{"id":61968,"date":"2021-03-23T20:19:24","date_gmt":"2021-03-23T19:19:24","guid":{"rendered":"https:\/\/www.planet3dnow.de\/cms\/?p=61968"},"modified":"2021-03-23T21:26:08","modified_gmt":"2021-03-23T20:26:08","slug":"announcing-oracle-cloud-compute-e4-platform-on-third-gen-amd-epyc-processors","status":"publish","type":"post","link":"https:\/\/www.planet3dnow.de\/cms\/61968-announcing-oracle-cloud-compute-e4-platform-on-third-gen-amd-epyc-processors\/","title":{"rendered":"Announcing Oracle Cloud Compute <span class=\"caps\">E4<\/span> platform on third gen <span class=\"caps\">AMD<\/span> <span class=\"caps\">EPYC<\/span> processors"},"content":{"rendered":"<p class=\"selectionShareable\">At Ora\u00adcle Open\u00adWorld 2018, we announ\u00adced a stra\u00adte\u00adgic part\u00adner\u00adship with <span class=\"caps\">AMD<\/span> and laun\u00adched the Ora\u00adcle <span class=\"caps\">E2<\/span> plat\u00adform on the first gene\u00adra\u00adti\u00adon of <span class=\"caps\">AMD<\/span> <span class=\"caps\">EPYC<\/span> pro\u00adces\u00adsors. Sin\u00adce that launch, we\u2019ve deploy\u00aded Ora\u00adcle <span class=\"caps\">E2<\/span> Com\u00adpu\u00adte ins\u00adtances to all of our com\u00admer\u00adcial regi\u00adons. Eigh\u00adte\u00aden months later, in April 2020, we built on this part\u00adner\u00adship and laun\u00adched the <span class=\"caps\">E3<\/span> plat\u00adform on the second gene\u00adra\u00adti\u00adon. Many of our enter\u00adpri\u00adse cus\u00adto\u00admers and Ora\u00adcle Cloud appli\u00adca\u00adti\u00adons are run\u00adning gene\u00adral-pur\u00adpo\u00adse workloads on <span class=\"caps\">AMD<\/span> <span class=\"caps\">EPYC<\/span> pro\u00adces\u00adsors. Today, I\u2019m hap\u00adpy to announ\u00adce that Ora\u00adcle is laun\u00adching the <span class=\"caps\">E4<\/span> plat\u00adform based on third-gene\u00adra\u00adti\u00adon <span class=\"caps\">AMD<\/span> <span class=\"caps\">EPYC<\/span> pro\u00adces\u00adsors. The <span class=\"caps\">E4<\/span> plat\u00adform includes both bare metal and fle\u00adxi\u00adble vir\u00adtu\u00adal machi\u00adnes (VMs).<\/p>\n<div class=\"adwcsengine\" data-id=\"NONE PROVIDED\" data-adgroupid=\"NONE PROVIDED\">&nbsp;<\/div>\n<p class=\"selectionShareable\">The\u00adse <span class=\"caps\">E4<\/span> stan\u00addard ins\u00adtances use 64 core pro\u00adces\u00adsors, with a base clock fre\u00adquen\u00adcy of 2.55 GHz and a max boost of up to 3.5 GHz. The bare metal <span class=\"caps\">E4<\/span> stan\u00addard Com\u00adpu\u00adte ins\u00adtance sup\u00adports 128 OCPUs (128 cores and 256 threads) with 256 <span class=\"caps\">MB<\/span> of <span class=\"caps\">L3<\/span> cache, 2 <span class=\"caps\">TB<\/span> of <span class=\"caps\">RAM<\/span>, and 100 Gbps of over\u00adall net\u00adwork band\u00adwidth. This con\u00adfi\u00adgu\u00adra\u00adti\u00adon is the hig\u00adhest core count for a bare metal ins\u00adtance on any public cloud. The memo\u00adry band\u00adwidth is well sui\u00adted for both gene\u00adral-pur\u00adpo\u00adse and high-band\u00adwidth workloads that requi\u00adre lar\u00adger and fas\u00adter memory.<\/p>\n<p class=\"selectionShareable\">As demons\u00adtra\u00adted in the fol\u00adlo\u00adwing per\u00adfor\u00admance bench\u00admarks, the <span class=\"caps\">E4<\/span> Stan\u00addard ins\u00adtances deli\u00adver up to a 15% increase in inte\u00adger per\u00adfor\u00admance, a 21% increase in floa\u00adting point per\u00adfor\u00admance, and a 24% increase in java per\u00adfor\u00admance, com\u00adpared to <span class=\"caps\">E3<\/span> Stan\u00addard ins\u00adtances. Also, the <span class=\"caps\">E4<\/span> ins\u00adtances pro\u00advi\u00adde three times the pri\u00adce per\u00adfor\u00admance rela\u00adti\u00adve to other gene\u00adral-pur\u00adpo\u00adse ins\u00adtances offe\u00adred by other cloud providers.