— Micro­soft, Dell Tech­no­lo­gies, HPE, Leno­vo, Meta, Ora­cle, Super­mi­cro and others adopt new AMD Instinct MI300X and MI300A data cen­ter AI acce­le­ra­tors for trai­ning and infe­rence solutions —

— AMD also laun­ches ROCm 6 soft­ware stack with signi­fi­cant per­for­mance opti­miza­ti­ons and new fea­tures for Lar­ge Lan­guage Models and Ryzen 8040 Series note­book pro­ces­sors for AI PCs —

SAN JOSE, Calif., Dec. 06, 2023 (GLOBE NEWSWIRE) — Today at the “Advan­cing AI” event, AMD (NASDAQ: AMD) was joi­n­ed by indus­try lea­ders inclu­ding Micro­soft, Meta, Ora­cle, Dell Tech­no­lo­gies, HPE, Leno­vo, Super­mi­cro, Aris­ta, Broad­com and Cis­co to show­ca­se how the­se com­pa­nies are working with AMD to deli­ver advan­ced AI solu­ti­ons span­ning from cloud to enter­pri­se and PCs. AMD laun­ched mul­ti­ple new pro­ducts at the event, inclu­ding the AMD Instinct™ MI300 Series data cen­ter AI acce­le­ra­tors, ROCm™ 6 open soft­ware stack with signi­fi­cant opti­miza­ti­ons and new fea­tures sup­port­ing Lar­ge Lan­guage Models (LLMs) and Ryzen™ 8040 Series pro­ces­sors with Ryzen AI.

“AI is the future of com­pu­ting and AMD is uni­que­ly posi­tio­ned to power the end-to-end infra­struc­tu­re that will defi­ne this AI era, from mas­si­ve cloud instal­la­ti­ons to enter­pri­se clus­ters and AI-enab­led intel­li­gent embedded devices and PCs,” said AMD Chair and CEO Dr. Lisa Su. “We are see­ing very strong demand for our new Instinct MI300 GPUs, which are the hig­hest-per­for­mance acce­le­ra­tors in the world for gene­ra­ti­ve AI. We are also buil­ding signi­fi­cant momen­tum for our data cen­ter AI solu­ti­ons with the lar­gest cloud com­pa­nies, the industry’s top ser­ver pro­vi­ders, and the most inno­va­ti­ve AI start­ups ꟷ who we are working clo­se­ly with to rapidly bring Instinct MI300 solu­ti­ons to mar­ket that will dra­ma­ti­cal­ly acce­le­ra­te the pace of inno­va­ti­on across the enti­re AI eco­sys­tem¹.”

Advan­cing Data Cen­ter AI from the Cloud to Enter­pri­se Data Cen­ters and Supercomputers
AMD was joi­n­ed by mul­ti­ple part­ners during the event to high­light the strong adop­ti­on and gro­wing momen­tum for the AMD Instinct data cen­ter AI accelerators.

