Scumbria your thoughts please.
Could it be the program he used?
Is this latency so T-bird can scale?
"Thunderbird Cachemem Performance (CPU)
Posted By Brian Neal
Monday, June 5, 2000 - 10:01:50 PM
Well, I just woke up from a short nap and I wanted to go ahead and put out some results that weren't initially included in our Thunderbird review. For starters, if you haven't done so already, please go take a look at Codename: Thunderbird.
Ok, as many of you have no doubt noticed, the Thunderbird's L2 cache performance is a bit interesting and actually a bit dissapointing as quite a few programs report the latency extremely high. The Calibrator graph on the third page of the review illustrates this perfectly. However, below I've produced the test results I received for cachemem, a cache latency/bandwidth analysis program created by LRMS. The first output is for the Thunderbird, and 1000 MHz K75s follow (with 1/3 and 2/5 caches, respectively).
Now, when looking at the results from Calibrator or Cachemem, it's important to remember that AMD is quoting a latency figure of 11 cycles (8 without L1 miss penalty). This results are substantially higher, with cachemem reporting 20 cycles. Now, as mentioned in the review, I don't know the exact reason for these results. At best I can speculate, and I think that the narrow (64-bit) cache connection is a likely culprit for at least some part of this reported latency. Judging from these results, it's also very easy to conclude that the L2 cache is running at half of the host CPU clock, but I was unable to get any confirmation to that effect.
So please take a look at this cachemem results if you are interested and also take note of the latency as reported by Calibrator. Perhaps the message board can work to find a definite answer one way or the other regarding these results.
Thunderbird 1000
Cache size/Memory speed info tool 2.21MMX - (c) LRMS, 1999 - DJGPP compiled
CPUID support detected...FPU TSC MMX
CPU clock: 1001.9 MHz
Using 32MB physical memory block
Now testing memory bandwidth (please wait some seconds)
Block of 1KB, Speed R/W: 9314.9 / 7943.0 MB/s (0 / 0 cycles)
Block of 2KB, Speed R/W: 8811.7 / 7272.2 MB/s (0 / 1 cycles)
Block of 4KB, Speed R/W: 9093.4 / 7452.6 MB/s (0 / 1 cycles)
Block of 8KB, Speed R/W: 9244.0 / 7545.4 MB/s (0 / 1 cycles)
Block of 16KB, Speed R/W: 9327.0 / 7589.0 MB/s (0 / 1 cycles)
Block of 32KB, Speed R/W: 9365.2 / 7613.6 MB/s (0 / 1 cycles)
Block of 64KB, Speed R/W: 9384.7 / 7629.6 MB/s (0 / 1 cycles)
Block of 128KB, Speed R/W: 3053.3 / 3045.4 MB/s (2 / 2 cycles)
Block of 256KB, Speed R/W: 3055.7 / 3046.1 MB/s (2 / 2 cycles)
Block of 512KB, Speed R/W: 998.5 / 335.1 MB/s (7 / 22 cycles)
Block of 1024KB, Speed R/W: 998.4 / 335.2 MB/s (7 / 22 cycles)
Block of 2048KB, Speed R/W: 968.1 / 337.8 MB/s (7 / 22 cycles)
Block of 4096KB, Speed R/W: 961.0 / 338.4 MB/s (7 / 22 cycles)
Block of 8192KB, Speed R/W: 957.1 / 338.8 MB/s (7 / 22 cycles)
Block of 16384KB, Speed R/W: 955.3 / 339.0 MB/s (8 / 22 cycles)
Block of 32768KB, Speed R/W: 954.6 / 338.9 MB/s (8 / 22 cycles)
Latency - Memory walk tests...
Null size: 4 cycles 1 cycles
steps: 4 8 16 32 64 128 256 512 1k 2k 4k (bytes)
Block of 1KB: 4 4 4 4 4 4 4 4 cycles
Block of 2KB: 4 4 4 4 4 4 4 4 4 cycles
Block of 4KB: 4 4 4 4 4 4 4 4 4 4 cycles
Block of 8KB: 4 4 4 4 4 4 4 4 4 4 4 cycles
Block of 16KB: 4 4 4 4 4 4 4 4 4 4 4 cycles
Block of 32KB: 4 4 4 4 4 4 4 4 4 4 4 cycles
Block of 64KB: 4 4 4 4 4 4 4 4 4 4 4 cycles
Block of 128KB: 4 5 9 10 20 20 20 20 20 20 20 cycles
Block of 256KB: 4 5 9 10 20 20 20 20 20 20 20 cycles
Block of 512KB: 16 30 55 100 161 162 164 167 176 190 200 cycles
Block of 1024KB: 16 30 55 100 161 162 164 167 176 190 200 cycles
Block of 2048KB: 16 30 55 100 162 163 165 169 178 197 227 cycles
Block of 4096KB: 16 30 55 100 162 163 165 170 179 198 233 cycles
Block of 8192KB: 16 30 55 100 162 163 165 170 180 199 236 cycles
Block of 16384KB: 16 30 55 100 162 163 165 170 180 200 238 cycles
Block of 32768KB: 16 30 55 100 162 163 166 170 180 200 239 cycles
Done.