<\/p>\n<p class=\"selectionShareable\">New pro\u00adces\u00adsors, bet\u00adter per\u00adfor\u00admance, and the same pri\u00adce as <span class=\"caps\">E3<\/span> Com\u00adpu\u00adte ins\u00adtances, com\u00adbi\u00adned with our fle\u00adxi\u00adble Com\u00adpu\u00adte approach, which enables you to gra\u00adnu\u00adlar\u00adly cus\u00adto\u00admi\u00adze the core counts and memo\u00adry of VMs, pro\u00advi\u00adde bet\u00adter&nbsp;value.<\/p>\n<p class=\"selectionShareable\">Many enter\u00adpri\u00adse cus\u00adto\u00admers for Ora\u00adcle appli\u00adca\u00adti\u00adons and gene\u00adral-pur\u00adpo\u00adse workloads have adopted our <span class=\"caps\">AMD<\/span> <span class=\"caps\">EPYC<\/span> pro\u00adces\u00adsor-based Com\u00adpu\u00adte ins\u00adtances. Our cus\u00adto\u00admers have alre\u00ada\u00addy suc\u00adcessful\u00adly deploy\u00aded AMD-based ins\u00adtances, inclu\u00adding the fol\u00adlo\u00adwing use&nbsp;cases:<\/p>\n<ul>\n<li>\n<p class=\"selectionShareable\">Video con\u00adfe\u00adren\u00adcing<\/p>\n<\/li>\n<li>\n<p class=\"selectionShareable\">Real-time video processing<\/p>\n<\/li>\n<li>\n<p class=\"selectionShareable\">Mas\u00adsi\u00adve\u00adly par\u00adal\u00adlel pro\u00adces\u00adsing and high-per\u00adfor\u00admance com\u00adpu\u00adting (<span class=\"caps\">HPC<\/span>)<\/p>\n<\/li>\n<li>\n<p class=\"selectionShareable\">Busi\u00adness-cri\u00adti\u00adcal applications<\/p>\n<\/li>\n<li>\n<p class=\"selectionShareable\">Web and appli\u00adca\u00adti\u00adon servers<\/p>\n<\/li>\n<li>\n<p class=\"selectionShareable\">Backend ser\u00advers for enter\u00adpri\u00adse applications<\/p>\n<\/li>\n<li>\n<p class=\"selectionShareable\">Gam\u00ading servers<\/p>\n<\/li>\n<li>\n<p class=\"selectionShareable\">In-memo\u00adry database<\/p>\n<\/li>\n<li>\n<p class=\"selectionShareable\">Caching fleets<\/p>\n<\/li>\n<li>\n<p class=\"selectionShareable\">App deve\u00adlo\u00adp\u00adment environments<\/p>\n<\/li>\n<\/ul>\n<p class=\"selectionShareable\">Ora\u00adcle cus\u00adto\u00admer Xact\u00adly, the lea\u00adder in reve\u00adnue intel\u00adli\u00adgence solu\u00adti\u00adons, uses Oracle\u2019s <span class=\"caps\">AMD<\/span> <span class=\"caps\">EPYC<\/span> pro\u00adces\u00adsor-based Compute.<\/p>\n<p class=\"selectionShareable\"><span class=\"dquo\">\u201c<\/span>Ora\u00adcle Cloud Infra\u00adstruc\u00adtu\u00adre (<span class=\"caps\">OCI<\/span>) is uni\u00adque in pro\u00advi\u00adding cus\u00adto\u00admers with the gra\u00adnu\u00adla\u00adri\u00adty and fle\u00adxi\u00adbi\u00adli\u00adty in com\u00adpu\u00adte resour\u00adces,\u201d said Ron Ras\u00admus\u00adsen, <span class=\"caps\">CTO<\/span> of Xact\u00adly. \u201cXact\u00adly has seen gre\u00adat, mea\u00adsu\u00adred bene\u00adfits from deploy\u00ading <span class=\"caps\">E3<\/span> in ope\u00adra\u00adting our mul\u00adti-ten\u00adant SaaS reve\u00adnue intel\u00adli\u00adgence solu\u00adti\u00adons. Our mul\u00adti\u00adten\u00adant SaaS solu\u00adti\u00adon is built on Java tech\u00adno\u00adlo\u00adgy, and we look for\u00adward to exten\u00adding the addi\u00adtio\u00adnal bene\u00adfits of <span class=\"caps\">E4<\/span> to our cus\u00adto\u00admers in the near future.