Loading...
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 | [ { "BriefDescription": "DRAM Activate Count", "EventCode": "0x1", "EventName": "UNC_M_ACT_COUNT", "PerPkg": "1", "PublicDescription": "Counts the number of DRAM Activate commands sent on this channel. Activate commands are issued to open up a page on the DRAM devices so that it can be read or written to with a CAS. One can calculate the number of Page Misses by subtracting the number of Page Miss precharges from the number of Activates.", "Unit": "iMC" }, { "BriefDescription": "DRAM RD_CAS and WR_CAS Commands.; All DRAM WR_CAS (w/ and w/out auto-pre)", "EventCode": "0x4", "EventName": "UNC_M_CAS_COUNT.ALL", "PerPkg": "1", "UMask": "0xf", "Unit": "iMC" }, { "BriefDescription": "DRAM RD_CAS and WR_CAS Commands.; All DRAM Reads (RD_CAS + Underfills)", "EventCode": "0x4", "EventName": "UNC_M_CAS_COUNT.RD", "PerPkg": "1", "UMask": "0x3", "Unit": "iMC" }, { "BriefDescription": "DRAM RD_CAS and WR_CAS Commands.; All DRAM RD_CAS (w/ and w/out auto-pre)", "EventCode": "0x4", "EventName": "UNC_M_CAS_COUNT.RD_REG", "PerPkg": "1", "UMask": "0x1", "Unit": "iMC" }, { "BriefDescription": "DRAM RD_CAS and WR_CAS Commands.; Underfill Read Issued", "EventCode": "0x4", "EventName": "UNC_M_CAS_COUNT.RD_UNDERFILL", "PerPkg": "1", "UMask": "0x2", "Unit": "iMC" }, { "BriefDescription": "DRAM RD_CAS and WR_CAS Commands.; All DRAM WR_CAS (both Modes)", "EventCode": "0x4", "EventName": "UNC_M_CAS_COUNT.WR", "PerPkg": "1", "UMask": "0xc", "Unit": "iMC" }, { "BriefDescription": "DRAM RD_CAS and WR_CAS Commands.; DRAM WR_CAS (w/ and w/out auto-pre) in Read Major Mode", "EventCode": "0x4", "EventName": "UNC_M_CAS_COUNT.WR_RMM", "PerPkg": "1", "UMask": "0x8", "Unit": "iMC" }, { "BriefDescription": "DRAM RD_CAS and WR_CAS Commands.; DRAM WR_CAS (w/ and w/out auto-pre) in Write Major Mode", "EventCode": "0x4", "EventName": "UNC_M_CAS_COUNT.WR_WMM", "PerPkg": "1", "UMask": "0x4", "Unit": "iMC" }, { "BriefDescription": "uclks", "EventName": "UNC_M_CLOCKTICKS", "PerPkg": "1", "PublicDescription": "Uncore Fixed Counter - uclks", "Unit": "iMC" }, { "BriefDescription": "DRAM Precharge All Commands", "EventCode": "0x6", "EventName": "UNC_M_DRAM_PRE_ALL", "PerPkg": "1", "PublicDescription": "Counts the number of times that the precharge all command was sent.", "Unit": "iMC" }, { "BriefDescription": "Number of DRAM Refreshes Issued", "EventCode": "0x5", "EventName": "UNC_M_DRAM_REFRESH.HIGH", "PerPkg": "1", "PublicDescription": "Counts the number of refreshes issued.", "UMask": "0x4", "Unit": "iMC" }, { "BriefDescription": "Number of DRAM Refreshes Issued", "EventCode": "0x5", "EventName": "UNC_M_DRAM_REFRESH.PANIC", "PerPkg": "1", "PublicDescription": "Counts the number of refreshes issued.", "UMask": "0x2", "Unit": "iMC" }, { "BriefDescription": "ECC Correctable Errors", "EventCode": "0x9", "EventName": "UNC_M_ECC_CORRECTABLE_ERRORS", "PerPkg": "1", "PublicDescription": "Counts the number of ECC errors detected and corrected by the iMC on this channel. This counter is only useful with ECC DRAM devices. This count will increment one time for each correction regardless of the number of bits corrected. The iMC can correct up to 4 bit errors in independent channel mode and 8 bit errors in lockstep mode.", "Unit": "iMC" }, { "BriefDescription": "Cycles in a Major Mode; Isoch Major Mode", "EventCode": "0x7", "EventName": "UNC_M_MAJOR_MODES.ISOCH", "PerPkg": "1", "PublicDescription": "Counts the total number of cycles spent in a major mode (selected by a filter) on the given channel. Major modea are channel-wide, and not a per-rank (or dimm or bank) mode.", "UMask": "0x8", "Unit": "iMC" }, { "BriefDescription": "Cycles in a Major Mode; Partial Major Mode", "EventCode": "0x7", "EventName": "UNC_M_MAJOR_MODES.PARTIAL", "PerPkg": "1", "PublicDescription": "Counts the total number of cycles spent in a major mode (selected by a filter) on the given channel. Major modea are channel-wide, and not a per-rank (or dimm or bank) mode.", "UMask": "0x4", "Unit": "iMC" }, { "BriefDescription": "Cycles in a Major Mode; Read Major Mode", "EventCode": "0x7", "EventName": "UNC_M_MAJOR_MODES.READ", "PerPkg": "1", "PublicDescription": "Counts the total number of cycles spent in a major mode (selected by a filter) on the given channel. Major modea are channel-wide, and not a per-rank (or dimm or bank) mode.", "UMask": "0x1", "Unit": "iMC" }, { "BriefDescription": "Cycles in a Major Mode; Write Major Mode", "EventCode": "0x7", "EventName": "UNC_M_MAJOR_MODES.WRITE", "PerPkg": "1", "PublicDescription": "Counts the total number of cycles spent in a major mode (selected by a filter) on the given channel. Major modea are channel-wide, and not a per-rank (or dimm or bank) mode.", "UMask": "0x2", "Unit": "iMC" }, { "BriefDescription": "Channel DLLOFF Cycles", "EventCode": "0x84", "EventName": "UNC_M_POWER_CHANNEL_DLLOFF", "PerPkg": "1", "PublicDescription": "Number of cycles when all the ranks in the channel are in CKE Slow (DLLOFF) mode.", "Unit": "iMC" }, { "BriefDescription": "Channel PPD Cycles", "EventCode": "0x85", "EventName": "UNC_M_POWER_CHANNEL_PPD", "PerPkg": "1", "PublicDescription": "Number of cycles when all the ranks in the channel are in PPD mode. If IBT=off is enabled, then this can be used to count those cycles. If it is not enabled, then this can count the number of cycles when that could have been taken advantage of.", "Unit": "iMC" }, { "BriefDescription": "CKE_ON_CYCLES by Rank; DIMM ID", "EventCode": "0x83", "EventName": "UNC_M_POWER_CKE_CYCLES.RANK0", "PerPkg": "1", "PublicDescription": "Number of cycles spent in CKE ON mode. The filter allows you to select a rank to monitor. If multiple ranks are in CKE ON mode at one time, the counter will ONLY increment by one rather than doing accumulation. Multiple counters will need to be used to track multiple ranks simultaneously. There is no distinction between the different CKE modes (APD, PPDS, PPDF). This can be determined based on the system programming. These events should commonly be used with Invert to get the number of cycles in power saving mode. Edge Detect is also useful here. Make sure that you do NOT use Invert with Edge Detect (this just confuses the system and is not necessary).", "UMask": "0x1", "Unit": "iMC" }, { "BriefDescription": "CKE_ON_CYCLES by Rank; DIMM ID", "EventCode": "0x83", "EventName": "UNC_M_POWER_CKE_CYCLES.RANK1", "PerPkg": "1", "PublicDescription": "Number of cycles spent in CKE ON mode. The filter allows you to select a rank to monitor. If multiple ranks are in CKE ON mode at one time, the counter will ONLY increment by one rather than doing accumulation. Multiple counters will need to be used to track multiple ranks simultaneously. There is no distinction between the different CKE modes (APD, PPDS, PPDF). This can be determined based on the system programming. These events should commonly be used with Invert to get the number of cycles in power saving mode. Edge Detect is also useful here. Make sure that you do NOT use Invert with Edge Detect (this just confuses the system and is not necessary).", "UMask": "0x2", "Unit": "iMC" }, { "BriefDescription": "CKE_ON_CYCLES by Rank; DIMM ID", "EventCode": "0x83", "EventName": "UNC_M_POWER_CKE_CYCLES.RANK2", "PerPkg": "1", "PublicDescription": "Number of cycles spent in CKE ON mode. The filter allows you to select a rank to monitor. If multiple ranks are in CKE ON mode at one time, the counter will ONLY increment by one rather than doing accumulation. Multiple counters will need to be used to track multiple ranks simultaneously. There is no distinction between the different CKE modes (APD, PPDS, PPDF). This can be determined based on the system programming. These events should commonly be used with Invert to get the