[{"data":1,"prerenderedAt":2440},["ShallowReactive",2],{"blog-post-nl-\u002Fblog\u002Fcranking-out-faster-tokens-for-fewer-dollars-amd-mi300x-vs-nvidia-h200":3,"blog-posts-sidebar-nl":2017},{"id":4,"title":5,"body":6,"categories":1997,"date":2005,"description":2006,"extension":2007,"image":2008,"meta":2009,"navigation":2010,"originalUrl":2011,"path":2012,"seo":2013,"slug":2014,"stem":2015,"__hash__":2016},"blogNl\u002Fblog\u002Fcranking-out-faster-tokens-for-fewer-dollars-amd-mi300x-vs-nvidia-h200.md","Snellere tokens voor minder geld: AMD MI300X tegenover NVIDIA H200",{"type":7,"value":8,"toc":1985},"minimark",[9,16,23,32,35,38,44,47,83,86,89,97,102,109,112,115,118,121,123,129,132,139,142,144,150,153,172,175,184,187,206,213,633,639,659,666,668,675,678,988,993,998,1000,1008,1011,1018,1483,1488,1502,1507,1509,1516,1818,1823,1829,1842,1847,1849,1855,1862,1879,1882,1887,1892,1894,1900,1906,1926,1929,1931,1937,1957,1959,1965,1970,1973,1978],[10,11,12],"p",{},[13,14,15],"strong",{},"Qwen3-32B op Paiton + AMD MI300X tegenover NVIDIA H200",[17,18,20],"h3",{"id":19},"_1-introductie",[13,21,22],{},"1. Introductie",[24,25,26],"blockquote",{},[10,27,28],{},[29,30,31],"em",{},"“Terwijl we modellen trainen voor lokale klanten en kritieke bedrijfsprocessen automatiseren en stroomlijnen, vonden we ook tijd om ons Paiton-framework met Qwen3-32B tot het uiterste te drijven.”",[10,33,34],{},"In de competitieve wereld van LLM's krijgt nieuwe hardware zoals de NVIDIA H200 vaak de meeste aandacht. Onze AMD MI300X-oplossing, geoptimaliseerd met Paiton, biedt echter tegen een aanzienlijk lagere prijs het beste van twee werelden: vergelijkbare of betere prestaties en aantrekkelijke kosten per miljoen tokens.",[36,37],"hr",{},[17,39,41],{"id":40},"_2-wat-we-hebben-getest",[13,42,43],{},"2. Wat we hebben getest",[10,45,46],{},"We testten het pas uitgebrachte Qwen3-32B-model op:",[48,49,50,57,63,69],"ul",{},[51,52,53,56],"li",{},[13,54,55],{},"AMD MI300X"," met de oudere drivers 6.3.1, dus zelfs niet met de recentere versie 6.4",[51,58,59,62],{},[13,60,61],{},"NVIDIA H200"," met de nieuwste stuurprogramma's\u002Ftoolchains",[51,64,65,68],{},[13,66,67],{},"Paiton",": ons framework voor concurrency en kernelfusie, geïntegreerd met vLLM 0.8.4",[51,70,71,74,75,79,80],{},[13,72,73],{},"Benchmarking"," met ",[76,77,78],"code",{},"python3 benchmark_serving.py",", in verschillende configuraties, met en zonder het argument ",[76,81,82],{},"--sharegpt-output-len=256",[10,84,85],{},"Ter referentie testten we ook een niet-geoptimaliseerde Stock MI300X-opstelling. De echte blikvanger is echter Paiton op de MI300X, onze softwarelaag voor een hogere throughput.",[10,87,88],{},"Hieronder vindt u een korte samenvatting van onze typische opdrachten:",[10,90,91],{},[13,92,93,94],{},"Zonder ",[76,95,96],{},"--sharegpt-output-len",[24,98,99],{},[10,100,101],{},"python3 benchmark_serving.py --backend vllm --model Qwen\u002FQwen3-32B --dataset-name sharegpt --dataset-path .\u002FShareGPT_V3_unfiltered_cleaned_split.json --num-prompts 32 --random-range-ratio 1.0 --host 0.0.0.0 --port 8888 --percentile-metrics ttft,tpot,itl,e2e",[10,103,104],{},[13,105,106,107],{},"Met ",[76,108,82],{},[10,110,111],{},"python3 benchmark_serving.py --backend vllm --model Qwen\u002FQwen3-32B --dataset-name sharegpt --dataset-path .\u002FShareGPT_V3_unfiltered_cleaned_split.json --num-prompts 32 --random-range-ratio 1.0 --host 0.0.0.0 --port 8000 --percentile-metrics ttft,tpot,itl,e2el --sharegpt-output-len 256",[10,113,114],{},"We hebben batchgroottes 1, 2, 4, 8, 16, 32, 64, 128 getest in elk scenario en omgevingsvariabelen gebruikt:",[10,116,117],{},"HIP_VISIBLE_DEVICES=1 vllm serve -tp 1 --swap-space 16 --port 8888 --disable-log-requests Qwen\u002FQwen3-32B --num-scheduler-steps 10",[10,119,120],{},"De met Paiton geoptimaliseerde modelruns gebruiken naast deze flags ook gespecialiseerde concurrencyoptimalisaties en kernels.",[36,122],{},[17,124,126],{"id":125},"_3-belangrijkste-cijfers-opnieuw-beter-dan-de-h200",[13,127,128],{},"3. Belangrijkste cijfers: opnieuw beter dan de H200",[10,130,131],{},"Hoewel de H200 volop in de belangstelling staat, evenaart of overtreft Paiton + MI300X de H200 bij meerdere batchgroottes, en dat tegen lagere totale hardwarekosten:",[10,133,134,135,138],{},"Een systeem met 8 H200 GPU's tegenover een systeem met 8 MI300X GPU's: ",[13,136,137],{},"$40.000 besparing"," met AMD.",[10,140,141],{},"Dat is niet klein. Als u ook het aantal tokens over de volledige levensduur van het systeem meerekent, vallen de kosten per miljoen tokens nog gunstiger uit voor de MI300X.",[36,143],{},[17,145,147],{"id":146},"_4-gedetailleerde-prestatietabellen",[13,148,149],{},"4. Gedetailleerde prestatietabellen",[10,151,152],{},"We laten twee sets gegevens zien:",[154,155,156,164],"ol",{},[51,157,158,161,162],{},[13,159,160],{},"Zonder"," de flag ",[76,163,96],{},[51,165,166,169,170],{},[13,167,168],{},"Met"," ",[76,171,82],{},[10,173,174],{},"Voor elke reeks tonen we throughput (requests\u002Fs, outputtokens\u002Fs en totale tokens\u002Fs) en latency (TTFT, TPOT, ITL en E2E).",[176,177,179],"h4",{"id":178},"_41-zonder-sharegpt-output-len",[13,180,181,182],{},"4.1. Zonder ",[76,183,96],{},[10,185,186],{},"We hebben drie configuraties vergeleken:",[48,188,189,195,201],{},[51,190,191,194],{},[13,192,193],{},"H200"," (nieuwste stuurprogramma's en Torch-stack)",[51,196,197,200],{},[13,198,199],{},"Stock MI300X"," (oudere 6.3.1-drivers)",[51,202,203,205],{},[13,204,67],{}," (onze gespecialiseerde gelijktijdigheid + kernelfusie)",[10,207,208],{},[13,209,210,211],{},"4.1.A. Throughput zonder ",[76,212,96],{},[214,215,216,248],"table",{},[217,218,219],"thead",{},[220,221,222,228,233,238,243],"tr",{},[223,224,225],"th",{},[13,226,227],{},"Batch",[223,229,230],{},[13,231,232],{},"Configuratie",[223,234,235],{},[13,236,237],{},"Requests\u002Fs",[223,239,240],{},[13,241,242],{},"Out Tok\u002Fs",[223,244,245],{},[13,246,247],{},"Total Tok\u002Fs",[249,250,251,268,283,300,316,331,348,364,379,396,412,426,443,459,473,490,506,520,537,553,568,585,601,616],"tbody",{},[220,252,253,257,259,262,265],{},[254,255,256],"td",{},"1",[254,258,193],{},[254,260,261],{},"0,33",[254,263,264],{},"39,48",[254,266,267],{},"43,46",[220,269,270,272,275,277,280],{},[254,271,256],{},[254,273,274],{},"MI300x (Stock)",[254,276,261],{},[254,278,279],{},"39,26",[254,281,282],{},"43,22",[220,284,285,287,289,292,295],{},[254,286,256],{},[254,288,67],{},[254,290,291],{},"0,43",[254,293,294],{},"51,14",[254,296,297],{},[13,298,299],{},"56,30",[220,301,302,305,307,310,313],{},[254,303,304],{},"2",[254,306,193],{},[254,308,309],{},"0,11",[254,311,312],{},"48,24",[254,314,315],{},"50,19",[220,317,318,320,322,325,328],{},[254,319,304],{},[254,321,274],{},[254,323,324],{},"0,10",[254,326,327],{},"45,89",[254,329,330],{},"47,75",[220,332,333,335,337,340,343],{},[254,334,304],{},[254,336,67],{},[254,338,339],{},"0,13",[254,341,342],{},"59,04",[254,344,345],{},[13,346,347],{},"61,43",[220,349,350,353,355,358,361],{},[254,351,352],{},"4",[254,354,193],{},[254,356,357],{},"0,21",[254,359,360],{},"67,46",[254,362,363],{},"71,11",[220,365,366,368,370,373,376],{},[254,367,352],{},[254,369,274],{},[254,371,372],{},"0,20",[254,374,375],{},"67,23",[254,377,378],{},"70,86",[220,380,381,383,385,388,391],{},[254,382,352],{},[254,384,67],{},[254,386,387],{},"0,26",[254,389,390],{},"86,65",[254,392,393],{},[13,394,395],{},"91,34",[220,397,398,401,403,406,409],{},[254,399,400],{},"8",[254,402,193],{},[254,404,405],{},"0,40",[254,407,408],{},"109,04",[254,410,411],{},"166,55",[220,413,414,416,418,420,423],{},[254,415,400],{},[254,417,274],{},[254,419,405],{},[254,421,422],{},"108,94",[254,424,425],{},"166,39",[220,427,428,430,432,435,438],{},[254,429,400],{},[254,431,67],{},[254,433,434],{},"0,52",[254,436,437],{},"141,50",[254,439,440],{},[13,441,442],{},"216,13",[220,444,445,448,450,453,456],{},[254,446,447],{},"16",[254,449,193],{},[254,451,452],{},"0,78",[254,454,455],{},"178,20",[254,457,458],{},"333,99",[220,460,461,463,465,467,470],{},[254,462,447],{},[254,464,274],{},[254,466,452],{},[254,468,469],{},"178,48",[254,471,472],{},"334,51",[220,474,475,477,479,482,485],{},[254,476,447],{},[254,478,67],{},[254,480,481],{},"1,00",[254,483,484],{},"229,26",[254,486,487],{},[13,488,489],{},"429,68",[220,491,492,495,497,500,503],{},[254,493,494],{},"32",[254,496,193],{},[254,498,499],{},"1,47",[254,501,502],{},"329,48",[254,504,505],{},"677,57",[220,507,508,510,512,514,517],{},[254,509,494],{},[254,511,274],{},[254,513,499],{},[254,515,516],{},"329,93",[254,518,519],{},"679,47",[220,521,522,524,526,529,532],{},[254,523,494],{},[254,525,67],{},[254,527,528],{},"1,87",[254,530,531],{},"417,91",[254,533,534],{},[13,535,536],{},"860,65",[220,538,539,542,544,547,550],{},[254,540,541],{},"64",[254,543,193],{},[254,545,546],{},"2,69",[254,548,549],{},"564,33",[254,551,552],{},"1226,38",[220,554,555,557,559,562,565],{},[254,556,541],{},[254,558,274],{},[254,560,561],{},"2,53",[254,563,564],{},"535,74",[254,566,567],{},"1157,55",[220,569,570,572,574,577,580],{},[254,571,541],{},[254,573,67],{},[254,575,576],{},"3,30",[254,578,579],{},"699,03",[254,581,582],{},[13,583,584],{},"1511,26",[220,586,587,590,592,595,598],{},[254,588,589],{},"128",[254,591,193],{},[254,593,594],{},"4,68",[254,596,597],{},"1021,46",[254,599,600],{},"2107,07",[220,602,603,605,607,610,613],{},[254,604,589],{},[254,606,274],{},[254,608,609],{},"4,57",[254,611,612],{},"995,06",[254,614,615],{},"2056,82",[220,617,618,620,622,625,628],{},[254,619,589],{},[254,621,67],{},[254,623,624],{},"5,33",[254,626,627],{},"1163,39",[254,629,630],{},[13,631,632],{},"2401,25",[10,634,635,638],{},[13,636,637],{},"Waarnemingen"," (geen sharegpt-output-len):",[48,640,641,651,656],{},[51,642,643,646,647,650],{},[13,644,645],{},"Paiton + MI300X"," leidt bij ",[13,648,649],{},"alle"," batchgroottes voor de totale throughput, zowel in requests\u002Fs als total tokens\u002Fs.",[51,652,653,655],{},[13,654,193],{}," en Stock MI300X liggen tot ongeveer batchgrootte 16 dicht bij elkaar. Daarna lopen de resultaten wat uiteen, terwijl Paiton een duidelijke sprong voorwaarts maakt.",[51,657,658],{},"Zelfs bij de grootste batchgrootte (128) ligt Paiton 15-20% voor op H200 in totaal aantal tokens\u002Fs.",[10,660,661],{},[662,663],"img",{"alt":664,"src":665},"","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Frequests_no_cap.jpg",[36,667],{},[10,669,670],{},[13,671,672,673],{},"4.1.B. Latency zonder ",[76,674,96],{},[10,676,677],{},"Hieronder tonen we de gemiddelden voor TTFT (Time-to-First-Token), TPOT (Time per Output Token), ITL (Inter-Token Latency) en E2E (End-to-End). Voor de leesbaarheid laten we de mediaan en p99 weg.",[214,679,680,712],{},[217,681,682],{},[220,683,684,688,692,697,702,707],{},[223,685,686],{},[13,687,227],{},[223,689,690],{},[13,691,232],{},[223,693,694],{},[13,695,696],{},"Gemiddelde TTFT (ms)",[223,698,699],{},[13,700,701],{},"Gemiddelde TPOT (ms)",[223,703,704],{},[13,705,706],{},"Gemiddelde ITL (ms)",[223,708,709],{},[13,710,711],{},"Gemiddelde E2E (ms)",[249,713,714,730,748,777,795,813,843,861,878,907,925,943,973],{},[220,715,716,718,720,722,725,727],{},[254,717,256],{},[254,719,193],{},[254,721,363],{},[254,723,724],{},"24,94",[254,726,724],{},[254,728,729],{},"3013,7",[220,731,732,734,737,740,743,745],{},[254,733,256],{},[254,735,736],{},"MI300x Stock",[254,738,739],{},"96,59",[254,741,742],{},"24,86",[254,744,742],{},[254,746,747],{},"3030,07",[220,749,750,754,758,763,768,772],{},[254,751,752],{},[13,753,256],{},[254,755,756],{},[13,757,67],{},[254,759,760],{},[13,761,762],{},"61,42",[254,764,765],{},[13,766,767],{},"19,19",[254,769,770],{},[13,771,767],{},[254,773,774],{},[13,775,776],{},"2325,87",[220,778,779,781,783,786,789,792],{},[254,780,304],{},[254,782,193],{},[254,784,785],{},"70,30",[254,787,788],{},"24,15",[254,790,791],{},"23,95",[254,793,794],{},"10690,70",[220,796,797,799,801,804,807,810],{},[254,798,304],{},[254,800,736],{},[254,802,803],{},"95,40",[254,805,806],{},"25,26",[254,808,809],{},"25,12",[254,811,812],{},"11235,03",[220,814,815,819,823,828,833,838],{},[254,816,817],{},[13,818,304],{},[254,820,821],{},[13,822,67],{},[254,824,825],{},[13,826,827],{},"66,48",[254,829,830],{},[13,831,832],{},"19,38",[254,834,835],{},[13,836,837],{},"19,46",[254,839,840],{},[13,841,842],{},"8697,51",[220,844,845,847,849,852,855,858],{},[254,846,352],{},[254,848,193],{},[254,850,851],{},"71,00",[254,853,854],{},"25,65",[254,856,857],{},"25,46",[254,859,860],{},"8409,96",[220,862,863,865,867,870,872,875],{},[254,864,352],{},[254,866,736],{},[254,868,869],{},"94,67",[254,871,854],{},[254,873,874],{},"25,49",[254,876,877],{},"8442,12",[220,879,880,884,888,892,897,902],{},[254,881,882],{},[13,883,352],{},[254,885,886],{},[13,887,67],{},[254,889,890],{},[13,891,378],{},[254,893,894],{},[13,895,896],{},"19,69",[254,898,899],{},[13,900,901],{},"19,65",[254,903,904],{},[13,905,906],{},"6506,55",[220,908,909,911,913,916,919,922],{},[254,910,400],{},[254,912,193],{},[254,914,915],{},"190,87",[254,917,918],{},"26,25",[254,920,921],{},"26,00",[254,923,924],{},"7283,43",[220,926,927,929,931,934,937,940],{},[254,928,400],{},[254,930,736],{},[254,932,933],{},"244,19",[254,935,936],{},"26,09",[254,938,939],{},"25,93",[254,941,942],{},"7317,94",[220,944,945,949,953,958,963,968],{},[254,946,947],{},[13,948,400],{},[254,950,951],{},[13,952,67],{},[254,954,955],{},[13,956,957],{},"205,81",[254,959,960],{},[13,961,962],{},"19,91",[254,964,965],{},[13,966,967],{},"19,90",[254,969,970],{},[13,971,972],{},"5634,23",[220,974,975,978,980,982,984,986],{},[254,976,977],{},"…",[254,979,977],{},[254,981,977],{},[254,983,977],{},[254,985,977],{},[254,987,977],{},[10,989,990],{},[29,991,992],{},"(Tabel ingekort voor leesbaarheid, maar de trend is consistent: Paiton reduceert de tijd tot het eerste token bij kleine batchgroottes en kan de E2E-latentie met een betekenisvolle marge verkorten bij batchgroottes uit het middensegment.)",[10,994,995],{},[662,996],{"alt":664,"src":997},"\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fe2e_no_cap.jpg",[36,999],{},[176,1001,1003],{"id":1002},"_42-met-sharegpt-output-len256",[13,1004,1005,1006],{},"4.2. Met ",[76,1007,82],{},[10,1009,1010],{},"Vervolgens bekijken we het scenario waarin de outputlengte op 256 tokens is vastgelegd. Dat helpt vaak bij concurrency en scheduling, omdat het model geen rekening meer hoeft te houden met variabele of onzekere antwoordlengtes.",[10,1012,1013],{},[13,1014,1015,1016],{},"4.2.A. Throughput met ",[76,1017,82],{},[214,1019,1020,1044],{},[217,1021,1022],{},[220,1023,1024,1028,1032,1036,1040],{},[223,1025,1026],{},[13,1027,227],{},[223,1029,1030],{},[13,1031,232],{},[223,1033,1034],{},[13,1035,237],{},[223,1037,1038],{},[13,1039,242],{},[223,1041,1042],{},[13,1043,247],{},[249,1045,1046,1061,1075,1100,1115,1129,1154,1169,1183,1208,1223,1238,1263,1278,1293,1318,1333,1348,1373,1388,1403,1428,1453,1468],{},[220,1047,1048,1050,1052,1055,1058],{},[254,1049,256],{},[254,1051,193],{},[254,1053,1054],{},"0,15",[254,1056,1057],{},"39,32",[254,1059,1060],{},"41,17",[220,1062,1063,1065,1067,1069,1072],{},[254,1064,256],{},[254,1066,736],{},[254,1068,1054],{},[254,1070,1071],{},"39,12",[254,1073,1074],{},"40,95",[220,1076,1077,1081,1085,1090,1095],{},[254,1078,1079],{},[13,1080,256],{},[254,1082,1083],{},[13,1084,67],{},[254,1086,1087],{},[13,1088,1089],{},"0,19",[254,1091,1092],{},[13,1093,1094],{},"49,75",[254,1096,1097],{},[13,1098,1099],{},"52,09",[220,1101,1102,1104,1106,1109,1112],{},[254,1103,304],{},[254,1105,193],{},[254,1107,1108],{},"0,30",[254,1110,1111],{},"76,36",[254,1113,1114],{},"81,73",[220,1116,1117,1119,1121,1123,1126],{},[254,1118,304],{},[254,1120,736],{},[254,1122,1108],{},[254,1124,1125],{},"76,30",[254,1127,1128],{},"81,66",[220,1130,1131,1135,1139,1144,1149],{},[254,1132,1133],{},[13,1134,304],{},[254,1136,1137],{},[13,1138,67],{},[254,1140,1141],{},[13,1142,1143],{},"0,39",[254,1145,1146],{},[13,1147,1148],{},"99,84",[254,1150,1151],{},[13,1152,1153],{},"106,86",[220,1155,1156,1158,1160,1163,1166],{},[254,1157,352],{},[254,1159,193],{},[254,1161,1162],{},"0,59",[254,1164,1165],{},"152,27",[254,1167,1168],{},"162,83",[220,1170,1171,1173,1175,1177,1180],{},[254,1172,352],{},[254,1174,736],{},[254,1176,1162],{},[254,1178,1179],{},"151,23",[254,1181,1182],{},"161,72",[220,1184,1185,1189,1193,1198,1203],{},[254,1186,1187],{},[13,1188,352],{},[254,1190,1191],{},[13,1192,67],{},[254,1194,1195],{},[13,1196,1197],{},"0,76",[254,1199,1200],{},[13,1201,1202],{},"194,67",[254,1204,1205],{},[13,1206,1207],{},"208,17",[220,1209,1210,1212,1214,1217,1220],{},[254,1211,400],{},[254,1213,193],{},[254,1215,1216],{},"1,14",[254,1218,1219],{},"291,00",[254,1221,1222],{},"455,11",[220,1224,1225,1227,1229,1232,1235],{},[254,1226,400],{},[254,1228,736],{},[254,1230,1231],{},"1,13",[254,1233,1234],{},"289,74",[254,1236,1237],{},"453,14",[220,1239,1240,1244,1248,1253,1258],{},[254,1241,1242],{},[13,1243,400],{},[254,1245,1246],{},[13,1247,67],{},[254,1249,1250],{},[13,1251,1252],{},"1,44",[254,1254,1255],{},[13,1256,1257],{},"369,11",[254,1259,1260],{},[13,1261,1262],{},"577,28",[220,1264,1265,1267,1269,1272,1275],{},[254,1266,447],{},[254,1268,193],{},[254,1270,1271],{},"2,08",[254,1273,1274],{},"531,34",[254,1276,1277],{},"996,58",[220,1279,1280,1282,1284,1287,1290],{},[254,1281,447],{},[254,1283,736],{},[254,1285,1286],{},"2,13",[254,1288,1289],{},"545,21",[254,1291,1292],{},"1022,13",[220,1294,1295,1299,1303,1308,1313],{},[254,1296,1297],{},[13,1298,447],{},[254,1300,1301],{},[13,1302,67],{},[254,1304,1305],{},[13,1306,1307],{},"2,63",[254,1309,1310],{},[13,1311,1312],{},"673,64",[254,1314,1315],{},[13,1316,1317],{},"1262,91",[220,1319,1320,1322,1324,1327,1330],{},[254,1321,494],{},[254,1323,193],{},[254,1325,1326],{},"3,75",[254,1328,1329],{},"959,10",[254,1331,1332],{},"1836,45",[220,1334,1335,1337,1339,1342,1345],{},[254,1336,494],{},[254,1338,736],{},[254,1340,1341],{},"3,72",[254,1343,1344],{},"951,17",[254,1346,1347],{},"1820,83",[220,1349,1350,1354,1358,1363,1368],{},[254,1351,1352],{},[13,1353,494],{},[254,1355,1356],{},[13,1357,67],{},[254,1359,1360],{},[13,1361,1362],{},"4,35",[254,1364,1365],{},[13,1366,1367],{},"1112,82",[254,1369,1370],{},[13,1371,1372],{},"2130,28",[220,1374,1375,1377,1379,1382,1385],{},[254,1376,541],{},[254,1378,193],{},[254,1380,1381],{},"6,35",[254,1383,1384],{},"1614,17",[254,1386,1387],{},"3186,41",[220,1389,1390,1392,1394,1397,1400],{},[254,1391,541],{},[254,1393,736],{},[254,1395,1396],{},"5,19",[254,1398,1399],{},"1328,16",[254,1401,1402],{},"2613,51",[220,1404,1405,1409,1413,1418,1423],{},[254,1406,1407],{},[13,1408,541],{},[254,1410,1411],{},[13,1412,67],{},[254,1414,1415],{},[13,1416,1417],{},"6,49",[254,1419,1420],{},[13,1421,1422],{},"1661,29",[254,1424,1425],{},[13,1426,1427],{},"3269,05",[220,1429,1430,1434,1438,1443,1448],{},[254,1431,1432],{},[13,1433,589],{},[254,1435,1436],{},[13,1437,193],{},[254,1439,1440],{},[13,1441,1442],{},"9,21",[254,1444,1445],{},[13,1446,1447],{},"2356,92",[254,1449,1450],{},[13,1451,1452],{},"4547,69",[220,1454,1455,1457,1459,1462,1465],{},[254,1456,589],{},[254,1458,736],{},[254,1460,1461],{},"8,18",[254,1463,1464],{},"2086,62",[254,1466,1467],{},"4032,20",[220,1469,1470,1472,1474,1477,1480],{},[254,1471,589],{},[254,1473,67],{},[254,1475,1476],{},"8,94",[254,1478,1479],{},"2278,53",[254,1481,1482],{},"4405,65",[10,1484,1485,1487],{},[13,1486,637],{}," (met sharegpt-output-len=256):",[48,1489,1490,1493],{},[51,1491,1492],{},"De prestaties gaan over de hele linie omhoog omdat het model het genereren van tokens voorspelbaarder kan plannen.",[51,1494,1495,1497,1498,1501],{},[13,1496,67],{}," breidt opnieuw de voorsprong van MI300x uit bij de meeste batchgroottes. Bij batch=64 en 128 liggen H200 en Paiton heel dicht bij elkaar, maar soms doet H200 een ",[29,1499,1500],{},"iets"," hogere aanvraagdoorvoer. Zelfs dan bevindt het totaal aantal tokens\u002Fs van Paiton zich in dezelfde marge of overtreft deze.",[10,1503,1504],{},[662,1505],{"alt":664,"src":1506},"\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Frequests_256_cap.jpg",[36,1508],{},[10,1510,1511],{},[13,1512,1513,1514],{},"4.2.B. Latency met ",[76,1515,82],{},[214,1517,1518,1550],{},[217,1519,1520],{},[220,1521,1522,1526,1530,1535,1540,1545],{},[223,1523,1524],{},[13,1525,227],{},[223,1527,1528],{},[13,1529,232],{},[223,1531,1532],{},[13,1533,1534],{},"TTFT (ms)",[223,1536,1537],{},[13,1538,1539],{},"TPOT (ms)",[223,1541,1542],{},[13,1543,1544],{},"ITL (ms)",[223,1546,1547],{},[13,1548,1549],{},"E2E (ms)",[249,1551,1552,1569,1586,1615,1632,1649,1678,1695,1712,1741,1758,1775,1804],{},[220,1553,1554,1556,1558,1561,1564,1566],{},[254,1555,256],{},[254,1557,193],{},[254,1559,1560],{},"143,90",[254,1562,1563],{},"24,96",[254,1565,1563],{},[254,1567,1568],{},"6509,40",[220,1570,1571,1573,1575,1578,1581,1583],{},[254,1572,256],{},[254,1574,736],{},[254,1576,1577],{},"176,79",[254,1579,1580],{},"24,97",[254,1582,1580],{},[254,1584,1585],{},"6543,64",[220,1587,1588,1592,1596,1601,1606,1610],{},[254,1589,1590],{},[13,1591,256],{},[254,1593,1594],{},[13,1595,67],{},[254,1597,1598],{},[13,1599,1600],{},"117,14",[254,1602,1603],{},[13,1604,1605],{},"19,71",[254,1607,1608],{},[13,1609,1605],{},[254,1611,1612],{},[13,1613,1614],{},"5144,32",[220,1616,1617,1619,1621,1624,1627,1629],{},[254,1618,304],{},[254,1620,193],{},[254,1622,1623],{},"145,96",[254,1625,1626],{},"25,72",[254,1628,1626],{},[254,1630,1631],{},"6703,84",[220,1633,1634,1636,1638,1641,1644,1646],{},[254,1635,304],{},[254,1637,736],{},[254,1639,1640],{},"171,51",[254,1642,1643],{},"25,64",[254,1645,1643],{},[254,1647,1648],{},"6708,87",[220,1650,1651,1655,1659,1664,1669,1673],{},[254,1652,1653],{},[13,1654,304],{},[254,1656,1657],{},[13,1658,67],{},[254,1660,1661],{},[13,1662,1663],{},"113,62",[254,1665,1666],{},[13,1667,1668],{},"19,66",[254,1670,1671],{},[13,1672,1668],{},[254,1674,1675],{},[13,1676,1677],{},"5126,54",[220,1679,1680,1682,1684,1687,1690,1692],{},[254,1681,352],{},[254,1683,193],{},[254,1685,