\n(4x Zen 5 and 8x Zen 5c)<\/td>
\n
\nDDR5 – 5600
\n
\n2 Memory Channels<\/td>
\n
\nDDR5 – 7500
\n
\n2 Memory Channels<\/td>
\nScale: 70300 MB\/s
\nAdd: 73600 MB\/s
\nTriad: 73000 MB\/s<\/td>
\nScale: 86626 MS\/s
\nAdd: 88192 MB\/s
\nTriad: 87655 MB\/s<\/td>
\n
\nL2 – 1 MB, 8-way, 14 clocks
\n
\nL3 – 16MB, 24 way, 47 clocks<\/td>
\n
\nL2 – 1 MB
\n
\nL3 – 24 MB<\/td>
\nL2 – 3 ns
\nL3 – 8 ns<\/td>
\nL2 – 3ns
\nL3 – 8 ns<\/td>
\n
\n12 cores
\n
\n2700 MHz<\/td>
\n
\n16 cores
\n
\n2900 MHz<\/td>
Following are the results from likwid-topology. This is a hybrid core with four Zen5 cores and eight Zen5c cores. I believe the first four cores are Zen5 and the remaining eight are Zen5c. <\/p>\n\n\n\n
--------------------------------------------------------------------------------\nCPU name:\tAMD Ryzen AI 9 HX 370 w\/ Radeon 890M \nCPU type:\tnil\nCPU stepping:\t0\n********************************************************************************\nHardware Thread Topology\n********************************************************************************\nSockets:\t\t1\nCores per socket:\t12\nThreads per core:\t2\n--------------------------------------------------------------------------------\nHWThread Thread Core Die Socket Available\n0 0 0 0 0 * \n1 0 1 0 0 * \n2 0 2 0 0 * \n3 0 3 0 0 * \n4 0 4 0 0 * \n5 0 5 0 0 * \n6 0 6 0 0 * \n7 0 7 0 0 * \n8 0 8 0 0 * \n9 0 9 0 0 * \n10 0 10 0 0 * \n11 0 11 0 0 * \n12 1 0 0 0 * \n13 1 1 0 0 * \n14 1 2 0 0 * \n15 1 3 0 0 * \n16 1 4 0 0 * \n17 1 5 0 0 * \n18 1 6 0 0 * \n19 1 7 0 0 * \n20 1 8 0 0 * \n21 1 9 0 0 * \n22 1 10 0 0 * \n23 1 11 0 0 * \n--------------------------------------------------------------------------------\nSocket 0:\t\t( 0 12 1 13 2 14 3 15 4 16 5 17 6 18 7 19 8 20 9 21 10 22 11 23 )\n--------------------------------------------------------------------------------\n********************************************************************************\nCache Topology\n********************************************************************************\nLevel:\t\t\t1\nSize:\t\t\t48 kB\nCache groups:\t\t( 0 12 ) ( 1 13 ) ( 2 14 ) ( 3 15 ) ( 4 16 ) ( 5 17 ) ( 6 18 ) ( 7 19 ) ( 8 20 ) ( 9 21 ) ( 10 22 ) ( 11 23 )\n--------------------------------------------------------------------------------\nLevel:\t\t\t2\nSize:\t\t\t1 MB\nCache groups:\t\t( 0 12 ) ( 1 13 ) ( 2 14 ) ( 3 15 ) ( 4 16 ) ( 5 17 ) ( 6 18 ) ( 7 19 ) ( 8 20 ) ( 9 21 ) ( 10 22 ) ( 11 23 )\n--------------------------------------------------------------------------------\nLevel:\t\t\t3\nSize:\t\t\t16 MB\nCache groups:\t\t( 0 12 1 13 2 14 3 15 ) ( 4 16 5 17 6 18 7 19 ) ( 8 20 9 21 10 22 11 23 )\n--------------------------------------------------------------------------------\n********************************************************************************\nNUMA Topology\n********************************************************************************\nNUMA domains:\t\t1\n--------------------------------------------------------------------------------\nDomain:\t\t\t0\nProcessors:\t\t( 0 12 1 13 2 14 3 15 4 16 5 17 6 18 7 19 8 20 9 21 10 22 11 23 )\nDistances:\t\t10\nFree memory:\t\t22667.5 MB\nTotal memory:\t\t27574.2 MB\n--------------------------------------------------------------------------------\n<\/code><\/pre>\n\n\n\nThe L3 cache amount may be incorrect as specifications suggest 24 MB of cache. Using lmbench suggests the L3 cache attached to first four cores is 16MB and the next groups have 8MB likely together even though topology above makes them separate.<\/p>\n\n\n\n