This system appears to have 2 cache levels (enabled).
L1 cache (64KB) speed (MB/s): Read=9365.2, Write=7613.6
L2 cache (256KB) speed (MB/s): Read=3053.3, Write=3045.4
Main memory speed (MB/s): Read=954.6, Write=338.9
Athlon 1000 (1/3)
Cache size/Memory speed info tool 2.22MMX - (c) LRMS, 1999 - DJGPP compiled
CPUID support detected... 'AuthenticAMD' with FPU TSC MMX
Family=6 Model=1 Step=2 Type=0 Chipset (Vendor/Device ID(Rev)): AMD/7006(23)
CPU clock: 998.2 MHz
Using 32MB physical memory block
Now testing memory bandwidth (please wait some seconds)
Block of 1KB, Speed R/W: 9095.3 / 7777.2 MB/s (0 / 0 cycles)
Block of 2KB, Speed R/W: 8603.4 / 7073.2 MB/s (0 / 1 cycles)
Block of 4KB, Speed R/W: 8892.3 / 7276.6 MB/s (0 / 1 cycles)
Block of 8KB, Speed R/W: 9035.9 / 7370.0 MB/s (0 / 1 cycles)
Block of 16KB, Speed R/W: 9076.0 / 7417.1 MB/s (0 / 1 cycles)
Block of 32KB, Speed R/W: 9147.7 / 7442.9 MB/s (0 / 1 cycles)
Block of 64KB, Speed R/W: 9123.8 / 7448.8 MB/s (0 / 1 cycles)
Block of 128KB, Speed R/W: 1986.2 / 795.4 MB/s (3 / 9 cycles)
Block of 256KB, Speed R/W: 1987.6 / 795.9 MB/s (3 / 9 cycles)
Block of 512KB, Speed R/W: 1954.9 / 793.2 MB/s (3 / 9 cycles)
Block of 1024KB, Speed R/W: 719.0 / 341.7 MB/s (10 / 22 cycles)
Block of 2048KB, Speed R/W: 647.1 / 313.7 MB/s (11 / 24 cycles)
Block of 4096KB, Speed R/W: 643.1 / 313.5 MB/s (11 / 24 cycles)
Block of 8192KB, Speed R/W: 641.3 / 312.0 MB/s (11 / 24 cycles)
Block of 16384KB, Speed R/W: 641.3 / 313.6 MB/s (11 / 24 cycles)
Block of 32768KB, Speed R/W: 638.6 / 313.5 MB/s (11 / 24 cycles)
Latency - Memory walk tests...
Null size: 4 cycles 1 cycles
steps: 4 8 16 32 64 128 256 512 1k 2k 4k (bytes)
Block of 1KB: 4 4 4 4 4 4 4 4 cycles
Block of 2KB: 4 4 4 4 4 4 4 4 4 cycles
Block of 4KB: 4 4 4 4 4 4 4 4 4 4 cycles
Block of 8KB: 4 4 4 4 4 4 4 4 4 4 4 cycles
Block of 16KB: 4 4 4 4 4 4 4 4 4 4 4 cycles
Block of 32KB: 4 4 4 4 4 4 4 4 4 4 4 cycles
Block of 64KB: 4 4 4 4 4 4 4 4 4 4 4 cycles
Block of 128KB: 5 7 17 23 30 30 31 31 32 33 36 cycles
Block of 256KB: 5 7 17 23 30 30 31 31 32 33 36 cycles
Block of 512KB: 5 7 18 23 32 32 32 32 34 35 39 cycles
Block of 1024KB: 17 32 57 104 172 171 168 167 169 173 180 cycles
Block of 2048KB: 18 35 62 115 194 198 202 204 210 220 241 cycles
Block of 4096KB: 19 37 65 121 201 201 205 210 217 230 258 cycles
Block of 8192KB: 20 38 66 123 205 206 206 211 219 234 263 cycles
Block of 16384KB: 20 38 66 123 206 207 208 212 219 235 264 cycles
Block of 32768KB: 20 38 67 123 206 207 208 212 220 236 265 cycles
Done.