1686],{},"145,09",[254,1688,1689],{},"25,80",[254,1691,1689],{},[254,1693,1694],{},"6723,31",[220,1696,1697,1699,1701,1704,1707,1709],{},[254,1698,352],{},[254,1700,736],{},[254,1702,1703],{},"170,48",[254,1705,1706],{},"25,88",[254,1708,1706],{},[254,1710,1711],{},"6768,63",[220,1713,1714,1718,1722,1727,1732,1736],{},[254,1715,1716],{},[13,1717,352],{},[254,1719,1720],{},[13,1721,67],{},[254,1723,1724],{},[13,1725,1726],{},"117,72",[254,1728,1729],{},[13,1730,1731],{},"20,16",[254,1733,1734],{},[13,1735,1731],{},[254,1737,1738],{},[13,1739,1740],{},"5257,58",[220,1742,1743,1745,1747,1750,1753,1755],{},[254,1744,400],{},[254,1746,193],{},[254,1748,1749],{},"263,55",[254,1751,1752],{},"26,56",[254,1754,1752],{},[254,1756,1757],{},"7035,76",[220,1759,1760,1762,1764,1767,1770,1772],{},[254,1761,400],{},[254,1763,736],{},[254,1765,1766],{},"320,44",[254,1768,1769],{},"26,45",[254,1771,1769],{},[254,1773,1774],{},"7064,58",[220,1776,1777,1781,1785,1790,1795,1799],{},[254,1778,1779],{},[13,1780,400],{},[254,1782,1783],{},[13,1784,67],{},[254,1786,1787],{},[13,1788,1789],{},"253,04",[254,1791,1792],{},[13,1793,1794],{},"20,75",[254,1796,1797],{},[13,1798,1794],{},[254,1800,1801],{},[13,1802,1803],{},"5544,18",[220,1805,1806,1808,1810,1812,1814,1816],{},[254,1807,977],{},[254,1809,977],{},[254,1811,977],{},[254,1813,977],{},[254,1815,977],{},[254,1817,977],{},[10,1819,1820],{},[29,1821,1822],{},"(Nogmaals, gedeeltelijke gegevens worden kortheidshalve weergegeven.)",[10,1824,1825,1828],{},[13,1826,1827],{},"Belangrijkste latencyresultaten",":",[48,1830,1831,1836,1839],{},[51,1832,1833,1835],{},[13,1834,67],{}," reduceert consequent de Time-to-First-Token (TTFT) bij kleine batchgroottes.",[51,1837,1838],{},"De gemiddelde E2E-latentie ziet een merkbare daling bij Paiton versus de standaard MI300x of H200, vooral in het batchbereik van 1–16.",[51,1840,1841],{},"Bij grotere batchgroottes neemt de latency vanzelf toe, maar Paiton helpt ze onder controle te houden.",[10,1843,1844],{},[662,1845],{"alt":664,"src":1846},"\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fe2e_256_cap.jpg",[36,1848],{},[17,1850,1852],{"id":1851},"_5-kosten-per-miljoen-tokens-echte-roi-voor-paiton-mi300x",[13,1853,1854],{},"5. Kosten per miljoen tokens: echte ROI voor Paiton + MI300x",[10,1856,1857,1858,1861],{},"Puur vanuit bedrijfsperspectief is het kostenverschil tussen het 8-GPU H200-systeem en het 8-GPU MI300x-systeem, ",[13,1859,1860],{},"$40.000",", aanzienlijk. Wanneer genormaliseerd op basis van het totale aantal verwerkte tokens (bijvoorbeeld gedurende de meerjarige levenscyclus van het systeem), is de berekening in het voordeel van AMD:",[48,1863,1864,1869],{},[51,1865,1866,1868],{},[13,1867,193],{}," produceert mogelijk een marginaal hogere doorvoer bij extreem grote batchgroottes, maar vereist een grotere financiële uitgave.",[51,1870,1871,1874,1875,1878],{},[13,1872,1873],{},"MI300x + Paiton"," voldoet aan of verslaat de H200 tegen een veel lagere hardwareprijs. Uw ",[13,1876,1877],{},"$ \u002F miljoen tokens","-kosten kunnen dus aanzienlijk lager zijn.",[10,1880,1881],{},"Bij grootschalige inferentie, met miljarden of biljoenen tokens per maand, verdient dat prijsverschil zich snel terug.",[10,1883,1884],{},[662,1885],{"alt":664,"src":1886},"\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fcost_per_million_tokens_simple.jpg",[10,1888,1889],{},[29,1890,1891],{},"\"We berekenen de kosten per miljoen tokens door de geschatte hardwarekosten van elk systeem te delen door de aanhoudende tokenthroughput bij een gematigd concurrencyniveau. In een echte deployment kan het exacte bedrag variëren naargelang uw gebruikspatroon, operationele overhead en gekozen batchgroottes. Deze cijfers maken de relatieve kostenefficiëntie van de H200- en MI300X-oplossing met Paiton wel duidelijk.\"",[36,1893],{},[17,1895,1897],{"id":1896},"_6-paiton-maakt-het-verschil",[13,1898,1899],{},"6. Paiton maakt het verschil",[10,1901,1902,1903,1905],{},"De ruwe mogelijkheden van AMD-silicium zijn indrukwekkend. ",[13,1904,67],{}," is onze eigen softwarelaag die concurrency, kernel launches en geheugengebruik optimaliseert:",[48,1907,1908,1914,1920],{},[51,1909,1910,1913],{},[13,1911,1912],{},"Kernelfusie",": Minimaliseert overhead door bewerkingen samen te voegen.",[51,1915,1916,1919],{},[13,1917,1918],{},"Adaptieve concurrency",": benut het HBM-geheugen van de GPU om pieken met meerdere aanvragen te verwerken.",[51,1921,1922,1925],{},[13,1923,1924],{},"Sterk met oudere drivers",": zelfs met versie 6.3.1 verslaan we de H200. We verwachten nog grotere verbeteringen bij de overstap naar 6.4 en later.",[10,1927,1928],{},"In vrijwel elke bovenstaande tabel presteert “MI300X + Paiton” beter dan “H200” en “Stock MI300X”. Dat is niet alleen aan de hardware te danken, maar aan de wisselwerking tussen Paiton en de krachtige geheugenarchitectuur van AMD.",[36,1930],{},[17,1932,1934],{"id":1933},"_7-belangrijkste-conclusies",[13,1935,1936],{},"7. Belangrijkste conclusies",[154,1938,1939,1945,1951],{},[51,1940,1941,1944],{},[13,1942,1943],{},"Paiton + MI300x"," overtreft (of voldoet aan) de NVIDIA H200 op Qwen3-32B in kleine tot middelgrote batchgroottes en houdt vaak ook goed stand bij grotere formaten.",[51,1946,1947,1950],{},[13,1948,1949],{},"$40K goedkoper voor een 8-GPU-systeem",": Dat is een echt verschil in kapitaaluitgaven, culminerend in betere kosten per miljoen tokens in veel praktijkscenario's.",[51,1952,1953,1956],{},[13,1954,1955],{},"Nog meer prestatiewinst in het vooruitzicht",": nieuwe AMD-drivers en verdere concurrencyoptimalisaties in Paiton zullen de prestatiekloof verder vergroten.",[36,1958],{},[17,1960,1962],{"id":1961},"_8-vooruitblik",[13,1963,1964],{},"8. Vooruitblik",[10,1966,1967],{},[29,1968,1969],{},"“Blijf ons volgen. We hebben veelbelovende verbeteringen voor FP8 en meer in de pijplijn.”",[10,1971,1972],{},"We bouwen verder op deze resultaten. We blijven Paiton verfijnen met geavanceerde quantisatiestrategieën, diepere optimalisatietechnieken en doorlopende verbeteringen voor AMD's MI300X-platform. Nu steeds meer ondernemingen kiezen voor grootschalige interne LLM-deployments, kan de combinatie van AMD MI300X-hardware en Paiton de kosten verlagen en tegelijk de prestaties verhogen.",[10,1974,1975],{},[13,1976,1977],{},"Bedankt voor het lezen. Neem gerust contact met ons op als u meer gegevens, een privédemo of een technische deep dive in ons concurrencymodel wilt.",[10,1979,1980],{},[29,1981,1982],{},[13,1983,1984],{},"Het Paiton-team",{"title":664,"searchDepth":1986,"depth":1986,"links":1987},2,[1988,1990,1991,1992,1993,1994,1995,1996],{"id":19,"depth":1989,"text":22},3,{"id":40,"depth":1989,"text":43},{"id":125,"depth":1989,"text":128},{"id":146,"depth":1989,"text":149},{"id":1851,"depth":1989,"text":1854},{"id":1896,"depth":1989,"text":1899},{"id":1933,"depth":1989,"text":1936},{"id":1961,"depth":1989,"text":1964},[1998,1999,67,2000,2001,193,2002,2003,67,2004],"Alle","Kunstmatige intelligentie","AI","AMD","MI300","NVidia","vLLM","2025-05-02T21:10:30","We vergelijken Qwen3-32B op een AMD MI300X, geoptimaliseerd met Paiton, met de NVIDIA H200. De resultaten tonen hoe Paiton vergelijkbare of betere prestaties kan leveren tegen lagere hardwarekosten en met aantrekkelijke kosten per miljoen