number of cycles in power saving mode. Edge Detect is also useful here. Make sure that you do NOT use Invert with Edge Detect (this just confuses the system and is not necessary).", "UMask": "0x4", "Unit": "iMC" }, { "BriefDescription": "CKE_ON_CYCLES by Rank; DIMM ID", "EventCode": "0x83", "EventName": "UNC_M_POWER_CKE_CYCLES.RANK3", "PerPkg": "1", "PublicDescription": "Number of cycles spent in CKE ON mode. The filter allows you to select a rank to monitor. If multiple ranks are in CKE ON mode at one time, the counter will ONLY increment by one rather than doing accumulation. Multiple counters will need to be used to track multiple ranks simultaneously. There is no distinction between the different CKE modes (APD, PPDS, PPDF). This can be determined based on the system programming. These events should commonly be used with Invert to get the number of cycles in power saving mode. Edge Detect is also useful here. Make sure that you do NOT use Invert with Edge Detect (this just confuses the system and is not necessary).", "UMask": "0x8", "Unit": "iMC" }, { "BriefDescription": "CKE_ON_CYCLES by Rank; DIMM ID", "EventCode": "0x83", "EventName": "UNC_M_POWER_CKE_CYCLES.RANK4", "PerPkg": "1", "PublicDescription": "Number of cycles spent in CKE ON mode. The filter allows you to select a rank to monitor. If multiple ranks are in CKE ON mode at one time, the counter will ONLY increment by one rather than doing accumulation. Multiple counters will need to be used to track multiple ranks simultaneously. There is no distinction between the different CKE modes (APD, PPDS, PPDF). This can be determined based on the system programming. These events should commonly be used with Invert to get the number of cycles in power saving mode. Edge Detect is also useful here. Make sure that you do NOT use Invert with Edge Detect (this just confuses the system and is not necessary).", "UMask": "0x10", "Unit": "iMC" }, { "BriefDescription": "CKE_ON_CYCLES by Rank; DIMM ID", "EventCode": "0x83", "EventName": "UNC_M_POWER_CKE_CYCLES.RANK5", "PerPkg": "1", "PublicDescription": "Number of cycles spent in CKE ON mode. The filter allows you to select a rank to monitor. If multiple ranks are in CKE ON mode at one time, the counter will ONLY increment by one rather than doing accumulation. Multiple counters will need to be used to track multiple ranks simultaneously. There is no distinction between the different CKE modes (APD, PPDS, PPDF). This can be determined based on the system programming. These events should commonly be used with Invert to get the number of cycles in power saving mode. Edge Detect is also useful here. Make sure that you do NOT use Invert with Edge Detect (this just confuses the system and is not necessary).", "UMask": "0x20", "Unit": "iMC" }, { "BriefDescription": "CKE_ON_CYCLES by Rank; DIMM ID", "EventCode": "0x83", "EventName": "UNC_M_POWER_CKE_CYCLES.RANK6", "PerPkg": "1", "PublicDescription": "Number of cycles spent in CKE ON mode. The filter allows you to select a rank to monitor. If multiple ranks are in CKE ON mode at one time, the counter will ONLY increment by one rather than doing accumulation. Multiple counters will need to be used to track multiple ranks simultaneously. There is no distinction between the different CKE modes (APD, PPDS, PPDF). This can be determined based on the system programming. These events should commonly be used with Invert to get the number of cycles in power saving mode. Edge Detect is also useful here. Make sure that you do NOT use Invert with Edge Detect (this just confuses the system and is not necessary).", "UMask": "0x40", "Unit": "iMC" }, { "BriefDescription": "CKE_ON_CYCLES by Rank; DIMM ID", "EventCode": "0x83", "EventName": "UNC_M_POWER_CKE_CYCLES.RANK7", "PerPkg": "1", "PublicDescription": "Number of cycles spent in CKE ON mode. The filter allows you to select a rank to monitor. If multiple