\u201d<\/p>\n<h2 class=\"selectionShareable\">Key highlights<\/h2>\n<p class=\"selectionShareable\"><span class=\"caps\">E4<\/span> ins\u00adtances con\u00adti\u00adnue the fle\u00adxi\u00adble infra\u00adstruc\u00adtu\u00adre approach that we estab\u00adlished with <span class=\"caps\">E3<\/span>. You\u2019re free to sel\u00adect the exact num\u00adber of OCPUs and amount of memo\u00adry that you need for a <span class=\"caps\">VM<\/span>, not forced to choo\u00adse from a fixed menu of 1, 2, 4, 8, or 16. You can launch any cus\u00adtom <span class=\"caps\">VM<\/span> size that meets your needs, such as a 3\u2011core, 6\u2011core, or 63-core <span class=\"caps\">VM<\/span> with any\u00adwhe\u00adre from 1 <span class=\"caps\">GB<\/span>\u20131 <span class=\"caps\">TB<\/span> of memory.<\/p>\n<p class=\"selectionShareable\">The\u00adse cus\u00adtom sizes mean lower cos\u00adts. <span class=\"caps\">E4<\/span> ins\u00adtances bill sepa\u00adra\u00adte\u00adly for the <span class=\"caps\">OCPU<\/span> and memo\u00adry resour\u00adces pro\u00advi\u00adsio\u00adned. Each <span class=\"caps\">OCPU<\/span> comes with its asso\u00adcia\u00adted simul\u00adta\u00adneous mul\u00adti\u00adth\u00adre\u00ada\u00adding unit and is pri\u00adced at <span class=\"caps\">US<\/span>$0.025, and memo\u00adry is pri\u00adced at <span class=\"caps\">US<\/span>$0.0015 per <span class=\"caps\">GB<\/span>\u2014the same pri\u00adces as our <span class=\"caps\">E3<\/span> Com\u00adpu\u00adte ins\u00adtances. An <span class=\"caps\">E4<\/span> ins\u00adtance with 1\u2011<span class=\"caps\">OCPU<\/span> ins\u00adtance and 16 <span class=\"caps\">GB<\/span> of <span class=\"caps\">RAM<\/span>&nbsp; has a pri\u00adce of <span class=\"caps\">US<\/span>$0.049, which is 23% less than <span class=\"caps\">X7<\/span> Stan\u00addard and 57\u201361% less than com\u00adpa\u00adra\u00adble ins\u00adtances offe\u00adred by other clouds.<\/p>\n<p class=\"selectionShareable\"><span class=\"caps\">E4<\/span> ins\u00adtances sup\u00adport reboot resi\u00adze, so you can easi\u00adly migra\u00adte from your exis\u00adting com\u00adpu\u00adte ins\u00adtances to an <span class=\"caps\">E4<\/span> ins\u00adtance of com\u00adpa\u00adti\u00adble&nbsp;shape.<\/p>\n<h2 class=\"selectionShareable\">Instance availability<\/h2>\n<p class=\"selectionShareable\">Table 1 shows the <span class=\"caps\">VM<\/span> and the bare metal ins\u00adtances that are available at launch. The\u00adse ins\u00adtances are available in the fol\u00adlo\u00adwing regi\u00adons, with plans to have ins\u00adtances wide\u00adly available in all com\u00admer\u00adcial regi\u00adons in the future:<\/p>\n<ul>\n<li>\n<p class=\"selectionShareable\"><span class=\"caps\">US<\/span> East (Ashb\u00adurn)<\/p>\n<\/li>\n<li>\n<p class=\"selectionShareable\"><span class=\"caps\">US<\/span> West (Phoe\u00adnix)<\/p>\n<\/li>\n<li>\n<p class=\"selectionShareable\">India west (Mum\u00adbai)<\/p>\n<\/li>\n<li>\n<p class=\"selectionShareable\">Switz\u00ader\u00adland North (Zurich)<\/p>\n<\/li>\n<li>\n<p class=\"selectionShareable\">Bra\u00adzil East (Sao&nbsp;Paulo)<\/p>\n<\/li>\n<li>\n<p class=\"selectionShareable\">Cana\u00adda Sou\u00adthe\u00adast (Mont\u00adre\u00adal)<\/p>\n<\/li>\n<li>\n<p class=\"selectionShareable\">Aus\u00adtra\u00adlia sou\u00adthe\u00adast (Mel\u00adbourne)<\/p>\n<\/li>\n<li>\n<p class=\"selectionShareable\">Cana\u00adda Sou\u00adthe\u00adast (Toron\u00adto)<\/p>\n<\/li>\n<\/ul>\n<p class=\"selectionShareable\"><strong>Table 1: <span class=\"caps\">E4<\/span> ins\u00adtances available at launch<\/strong><\/p>\n<table border=\"1\" width=\"544\" cellspacing=\"1\" cellpadding=\"1\">\n<tbody>\n<tr>\n<td>Ins\u00adtance<\/td>\n<td><span class=\"caps\">OCPU<\/span><\/td>\n<td>Memo\u00adry<\/td>\n<td>Extra sto\u00adrage<\/td>\n<td>Net\u00adwork<\/td>\n<\/tr>\n<tr>\n<td><span class=\"caps\">BM<\/span>.Standard.<span class=\"caps\">E4<\/span>.128<\/td>\n<td>128<\/td>\n<td>2,048 <span class=\"caps\">GB<\/span><\/td>\n<td>Up to 1 <span class=\"caps\">PB<\/span> of remo\u00adte block storage<\/td>\n<td>100 Gbps<\/td>\n<\/tr>\n<tr>\n<td><span class=\"caps\">VM<\/span>.Standard.