tokens.","md","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002F3ac59a73-2466-4422-b7e5-ef2e4a8ca58e.webp",{},true,"https:\u002F\u002Feliovp.com\u002Fcranking-out-faster-tokens-for-fewer-dollars-amd-mi300x-vs-nvidia-h200\u002F","\u002Fblog\u002Fcranking-out-faster-tokens-for-fewer-dollars-amd-mi300x-vs-nvidia-h200",{"title":5,"description":2006},"cranking-out-faster-tokens-for-fewer-dollars-amd-mi300x-vs-nvidia-h200","blog\u002Fcranking-out-faster-tokens-for-fewer-dollars-amd-mi300x-vs-nvidia-h200","gP9_8cgDONK3FkuaJ0ePhT_I6Qre7OF1lsW55Duw_8M",[2018,2034,2048,2078,2090,2112,2130,2149,2168,2186,2202,2214,2229,2244,2254,2263,2271,2286,2298,2309,2320,2322,2335,2345,2358,2369,2379,2390,2399,2411,2422,2431],{"path":2019,"title":2020,"description":2021,"date":2022,"slug":2023,"image":2024,"originalUrl":2025,"categories":2026},"\u002Fblog\u002Fpaiton-qwen38-radeon-ai-pro-r9700","Paiton: 21% meer Qwen3.8-doorvoer op Radeon AI PRO R9700","Paiton draait AMD's Qwen3.8 27B op één Radeon AI PRO R9700 met 39,77 outputtokens per seconde. Dat levert 21% meer throughput en 17,4% lagere gemodelleerde kosten op.","2026-09-04T09:00:00","paiton-qwen38-radeon-ai-pro-r9700","\u002Fasset\u002Fimages\u002Fblog\u002Fpaiton-r9700\u002F00-featured-paiton-r9700.webp","https:\u002F\u002Feliovp.com\u002Fblog\u002Fpaiton-qwen38-radeon-ai-pro-r9700",[67,1999,2027,2028,2029,2030,2031,2032,2004,2033],"AMD Radeon","AI-inferentie","GPU-prestaties","Inferentielatentie","Inferentie-optimalisatie","Grote taalmodellen","Kostenefficiëntie",{"path":2035,"title":2036,"description":2037,"date":2038,"slug":2039,"image":2040,"originalUrl":2041,"categories":2042},"\u002Fblog\u002Fai-data-center-power-requirements-gpu-per-megawatt","Stroomvereisten voor AI-datacenters: de GPU-per-MW-illusie","Waarom vermelden offertes voor AI-datacenters verschillende GPU-capaciteiten? Ontdek hoe PUE, piekbelasting, opslag, netwerken en koeling bepalen hoeveel compute werkelijk inzetbaar is.","2026-07-27T23:52:00","ai-data-center-power-requirements-gpu-per-megawatt","\u002Fasset\u002Fimages\u002Fblog\u002Fai-data-center-power-requirements-gpu-per-megawatt\u002Fgpu-per-megawatt-illusion.webp",null,[1998,2043,2044,2045,2046,2047],"AI-infrastructuur","Datacenters","ModFlex","HPC","AMD Helios",{"path":2049,"title":2050,"description":2051,"date":2052,"slug":2053,"image":2054,"originalUrl":2055,"categories":2056},"\u002Fblog\u002Fpaiton-returns-to-its-diffusion-roots-optimizing-wan2-2-t2v-a14b-on-amd-mi355x","Paiton keert terug naar zijn diffusiewortels: optimalisatie van Wan2.2-T2V-A14B op AMD MI355X","Toen we begonnen met het bouwen van Paiton, was een van onze eerste aandachtsgebieden het optimaliseren van diffusiemodellen. Stable Diffusion XL was een van de eerste grote modellen waarin we lieten zien dat gefuseerde operators, efficiënte uitvoering en hardwarebewuste kernels een echt verschil konden maken. Nu keren we terug naar die oorsprong. Met de groeiende belangstelling voor het genereren van tekst-naar-video, ...","2026-06-10T14:04:04","paiton-returns-to-its-diffusion-roots-optimizing-wan2-2-t2v-a14b-on-amd-mi355x","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fpaitonwan2.webp","https:\u002F\u002Feliovp.com\u002Fpaiton-returns-to-its-diffusion-roots-optimizing-wan2-2-t2v-a14b-on-amd-mi355x\u002F",[1998,1999,67,2057,2001,2058,2059,2060,2061,2062,2063,2064,2065,2066,2067,2068,2069,2003,2070,2071,67,2072,2073,2074,2075,2076,2077],"14B","B200","Benchmarks","Blackwell","Compute","Diffusie","Eliovp","Generatieve AI","GPU","Hardware","Inferentie","Instinct","MI355x","On-premises","Optimalisatie","Soevereine AI","T2V","Tekst-naar-video","Tuning","Video-generatie","Wan2.2",{"path":2079,"title":2080,"description":2081,"date":2082,"slug":2083,"image":2084,"originalUrl":2085,"categories":2086},"\u002Fblog\u002Ffrom-the-attic-to-the-front-page-eliovp-recognized-as-a-pioneer-in-chip-optimization-data-center-infrastructure","Van zolder tot voorpagina: ElioVP erkend als pionier op het gebied van chipoptimalisatie en datacenterinfrastructuur","Het waren bijzondere weken voor het team van Eliovp. Ons bedrijf haalde de voorpagina van De Tijd, de toonaangevende zakenkrant van België. Ons verhaal begon bij een oprichter die op zolder aan hardware sleutelde en groeide uit tot een onderneming met 215 miljoen euro omzet.","2026-02-10T20:48:12","from-the-attic-to-the-front-page-eliovp-recognized-as-a-pioneer-in-chip-optimization-data-center-infrastructure","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fphysicalnewspaper.webp","https:\u002F\u002Feliovp.com\u002Ffrom-the-attic-to-the-front-page-eliovp-recognized-as-a-pioneer-in-chip-optimization-data-center-infrastructure\u002F",[1998,1999,2087,2088,2001,2089,2087,2003],"Modulaire DC","Niet gecategoriseerd","De Tijd",{"path":2091,"title":2092,"description":2093,"date":2094,"slug":2095,"image":2096,"originalUrl":2097,"categories":2098},"\u002Fblog\u002Fprivacy-is-geen-it-probleem-meer-het-is-een-strategische-prioriteit","Privacy is geen IT-probleem meer, het is een strategische prioriteit","Privacyrisico's, psychologische valkuilen en de operationele realiteit van generatieve AI in de Benelux. Waarom bedrijven gevoelige gegevens, governance en lokale AI strategisch moeten benaderen.","2026-01-29T13:51:11","privacy-is-geen-it-probleem-meer-het-is-een-strategische-prioriteit","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fheaderimage.webp","https:\u002F\u002Feliovp.com\u002Fprivacy-is-geen-it-probleem-meer-het-is-een-strategische-prioriteit\u002F",[1998,1999,2099,2088,2100,2101,2102,2103,2104,2105,2106,2107,2063,2108,2064,2109,2110,2111],"Trending","AI Act","Antropomorfisme","AVG","Benelux","ChatGPT","Cyberbeveiliging","Databeheer","Gegevensbeveiliging","GDPR","Microsoft Copilot","Privacy","Shadow AI",{"path":2113,"title":2114,"description":2115,"date":2116,"slug":2117,"image":2118,"originalUrl":2119,"categories":2120},"\u002Fblog\u002Fitsme-bij-ons-is-het-its-not-me-en-dit-is-waarom","Itsme? Bij ons is het “it's not me”, en dit is waarom.","Waarom Eliovp itsme niet gebruikt voor medewerkers en interne processen: een analyse van datasoevereiniteit, cloudjurisdictie, metadata, eigendomsstructuur en technische risico's.","2025-11-27T09:32:14","itsme-bij-ons-is-het-its-not-me-en-dit-is-waarom","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Ffrontimage.webp","https:\u002F\u002Feliovp.com\u002Fitsme-bij-ons-is-het-its-not-me-en-dit-is-waarom\u002F",[1998,2121,2122,2123,2105,2124,2125,2126,2108,2127,2128,2129,2110],"AWS","Belgian Mobile ID","CLOUD Act","Datasoevereiniteit","Digitale identiteit","eIDAS","itsme","Liberty Global","MyGov.be",{"path":2131,"title":2132,"description":2133,"date":2134,"slug":2135,"image":2136,"originalUrl":2137,"categories":2138},"\u002Fblog\u002Ffield-report-the-reality-of-building-agentic-ai-in-2025","Praktijkrapport: de realiteit van Agentic AI bouwen in 2025","De verwachtingen rond Agentic AI staan vaak ver af van de technische realiteit. Dit praktijkrapport beschrijft wat er nodig is om betrouwbare, lokale AI-agents voor bedrijfsworkflows te bouwen, van modulaire architectuur en observability tot modeltraining, VRAM en soevereine infrastructuur.","2025-11-25T14:03:39","field-report-the-reality-of-building-agentic-ai-in-2025","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Ffieldreport.webp","https:\u002F\u002Feliovp.com\u002Ffield-report-the-reality-of-building-agentic-ai-in-2025\u002F",[1998,1999,2139,2099,2140,2141,2142,2143,2144,2145,2146,2147,2072,2148],"Oplossingen","Agentic