ranks are in CKE ON mode at one time, the counter will ONLY increment by one rather than doing accumulation. Multiple counters will need to be used to track multiple ranks simultaneously. There is no distinction between the different CKE modes (APD, PPDS, PPDF). This can be determined based on the system programming. These events should commonly be used with Invert to get the number of cycles in power saving mode. Edge Detect is also useful here. Make sure that you do NOT use Invert with Edge Detect (this just confuses the system and is not necessary).", "UMask": "0x80", "Unit": "iMC" }, { "BriefDescription": "Critical Throttle Cycles", "EventCode": "0x86", "EventName": "UNC_M_POWER_CRITICAL_THROTTLE_CYCLES", "PerPkg": "1", "PublicDescription": "Counts the number of cycles when the iMC is in critical thermal throttling. When this happens, all traffic is blocked. This should be rare unless something bad is going on in the platform. There is no filtering by rank for this event.", "Unit": "iMC" }, { "BriefDescription": "Clock-Enabled Self-Refresh", "EventCode": "0x43", "EventName": "UNC_M_POWER_SELF_REFRESH", "PerPkg": "1", "PublicDescription": "Counts the number of cycles when the iMC is in self-refresh and the iMC still has a clock. This happens in some package C-states. For example, the PCU may ask the iMC to enter self-refresh even though some of the cores are still processing. One use of this is for Monroe technology. Self-refresh is required during package C3 and C6, but there is no clock in the iMC at this time, so it is not possible to count these cases.", "Unit": "iMC" }, { "BriefDescription": "Throttle Cycles for Rank 0; DIMM ID", "EventCode": "0x41", "EventName": "UNC_M_POWER_THROTTLE_CYCLES.RANK0", "PerPkg": "1", "PublicDescription": "Counts the number of cycles while the iMC is being throttled by either thermal constraints or by the PCU throttling. It is not possible to distinguish between the two. This can be filtered by rank. If multiple ranks are selected and are being throttled at the same time, the counter will only increment by 1.", "UMask": "0x1", "Unit": "iMC" }, { "BriefDescription": "Throttle Cycles for Rank 0; DIMM ID", "EventCode": "0x41", "EventName": "UNC_M_POWER_THROTTLE_CYCLES.RANK1", "PerPkg": "1", "PublicDescription": "Counts the number of cycles while the iMC is being throttled by either thermal constraints or by the PCU throttling. It is not possible to distinguish between the two. This can be filtered by rank. If multiple ranks are selected and are being throttled at the same time, the counter will only increment by 1.", "UMask": "0x2", "Unit": "iMC" }, { "BriefDescription": "Throttle Cycles for Rank 0; DIMM ID", "EventCode": "0x41", "EventName": "UNC_M_POWER_THROTTLE_CYCLES.RANK2", "PerPkg": "1", "PublicDescription": "Counts the number of cycles while the iMC is being throttled by either thermal constraints or by the PCU throttling. It is not possible to distinguish between the two. This can be filtered by rank. If multiple ranks are selected and are being throttled at the same time, the counter will only increment by 1.", "UMask": "0x4", "Unit": "iMC" }, { "BriefDescription": "Throttle Cycles for Rank 0; DIMM ID", "EventCode": "0x41", "EventName": "UNC_M_POWER_THROTTLE_CYCLES.RANK3", "PerPkg": "1", "PublicDescription": "Counts the number of cycles while the iMC is being throttled by either thermal constraints or by the PCU throttling. It is not possible to distinguish between the two. This can be filtered by rank. If multiple ranks are selected and are being throttled at the same time, the counter will only increment by 1.", "UMask": "0x8", "Unit": "iMC" }, { "BriefDescription": "Throttle Cycles for Rank 0; DIMM ID", "EventCode": "0x41", "EventName": "UNC_M_POWER_THROTTLE_CYCLES.RANK4", "PerPkg": "1", "PublicDescription": "Counts the number of cycles while the iMC is being throttled by either thermal