<span class=\"caps\">E4<\/span>.Flex<\/td>\n<td>1\u201364<\/td>\n<td>1\u201364 <span class=\"caps\">GB<\/span> per <span class=\"caps\">OCPU<\/span> up to 1024&nbsp;<span class=\"caps\">GB<\/span><\/td>\n<td>Up to 1 <span class=\"caps\">PB<\/span> of remo\u00adte block storage<\/td>\n<td>1 Gbps per&nbsp;<span class=\"caps\">OCPU<\/span><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p class=\"selectionShareable\"><span class=\"caps\">OCPU<\/span> refers to a sin\u00adgle core and its asso\u00adcia\u00adted thread. We pro\u00advi\u00adde 1 Gbps per <span class=\"caps\">OCPU<\/span>, up to a maxi\u00admum of 40&nbsp;Gbps.<\/p>\n<h2 class=\"selectionShareable\">Performance<\/h2>\n<p class=\"selectionShareable\">We com\u00adpared the <span class=\"caps\">AMD<\/span> <span class=\"caps\">EPYC<\/span> CPU-based ins\u00adtances to our cur\u00adrent x86 stan\u00addard alter\u00adna\u00adti\u00adves. Table 2 shows detail\u00aded configurations.<\/p>\n<p class=\"selectionShareable\"><strong>Table 2: Com\u00adpu\u00adte con\u00adfi\u00adgu\u00adra\u00adti\u00adons for per\u00adfor\u00admance&nbsp;tests<\/strong><\/p>\n<table border=\"1\" cellspacing=\"1\" cellpadding=\"1\">\n<tbody>\n<tr>\n<td>Sys\u00adtem Configuration<\/td>\n<td><span class=\"caps\">X7<\/span> Sys\u00adtem<\/td>\n<td><span class=\"caps\">E3<\/span> Sys\u00adtem<\/td>\n<td><span class=\"caps\">E4<\/span> Sys\u00adtem<\/td>\n<\/tr>\n<tr>\n<td><span class=\"caps\">CPU<\/span><\/td>\n<td>Two x86 pro\u00adces\u00adsors, 26 cores per socket @ 2.0&nbsp;GHz<\/td>\n<td>Two <span class=\"caps\">AMD<\/span> <span class=\"caps\">EPYC<\/span> 7742, 64 cores per socket @ 2.25 GHz Base, and @3.<span class=\"caps\">4G<\/span>&nbsp;Turbo<\/td>\n<td>Two <span class=\"caps\">AMD<\/span> <span class=\"caps\">EPYC<\/span> <span class=\"caps\">7J13<\/span>, 64 cores per socket @ 2.55 GHz Base, and @3.<span class=\"caps\">5G<\/span>&nbsp;Turbo<\/td>\n<\/tr>\n<tr>\n<td>Memo\u00adry<\/td>\n<td>786 <span class=\"caps\">GB<\/span>&nbsp;<span class=\"caps\">DDR4<\/span><\/td>\n<td>2,048 <span class=\"caps\">GB<\/span>&nbsp;<span class=\"caps\">DDR4<\/span><\/td>\n<td>2,048 <span class=\"caps\">GB<\/span>&nbsp;<span class=\"caps\">DDR4<\/span><\/td>\n<\/tr>\n<tr>\n<td>Net\u00adwork<\/td>\n<td>Two 25&nbsp;Gbps<\/td>\n<td>Two 50&nbsp;Gbps<\/td>\n<td>Two 50&nbsp;Gbps<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p class=\"selectionShareable\">We ran per\u00adfor\u00admance tests to exer\u00adcise the <span class=\"caps\">CPU<\/span> per\u00adfor\u00admance, floa\u00adting point per\u00adfor\u00admance, and memo\u00adry sub\u00adsys\u00adtem per\u00adfor\u00admance (see Table 3). We ran tests on ven\u00addor-recom\u00admen\u00added pro\u00adprie\u00adta\u00adry com\u00adpi\u00adlers and Ora\u00adcle Linux ope\u00adra\u00adting sys\u00adtems. We ran the tests seve\u00adral times and aver\u00ada\u00adged the results. All <span class=\"caps\">SPEC<\/span> num\u00adbers are estimates.<\/p>\n<p class=\"selectionShareable\"><strong>Table 3: Per\u00adfor\u00admance tests and bench\u00admark targets<\/strong><\/p>\n<table border=\"1\" cellspacing=\"1\" cellpadding=\"1\">\n<tbody>\n<tr>\n<td>Test<\/td>\n<td>Bench\u00admark target<\/td>\n<\/tr>\n<tr>\n<td>SPE\u00adCra\u00adte 2017 Integer<\/td>\n<td>Inte\u00adger performance<\/td>\n<\/tr>\n<tr>\n<td>SPE\u00adCra\u00adte 2017 Floa\u00adting&nbsp;Point<\/td>\n<td>Floa\u00adting point performance<\/td>\n<\/tr>\n<tr>\n<td><span class=\"caps\">STREAM<\/span> Tri\u00adad<\/td>\n<td>Memo\u00adry sub\u00adsys\u00adtem performance<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p class=\"selectionShareable\">The fol\u00adlo\u00adwing graphs show how the <span class=\"caps\">AMD<\/span> <span class=\"caps\">E4<\/span> ins\u00adtances com\u00adpared against our <span class=\"caps\">X7<\/span> and <span class=\"caps\">E3<\/span> Com\u00adpu\u00adte instances.