AI","AI-techniek","AI-strategie","Autonome agenten","Bedrijfs-AI","Lokale LLM","Modelverfijning","AI op locatie","VRAM-optimalisatie",{"path":2150,"title":2151,"description":2152,"date":2153,"slug":2154,"image":2155,"originalUrl":2156,"categories":2157},"\u002Fblog\u002Fthe-synthetic-unicorn-bubble","De synthetische unicornzeepbel","Hoe AI-neocloudinfrastructuur circulaire kasstromen omzet in schijngroei. Tussen 2023 en 2025 stroomden miljarden aan durfkapitaal naar het neocloudecosysteem. Onze analyse onderzoekt hoe circulaire financiering, vaporware-infrastructuur en roofzuchtige contracten onderliggende activawaarden kunstmatig kunnen opbouwen.","2025-11-24T19:22:28","the-synthetic-unicorn-bubble","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fsyntheticunicorn.webp","https:\u002F\u002Feliovp.com\u002Fthe-synthetic-unicorn-bubble\u002F",[1998,1999,2099,2158,2159,2160,2161,2162,2163,2164,2165,2166,2167],"AI Infrastructure","AI Neocloud","Circulaire financiering","GPU Cloud","Beleggingsrisico's","Opstartwaardering","Synthetische bubbel","Technische analyse","Vaporware","Durfkapitaal",{"path":2169,"title":2170,"description":2171,"date":2172,"slug":2173,"image":2174,"originalUrl":2175,"categories":2176},"\u002Fblog\u002Fbuilding-the-engine-for-the-ai-race-the-4-month-path-to-nvidia-gb300-nvl72-power","De motor voor de AI-race bouwen: in vier maanden klaar voor NVIDIA GB300 NVL72","Bij AI-infrastructuur is snelheid een concurrentieel voordeel. Traditionele datacenterprojecten duren vaak 18 tot 24 maanden. Onze modulaire aanpak levert in vier maanden een operationele faciliteit die specifiek is ontworpen voor de NVIDIA GB300 NVL72.","2025-11-20T14:10:19","building-the-engine-for-the-ai-race-the-4-month-path-to-nvidia-gb300-nvl72-power","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fsuperpodmodflexfrontimage.webp","https:\u002F\u002Feliovp.com\u002Fbuilding-the-engine-for-the-ai-race-the-4-month-path-to-nvidia-gb300-nvl72-power\u002F",[1998,2087,2088,2177,2043,2178,2179,2180,2181,2182,2183,2184,2185],"150 kW rek","DLC","Hoge dichtheid","Vloeistofkoeling","Modulair datacenter","NVIDIA Blackwell Ultra","NVIDIA GB300","NVL72","Snelle implementatie",{"path":2187,"title":2188,"description":2189,"date":2190,"slug":2191,"image":2192,"originalUrl":2193,"categories":2194},"\u002Fblog\u002Fwhy-cuda-translation-wont-unlock-amds-real-potential","Waarom de “CUDA”-vertaling het echte potentieel van AMD niet zal ontsluiten","Om de paar jaar verschijnt een oplossing met dezelfde belofte: behoud uw CUDA-code, wissel van toolchain en draai plots op verschillende GPU-platforms. Dat kan nuttig zijn voor compatibiliteit, maar ontsluit niet automatisch AMD's echte prestatiepotentieel. Daarvoor zijn AMD-first kernels, libraries en tuning nodig.","2025-11-12T14:48:37","why-cuda-translation-wont-unlock-amds-real-potential","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fchatgpt-image-nov-11-2025-09_16_10-pm-1.webp","https:\u002F\u002Feliovp.com\u002Fwhy-cuda-translation-wont-unlock-amds-real-potential\u002F",[1998,1999,67,2088,55,1999,2195,2196,2197,2198,2199,2200,67,2201],"CUDA-vertaling","FP8","GPU-optimalisatie","Hoogwaardige computers","HIP","Kerneltuning","ROCm",{"path":2203,"title":2204,"description":2205,"date":2206,"slug":2207,"image":2208,"originalUrl":2209,"categories":2210},"\u002Fblog\u002Fpaiton-the-simplest-way-to-supercharge-ai-inference","Paiton: de eenvoudigste manier om AI-inferentie een boost te geven","Laten we eerlijk zijn: marketing is niet onze sterkste kant. We namen nooit extern kapitaal aan, verbrandden geen budget aan advertentiecampagnes en bouwden geen verkoopleger. We bouwen gewoon technologie die werkt. Paiton versnelt uw bestaande inferentiestack zonder migratie, nieuwe API's of modelconversie.","2025-11-11T10:31:22","paiton-the-simplest-way-to-supercharge-ai-inference","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fpaiton-powaaah.webp","https:\u002F\u002Feliovp.com\u002Fpaiton-the-simplest-way-to-supercharge-ai-inference\u002F",[1998,1999,67,2028,2211,55,2033,2212,2031,2200,67,2213,2004],"AMD Instinct","Hoge throughput","SGLang",{"path":2215,"title":2216,"description":2217,"date":2218,"slug":2219,"image":2220,"originalUrl":2221,"categories":2222},"\u002Fblog\u002Fstop-overpaying-paiton-mi300x-moe-beats-h200-b200-on-1m-tokens","Stop met te veel betalen: Paiton MI300X MoE verslaat H200\u002FB200 in kosten per 1 miljoen tokens","We benchmarkten Paiton's nieuwe MoE-ondersteuning met Qwen\u002FQwen3-30B-A3B-Instruct-2507 en vergeleken de inferentieprestaties van verschillende opstellingen. Elke configuratie draaide vijf keer per batchgrootte. We rapporteren het gemiddelde van die runs op basis van realistische conversationele workloads.","2025-09-26T13:36:18","stop-overpaying-paiton-mi300x-moe-beats-h200-b200-on-1m-tokens","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fhulkvshulkpaitonwins.webp","https:\u002F\u002Feliovp.com\u002Fstop-overpaying-paiton-mi300x-moe-beats-h200-b200-on-1m-tokens\u002F",[1998,1999,67,2223,55,2224,2031,2225,2226,2227,61,67,2228],"AI-benchmarks","Kosten per token","Mixture of Experts","MoE","NVIDIA B200","Qwen3",{"path":2230,"title":2231,"description":2232,"date":2233,"slug":2234,"image":2235,"originalUrl":2236,"categories":2237},"\u002Fblog\u002Fagentic-ai-but-make-it-local-from-inbox-to-insight-to-action-en","Agentic AI, maar dan lokaal: van inbox naar inzicht en actie","We bouwen productieklare, local-first AI-agents die aansluiten op uw bestaande e-mailomgeving. Ze maken tickets aan, classificeren berichten, lezen documenten, herkennen facturen en offertes, analyseren beelden en sturen gestructureerde rapporten naar uw systemen, zonder verplichte afhankelijkheid van externe AI-API's.","2025-09-16T13:09:00","agentic-ai-but-make-it-local-from-inbox-to-insight-to-action-en","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Ffrontfotoblog.webp","https:\u002F\u002Feliovp.com\u002Fagentic-ai-but-make-it-local-from-inbox-to-insight-to-action-en\u002F",[1998,1999,2139,2088,2140,2238,2239,2240,2241,2145,2147,2072,2242,2243],"Schadedetectie","Documentverwerking","E-mailautomatisering","Factuurextractie","Ticketautomatisering","Workflowautomatisering",{"path":2245,"title":2246,"description":2247,"date":2248,"slug":2249,"image":2250,"originalUrl":2251,"categories":2252},"\u002Fblog\u002Fmi300x-fp8-data-parallel-benchmarks-8-64-gpus-h200-left-behind-b200-within-reach","MI300X FP8 data-parallelle benchmarks (8 tot 64 GPU's): H200 achter zich, B200 binnen bereik","We schalen Llama-3.1-8B-Instruct-FP8-KV met Paiton op een gepartitioneerde AMD MI300X-server van 8 fysieke GPU's naar 64 virtuele GPU's. De resultaten tonen sterke multi-tenant throughput, veel lagere latency en een bijna gelijkspel met NVIDIA's B200.","2025-07-31T13:32:57","mi300x-fp8-data-parallel-benchmarks-8-64-gpus-h200-left-behind-b200-within-reach","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fimage-2-1.webp","https:\u002F\u002Feliovp.com\u002Fmi300x-fp8-data%e2%80%91parallel-benchmarks-8-64-gpus-h200-left-behind-b200-within-reach\u002F",[1998,1999,67,2088,2000,2001,2058,193,2253,2069,2003,67,2004],"MI300X",{"path":2255,"title":2256,"description":2257,"date":2258,"slug":2259,"image":2260,"originalUrl":2261,"categories":2262},"\u002Fblog\u002Fapplicable-ai-for-businesses","Toepasbare