constraints or by the PCU throttling. It is not possible to distinguish between the two. This can be filtered by rank. If multiple ranks are selected and are being throttled at the same time, the counter will only increment by 1.", "UMask": "0x10", "Unit": "iMC" }, { "BriefDescription": "Throttle Cycles for Rank 0; DIMM ID", "EventCode": "0x41", "EventName": "UNC_M_POWER_THROTTLE_CYCLES.RANK5", "PerPkg": "1", "PublicDescription": "Counts the number of cycles while the iMC is being throttled by either thermal constraints or by the PCU throttling. It is not possible to distinguish between the two. This can be filtered by rank. If multiple ranks are selected and are being throttled at the same time, the counter will only increment by 1.", "UMask": "0x20", "Unit": "iMC" }, { "BriefDescription": "Throttle Cycles for Rank 0; DIMM ID", "EventCode": "0x41", "EventName": "UNC_M_POWER_THROTTLE_CYCLES.RANK6", "PerPkg": "1", "PublicDescription": "Counts the number of cycles while the iMC is being throttled by either thermal constraints or by the PCU throttling. It is not possible to distinguish between the two. This can be filtered by rank. If multiple ranks are selected and are being throttled at the same time, the counter will only increment by 1.", "UMask": "0x40", "Unit": "iMC" }, { "BriefDescription": "Throttle Cycles for Rank 0; DIMM ID", "EventCode": "0x41", "EventName": "UNC_M_POWER_THROTTLE_CYCLES.RANK7", "PerPkg": "1", "PublicDescription": "Counts the number of cycles while the iMC is being throttled by either thermal constraints or by the PCU throttling. It is not possible to distinguish between the two. This can be filtered by rank. If multiple ranks are selected and are being throttled at the same time, the counter will only increment by 1.", "UMask": "0x80", "Unit": "iMC" }, { "BriefDescription": "Read Preemption Count; Read over Read Preemption", "EventCode": "0x8", "EventName": "UNC_M_PREEMPTION.RD_PREEMPT_RD", "PerPkg": "1", "PublicDescription": "Counts the number of times a read in the iMC preempts another read or write. Generally reads to an open page are issued ahead of requests to closed pages. This improves the page hit rate of the system. However, high priority requests can cause pages of active requests to be closed in order to get them out. This will reduce the latency of the high-priority request at the expense of lower bandwidth and increased overall average latency.", "UMask": "0x1", "Unit": "iMC" }, { "BriefDescription": "Read Preemption Count; Read over Write Preemption", "EventCode": "0x8", "EventName": "UNC_M_PREEMPTION.RD_PREEMPT_WR", "PerPkg": "1", "PublicDescription": "Counts the number of times a read in the iMC preempts another read or write. Generally reads to an open page are issued ahead of requests to closed pages. This improves the page hit rate of the system. However, high priority requests can cause pages of active requests to be closed in order to get them out. This will reduce the latency of the high-priority request at the expense of lower bandwidth and increased overall average latency.", "UMask": "0x2", "Unit": "iMC" }, { "BriefDescription": "DRAM Precharge commands.; Precharge due to timer expiration", "EventCode": "0x2", "EventName": "UNC_M_PRE_COUNT.PAGE_CLOSE", "PerPkg": "1", "PublicDescription": "Counts the number of DRAM Precharge commands sent on this channel.", "UMask": "0x2", "Unit": "iMC" }, { "BriefDescription": "DRAM Precharge commands.; Precharges due to page miss", "EventCode": "0x2", "EventName": "UNC_M_PRE_COUNT.PAGE_MISS", "PerPkg": "1", "PublicDescription": "Counts the number of DRAM Precharge commands sent on this channel.", "UMask": "0x1", "Unit": "iMC" }, { "BriefDescription": "Read Pending Queue Full Cycles", "EventCode": "0x12", "EventName": "UNC_M_RPQ_CYCLES_FULL", "PerPkg": "1", "PublicDescription": "Counts the number of cycles when the Read Pending Queue is