<\/p>\n<figure id=\"attachment_62091\" aria-describedby=\"caption-attachment-62091\" style=\"width: 467px\" class=\"wp-caption aligncenter\"><a href=\"https:\/\/www.planet3dnow.de\/cms\/wp-content\/uploads\/2021\/03\/e4_blog1.jpg\"><img loading=\"lazy\" decoding=\"async\" class=\"size-full wp-image-62091\" src=\"https:\/\/www.planet3dnow.de\/cms\/wp-content\/uploads\/2021\/03\/e4_blog1.jpg\" alt width=\"467\" height=\"214\" srcset=\"https:\/\/www.planet3dnow.de\/cms\/wp-content\/uploads\/2021\/03\/e4_blog1.jpg 467w, https:\/\/www.planet3dnow.de\/cms\/wp-content\/uploads\/2021\/03\/e4_blog1-300x137.jpg 300w\" sizes=\"auto, (max-width: 467px) 100vw, 467px\"><\/a><figcaption id=\"caption-attachment-62091\" class=\"wp-caption-text\">Figu\u00adre 1: Nor\u00adma\u00adli\u00adzed bare metal per\u00adfor\u00admance results<\/figcaption><\/figure>\n<p class=\"selectionShareable\">The third-gene\u00adra\u00adti\u00adon <span class=\"caps\">AMD<\/span> <span class=\"caps\">EPYC<\/span> pro\u00adces\u00adsor-based ser\u00adver per\u00adforms well com\u00adpared to our exis\u00adting port\u00adfo\u00adlio of bare metal Com\u00adpu\u00adte ins\u00adtances. The <span class=\"caps\">E4<\/span> Stan\u00addard ins\u00adtances deli\u00adver up to a 15% increase in inte\u00adger per\u00adfor\u00admance and a 9% increase in floa\u00adting point per\u00adfor\u00admance com\u00adpared to the <span class=\"caps\">E3<\/span> ins\u00adtances. Com\u00adpared to the <span class=\"caps\">X7<\/span> stan\u00addard ins\u00adtance, the <span class=\"caps\">E4<\/span> ins\u00adtance deli\u00advers a 215% increase in inte\u00adger per\u00adfor\u00admance, a 160% increase in floa\u00adting point per\u00adfor\u00admance, and a 140% increase in <span class=\"caps\">STREAM<\/span> Tri\u00adad bandwidth.<\/p>\n<p class=\"selectionShareable\">Figu\u00adre 2 shows the results of the bench\u00admarks on VMs. All the tests were run on 2\u2011core VMs.<\/p>\n<figure id=\"attachment_62090\" aria-describedby=\"caption-attachment-62090\" style=\"width: 409px\" class=\"wp-caption aligncenter\"><a href=\"https:\/\/www.planet3dnow.de\/cms\/wp-content\/uploads\/2021\/03\/blog_pic2.jpg\"><img loading=\"lazy\" decoding=\"async\" class=\"size-full wp-image-62090\" src=\"https:\/\/www.planet3dnow.de\/cms\/wp-content\/uploads\/2021\/03\/blog_pic2.jpg\" alt width=\"409\" height=\"232\" srcset=\"https:\/\/www.planet3dnow.de\/cms\/wp-content\/uploads\/2021\/03\/blog_pic2.jpg 409w, https:\/\/www.planet3dnow.de\/cms\/wp-content\/uploads\/2021\/03\/blog_pic2-300x170.jpg 300w\" sizes=\"auto, (max-width: 409px) 100vw, 409px\"><\/a><figcaption id=\"caption-attachment-62090\" class=\"wp-caption-text\">Figu\u00adre 2: Nor\u00adma\u00adli\u00adzed <span class=\"caps\">VM<\/span> per\u00adfor\u00admance results<\/figcaption><\/figure>\n<p class=\"selectionShareable\">The ser\u00adver also per\u00adforms well com\u00adpared to our exis\u00adting <span class=\"caps\">VM<\/span> ins\u00adtances on this set of stan\u00addar\u00addi\u00adzed bench\u00admarks. The <span class=\"caps\">E4<\/span> stan\u00addard ins\u00adtances deli\u00adver\u00aded a 15% impro\u00adve\u00adment in inte\u00adger per\u00adfor\u00admance per core and 21% impro\u00adve\u00adment in floa\u00adting point per\u00adfor\u00admance per core com\u00adpared to <span class=\"caps\">E3<\/span> based ins\u00adtances. More nota\u00adble is the com\u00adpa\u00adri\u00adson against <span class=\"caps\">X7<\/span> ins\u00adtances. The <span class=\"caps\">E4<\/span> stan\u00addard ins\u00adtances deli\u00adver\u00aded a 67% impro\u00adve\u00adment in inte\u00adger per\u00adfor\u00admance, 97% impro\u00adve\u00adment in Floa\u00adting point per\u00adfor\u00admance, and a 108% increase in memo\u00adry per\u00adfor\u00admance over an <span class=\"caps\">X7<\/span> instance.