At Ora­cle Open­World 2018, we announ­ced a stra­te­gic part­ner­ship with AMD and laun­ched the Ora­cle E2 plat­form on the first gene­ra­ti­on of AMD EPYC pro­ces­sors. Sin­ce that launch, we’ve deploy­ed Ora­cle E2 Com­pu­te ins­tances to all of our com­mer­cial regi­ons. Eigh­te­en months later, in April 2020, we built on this part­ner­ship and laun­ched the E3 plat­form on the second gene­ra­ti­on. Many of our enter­pri­se cus­to­mers and Ora­cle Cloud appli­ca­ti­ons are run­ning gene­ral-pur­po­se workloads on AMD EPYC pro­ces­sors. Today, I’m hap­py to announ­ce that Ora­cle is laun­ching the E4 plat­form based on third-gene­ra­ti­on AMD EPYC pro­ces­sors. The E4 plat­form includes both bare metal and fle­xi­ble vir­tu­al machi­nes (VMs).

The­se E4 stan­dard ins­tances use 64 core pro­ces­sors, with a base clock fre­quen­cy of 2.55 GHz and a max boost of up to 3.5 GHz. The bare metal E4 stan­dard Com­pu­te ins­tance sup­ports 128 OCPUs (128 cores and 256 threads) with 256 MB of L3 cache, 2 TB of RAM, and 100 Gbps of over­all net­work band­width. This con­fi­gu­ra­ti­on is the hig­hest core count for a bare metal ins­tance on any public cloud. The memo­ry band­width is well sui­ted for both gene­ral-pur­po­se and high-band­width workloads that requi­re lar­ger and fas­ter memory.

As demons­tra­ted in the fol­lo­wing per­for­mance bench­marks, the E4 Stan­dard ins­tances deli­ver up to a 15% increase in inte­ger per­for­mance, a 21% increase in floa­ting point per­for­mance, and a 24% increase in java per­for­mance, com­pared to E3 Stan­dard ins­tances. Also, the E4 ins­tances pro­vi­de three times the pri­ce per­for­mance rela­ti­ve to other gene­ral-pur­po­se ins­tances offe­red by other cloud providers.

New pro­ces­sors, bet­ter per­for­mance, and the same pri­ce as E3 Com­pu­te ins­tances, com­bi­ned with our fle­xi­ble Com­pu­te approach, which enables you to gra­nu­lar­ly cus­to­mi­ze the core counts and memo­ry of VMs, pro­vi­de bet­ter value.

Many enter­pri­se cus­to­mers for Ora­cle appli­ca­ti­ons and gene­ral-pur­po­se workloads have adopted our AMD EPYC pro­ces­sor-based Com­pu­te ins­tances. Our cus­to­mers have alre­a­dy suc­cessful­ly deploy­ed AMD-based ins­tances, inclu­ding the fol­lo­wing use cases:

• Video con­fe­ren­cing

• Real-time video processing

• Mas­si­ve­ly par­al­lel pro­ces­sing and high-per­for­mance com­pu­ting (HPC)

• Busi­ness-cri­ti­cal applications

• Web and appli­ca­ti­on servers

• Backend ser­vers for enter­pri­se applications

• Gam­ing servers

• In-memo­ry database

• Caching fleets

• App deve­lo­p­ment environments

Ora­cle cus­to­mer Xact­ly, the lea­der in reve­nue intel­li­gence solu­ti­ons, uses Oracle’s AMD EPYC pro­ces­sor-based Compute.

“Ora­cle Cloud Infra­struc­tu­re (OCI) is uni­que in pro­vi­ding cus­to­mers with the gra­nu­la­ri­ty and fle­xi­bi­li­ty in com­pu­te resour­ces,” said Ron Ras­mus­sen, CTO of Xact­ly. “Xact­ly has seen gre­at, mea­su­red bene­fits from deploy­ing E3 in ope­ra­ting our mul­ti-ten­ant SaaS reve­nue intel­li­gence solu­ti­ons. Our mul­ti­ten­ant SaaS solu­ti­on is built on Java tech­no­lo­gy, and we look for­ward to exten­ding the addi­tio­nal bene­fits of E4 to our cus­to­mers in the near future.”