full. When the RPQ is full, the HA will not be able to issue any additional read requests into the iMC. This count should be similar count in the HA which tracks the number of cycles that the HA has no RPQ credits, just somewhat smaller to account for the credit return overhead. We generally do not expect to see RPQ become full except for potentially during Write Major Mode or while running with slow DRAM. This event only tracks non-ISOC queue entries.", "Unit": "iMC" }, { "BriefDescription": "Read Pending Queue Not Empty", "EventCode": "0x11", "EventName": "UNC_M_RPQ_CYCLES_NE", "PerPkg": "1", "PublicDescription": "Counts the number of cycles that the Read Pending Queue is not empty. This can then be used to calculate the average occupancy (in conjunction with the Read Pending Queue Occupancy count). The RPQ is used to schedule reads out to the memory controller and to track the requests. Requests allocate into the RPQ soon after they enter the memory controller, and need credits for an entry in this buffer before being sent from the HA to the iMC. They deallocate after the CAS command has been issued to memory. This filter is to be used in conjunction with the occupancy filter so that one can correctly track the average occupancies for schedulable entries and scheduled requests.", "Unit": "iMC" }, { "BriefDescription": "Read Pending Queue Allocations", "EventCode": "0x10", "EventName": "UNC_M_RPQ_INSERTS", "PerPkg": "1", "PublicDescription": "Counts the number of allocations into the Read Pending Queue. This queue is used to schedule reads out to the memory controller and to track the requests. Requests allocate into the RPQ soon after they enter the memory controller, and need credits for an entry in this buffer before being sent from the HA to the iMC. They deallocate after the CAS command has been issued to memory. This includes both ISOCH and non-ISOCH requests.", "Unit": "iMC" }, { "BriefDescription": "Read Pending Queue Occupancy", "EventCode": "0x80", "EventName": "UNC_M_RPQ_OCCUPANCY", "PerPkg": "1", "PublicDescription": "Accumulates the occupancies of the Read Pending Queue each cycle. This can then be used to calculate both the average occupancy (in conjunction with the number of cycles not empty) and the average latency (in conjunction with the number of allocations). The RPQ is used to schedule reads out to the memory controller and to track the requests. Requests allocate into the RPQ soon after they enter the memory controller, and need credits for an entry in this buffer before being sent from the HA to the iMC. They deallocate after the CAS command has been issued to memory.", "Unit": "iMC" }, { "BriefDescription": "Write Pending Queue Full Cycles", "EventCode": "0x22", "EventName": "UNC_M_WPQ_CYCLES_FULL", "PerPkg": "1", "PublicDescription": "Counts the number of cycles when the Write Pending Queue is full. When the WPQ is full, the HA will not be able to issue any additional read requests into the iMC. This count should be similar count in the HA which tracks the number of cycles that the HA has no WPQ credits, just somewhat smaller to account for the credit return overhead.", "Unit": "iMC" }, { "BriefDescription": "Write Pending Queue Not Empty", "EventCode": "0x21", "EventName": "UNC_M_WPQ_CYCLES_NE", "PerPkg": "1", "PublicDescription": "Counts the number of cycles that the Write Pending Queue is not empty. This can then be used to calculate the average queue occupancy (in conjunction with the WPQ Occupancy Accumulation count). The WPQ is used to schedule write out to the memory controller and to track the writes. Requests allocate into the WPQ soon after they enter the memory controller, and need credits for an entry in this buffer before being sent from the HA to the iMC. They deallocate after being issued to DRAM. Write requests themselves are able to complete (from the perspective of the rest of the system) as