AI voor bedrijven","Bij Eliovp blijven we innoveren om praktische oplossingen te bouwen. Een van onze grootste sterktes is het vermogen om buiten de gebaande paden te denken. Daarom ontwikkelen we toepasbare AI-oplossingen die dagelijks bruikbaar zijn en precies aansluiten op de behoeften van onze klanten.","2025-07-09T21:35:30","applicable-ai-for-businesses","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fscherm_afbeelding-2025-07-09-om-23.30.17.webp","https:\u002F\u002Feliovp.com\u002Fapplicable-ai-for-businesses\u002F",[1998,1999,2139,2140,2238,2239,2240,2241,2145,2147,2072,2242,2243],{"path":2264,"title":2265,"description":2266,"date":2267,"slug":2268,"image":664,"originalUrl":2269,"categories":2270},"\u002Fblog\u002Fintroducing-paitons-free-evaluation-models","Maak kennis met de gratis evaluatiemodellen van Paiton","AI verandert elke sector in hoog tempo, maar grote modellen efficiënt uitvoeren blijft een aanzienlijke technische en financiële uitdaging. ElioVP specialiseert zich in optimalisatie voor AMD-accelerators en helpt organisaties het volledige potentieel van hun hardware te benutten. Met onze gratis evaluatiemodellen kunt u de optimalisaties van Paiton vooraf op uw eigen workloads testen.","2025-07-07T11:26:13","introducing-paitons-free-evaluation-models","https:\u002F\u002Feliovp.com\u002Fintroducing-paitons-free-evaluation-models\u002F",[1998,1999,67],{"path":2272,"title":2273,"description":2274,"date":2275,"slug":2276,"image":2277,"originalUrl":2278,"categories":2279},"\u002Fblog\u002Fpaiton-dramatically-faster-startup-and-performance-for-llama-3-1-405b","Paiton: veel sneller opstarten en betere prestaties voor Llama-3.1-405B","Met Paiton streven we niet alleen naar maximale inferentiesnelheden. We verbeteren de volledige levenscyclus van de deployment van grote taalmodellen. Onze nieuwste test combineert AMD MI300X-GPU's met het omvangrijke Llama-3.1-405B-Instruct-FP8-KV-model en levert baanbrekende resultaten op voor zowel opstartsnelheid als runtimeprestaties.","2025-06-12T20:15:23","paiton-dramatically-faster-startup-and-performance-for-llama-3-1-405b","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fservingscreenshot.webp","https:\u002F\u002Feliovp.com\u002Fpaiton-dramatically-faster-startup-and-performance-for-llama-3-1-405b\u002F",[1998,1999,67,2088,2028,55,2280,2196,2281,2282,2283,67,2284,2285],"Koude start","Grafiekcompilatie","Llama 3.1 405B","LLM-optimalisatie","Opstartlatentie","Tensor-parallellisme",{"path":2287,"title":2288,"description":2289,"date":2290,"slug":2291,"image":2292,"originalUrl":2293,"categories":2294},"\u002Fblog\u002Fpaiton-fp8-beats-nvidias-h200-on-amds-mi300x","Paiton FP8 verslaat NVIDIA's H200 op AMD's MI300X","AI evolueert razendsnel en efficiënte inferentie is essentieel om krachtige modellen in praktijksituaties in te zetten. Paiton levert nu sterkere FP8-inferentieprestaties op AMD MI300X-GPU's dan NVIDIA's H200, dankzij zorgvuldig geoptimaliseerde kernels.","2025-06-08T19:12:40","paiton-fp8-beats-nvidias-h200-on-amds-mi300x","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fblognewfp8.webp","https:\u002F\u002Feliovp.com\u002Fpaiton-fp8-beats-nvidias-h200-on-amds-mi300x\u002F",[1998,1999,67,2088,55,2295,2144,2064,2029,2030,2032,2282,2296,2297],"Koude startoptimalisatie","Model serving","vLLM-optimalisatie",{"path":2299,"title":2300,"description":2301,"date":2302,"slug":2303,"image":2304,"originalUrl":2305,"categories":2306},"\u002Fblog\u002Fmi300x-vs-h200-vs-rx-7900-xtx-vs-tenstorrent-n300s-with-vllm","MI300X versus H200 versus RX 7900 XTX versus Tenstorrent n300s met vLLM","Nu grote taalmodellen een fundamenteel onderdeel worden van moderne toepassingen, is de juiste server voor deployment belangrijker dan ooit. We vergelijken AMD MI300X, NVIDIA H200, AMD RX 7900 XTX en Tenstorrent n300s met vLLM op throughput en kosten.","2025-05-09T14:03:58","mi300x-vs-h200-vs-rx-7900-xtx-vs-tenstorrent-n300s-with-vllm","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fcomparisontenstor.webp","https:\u002F\u002Feliovp.com\u002Fmi300x-vs-h200-vs-rx-7900-xtx-vs-tenstorrent-n300s-with-vllm\u002F",[1998,1999,67,2139,2088,2001,2253,2003,2307,2308],"RX7900XTX","tenstorrent",{"path":2310,"title":2311,"description":2312,"date":2313,"slug":2314,"image":2315,"originalUrl":2316,"categories":2317},"\u002Fblog\u002Fclusterpl-empowering-gpu-cluster-investors-with-real-world-financial-insights","ClusterP&L: financiële inzichten voor investeerders in GPU-clusters","Eliovp ontwerpt, bouwt en optimaliseert al jaren GPU-clusters in heel Europa. Omdat klanten ons telkens opnieuw vroegen om een P&L-model voor hun investering, bundelden we onze praktijkkennis in ClusterP&L: een toegankelijke, voortdurend bijgewerkte tool voor financiële prognoses van GPU-infrastructuur.","2025-05-03T10:52:22","clusterpl-empowering-gpu-cluster-investors-with-real-world-financial-insights","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fcomparisonscenarios.webp","https:\u002F\u002Feliovp.com\u002Fclusterpl-empowering-gpu-cluster-investors-with-real-world-financial-insights\u002F",[1998,1999,2087,2139,2058,193,2318,2003,2319],"MI325x","pnl-rekenmachine",{"path":2012,"title":5,"description":2006,"date":2005,"slug":2014,"image":2008,"originalUrl":2011,"categories":2321},[1998,1999,67,2000,2001,193,2002,2003,67,2004],{"path":2323,"title":2324,"description":2325,"date":2326,"slug":2327,"image":2328,"originalUrl":2329,"categories":2330},"\u002Fblog\u002Fpower-meets-precision-high-density-modular-data-center-for-nvidia-nvl-deployments-1-2-mw","Waar vermogen en precisie samenkomen: modulair high-density datacenter voor NVIDIA NVL-deployments (1 tot 2 MW)","Eliovp ontwikkelt modulaire high-density infrastructuur voor AI-workloads van Blackwell-klasse. Het platform ondersteunt NVIDIA NVL-configuraties van NVL4 tot NVL72, van gedistribueerde edge-inferentie tot modeltraining op hyperscale, met 1 tot 2 MW IT-vermogen per module.","2025-05-02T14:09:59","power-meets-precision-high-density-modular-data-center-for-nvidia-nvl-deployments-1-2-mw","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Feliovp_critical-1mw-pod_rev-2_transparent.webp","https:\u002F\u002Feliovp.com\u002Fpower-meets-precision-high-density-modular-data-center-for-nvidia-nvl-deployments-1-2-mw\u002F",[1998,2087,2331,2158,2179,2046,2180,2181,2332,2333,2184,2334],"1-2MW datacenter","NVIDIA Blackwell","NVIDIA NVL","Precisiekoeling",{"path":2336,"title":2337,"description":2338,"date":2339,"slug":2340,"image":2341,"originalUrl":2342,"categories":2343},"\u002Fblog\u002Fexamining-ai-agents-in-the-medical-field-ai-that-speaks-dicom","AI-agents in de medische wereld: AI die DICOM spreekt","Eliovp ontwikkelde een medische AI-agent die rechtstreeks met DICOM-servers in ziekenhuizen communiceert. De assistent kan patiënten en onderzoeken opzoeken, relevante beelden ophalen en medische beeldanalyse ondersteunen binnen de lokale ziekenhuisinfrastructuur.","2025-04-11T14:45:48","examining-ai-agents-in-the-medical-field-ai-that-speaks-dicom","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fhealthcareblog-1.webp","https:\u002F\u002Feliovp.com\u002Fexamining-ai-agents-in-the-medical-field-ai-that-speaks-dicom\u002F",[1998,1999,2139,2088,2000,2001,2344],"Zorg",{"path":2346,"title":2347,"description":2348,"date":2349,"slug":2350,"image":2351,"originalUrl":2352,"categories":2353},"\u002Fblog\u002Feliovp-bv-your-trusted-partner-for-supply-chain-resilience-amidst-new-u-s-tariffs","Eliovp