Intel Pentium MMX
- L1 Data cache = 16 KB. 32 B/line. 4-Way
- L1 Instruction cache = 16 KB. 32 B/line, 4-WAY.
- L1 B/W (Parallel Random Read) = 0.62 cycles per one access
Intel Pentium MMX 200 MHz
Intel Pentium MMX 200 MHz (100 * 2) + 128 MB (PC-133 CL2 @ 100MHz). via MVP3, L2 512 KB.
- L2 cache size = 512 KB. direct-mapped. 32-byte line size. Latency = 2 + 50 ns
- RAM latency = 2 + 50 ns + 54 ns (for open RAM page). Open RAM page size is 2 KB?
- RAM latency = 2 + 50 ns + 54 ns + 60 ns (for new RAM page)
4 MB pages mode (linux)
TLB size = 64 items.
Size Latency Increase Description
16 K 2
32 K 2 + 25 ns 25 ns + 50 ns (L2 cache)
64 K 2 + 38 ns 13 ns
128 K 2 + 44 ns 6 ns
256 K 2 + 47 ns 3 ns
512 K 2 + 49 ns 2 ns
1 M 2 + 110 ns 61 ns + 114 ns (RAM)
2 M 2 + 139 ns 29 ns
4 M 2 + 151 ns 12 ns
8 M 2 + 157 ns 6 ns
16 M 2 + 160 ns 3 ns
32 M 2 + 162 ns 2 ns
64 M 2 + 163 ns 1 ns
- L2->L1 B/W (Random Read) = 12 cycles (2 + 50 ns) per cache line (32-bytes) = 533 MB/s
- L2->L1 B/W (Read, 32 bytes step) = 12 cycles (2 + 50 ns) cycles per cache line (32-bytes) = 533 MB/s
- RAM Read B/W (Random Read) = 2 + 164 ns / cache line = 184 MB/s
- RAM Read B/W (Read, 4 Bytes step) = 170 MB/s
- RAM Read B/W (Read, 32 Bytes step) = 2 + 104 ns / cache line (32-bytes line) = 280 MB/s
- L1/L2/RAM Write = 10 cycles (50 ns) per write = 80 MB/s
4 KB pages mode (linux)
TLB size = 64 items. TLB miss penalty = 4 cycles + 100 ns
Size Latency Increase Description
16 K 2
32 K 2 + 25 ns 25 ns + 50 ns (L2 cache)
64 K 2 + 37 ns 12 ns
128 K 2 + 54 ns 17 ns
256 K 2 + 68 ns 14 ns
512 K 4 + 140 ns 2 + 72 ns + 4 + 100 ns (TLB miss to L2)
1 M 5 + 190 ns 1 + 50 ns + 114 ns (RAM)
2 M 6 + 223 ns 1 + 33 ns
4 M 6 + 245 ns + 22 ns
8 M 6 + 258 ns 13 ns
16 M 6 + 265 ns 7 ns
32 M 6 + 273 ns 8 ns
64 M 6 + 284 ns 11 ns + (Page walk to RAM) ?
Intel Pentium MMX 166 MHz
Intel Pentium MMX 166 MHz + 128 MB (PC-133 CL2 @ 66MHz) (only first 64 were used). i430TX.
- L2 cache size = 512 KB. direct-mapped? 32-byte line size. Latency = 2 + 60 ns
- L1/L2/RAM Write = 7.5 cycles (45 ns) per write. 90 MB / s
MISC
- 4-bytes range cross penalty = 3 cycles.
- Branch misprediction penalty = 4 cycles.
Links
Intel Pentium MMX at Wikipedia