This system appears to have 2 cache levels (enabled).
L1 cache (64KB) speed (MB/s): Read=9147.7, Write=7442.9
L2 cache (512KB) speed (MB/s): Read=1987.6, Write=795.9
Main memory speed (MB/s): Read=638.6, Write=313.5
Athlon 1000 (2/5)
Cache size/Memory speed info tool 2.22MMX - (c) LRMS, 1999 - DJGPP compiled
CPUID support detected... 'AuthenticAMD' with FPU TSC MMX
Family=6 Model=1 Step=2 Type=0 Chipset (Vendor/Device ID(Rev)): AMD/7006(23)
CPU clock: 998.2 MHz
Using 32MB physical memory block
Now testing memory bandwidth (please wait some seconds)
Block of 1KB, Speed R/W: 9095.1 / 7778.1 MB/s (0 / 0 cycles)
Block of 2KB, Speed R/W: 8604.7 / 7132.1 MB/s (0 / 1 cycles)
Block of 4KB, Speed R/W: 8902.3 / 7277.5 MB/s (0 / 1 cycles)
Block of 8KB, Speed R/W: 9036.6 / 7341.0 MB/s (0 / 1 cycles)
Block of 16KB, Speed R/W: 9112.3 / 7417.0 MB/s (0 / 1 cycles)
Block of 32KB, Speed R/W: 9151.0 / 7410.9 MB/s (0 / 1 cycles)
Block of 64KB, Speed R/W: 9162.8 / 7448.9 MB/s (0 / 1 cycles)
Block of 128KB, Speed R/W: 2383.1 / 917.9 MB/s (3 / 8 cycles)
Block of 256KB, Speed R/W: 2384.9 / 918.7 MB/s (3 / 8 cycles)
Block of 512KB, Speed R/W: 2369.9 / 911.9 MB/s (3 / 8 cycles)
Block of 1024KB, Speed R/W: 735.8 / 342.0 MB/s (10 / 22 cycles)
Block of 2048KB, Speed R/W: 649.7 / 311.9 MB/s (11 / 24 cycles)
Block of 4096KB, Speed R/W: 643.1 / 313.8 MB/s (11 / 24 cycles)
Block of 8192KB, Speed R/W: 640.3 / 312.2 MB/s (11 / 24 cycles)
Block of 16384KB, Speed R/W: 639.4 / 313.4 MB/s (11 / 24 cycles)
Block of 32768KB, Speed R/W: 639.5 / 313.7 MB/s (11 / 24 cycles)
Latency - Memory walk tests...
Null size: 4 cycles 1 cycles
steps: 4 8 16 32 64 128 256 512 1k 2k 4k (bytes)
Block of 1KB: 4 4 4 4 4 4 4 4 cycles
Block of 2KB: 4 4 4 4 4 4 4 4 4 cycles
Block of 4KB: 4 4 4 4 4 4 4 4 4 4 cycles
Block of 8KB: 4 4 4 4 4 4 4 4 4 4 4 cycles
Block of 16KB: 4 4 4 4 4 4 4 4 4 4 4 cycles
Block of 32KB: 4 4 4 4 4 4 4 4 4 4 4 cycles
Block of 64KB: 4 4 4 4 4 4 4 4 4 4 4 cycles
Block of 128KB: 5 7 16 20 28 28 28 28 29 30 33 cycles
Block of 256KB: 5 7 16 20 28 28 28 28 29 30 33 cycles
Block of 512KB: 6 7 16 21 29 29 29 30 30 32 34 cycles
Block of 1024KB: 17 32 57 103 171 171 167 166 167 171 180 cycles
Block of 2048KB: 18 35 62 115 193 199 202 204 210 219 246 cycles
Block of 4096KB: 20 37 65 120 200 201 204 209 217 230 257 cycles
Block of 8192KB: 20 38 66 123 204 205 207 211 219 233 264 cycles
Block of 16384KB: 20 38 66 123 206 205 208 211 219 234 265 cycles
Block of 32768KB: 20 38 67 123 206 207 209 212 219 235 266 cycles
Done.
This system appears to have 2 cache levels (enabled).
L1 cache (64KB) speed (MB/s): Read=9151.0, Write=7410.9
L2 cache (512KB) speed (MB/s): Read=2384.9, Write=918.7
Main memory speed (MB/s): Read=639.5, Write=313.7
" |
