{"id":2688,"date":"2026-07-12T21:17:42","date_gmt":"2026-07-12T21:17:42","guid":{"rendered":"https:\/\/mvermeulen.org\/perf\/?p=2688"},"modified":"2026-07-12T21:17:44","modified_gmt":"2026-07-12T21:17:44","slug":"performance-measurements-on-four-systems","status":"publish","type":"post","link":"https:\/\/mvermeulen.org\/perf\/2026\/07\/12\/performance-measurements-on-four-systems\/","title":{"rendered":"Performance Measurements on four systems"},"content":{"rendered":"\n

With help of AI, I created a new “roofline” measurement that tries to write microbenchmarks to run system parameters. I have compared four CPU systems below. In some cases there might be issues with the microbenchmarks created that enable prefetching\/optimization but still a useful overall comparison<\/p>\n\n\n\n\n\n\n\t\n\n\t\n\t\n\t\n\t\n\t\n\t\n\t\n\t\n\t\n\t\n\t\n\t\n\t\n\t\n\t\n\t\n\t\n\t\n\t\n\t\n\t\n\t\n\t\n\t\n\t\n\t\n\t\n\t\n\t\n\t\n\t\n\t
Metric<\/th>Ryzen 9950 + RX 7600 XT<\/th>Ryzen AI 9 HP 370 + 890M<\/th>Ryzen AI 395 + Radeon 8060S<\/th>ARM CP 8180<\/th>\n<\/tr>\n<\/thead>\n
Cores<\/td>16<\/td>12<\/td>16<\/td>12<\/td>\n<\/tr>\n
Threads<\/td>32<\/td>24<\/td>32<\/td>12<\/td>\n<\/tr>\n
L1d<\/td>16x 48k = 768k<\/td>12x 48k = 576k<\/td>16x 48k = 768k<\/td>4x 32k, 8x 64k = 640k<\/td>\n<\/tr>\n
L1i<\/td>16x 32k = 512k<\/td>12x 32k = 384k<\/td>16x 32k = 512k<\/td>4x 32k, 8x 64k = 640k<\/td>\n<\/tr>\n
L2<\/td>16x 1024k = 16M<\/td>12x 1024k = 12M<\/td>16 x 1024k = 16M<\/td>8x 512k = 4M<\/td>\n<\/tr>\n
L3<\/td>2x 32M = 64M<\/td>16M + 8M = 24M<\/td>2 x 32M = 64M<\/td>1x 12M = 12M<\/td>\n<\/tr>\n
Memory<\/td>128 GB<\/td>96 GB<\/td>128 GB<\/td>64 GB<\/td>\n<\/tr>\n
Best CPU Core<\/td>Zen5 5756 MHz<\/td>Zen 5 5158 MHz<\/td>Zen5 5188 MHz<\/td>Cortex-A720 2600 MHz<\/td>\n<\/tr>\n
FP64 FLOPS<\/td>45.40 GFLOPS<\/td>41.00 GFLOPS<\/td>41.02 GFLOPS<\/td>10.34 GFLOPS<\/td>\n<\/tr>\n
FP32 FLOPS<\/td>45.67 GFLOPS<\/td>40.96 GFLOPS<\/td>40.99 GFLOPS<\/td>20.73 GFLOPS<\/td>\n<\/tr>\n
FP16 FLOPS<\/td>6.73 GFLOPS<\/td>6.12 GFLOPS<\/td>6.04 GFLOPS<\/td>10.36 GFLOPS<\/td>\n<\/tr>\n
L1 Triad Bandwidth<\/td>223.33 GB\/s<\/td>207.39 GB\/s<\/td>202.87 GB\/s<\/td>52.30 GB\/s<\/td>\n<\/tr>\n
L2 Triad Bandwidth<\/td>153.51 GB\/s<\/td>131.58 GB\/s<\/td>129.74 GB\/s<\/td>54.36 GB\/s<\/td>\n<\/tr>\n
L3 Triad Bandwidth<\/td>66.12 GB\/s<\/td>73.21 GB\/s<\/td>69.30 GB\/s<\/td>46.84 GB\/s<\/td>\n<\/tr>\n
DRAM Bandwidth<\/td>34.98 GB\/s<\/td>44.74 GB\/s<\/td>39.81 GB\/s<\/td>21.60 GB\/s<\/td>\n<\/tr>\n
L1 Latency<\/td>0.698 ns<\/td>0.781 ns<\/td>0.832 ns<\/td>1.544 ns<\/td>\n<\/tr>\n
L2 Latency<\/td>8.461 ns<\/td>4.140 ns<\/td>6.937 ns<\/td>3.731 ns<\/td>\n<\/tr>\n
L3 Latency<\/td>11.971 ns<\/td>11.543 ns<\/td>13.326 ns<\/td>6.143 ns<\/td>\n<\/tr>\n
