diff --git a/blake32.cu b/blake32.cu
index e7aae76..65df026 100644
--- a/blake32.cu
+++ b/blake32.cu
@@ -39,8 +39,6 @@ extern "C" void blake256hash(void *output, const void *input, int8_t rounds = 14
 
 #include "cuda_helper.h"
 
-#define MAXU 0xffffffffU
-
 #if PRECALC64
 __constant__ uint32_t _ALIGN(32) d_data[12];
 #else
@@ -58,7 +56,7 @@ static uint32_t *h_resNonce[8];
 
 /* max count of found nonces in one call */
 #define NBN 2
-static uint32_t extra_results[NBN] = { MAXU };
+static uint32_t extra_results[NBN] = { UINT32_MAX };
 
 /* prefer uint32_t to prevent size conversions = speed +5/10 % */
 __constant__
@@ -250,7 +248,7 @@ uint32_t blake256_cpu_hash_80(const int thr_id, const uint32_t threads, const ui
 	const uint32_t crcsum, const int8_t rounds)
 {
 	const int threadsperblock = TPB;
-	uint32_t result = MAXU;
+	uint32_t result = UINT32_MAX;
 
 	dim3 grid((threads + threadsperblock-1)/threadsperblock);
 	dim3 block(threadsperblock);
@@ -339,7 +337,7 @@ static uint32_t blake256_cpu_hash_16(const int thr_id, const uint32_t threads, c
 	const int8_t rounds)
 {
 	const int threadsperblock = TPB;
-	uint32_t result = MAXU;
+	uint32_t result = UINT32_MAX;
 
 	dim3 grid((threads + threadsperblock-1)/threadsperblock);
 	dim3 block(threadsperblock);
@@ -404,12 +402,12 @@ extern "C" int scanhash_blake256(int thr_id, uint32_t *pdata, const uint32_t *pt
 	int rc = 0;
 
 #if NBN > 1
-	if (extra_results[0] != MAXU) {
+	if (extra_results[0] != UINT32_MAX) {
 		// possible extra result found in previous call
 		if (first_nonce <= extra_results[0] && max_nonce >= extra_results[0]) {
 			pdata[19] = extra_results[0];
 			*hashes_done = pdata[19] - first_nonce + 1;
-			extra_results[0] = MAXU;
+			extra_results[0] = UINT32_MAX;
 			rc = 1;
 			goto exit_scan;
 		}
@@ -455,7 +453,7 @@ extern "C" int scanhash_blake256(int thr_id, uint32_t *pdata, const uint32_t *pt
 		// GPU FULL HASH
 		blake256_cpu_hash_80(thr_id, throughput, pdata[19], targetHigh, crcsum, blakerounds);
 #endif
-		if (foundNonce != MAXU)
+		if (foundNonce != UINT32_MAX)
 		{
 			uint32_t vhashcpu[8];
 			uint32_t Htarg = (uint32_t)targetHigh;
@@ -472,7 +470,7 @@ extern "C" int scanhash_blake256(int thr_id, uint32_t *pdata, const uint32_t *pt
 				pdata[19] = foundNonce;
 				rc = 1;
 
-				if (extra_results[0] != MAXU) {
+				if (extra_results[0] != UINT32_MAX) {
 					// Rare but possible if the throughput is big
 					be32enc(&endiandata[19], extra_results[0]);
 
@@ -481,7 +479,7 @@ extern "C" int scanhash_blake256(int thr_id, uint32_t *pdata, const uint32_t *pt
 						applog(LOG_NOTICE, "GPU found more than one result " CL_GRN "yippee!");
 						rc = 2;
 					} else {
-						extra_results[0] = MAXU;
+						extra_results[0] = UINT32_MAX;
 					}
 				}
 
diff --git a/ccminer.cpp b/ccminer.cpp
index cfd72eb..e24fc97 100644
--- a/ccminer.cpp
+++ b/ccminer.cpp
@@ -1122,8 +1122,12 @@ static void *miner_thread(void *userdata)
 			max64 = max(minmax-1, max64);
 		}
 
-		if (opt_debug)
-			applog(LOG_DEBUG, "GPU #%d: start=%08x range=%llx", device_map[thr_id], *nonceptr, max64);
+		// we can't scan more than uint capacity
+		max64 = min(UINT32_MAX, max64);
+
+		//if (opt_debug)
+		//	applog(LOG_DEBUG, "GPU #%d: start=%08x end=%08x max64=%llx",
+		//		device_map[thr_id], *nonceptr, end_nonce, max64);
 
 		start_nonce = *nonceptr;
 
@@ -1137,7 +1141,7 @@ static void *miner_thread(void *userdata)
 			range.data = hashlog_get_scan_range(work.job_id);
 			if (range.data && !opt_benchmark) {
 				bool stall = false;
-				if (range.scanned[0] == 1 && range.scanned[1] == 0xFFFFFFFFUL) {
+				if (range.scanned[0] == 1 && range.scanned[1] == UINT32_MAX) {
 					applog(LOG_WARNING, "detected a rescan of fully scanned job!");
 				} else if (range.scanned[0] > 0 && range.scanned[1] > 0 && range.scanned[1] < 0xFFFFFFF0UL) {
 					/* continue scan the end */
@@ -1163,18 +1167,24 @@ static void *miner_thread(void *userdata)
 			}
 		}
 
+		/* never let small ranges at end */
+		if (end_nonce >= UINT32_MAX - 256)
+			end_nonce = UINT32_MAX;
+
 		if ((max64 + start_nonce) >= end_nonce)
 			max_nonce = end_nonce;
 		else
 			max_nonce = (uint32_t) (max64 + start_nonce);
 
-		/* never let small ranges at end */
-		if (max_nonce >= UINT32_MAX - 256)
-			max_nonce = UINT32_MAX;
+		// todo: keep it rounded for gpu threads ?
 
 		work.scanned_from = start_nonce;
 		(*nonceptr) = start_nonce;
 
+		if (opt_debug)
+			applog(LOG_DEBUG, "GPU #%d: start=%08x end=%08x range=%08x",
+				device_map[thr_id], start_nonce, max_nonce, (max_nonce-start_nonce));
+
 		hashes_done = 0;
 continue_scan:
 		gettimeofday(&tv_start, NULL);
diff --git a/stats.cpp b/stats.cpp
index 6204006..ad60831 100644
--- a/stats.cpp
+++ b/stats.cpp
@@ -28,7 +28,7 @@ extern int device_map[8];
 void stats_remember_speed(int thr_id, uint32_t hashcount, double hashrate, uint8_t found)
 {
 	uint64_t gpu = device_map[thr_id];
-	uint64_t key = (gpu << 56) + (uid++ % UINT_MAX);
+	uint64_t key = (gpu << 56) + (uid++ % UINT32_MAX);
 	stats_data data;
 	// to enough hashes to give right stats
 	if (hashcount < 1000 || hashrate < 0.01)