BV: uw partner voor een veerkrachtige supply chain bij nieuwe Amerikaanse invoerheffingen","Het internationale handelslandschap verandert snel. Nieuwe Amerikaanse invoerheffingen zorgen voor onzekerheid, vooral bij bedrijven die afhankelijk zijn van high-performance computing en AI-infrastructuur. Eliovp helpt klanten hun supply chain veerkrachtig, betrouwbaar en kostenefficiënt te houden.","2025-04-04T10:01:27","eliovp-bv-your-trusted-partner-for-supply-chain-resilience-amidst-new-u-s-tariffs","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Ftariffsshipping.webp","https:\u002F\u002Feliovp.com\u002Feliovp-bv-your-trusted-partner-for-supply-chain-resilience-amidst-new-u-s-tariffs\u002F",[1998,2099,2000,2001,2354,2355,2356,2357],"importeren","Taiwan","Tarieven","Trump",{"path":2359,"title":2360,"description":2361,"date":2362,"slug":2363,"image":2364,"originalUrl":2365,"categories":2366},"\u002Fblog\u002Fwhy-ai-agents-are-the-future","Waarom AI-agenten de toekomst zijn","AI-agenten integreren met uw bestaande ERP-, CRM- en marketingplatforms. Ze vullen uw huidige systemen aan, nemen intelligentere beslissingen en schalen mee met uw organisatie. Zo verlaagt u kosten, beperkt u fouten en maakt u medewerkers vrij voor werk met meer toegevoegde waarde.","2025-03-23T22:06:59","why-ai-agents-are-the-future","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Ferp2.jpeg","https:\u002F\u002Feliovp.com\u002Fwhy-ai-agents-are-the-future\u002F",[1998,1999,2139,2000,2367,2368],"AI-agenten","ERP",{"path":2370,"title":2371,"description":2372,"date":2373,"slug":2374,"image":2375,"originalUrl":2376,"categories":2377},"\u002Fblog\u002Fthe-rise-of-open-source-ai-model-optimization","De opkomst van open-source AI-modeloptimalisatie","In het snel evoluerende AI-landschap zijn opensourceoplossingen belangrijke aanjagers van innovatie en betere prestaties. Deze door de gemeenschap gedragen platforms maken geavanceerde technologie breder toegankelijk, stimuleren samenwerking en versnellen de optimalisatie van AI-modellen.","2025-03-22T20:59:23","the-rise-of-open-source-ai-model-optimization","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Friseofopensource.jpeg","https:\u002F\u002Feliovp.com\u002Fthe-rise-of-open-source-ai-model-optimization\u002F",[1998,1999,2099,2378,2001,2065,2003],"AI-nieuws",{"path":2380,"title":2381,"description":2382,"date":2383,"slug":2384,"image":2385,"originalUrl":2386,"categories":2387},"\u002Fblog\u002Fintroducing-our-benchmarking-tool-powered-by-dstack","Maak kennis met onze benchmarktool, gebouwd op dstack","Betrouwbare benchmarks zijn essentieel om AI-modellen en software te optimaliseren. Onze nieuwe tool integreert met dstack en maakt geautomatiseerde, reproduceerbare en schaalbare benchmarkworkflows mogelijk, zowel op lokale GPU's als in de cloud.","2025-03-20T14:21:59","introducing-our-benchmarking-tool-powered-by-dstack","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fbenchmarktool.jpeg","https:\u002F\u002Feliovp.com\u002Fintroducing-our-benchmarking-tool-powered-by-dstack\u002F",[1998,1999,67,2000,2001,2388,2389,2253,67],"benchmark","LLM",{"path":2391,"title":2392,"description":2393,"date":2394,"slug":2395,"image":2396,"originalUrl":2397,"categories":2398},"\u002Fblog\u002Foptimizing-qwq-32b-by-qwen-amd-mi300x-vs-nvidia-h200","QwQ-32B optimaliseren (door Qwen): AMD MI300X versus NVIDIA H200","De meeste LLM-benchmarks richten zich op afgeleiden van Llama of DeepSeek. Met QwQ-32B verbreden we die vergelijking en testen we hoe Paiton de AMD MI300X optimaliseert tegenover NVIDIA's H200, van throughput en concurrency tot TTFT en end-to-end-latency.","2025-03-19T21:41:44","optimizing-qwq-32b-by-qwen-amd-mi300x-vs-nvidia-h200","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fpaiton4.jpeg","https:\u002F\u002Feliovp.com\u002Foptimizing-qwq-32b-by-qwen-amd-mi300x-vs-nvidia-h200\u002F",[1998,1999,67],{"path":2400,"title":2401,"description":2402,"date":2403,"slug":2404,"image":2405,"originalUrl":2406,"categories":2407},"\u002Fblog\u002Feliovp-featured-on-amd-tech-talk-podcast","Eliovp te gast in de AMD Tech Talk-podcast","Eliovp was onlangs te gast in de AMD Tech Talk-podcast. Onze CEO Elio Van Puyvelde sprak met Jim Greene over het ontstaan van Eliovp, de passie en expertise achter het bedrijf en de geïntegreerde oplossingen die we vandaag aanbieden.","2025-03-19T07:53:39","eliovp-featured-on-amd-tech-talk-podcast","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Ftechtalkjimgreene.jpeg","https:\u002F\u002Feliovp.com\u002Feliovp-featured-on-amd-tech-talk-podcast\u002F",[1998,2001,2408,2409,2410],"Jim Greene","Podcast","Tech Talk",{"path":2412,"title":2413,"description":2414,"date":2415,"slug":2416,"image":2417,"originalUrl":2418,"categories":2419},"\u002Fblog\u002Ffurther-optimizing-amd-powered-inference-with-paiton","AMD-inferentie verder optimaliseren met Paiton","Paiton richt zich volledig op inferentieoptimalisatie voor AMD. Ons recentste werk levert voor DeepSeek R1 Distill Llama 8B een 10 tot 15% hogere throughput, een betere time-to-first-token en stabielere prestaties bij kleinere batchgroottes.","2025-03-13T06:18:30","further-optimizing-amd-powered-inference-with-paiton","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fpaitonpost3.webp","https:\u002F\u002Feliovp.com\u002Ffurther-optimizing-amd-powered-inference-with-paiton\u002F",[1998,1999,67,2001,2420,2421,193,2253,2318,67,2004],"Deepseek","H100",{"path":2423,"title":2424,"description":2425,"date":2426,"slug":2427,"image":2428,"originalUrl":2429,"categories":2430},"\u002Fblog\u002Fa-first-look-at-paiton-in-action-deepseek-r1-distill-llama-3-1-8b","Een eerste blik op Paiton in actie: Deepseek R1 Distill Llama 3.1 8B","Een eerste blik op hoe Paiton Deepseek R1 Distill Llama 3.1 8B optimaliseert voor de AMD MI300X. We vergelijken een met Paiton geoptimaliseerd model met de standaardversie en meten throughput, latency en end-to-end prestaties bij uiteenlopende batchgroottes.","2025-01-31T09:11:02","a-first-look-at-paiton-in-action-deepseek-r1-distill-llama-3-1-8b","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fpaitonpost2.webp","https:\u002F\u002Feliovp.com\u002Fa-first-look-at-paiton-in-action-deepseek-r1-distill-llama-3-1-8b\u002F",[1998,1999,67,2001,2420,2421,193,2253,2318,67,2004],{"path":2432,"title":2433,"description":2434,"date":2435,"slug":2436,"image":2437,"originalUrl":2438,"categories":2439},"\u002Fblog\u002Fai-model-optimization-with-paiton","AI-modeloptimalisatie met Paiton","In de snel evoluerende wereld van artificiële intelligentie zijn efficiënte modellen en sterke prestaties cruciaal. Met Paiton biedt ElioVP doorgedreven optimalisatie voor AI-modellen. Door de modelarchitectuur te compileren en onze eigen kernels te gebruiken, maakt Paiton snellere inferentie en een lager resourcegebruik op AMD GPU's mogelijk.","2025-01-30T19:53:25","ai-model-optimization-with-paiton","\u002Fasset\u002Fimages\u002Fblog\u002Fimported\u002Fpaitonpost1.webp","https:\u002F\u002Feliovp.com\u002Fai-model-optimization-with-paiton\u002F",[1998,1999,67,2001,2421,193,2253,2318,67,2004],1788530428990]