<\/p>\n<h2 class=\"selectionShareable\">Java benchmarks<\/h2>\n<p class=\"selectionShareable\">Apart from the stan\u00addar\u00addi\u00adzed bench\u00admarks, we also ran our inter\u00adnal Java bench\u00admark to mea\u00adsu\u00adre ser\u00adver-side Java per\u00adfor\u00admance. We ran the Java tests with an empha\u00adsis on the midd\u00adle-tier per\u00adfor\u00admance. We ran the tests seve\u00adral times and aver\u00ada\u00adged the results.<\/p>\n<p class=\"selectionShareable\"><strong>Table 4: Java per\u00adfor\u00admance test and bench\u00admark target<\/strong><\/p>\n<table border=\"1\" cellspacing=\"1\" cellpadding=\"1\">\n<tbody>\n<tr>\n<td>Test<\/td>\n<td>Bench\u00admark target<\/td>\n<\/tr>\n<tr>\n<td>Ser\u00adver-Side Java Benchmark<\/td>\n<td>Midd\u00adle-tier performance<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p class=\"selectionShareable\">Figu\u00adre 3 shows the results of run\u00adning the Java bench\u00admark tests on <span class=\"caps\">X7<\/span>, <span class=\"caps\">E3<\/span>, and <span class=\"caps\">E4<\/span> bare metal instances.<\/p>\n<figure id=\"attachment_62089\" aria-describedby=\"caption-attachment-62089\" style=\"width: 406px\" class=\"wp-caption aligncenter\"><a href=\"https:\/\/www.planet3dnow.de\/cms\/wp-content\/uploads\/2021\/03\/jops_pic1.jpg\"><img loading=\"lazy\" decoding=\"async\" class=\"size-full wp-image-62089\" src=\"https:\/\/www.planet3dnow.de\/cms\/wp-content\/uploads\/2021\/03\/jops_pic1.jpg\" alt width=\"406\" height=\"271\" srcset=\"https:\/\/www.planet3dnow.de\/cms\/wp-content\/uploads\/2021\/03\/jops_pic1.jpg 406w, https:\/\/www.planet3dnow.de\/cms\/wp-content\/uploads\/2021\/03\/jops_pic1-300x200.jpg 300w\" sizes=\"auto, (max-width: 406px) 100vw, 406px\"><\/a><figcaption id=\"caption-attachment-62089\" class=\"wp-caption-text\">Figu\u00adre 3: Nor\u00adma\u00adli\u00adzed bare metal per\u00adfor\u00admance results<\/figcaption><\/figure>\n<p class=\"selectionShareable\">The <span class=\"caps\">E4<\/span> Stan\u00addard ins\u00adtance deli\u00adver\u00aded a 13% increase in Java per\u00adfor\u00admance com\u00adpared to <span class=\"caps\">E3<\/span> ins\u00adtances and a 263% increase com\u00adpared to <span class=\"caps\">X7<\/span> instances.<\/p>\n<p class=\"selectionShareable\">Figu\u00adre 4 shows the results of run\u00adning the Java bench\u00admark tests on VMs. All the Java tests were run on 8\u2011core VMs becau\u00adse of mini\u00admum memo\u00adry requirements.<\/p>\n<figure id=\"attachment_62088\" aria-describedby=\"caption-attachment-62088\" style=\"width: 403px\" class=\"wp-caption aligncenter\"><a href=\"https:\/\/www.planet3dnow.de\/cms\/wp-content\/uploads\/2021\/03\/jops_pic2.jpg\"><img loading=\"lazy\" decoding=\"async\" class=\"size-full wp-image-62088\" src=\"https:\/\/www.planet3dnow.de\/cms\/wp-content\/uploads\/2021\/03\/jops_pic2.jpg\" alt width=\"403\" height=\"216\" srcset=\"https:\/\/www.planet3dnow.de\/cms\/wp-content\/uploads\/2021\/03\/jops_pic2.jpg 403w, https:\/\/www.planet3dnow.de\/cms\/wp-content\/uploads\/2021\/03\/jops_pic2-300x161.jpg 300w\" sizes=\"auto, (max-width: 403px) 100vw, 403px\"><\/a><figcaption id=\"caption-attachment-62088\" class=\"wp-caption-text\">Figu\u00adre 4: Nor\u00adma\u00adli\u00adzed <span class=\"caps\">VM<\/span> per\u00adfor\u00admance results<\/figcaption><\/figure>\n<p class=\"selectionShareable\">The <span class=\"caps\">E4<\/span> <span class=\"caps\">VM<\/span> ins\u00adtances deli\u00adver\u00aded a 24% increase in Java per\u00adfor\u00admance com\u00adpared to <span class=\"caps\">E3<\/span> and a 71% increase in Java per\u00adfor\u00admance com\u00adpared to <span class=\"caps\">X7<\/span> instances.