t(x), int16_t(y), r, g, b);
-    } else {
-      int x = pix % _panelWidth;
-      int y = pix / _panelWidth;
-      display->drawPixelRGB888(int16_t(x), int16_t(y), r, g, b);
-    }
+  if (virtualDisp != nullptr) {
+    int x = pix % _panelWidth; // TODO: check if using & and shift would be faster here, it limits to power-of-2 widths though
+    int y = pix / _panelWidth;
+    virtualDisp->drawPixelRGB888(int16_t(x), int16_t(y), r, g, b);
+  } else {
+    int x = pix % _panelWidth;
+    int y = pix / _panelWidth;
+    display->drawPixelRGB888(int16_t(x), int16_t(y), r, g, b);
   }
 }
 
 uint32_t BusHub75Matrix::getPixelColor(unsigned pix) const {
   if (!_valid) return IS_BLACK; // note: no need to check pix >= _len as that is checked in containsPixel()
-  if (_ledBuffer)
-    return uint32_t(_ledBuffer[pix]);  // fastled-slim already returns RGB, no need to mask out the upper byte
-  else
-    return getBitFromArray(_ledsDirty, pix) ? IS_DARKGREY: IS_BLACK;   // just a hack - we only know if the pixel is black or not
+  return getBitFromArray(_ledsDirty, pix) ? IS_DARKGREY: IS_BLACK;   // just a hack - we only know if the pixel is black or not
 }
 
 void BusHub75Matrix::setBrightness(uint8_t b) {
@@ -1153,25 +1135,8 @@ void BusHub75Matrix::setBrightness(uint8_t b) {
 }
 
 void BusHub75Matrix::show(void) {
-  if (!_valid) return;
-  if (_ledBuffer) {
-    // write out buffered LEDs
-    unsigned height = _isVirtual ? virtualDisp->height() : display->height();
-    unsigned width = _panelWidth;
-
-    //while(!previousBufferFree) delay(1);   // experimental - Wait before we allow any writing to the buffer. Stop flicker.
-    size_t pix = 0; // running pixel index
-    for (int y=0; y<height; y++) for (int x=0; x<width; x++) {
-      if (getBitFromArray(_ledsDirty, pix) == true) {        // only repaint the "dirty"  pixels
-        CRGB c = _ledBuffer[pix];
-        //c.nscale8_video(_bri); // apply brightness
-        if (_isVirtual) virtualDisp->drawPixelRGB888(int16_t(x), int16_t(y), c.r, c.g, c.b);
-        else                display->drawPixelRGB888(int16_t(x), int16_t(y), c.r, c.g, c.b);
-      }
-      pix++;
-    }
-    setBitArray(_ledsDirty, _len, false);  // buffer shown - reset all dirty bits
-  }
+  // show() is a no-op: pixels were already written directly in setPixelColor(),
+  // the DMA driver handles continuous refresh asynchronously
 }
 
 void BusHub75Matrix::cleanup() {
@@ -1189,7 +1154,6 @@ void BusHub75Matrix::cleanup() {
   #else  // runtime reconfiguration is not working on -S3, request reboot from user instead
     errorFlag = ERR_REBOOT_NEEDED;
   #endif
-  if (_ledBuffer != nullptr) d_free(_ledBuffer); _ledBuffer = nullptr;
   if (_ledsDirty != nullptr) d_free(_ledsDirty); _ledsDirty = nullptr;
 }
 
diff --git a/wled00/const.h b/wled00/const.h
index 04ff8ded61..56c240eb3c 100644
--- a/wled00/const.h
+++ b/wled00/const.h
@@ -546,7 +546,7 @@ static_assert(WLED_MAX_BUSSES <= 32, "WLED_MAX_BUSSES exceeds hard limit");
   #elif defined(CONFIG_IDF_TARGET_ESP32S2)
     #define MAX_LEDS 2048 //due to memory constraints S2
   #else
-    #define MAX_LEDS 16384
+    #define MAX_LEDS 16384 // note: S3 can take more but 24576 is already unstable using HUB75 (DMA buffers)
   #endif
 #endif
 
diff --git a/wled00/util.cpp b/wled00/util.cpp
index 20c934b0f1..fa194c9572 100644
--- a/wled00/util.cpp
+++ b/wled00/util.cpp
@@ -906,8 +906,8 @@ void *allocate_buffer(size_t size, uint32_t type) {
   buffer = d_malloc(size);
   #else
   if (type & BFRALLOC_PREFER_DRAM) {
-    if (getContiguousFreeHeap() < 3*(MIN_HEAP_SIZE/2) + size && size > PSRAM_THRESHOLD)
-      buffer = p_malloc(size); // prefer PSRAM for large allocations & when DRAM is low
+    if ((getContiguousFreeHeap() < 3*(MIN_HEAP_SIZE/2) + size && size > PSRAM_THRESHOLD) || size > 2*PSRAM_THRESHOLD)
+      buffer = p_malloc(size); // prefer PSRAM for large allocations & when DRAM is low or when buffer is huge
     else
       buffer = d_malloc(size); // allocate in DRAM if enough free heap is available, PSRAM as fallback
   }