Key highlights

E4 ins­tances con­ti­nue the fle­xi­ble infra­struc­tu­re approach that we estab­lished with E3. You’re free to sel­ect the exact num­ber of OCPUs and amount of memo­ry that you need for a VM, not forced to choo­se from a fixed menu of 1, 2, 4, 8, or 16. You can launch any cus­tom VM size that meets your needs, such as a 3‑core, 6‑core, or 63-core VM with any­whe­re from 1 GB–1 TB of memory.

The­se cus­tom sizes mean lower cos­ts. E4 ins­tances bill sepa­ra­te­ly for the OCPU and memo­ry resour­ces pro­vi­sio­ned. Each OCPU comes with its asso­cia­ted simul­ta­neous mul­ti­th­re­a­ding unit and is pri­ced at US$0.025, and memo­ry is pri­ced at US$0.0015 per GB—the same pri­ces as our E3 Com­pu­te ins­tances. An E4 ins­tance with 1‑OCPU ins­tance and 16 GB of RAM  has a pri­ce of US$0.049, which is 23% less than X7 Stan­dard and 57–61% less than com­pa­ra­ble ins­tances offe­red by other clouds.

E4 ins­tances sup­port reboot resi­ze, so you can easi­ly migra­te from your exis­ting com­pu­te ins­tances to an E4 ins­tance of com­pa­ti­ble shape.

Instance availability

Table 1 shows the VM and the bare metal ins­tances that are available at launch. The­se ins­tances are available in the fol­lo­wing regi­ons, with plans to have ins­tances wide­ly available in all com­mer­cial regi­ons in the future:

• US East (Ashb­urn)

• US West (Phoe­nix)

• India west (Mum­bai)

• Switz­er­land North (Zurich)

• Bra­zil East (Sao Paulo)

• Cana­da Sou­the­ast (Mont­re­al)

• Aus­tra­lia sou­the­ast (Mel­bourne)

• Cana­da Sou­the­ast (Toron­to)

Table 1: E4 ins­tances available at launch

Ins­tance	OCPU	Memo­ry	Extra sto­rage	Net­work
BM.Standard.E4.128	128	2,048 GB	Up to 1 PB of remo­te block storage	100 Gbps
VM.Standard.E4.Flex	1–64	1–64 GB per OCPU up to 1024 GB	Up to 1 PB of remo­te block storage	1 Gbps per OCPU

OCPU refers to a sin­gle core and its asso­cia­ted thread. We pro­vi­de 1 Gbps per OCPU, up to a maxi­mum of 40 Gbps.

Performance

We com­pared the AMD EPYC CPU-based ins­tances to our cur­rent x86 stan­dard alter­na­ti­ves. Table 2 shows detail­ed configurations.

Table 2: Com­pu­te con­fi­gu­ra­ti­ons for per­for­mance tests

Sys­tem Configuration	X7 Sys­tem	E3 Sys­tem	E4 Sys­tem
CPU	Two x86 pro­ces­sors, 26 cores per socket @ 2.0 GHz	Two AMD EPYC 7742, 64 cores per socket @ 2.25 GHz Base, and @3.4G Turbo	Two AMD EPYC 7J13, 64 cores per socket @ 2.55 GHz Base, and @3.5G Turbo
Memo­ry	786 GB DDR4	2,048 GB DDR4	2,048 GB DDR4
Net­work	Two 25 Gbps	Two 50 Gbps	Two 50 Gbps

We ran per­for­mance tests to exer­cise the CPU per­for­mance, floa­ting point per­for­mance, and memo­ry sub­sys­tem per­for­mance (see Table 3). We ran tests on ven­dor-recom­men­ded pro­prie­ta­ry com­pi­lers and Ora­cle Linux ope­ra­ting sys­tems. We ran the tests seve­ral times and aver­a­ged the results. All SPEC num­bers are estimates.

