diff --git a/src/cyclic_load_bytes.asm b/src/cyclic_load_bytes.asm
new file mode 100644
index 0000000..9e94a8a
--- /dev/null
+++ b/src/cyclic_load_bytes.asm
@@ -0,0 +1,22 @@
+global cyclic_load_bytes
+
+section .text
+
+; rdi - buffer
+; rsi - byte_count
+; rdx - mask
+cyclic_load_bytes:
+	xor rcx, rcx
+	align 64
+.loop:
+	mov r8, rcx
+	and r8, rdx
+	add r8, rdi
+	vmovdqu ymm0, [r8]
+	vmovdqu ymm0, [r8 + 32]
+	vmovdqu ymm0, [r8 + 64]
+	vmovdqu ymm0, [r8 + 96]
+	add rcx, 128
+	cmp rcx, rsi
+	jb .loop
+	ret
diff --git a/src/cyclic_load_bytes.h b/src/cyclic_load_bytes.h
new file mode 100644
index 0000000..2156604
--- /dev/null
+++ b/src/cyclic_load_bytes.h
@@ -0,0 +1,3 @@
+#include <stdint.h>
+
+void cyclic_load_bytes(uint8_t *buffer, uint64_t byte_count, uint64_t mask);
diff --git a/src/repetition_tester.c b/src/repetition_tester.c
index 2443a53..7e668ff 100644
--- a/src/repetition_tester.c
+++ b/src/repetition_tester.c
@@ -165,6 +165,12 @@ static float divide_safe(float a, float b) {
 	return b != 0 ? a / b : 0;
 }
 
+float repetitor_get_best_bandwidth(struct repetitor *repetitor) {
+	uint64_t min_time_taken = repetitor->min_time_taken;
+	uint64_t min_byte_count = repetitor->min_byte_count;
+	return bytes_to_gb_s(repetitor, min_byte_count, min_time_taken);
+}
+
 void repetitor_print_results(struct repetitor *repetitor) {
 	uint64_t avg_time_taken = repetitor->total_time_taken/repetitor->repetition_count;
 	uint64_t min_time_taken = repetitor->min_time_taken;
diff --git a/src/tests/cache_size.c b/src/tests/cache_size.c
new file mode 100644
index 0000000..a892401
--- /dev/null
+++ b/src/tests/cache_size.c
@@ -0,0 +1,75 @@
+#include "repetition_tester.c"
+#include "cyclic_load_bytes.h"
+#include <sys/mman.h>
+
+struct testcase {
+	char *name;
+	uint64_t mask;
+};
+
+struct testcase cases[] = {
+	{ .name = "load 4KiB"  , .mask = 0b111111111111 },
+	{ .name = "load 8KiB"  , .mask = 0b1111111111111 },
+	{ .name = "load 16KiB" , .mask = 0b11111111111111 },
+	{ .name = "load 32KiB" , .mask = 0b111111111111111 },
+	{ .name = "load 64KiB" , .mask = 0b1111111111111111 },
+	{ .name = "load 128KiB", .mask = 0b11111111111111111 },
+
+	{ .name = "load 512KiB", .mask = 0b1111111111111111111 },
+	{ .name = "load 1MiB"  , .mask = 0b11111111111111111111 },
+	{ .name = "load 2MiB"  , .mask = 0b111111111111111111111 },
+	{ .name = "load 4MiB"  , .mask = 0b1111111111111111111111 },
+	{ .name = "load 8MiB"  , .mask = 0b11111111111111111111111 },
+	{ .name = "load 16MiB" , .mask = 0b111111111111111111111111 },
+	{ .name = "load 32MiB" , .mask = 0b1111111111111111111111111 },
+	{ .name = "load 64MiB" , .mask = 0b11111111111111111111111111 },
+	{ .name = "load 128MiB", .mask = 0b111111111111111111111111111 },
+
+	{ .name = "load 1GiB"  , .mask = 0b111111111111111111111111111111 },
+};
+
+int main() {
+	struct repetitor repetitor = {};
+	repetitor_init(&repetitor);
+	printf("CPU Frequency: %ldHz (~%.2fGHz)\n", repetitor.cpu_freq, (float)repetitor.cpu_freq/(1000*1000*1000));
+
+	uint64_t byte_count = 1024 * 1024 * 1024;
+	if (byte_count % 4096) {
+		printf("ERROR: Size of buffer is not page aligned\n");
+		return -1;
+	}
+
+	uint8_t *buffer = mmap(0, byte_count, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
+	if (buffer == NULL) {
+		printf("ERROR: Failed to allocate buffer\n");
+		return -1;
+	}
+
+	if ((size_t)buffer % 64 != 0) {
+		printf("ERROR: Allocated buffer is not cache line aligned, it is %ld\n", (size_t)buffer % 64);
+		return -1;
+	}
+
+	// Touch pages so they would be mapped in, to avoid page faults during tests
+	for (uint64_t i = 0; i < byte_count; i += 4096) {
+		buffer[i] = (uint8_t)i;
+	}
+
+	for (int i = 0; i < ARRAY_LEN(cases); i++) {
+		struct testcase *testcase = &cases[i];
+		repetitor_clear(&repetitor);
+		while (repetitor_repeat(&repetitor, 2)) {
+			repetitor_start(&repetitor);
+			repetitor_measure_start(&repetitor);
+			cyclic_load_bytes(buffer, byte_count, testcase->mask);
+			repetitor_measure_stop(&repetitor, byte_count);
+			repetitor_stop(&repetitor);
+		}
+		repetitor_print_results_label(&repetitor, testcase->name);
+		// printf("%ld;%f\n", testcase->mask, repetitor_get_best_bandwidth(&repetitor));
+	}
+
+	munmap(buffer, byte_count);
+
+	return 0;
+}
diff --git a/src/tests/load_uop.c b/src/tests/load_uop.c
index 4d7d149..4c85231 100644
--- a/src/tests/load_uop.c
+++ b/src/tests/load_uop.c
@@ -1,20 +1,20 @@
 #include "repetition_tester.c"
 #include "load_uop.h"
 
+typedef void (*test_cb)(uint8_t *buffer, uint64_t byte_count);
+struct testcase {
+	char *name;
+	test_cb cb;
+};
+
+struct testcase cases[] = {
+	{ .name = "mov_load_x1()", .cb = mov_load_x1 },
+	{ .name = "mov_load_x2()", .cb = mov_load_x2 },
+	{ .name = "mov_load_x3()", .cb = mov_load_x3 },
+	{ .name = "mov_load_x4()", .cb = mov_load_x4 },
+};
+
 int main() {
-	typedef void (*test_cb)(uint8_t *buffer, uint64_t byte_count);
-	struct testcase {
-		char *name;
-		test_cb cb;
-	};
-
-	struct testcase cases[] = {
-		{ .name = "mov_load_x1()", .cb = mov_load_x1 },
-		{ .name = "mov_load_x2()", .cb = mov_load_x2 },
-		{ .name = "mov_load_x3()", .cb = mov_load_x3 },
-		{ .name = "mov_load_x4()", .cb = mov_load_x4 },
-	};
-
 	struct repetitor repetitor = {};
 	repetitor_init(&repetitor);
 	printf("CPU Frequency: %ldHz (~%.2fGHz)\n", repetitor.cpu_freq, (float)repetitor.cpu_freq/(1000*1000*1000));