<\/p>\n<h2 class=\"selectionShareable\">Competitive benchmarks<\/h2>\n<p class=\"selectionShareable\">We com\u00adpared our offe\u00adrings to what\u2019s available from <span class=\"caps\">AWS<\/span>. As shown in the fol\u00adlo\u00adwing gra\u00adphic, the <span class=\"caps\">E4<\/span> based ins\u00adtances out\u00adper\u00adfor\u00admed the <span class=\"caps\">AWS<\/span> ins\u00adtances in abso\u00adlu\u00adte per\u00adfor\u00admance and pri\u00adce-per\u00adfor\u00admance, with three times the pri\u00adce per\u00adfor\u00admance of <span class=\"caps\">AWS<\/span> ins\u00adtances across the\u00adse benchmarks.<\/p>\n<figure id=\"attachment_62087\" aria-describedby=\"caption-attachment-62087\" style=\"width: 360px\" class=\"wp-caption aligncenter\"><a href=\"https:\/\/www.planet3dnow.de\/cms\/wp-content\/uploads\/2021\/03\/aws_compare1.jpg\"><img loading=\"lazy\" decoding=\"async\" class=\"size-full wp-image-62087\" src=\"https:\/\/www.planet3dnow.de\/cms\/wp-content\/uploads\/2021\/03\/aws_compare1.jpg\" alt width=\"360\" height=\"251\" srcset=\"https:\/\/www.planet3dnow.de\/cms\/wp-content\/uploads\/2021\/03\/aws_compare1.jpg 360w, https:\/\/www.planet3dnow.de\/cms\/wp-content\/uploads\/2021\/03\/aws_compare1-300x209.jpg 300w\" sizes=\"auto, (max-width: 360px) 100vw, 360px\"><\/a><figcaption id=\"caption-attachment-62087\" class=\"wp-caption-text\">Figu\u00adre 5: Per\u00adfor\u00admance of <span class=\"caps\">OCI<\/span> <span class=\"caps\">E4<\/span> ver\u00adsus <span class=\"caps\">AWS<\/span> instances<\/figcaption><\/figure>\n<figure id=\"attachment_62086\" aria-describedby=\"caption-attachment-62086\" style=\"width: 574px\" class=\"wp-caption aligncenter\"><a href=\"https:\/\/www.planet3dnow.de\/cms\/wp-content\/uploads\/2021\/03\/aws_compare2.jpg\"><img loading=\"lazy\" decoding=\"async\" class=\"size-full wp-image-62086\" src=\"https:\/\/www.planet3dnow.de\/cms\/wp-content\/uploads\/2021\/03\/aws_compare2.jpg\" alt width=\"574\" height=\"253\" srcset=\"https:\/\/www.planet3dnow.de\/cms\/wp-content\/uploads\/2021\/03\/aws_compare2.jpg 574w, https:\/\/www.planet3dnow.de\/cms\/wp-content\/uploads\/2021\/03\/aws_compare2-300x132.jpg 300w\" sizes=\"auto, (max-width: 574px) 100vw, 574px\"><\/a><figcaption id=\"caption-attachment-62086\" class=\"wp-caption-text\">Figu\u00adre 6: Pri\u00adce per\u00adfor\u00admance of <span class=\"caps\">OCI<\/span> <span class=\"caps\">E4<\/span> ver\u00adsus&nbsp;<span class=\"caps\">AWS<\/span><\/figcaption><\/figure>\n<h2 class=\"selectionShareable\">Conclusion<\/h2>\n<p class=\"selectionShareable\">You can bene\u00adfit from this per\u00adfor\u00admance by deploy\u00ading and migra\u00adting your workloads to the\u00adse Com\u00adpu\u00adte ins\u00adtances today. To learn more about <span class=\"caps\">AMD<\/span> <span class=\"caps\">EPYC<\/span> cloud solu\u00adti\u00adons for <span class=\"caps\">OCI<\/span>,&nbsp;<a href=\"https:\/\/youtu.be\/X3nBIBbqqG4\" target=\"_blank\" rel=\"noopener\" name=\"cta=External Link Click\">watch the video<\/a>&nbsp;from our <span class=\"caps\">VP<\/span> Ross Brown. For more infor\u00adma\u00adti\u00adon on the <span class=\"caps\">E4<\/span> Stan\u00addard and fle\u00adxi\u00adble ins\u00adtances, you can access&nbsp;<a href=\"https:\/\/docs.oracle.com\/en-us\/iaas\/Content\/Compute\/References\/computeshapes.htm\" target=\"_blank\" rel=\"noopener\" name=\"cta=External Link Click\">our docu\u00admen\u00adta\u00adti\u00adon<\/a>. For more Ora\u00adcle Cloud Infra\u00adstruc\u00adtu\u00adre solu\u00adti\u00adons for <span class=\"caps\">AMD<\/span> EPYC-powered Ora\u00adcle Cloud ins\u00adtances, you can visit the&nbsp;<a href=\"http:\/\/www.amd.com\/cloud\/oracle\/solutions\" target=\"_blank\" rel=\"noopener\" name=\"cta=External Link Click\"><span class=\"caps\">AMD<\/span> tech docs and white paper libra\u00adry<\/a>.