DRAM Latency<\/td>81.349 ns<\/td>100.16 ns<\/td>81.067 ns<\/td>156.623 ns<\/td>\n<\/tr>\n
Core-Core Latency hyperthread<\/td>79.86 ns<\/td>26.23 ns<\/td>106.63 ns<\/td>n\/a<\/td>\n<\/tr>\n
Core-Core Latency Perf Cores<\/td>20.35 ns<\/td>35.42 ns<\/td>32.19 ns<\/td>157.59 ns<\/td>\n<\/tr>\n
Core-Core Latency Cross-CCD<\/td>50.01 ns<\/td>198.06 ns<\/td>70.96 ns<\/td>166.59 ns<\/td>\n<\/tr>\n
Best GPU Core<\/td>RX 7600 XT – HIP<\/td>890M – HIP<\/td>8060S – HIP<\/td>n\/a<\/td>\n<\/tr>\n
FP16 FLOPS<\/td>6465.80 GFLOPS<\/td>4307.66 GFLOPS<\/td>11290.44 GFLOPS<\/td>n\/a<\/td>\n<\/tr>\n
FP32 FLOPS<\/td>5258.37 GFLOPS<\/td>2792.52 GFLOPS<\/td>7302.15 GFLOPS<\/td>n\/a<\/td>\n<\/tr>\n
FP64 FLOPS<\/td>303.55 GFLOPS<\/td>184.57 GFLOPS<\/td>460.54 GFLOPS<\/td>n\/a<\/td>\n<\/tr>\n
Mem Triad<\/td>303.55 GFLOPS<\/td>79.69 GB\/s<\/td>231.22 GB\/s<\/td>n\/a<\/td>\n<\/tr>\n
H2D Copy (Pinned)<\/td>14.17 GB\/s<\/td>39.26 GB\/s<\/td>82.82 GB\/s<\/td>n\/a<\/td>\n<\/tr>\n
D2H Copy (Pinned)<\/td>14.29 GB\/s<\/td>38.10 GB\/s<\/td>64.62 GB\/s<\/td>n\/a<\/td>\n<\/tr>\n
Kernel Launch Latency (sync)<\/td>17.10 us<\/td>6.96 us<\/td>7.12 us<\/td>n\/a<\/td>\n<\/tr>\n
Kernel Launch Latency (async)<\/td>2.60 us<\/td>1.51 us<\/td>1.73 us<\/td>n\/a<\/td>\n<\/tr>\n
Event Sync Latency<\/td>15.07 us<\/td>4.99 us<\/td>5.03 us<\/td>n\/a<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n","protected":false},"excerpt":{"rendered":"

With help of AI, I created a new “roofline” measurement that tries to write microbenchmarks to run system parameters. I have compared four CPU systems below. In some cases there might be issues with the microbenchmarks created that enable prefetching\/optimization …<\/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":[9],"tags":[43,23,40,36],"class_list":["post-2688","post","type-post","status-publish","format-standard","hentry","category-hardware","tag-aarch64","tag-benchmarks","tag-ryzen-ai-9-hx-370","tag-zen5"],"_links":{"self":[{"href":"https:\/\/mvermeulen.org\/perf\/wp-json\/wp\/v2\/posts\/2688","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=2688"}],"version-history":[{"count":1,"href":"https:\/\/mvermeulen.org\/perf\/wp-json\/wp\/v2\/posts\/2688\/revisions"}],"predecessor-version":[{"id":2689,"href":"https:\/\/mvermeulen.org\/perf\/wp-json\/wp\/v2\/posts\/2688\/revisions\/2689"}],"wp:attachment":[{"href":"https:\/\/mvermeulen.org\/perf\/wp-json\/wp\/v2\/media?parent=2688"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/mvermeulen.org\/perf\/wp-json\/wp\/v2\/categories?post=2688"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/mvermeulen.org\/perf\/wp-json\/wp\/v2\/tags?post=2688"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}