This hybrid SOC shows up in the following coremark scaling comparison as shown in the graph below. There are several different regions<\/p>\n\n\n\n
\n- From 1 to 4 cores we compare Zen4 cores against Zen5 cores. The coremark value for 4 cores is ~12% ahead.<\/li>\n\n\n\n
- From 5 to 8 cores, we now have Zen5 + Zen5C cores against Zen4 cores. The coremark value for 8 cores is ~7% behind<\/li>\n\n\n\n
- From 9 to 12 cores, we use all the cores on HX 370 and start using SMT for the 7840. The coremark value for 12 cores is 6% ahead<\/li>\n\n\n\n
- From 13 to 16 cores we go to using SMT for all the Zen5 cores and not-SMT for Zen5C cores. The 7840 moves to fully SMT. The coremark value for 16 cores is 11% ahead<\/li>\n\n\n\n
- From 17 to 24 cores, we go to adding SMT for Zen5C cores. The overall coremark using all cores (24 vs 16) is 21% ahead.<\/li>\n<\/ul>\n\n\n\n
This suggests for coremark and other workloads there will be different regions where combinations of SMT and Zen5 vs Zen5C cores will create interesting comparisons between the systems.<\/p>\n\n\n\n![\"\"](\"https:\/\/mvermeulen.org\/perf\/wp-content\/uploads\/sites\/7\/2024\/10\/coremark.png\")
<\/figure>\n\n\n\nThe tabular version of coremark including performance counters is shown below.<\/p>\n\n\n\n
\n\n\n\tCores<\/th> Coremark HX 370<\/th> Coremark 7840<\/th> Scaling HX 370<\/th> Scaling 7840<\/th> Retiring HX 370<\/th> Frontend HX 370<\/th> Backend HX 370<\/th> Speculation HX 370<\/th> SMT-contention HX 370<\/th> Retiring 7840<\/th> Frontend 7840<\/th> Backend 7840<\/th> Speculation 7840<\/th> SMT-contention 7840<\/th>\n<\/tr>\n<\/thead>\n \n\n\t1<\/td> 48245<\/td> 43881<\/td> 100%<\/td> 100%<\/td> 44.2%<\/td> 25.2%<\/td> 62.0%<\/td> 2.0%<\/td> 0.0%<\/td> 43.9%<\/td> 12.4%<\/td> 43.0%<\/td> 0.7%<\/td> 0.0%<\/td>\n<\/tr>\n \n\t2<\/td> 96106<\/td> 85758<\/td> 100%<\/td> 98%<\/td> 44.0%<\/td> 25.5%<\/td> 61.8%<\/td> 2.0%<\/td> 0.0%<\/td> 43.9%<\/td> 12.4%<\/td> 43.1%<\/td> 0.7%<\/td> 0.0%<\/td>\n<\/tr>\n \n\t3<\/td> 144147<\/td> 128841<\/td> 100%<\/td> 98%<\/td> 44.0%<\/td> 25.5%<\/td> 61.8%<\/td> 2.0%<\/td> 0.0%<\/td> 43.6%<\/td> 13.0%<\/td> 42.7%<\/td> 0.7%<\/td> 0.0%<\/td>\n<\/tr>\n \n\t4<\/td> 192537<\/td> 171061<\/td> 100%<\/td> 97%<\/td> 44.1%<\/td> 25.4%<\/td> 61.9%<\/td> 2.0%<\/td> 0.0%<\/td> 43.9%<\/td> 12.3%<\/td> 43.1%<\/td> 0.7%<\/td> 0.0%<\/td>\n<\/tr>\n \n\t5<\/td> 214223<\/td> 210368<\/td> 89%<\/td> 96%<\/td> 44.0%<\/td> 25.5%<\/td> 61.8%<\/td> 2.0%<\/td> 0.0%<\/td> 43.9%<\/td> 12.3%<\/td> 43.1%<\/td> 0.7%<\/td> 0.0%<\/td>\n<\/tr>\n \n\t6<\/td> 227532<\/td> 251705<\/td> 79%<\/td> 96%<\/td> 44.0%<\/td> 25.4%<\/td> 61.9%<\/td> 2.0%<\/td> 0.0%<\/td> 43.2%<\/td> 12.9%<\/td> 43.2%<\/td> 0.7%<\/td> 0.0%<\/td>\n<\/tr>\n \n\t7<\/td> 260811<\/td> 281369<\/td> 77%<\/td> 92%<\/td> 44.0%<\/td> 25.7%<\/td> 61.7%<\/td> 2.0%<\/td> 0.0%<\/td> 43.3%<\/td> 12.2%<\/td> 43.7%<\/td> 0.7%<\/td> 0.0%<\/td>\n<\/tr>\n \n\t8<\/td> 297002<\/td> 319098<\/td> 77%<\/td> 91%<\/td> 44.1%<\/td> 25.3%<\/td> 61.9%<\/td> 2.0%<\/td> 0.0%<\/td> 42.7%<\/td> 12.8%<\/td> 43.8%<\/td> 0.7%<\/td> 0.0%<\/td>\n<\/tr>\n \n\t9<\/td> 325417<\/td> 334602<\/td> 75%<\/td> 