Table 3: Per­for­mance tests and bench­mark targets

Test	Bench­mark target
SPE­Cra­te 2017 Integer	Inte­ger performance
SPE­Cra­te 2017 Floa­ting Point	Floa­ting point performance
STREAM Tri­ad	Memo­ry sub­sys­tem performance

The fol­lo­wing graphs show how the AMD E4 ins­tances com­pared against our X7 and E3 Com­pu­te instances.

The third-gene­ra­ti­on AMD EPYC pro­ces­sor-based ser­ver per­forms well com­pared to our exis­ting port­fo­lio of bare metal Com­pu­te ins­tances. The E4 Stan­dard ins­tances deli­ver up to a 15% increase in inte­ger per­for­mance and a 9% increase in floa­ting point per­for­mance com­pared to the E3 ins­tances. Com­pared to the X7 stan­dard ins­tance, the E4 ins­tance deli­vers a 215% increase in inte­ger per­for­mance, a 160% increase in floa­ting point per­for­mance, and a 140% increase in STREAM Tri­ad bandwidth.

Figu­re 2 shows the results of the bench­marks on VMs. All the tests were run on 2‑core VMs.

The ser­ver also per­forms well com­pared to our exis­ting VM ins­tances on this set of stan­dar­di­zed bench­marks. The E4 stan­dard ins­tances deli­ver­ed a 15% impro­ve­ment in inte­ger per­for­mance per core and 21% impro­ve­ment in floa­ting point per­for­mance per core com­pared to E3 based ins­tances. More nota­ble is the com­pa­ri­son against X7 ins­tances. The E4 stan­dard ins­tances deli­ver­ed a 67% impro­ve­ment in inte­ger per­for­mance, 97% impro­ve­ment in Floa­ting point per­for­mance, and a 108% increase in memo­ry per­for­mance over an X7 instance.

Java benchmarks

Apart from the stan­dar­di­zed bench­marks, we also ran our inter­nal Java bench­mark to mea­su­re ser­ver-side Java per­for­mance. We ran the Java tests with an empha­sis on the midd­le-tier per­for­mance. We ran the tests seve­ral times and aver­a­ged the results.

Table 4: Java per­for­mance test and bench­mark target

Test	Bench­mark target
Ser­ver-Side Java Benchmark	Midd­le-tier performance

Figu­re 3 shows the results of run­ning the Java bench­mark tests on X7, E3, and E4 bare metal instances.

The E4 Stan­dard ins­tance deli­ver­ed a 13% increase in Java per­for­mance com­pared to E3 ins­tances and a 263% increase com­pared to X7 instances.

Figu­re 4 shows the results of run­ning the Java bench­mark tests on VMs. All the Java tests were run on 8‑core VMs becau­se of mini­mum memo­ry requirements.

The E4 VM ins­tances deli­ver­ed a 24% increase in Java per­for­mance com­pared to E3 and a 71% increase in Java per­for­mance com­pared to X7 instances.

Competitive benchmarks

We com­pared our offe­rings to what’s available from AWS. As shown in the fol­lo­wing gra­phic, the E4 based ins­tances out­per­for­med the AWS ins­tances in abso­lu­te per­for­mance and pri­ce-per­for­mance, with three times the pri­ce per­for­mance of AWS ins­tances across the­se benchmarks.

Conclusion

You can bene­fit from this per­for­mance by deploy­ing and migra­ting your workloads to the­se Com­pu­te ins­tances today. To learn more about AMD EPYC cloud solu­ti­ons for OCI, watch the video from our VP Ross Brown. For more infor­ma­ti­on on the E4 Stan­dard and fle­xi­ble ins­tances, you can access our docu­men­ta­ti­on. For more Ora­cle Cloud Infra­struc­tu­re solu­ti­ons for AMD EPYC-powered Ora­cle Cloud ins­tances, you can visit the AMD tech docs and white paper libra­ry.