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>At Ora\u00adcle Open\u00adWorld 2018, we announ\u00adced a stra\u00adte\u00adgic part\u00adner\u00adship with <span class=\"caps\">AMD<\/span> and laun\u00adched the Ora\u00adcle <span class=\"caps\">E2<\/span> plat\u00adform on the first gene\u00adra\u00adti\u00adon of <span class=\"caps\">AMD<\/span> <span class=\"caps\">EPYC<\/span> pro\u00adces\u00adsors. Sin\u00adce that launch, we\u2019ve deploy\u00aded Ora\u00adcle <span class=\"caps\">E2<\/span> Com\u00adpu\u00adte ins\u00adtances to all of our com\u00admer\u00adcial regi\u00adons. Eigh\u00adte\u00aden months later, in April 2020, we built on this part\u00adner\u00adship and laun\u00adched the <span class=\"caps\">E3<\/span> plat\u00adform on the second gene\u00adra\u00adti\u00adon. Many of our enter\u00adpri\u00adse cus\u00adto\u00admers and Ora\u00adcle Cloud appli\u00adca\u00adti\u00adons are run\u00adning gene\u00adral-pur\u00adpo\u00adse workloads on <span class=\"caps\">AMD<\/span> <span class=\"caps\">EPYC<\/span> pro\u00adces\u00adsors. Today, I\u2019m hap\u00adpy to announ\u00adce that Ora\u00adcle is laun\u00adching the <span class=\"caps\">E4<\/span> plat\u00adform based on third-gene\u00adra\u00adti\u00adon <span class=\"caps\">AMD<\/span> <span class=\"caps\">EPYC<\/span> pro\u00adces\u00adsors. The <span class=\"caps\">E4<\/span> plat\u00adform includes both bare metal and fle\u00adxi\u00adble vir\u00adtu\u00adal machi\u00adnes (VMs). (\u2026) <a class=\"moretag\" href=\"https:\/\/www.planet3dnow.de\/cms\/61968-announcing-oracle-cloud-compute-e4-platform-on-third-gen-amd-epyc-processors\/\">Wei\u00adter\u00adle\u00adsen&nbsp;\u00bb<\/a><\/p>\n","protected":false},"author":593,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"wp_typography_post_enhancements_disabled":false,"ngg_post_thumbnail":0,"footnotes":""},"categories":[22],"tags":[966,2324,406],"class_list":["post-61968","post","type-post","status-publish","format-standard","hentry","category-pressemitteilungen","tag-amd","tag-epyc-7003","tag-oracle","entry"],"share_on_mastodon":{"url":"","error":""},"_links":{"self":[{"href":"https:\/\/www.planet3dnow.de\/cms\/wp-json\/wp\/v2\/posts\/61968","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.planet3dnow.de\/cms\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.planet3dnow.de\/cms\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.planet3dnow.de\/cms\/wp-json\/wp\/v2\/users\/593"}],"replies":[{"embeddable":true,"href":"https:\/\/www.planet3dnow.de\/cms\/wp-json\/wp\/v2\/comments?post=61968"}],"version-history":[{"count":2,"href":"https:\/\/www.planet3dnow.de\/cms\/wp-json\/wp\/v2\/posts\/61968\/revisions"}],"predecessor-version":[{"id":62092,"href":"https:\/\/www.planet3dnow.de\/cms\/wp-json\/wp\/v2\/posts\/61968\/revisions\/62092"}],"wp:attachment":[{"href":"https:\/\/www.planet3dnow.de\/cms\/wp-json\/wp\/v2\/media?parent=61968"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.planet3dnow.de\/cms\/wp-json\/wp\/v2\/categories?post=61968"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.planet3dnow.de\/cms\/wp-json\/wp\/v2\/tags?post=61968"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}