85%<\/td> 44.1%<\/td> 25.3%<\/td> 62.0%<\/td> 2.0%<\/td> 0.0%<\/td> 40.2%<\/td> 15.9%<\/td> 36.3%<\/td> 0.6%<\/td> 7.1%<\/td>\n<\/tr>\n \n\t10<\/td> 347636<\/td> 347246<\/td> 72%<\/td> 79%<\/td> 44.0%<\/td> 25.3%<\/td> 61.9%<\/td> 2.0%<\/td> 0.0%<\/td> 38.4%<\/td> 17.8%<\/td> 30.2%<\/td> 0.5%<\/td> 13.1%<\/td>\n<\/tr>\n \n\t11<\/td> 380587<\/td> 359402<\/td> 72%<\/td> 74%<\/td> 44.0%<\/td> 25.5%<\/td> 61.8%<\/td> 2.0%<\/td> 0.0%<\/td> 36.9%<\/td> 19.6%<\/td> 25.3%<\/td> 0.5%<\/td> 17.8%<\/td>\n<\/tr>\n \n\t12<\/td> 413575<\/td> 363288<\/td> 71%<\/td> 69%<\/td> 44.0%<\/td> 25.4%<\/td> 61.9%<\/td> 2.0%<\/td> 0.0%<\/td> 35.5%<\/td> 21.1%<\/td> 21.6%<\/td> 0.4%<\/td> 21.3%<\/td>\n<\/tr>\n \n\t13<\/td> 426123<\/td> 362144<\/td> 68%<\/td> 63%<\/td> 42.1%<\/td> 28.2%<\/td> 52.9%<\/td> 1.8%<\/td> 8.3%<\/td> 34.4%<\/td> 22.4%<\/td> 18.5%<\/td> 0.4%<\/td> 24.3%<\/td>\n<\/tr>\n \n\t14<\/td> 446379<\/td> 377767<\/td> 66%<\/td> 61%<\/td> 40.5%<\/td> 30.6%<\/td> 45.6%<\/td> 1.6%<\/td> 15.1%<\/td> 33.1%<\/td> 24.4%<\/td> 15.2%<\/td> 0.4%<\/td> 26.9%<\/td>\n<\/tr>\n \n\t15<\/td> 452134<\/td> 397145<\/td> 62%<\/td> 60%<\/td> 39.5%<\/td> 32.2%<\/td> 40.6%<\/td> 1.4%<\/td> 19.7%<\/td> 32.2%<\/td> 25.3%<\/td> 12.0%<\/td> 0.3%<\/td> 30.2%<\/td>\n<\/tr>\n \n\t16<\/td> 464431<\/td> 418462<\/td> 60%<\/td> 60%<\/td> 38.3%<\/td> 33.7%<\/td> 35.8%<\/td> 1.3%<\/td> 24.2%<\/td> 31.1%<\/td> 26.0%<\/td> 9.5%<\/td> 0.3%<\/td> 33.1%<\/td>\n<\/tr>\n \n\t17<\/td> 476416<\/td> <\/td> 58%<\/td> <\/td> 37.9%<\/td> 34.4%<\/td> 33.5%<\/td> 1.2%<\/td> 26.3%<\/td> <\/td> <\/td> <\/td> <\/td> <\/td>\n<\/tr>\n \n\t18<\/td> 489001<\/td> <\/td> 56%<\/td> <\/td> 37.2%<\/td> 35.0%<\/td> 31.2%<\/td> 1.2%<\/td> 28.7%<\/td> <\/td> <\/td> <\/td> <\/td> <\/td>\n<\/tr>\n \n\t19<\/td> 484655<\/td> <\/td> 53%<\/td> <\/td> 36.6%<\/td> 35.4%<\/td> 29.2%<\/td> 1.1%<\/td> 30.9%<\/td> <\/td> <\/td> <\/td> <\/td> <\/td>\n<\/tr>\n \n\t20<\/td> 495826<\/td> <\/td> 51%<\/td> <\/td> 36.5%<\/td> 36.5%<\/td> 26.3%<\/td> 1.0%<\/td> 33.1%<\/td> <\/td> <\/td> <\/td> <\/td> <\/td>\n<\/tr>\n \n\t21<\/td> 501457<\/td> <\/td> 49%<\/td> <\/td> 35.7%<\/td> 37.3%<\/td> 23.9%<\/td> 1.0%<\/td> 35.5%<\/td> <\/td> <\/td> <\/td> <\/td> <\/td>\n<\/tr>\n \n\t22<\/td> 510946<\/td> <\/td> 48%<\/td> <\/td> 35.1%<\/td> 37.7%<\/td> 22.0%<\/td> 0.9%<\/td> 37.6%<\/td> <\/td> <\/td> <\/td> <\/td> <\/td>\n<\/tr>\n \n\t23<\/td> 544895<\/td> <\/td> 49%<\/td> <\/td> 34.7%<\/td> 38.5%<\/td> 19.5%<\/td> 0.8%<\/td> 39.8%<\/td> <\/td> <\/td> <\/td> <\/td> <\/td>\n<\/tr>\n \n\t24<\/td> 563477<\/td> <\/td> 49%<\/td> <\/td> 34.0%<\/td> 38.2%<\/td> 19.4%<\/td> 0.8%<\/td> 40.9%<\/td> <\/td> <\/td> <\/td> <\/td> <\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n\n\n\nI also measured stream and it looks ~15% faster than my 7840 system.<\/p>\n\n\n\n
-------------------------------------------------------------\nSTREAM version $Revision: 5.10 $\n-------------------------------------------------------------\nThis system uses 8 bytes per array element.\n-------------------------------------------------------------\nArray size = 100000000 (elements), Offset = 0 (elements)\nMemory per array = 762.9 MiB (= 0.7 GiB).\nTotal memory required = 2288.8 MiB (= 2.2 GiB).\nEach kernel will be executed 100 times.\n The *best* time for each kernel (excluding the first iteration)\n will be used to compute the reported bandwidth.