soon they have 'posted' to the iMC. This is not to be confused with actually performing the write to DRAM. Therefore, the average latency for this queue is actually not useful for deconstruction intermediate write latencies.", "Unit": "iMC" }, { "BriefDescription": "Write Pending Queue Allocations", "EventCode": "0x20", "EventName": "UNC_M_WPQ_INSERTS", "PerPkg": "1", "PublicDescription": "Counts the number of allocations into the Write Pending Queue. This can then be used to calculate the average queuing latency (in conjunction with the WPQ occupancy count). The WPQ is used to schedule write out to the memory controller and to track the writes. Requests allocate into the WPQ soon after they enter the memory controller, and need credits for an entry in this buffer before being sent from the HA to the iMC. They deallocate after being issued to DRAM. Write requests themselves are able to complete (from the perspective of the rest of the system) as soon they have 'posted' to the iMC.", "Unit": "iMC" }, { "BriefDescription": "Write Pending Queue Occupancy", "EventCode": "0x81", "EventName": "UNC_M_WPQ_OCCUPANCY", "PerPkg": "1", "PublicDescription": "Accumulates the occupancies of the Write Pending Queue each cycle. This can then be used to calculate both the average queue occupancy (in conjunction with the number of cycles not empty) and the average latency (in conjunction with the number of allocations). The WPQ is used to schedule write out to the memory controller and to track the writes. Requests allocate into the WPQ soon after they enter the memory controller, and need credits for an entry in this buffer before being sent from the HA to the iMC. They deallocate after being issued to DRAM. Write requests themselves are able to complete (from the perspective of the rest of the system) as soon they have 'posted' to the iMC. This is not to be confused with actually performing the write to DRAM. Therefore, the average latency for this queue is actually not useful for deconstruction intermediate write latencies. So, we provide filtering based on if the request has posted or not. By using the 'not posted' filter, we can track how long writes spent in the iMC before completions were sent to the HA. The 'posted' filter, on the other hand, provides information about how much queueing is actually happening in the iMC for writes before they are actually issued to memory. High average occupancies will generally coincide with high write major mode counts.", "Unit": "iMC" }, { "BriefDescription": "Write Pending Queue CAM Match", "EventCode": "0x23", "EventName": "UNC_M_WPQ_READ_HIT", "PerPkg": "1", "PublicDescription": "Counts the number of times a request hits in the WPQ (write-pending queue). The iMC allows writes and reads to pass up other writes to different addresses. Before a read or a write is issued, it will first CAM the WPQ to see if there is a write pending to that address. When reads hit, they are able to directly pull their data from the WPQ instead of going to memory. Writes that hit will overwrite the existing data. Partial writes that hit will not need to do underfill reads and will simply update their relevant sections.", "Unit": "iMC" }, { "BriefDescription": "Write Pending Queue CAM Match", "EventCode": "0x24", "EventName": "UNC_M_WPQ_WRITE_HIT", "PerPkg": "1", "PublicDescription": "Counts the number of times a request hits in the WPQ (write-pending queue). The iMC allows writes and reads to pass up other writes to different addresses. Before a read or a write is issued, it will first CAM the WPQ to see if there is a write pending to that address. When reads hit, they are able to directly pull their data from the WPQ instead of going to memory. Writes that hit will overwrite the existing data. Partial writes that hit will not need to do underfill reads and will simply update their relevant sections.", "Unit": "iMC" } ] |