\n-------------------------------------------------------------\nNumber of Threads requested = 2\nNumber of Threads counted = 2\n-------------------------------------------------------------\nYour clock granularity\/precision appears to be 1 microseconds.\nEach test below will take on the order of 31409 microseconds.\n (= 31409 clock ticks)\nIncrease the size of the arrays if this shows that\nyou are not getting at least 20 clock ticks per test.\n-------------------------------------------------------------\nWARNING -- The above is only a rough guideline.\nFor best results, please be sure you know the\nprecision of your system timer.\n-------------------------------------------------------------\nFunction Best Rate MB\/s Avg time Min time Max time\nCopy: 86725.2 0.018665 0.018449 0.021070\nScale: 86626.7 0.018713 0.018470 0.020643\nAdd: 88192.8 0.027540 0.027213 0.031095\nTriad: 87655.3 0.027729 0.027380 0.031028\n-------------------------------------------------------------\nSolution Validates: avg error less than 1.000000e-13 on all three arrays\n-------------------------------------------------------------\n<\/code><\/pre>\n\n\n\nHere<\/a> is a phoronix article comparing Ryzen AI 9 HX 370 with a variety of laptop systems. The overall geomean is ~10% but there is a wider variety between tests. Can be interesting to puzzle out why some of the differences. It is also likely that the power points used for the laptop comparisons in the phoronix article are less since I see lower scores e.g. coremark or different gaps than what I see with the same benchmark. So will need to puzzle out some of the SOC\/power choices.<\/p>\n\n\n\n<\/p>\n","protected":false},"excerpt":{"rendered":"
I have a new AMD performance machine for experiments. The processor is a Ryzen AI 9 HX 370 in a Beelink SER9 mini-PC. Following are some of the major parameters.in comparison with my Ryzen 7840HS comparison machine. Following are the …<\/span> Continue reading →<\/span><\/a><\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[11,9],"tags":[12,38,40,10,36],"class_list":["post-2573","post","type-post","status-publish","format-standard","hentry","category-experiment","category-hardware","tag-7840hs","tag-coremark","tag-ryzen-ai-9-hx-370","tag-stream","tag-zen5"],"_links":{"self":[{"href":"https:\/\/mvermeulen.org\/perf\/wp-json\/wp\/v2\/posts\/2573","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/mvermeulen.org\/perf\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/mvermeulen.org\/perf\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/mvermeulen.org\/perf\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/mvermeulen.org\/perf\/wp-json\/wp\/v2\/comments?post=2573"}],"version-history":[{"count":7,"href":"https:\/\/mvermeulen.org\/perf\/wp-json\/wp\/v2\/posts\/2573\/revisions"}],"predecessor-version":[{"id":2596,"href":"https:\/\/mvermeulen.org\/perf\/wp-json\/wp\/v2\/posts\/2573\/revisions\/2596"}],"wp:attachment":[{"href":"https:\/\/mvermeulen.org\/perf\/wp-json\/wp\/v2\/media?parent=2573"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/mvermeulen.org\/perf\/wp-json\/wp\/v2\/categories?post=2573"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/mvermeulen.org\/perf\/wp-json\/wp\/v2\/tags?post=2573"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}