• Micro­soft detail­ed how it is deploy­ing AMD Instinct MI300X acce­le­ra­tors to power the new Azu­re ND MI300x v5 Vir­tu­al Machi­ne (VM) series opti­mi­zed for AI workloads.
• Meta shared that the com­pa­ny is adding AMD Instinct MI300X acce­le­ra­tors to its data cen­ters in com­bi­na­ti­on with ROCm 6 to power AI infe­ren­cing workloads and reco­gni­zed the ROCm 6 opti­miza­ti­ons AMD has done on the Llama 2 fami­ly of models.
• Ora­cle unvei­led plans to offer OCI bare metal com­pu­te solu­ti­ons fea­turing AMD Instinct MI300X acce­le­ra­tors as well as plans to include AMD Instinct MI300X acce­le­ra­tors in their upco­ming gene­ra­ti­ve AI service.
• The lar­gest data cen­ter infra­struc­tu­re pro­vi­ders announ­ced plans to inte­gra­te AMD Instinct MI300 acce­le­ra­tors across their pro­duct port­fo­li­os. Dell announ­ced the inte­gra­ti­on of AMD Instinct MI300X acce­le­ra­tors with their PowerEdge XE9680 ser­ver solu­ti­on to deli­ver ground­brea­king per­for­mance for gene­ra­ti­ve AI workloads in a modu­lar and sca­lable for­mat for cus­to­mers. HPE announ­ced plans to bring AMD Instinct MI300 acce­le­ra­tors to its enter­pri­se and HPC offe­rings. Leno­vo shared plans to bring AMD Instinct MI300X acce­le­ra­tors to the Leno­vo Think­Sys­tem plat­form to deli­ver AI solu­ti­ons across indus­tries inclu­ding retail, manu­fac­tu­ring, finan­cial ser­vices and health­ca­re. Super­mi­cro announ­ced plans to offer AMD Instinct MI300 GPUs across their AI solu­ti­ons port­fo­lio. Asus, Giga­byte, Ingra­sys, Inven­tec, QCT, Wist­ron and Wiwynn also all plan to offer solu­ti­ons powered by AMD Instinct MI300 accelerators.
• Spe­cia­li­zed AI cloud pro­vi­ders inclu­ding Ali­gned, Arkon Ener­gy, Cir­ras­ca­le, Cru­soe, Den­vr Data­works and Ten­sor­wa­ves all plan to pro­vi­de offe­rings that will expand access to AMD Instinct MI300X GPUs for deve­lo­pers and AI startups.

Brin­ging an Open, Pro­ven and Rea­dy AI Soft­ware Plat­form to Market
AMD high­ligh­ted signi­fi­cant pro­gress expan­ding the soft­ware eco­sys­tem sup­port­ing AMD Instinct data cen­ter accelerators.

• AMD unvei­led the latest ver­si­on of the open-source soft­ware stack for AMD Instinct GPUs, ROCm 6, which has been opti­mi­zed for gene­ra­ti­ve AI, par­ti­cu­lar­ly lar­ge lan­guage models. ROCm 6 boasts sup­port for new data types, advan­ced graph and ker­nel opti­miza­ti­ons, opti­mi­zed libra­ri­es and sta­te of the art atten­ti­on algo­rith­ms, which tog­e­ther with MI300X deli­ver an ~8x per­for­mance increase for over­all laten­cy in text gene­ra­ti­on on Llama 2 com­pared to ROCm 5 run­ning on the MI250.²
• Dat­ab­ricks, Essen­ti­al AI and Lami­ni, three AI start­ups buil­ding emer­ging models and AI solu­ti­ons, joi­n­ed AMD on stage to dis­cuss how they’re lever­aging AMD Instinct MI300X acce­le­ra­tors and the open ROCm 6 soft­ware stack to deli­ver dif­fe­ren­tia­ted AI solu­ti­ons for enter­pri­se customers.
• Ope­nAI is adding sup­port for AMD Instinct acce­le­ra­tors to Tri­ton 3.0, pro­vi­ding out-of-the-box sup­port for AMD acce­le­ra­tors that will allow deve­lo­pers to work at a hig­her level of abs­trac­tion on AMD hardware.

Read here for more infor­ma­ti­on about AMD Instinct MI300 Series acce­le­ra­tors, ROCm 6 and the gro­wing eco­sys­tem of AMD-powered AI solutions.

Con­tin­ued Lea­der­ship in Advan­cing AI PCs
With mil­li­ons of AI PCs ship­ped to date, AMD announ­ced new lea­der­ship mobi­le pro­ces­sors with the launch of the latest AMD Ryzen 8040 Series pro­ces­sors that deli­ver robust AI com­pu­te capa­bi­li­ty. AMD also laun­ched Ryzen AI 1.0 Soft­ware, a soft­ware stack that enables deve­lo­pers to easi­ly deploy apps that use pre­trai­ned models to add AI capa­bi­li­ties for Win­dows appli­ca­ti­ons. AMD also dis­c­lo­sed that the upco­ming next-gen “Strix Point” CPUs, plan­ned to launch in 2024, will include the AMD XDNA™ 2 archi­tec­tu­re desi­gned to deli­ver more than a 3x increase in AI com­pu­te per­for­mance com­pared to the pri­or gene­ra­ti­on³ that will enable new gene­ra­ti­ve AI expe­ri­en­ces. Micro­soft also joi­n­ed to dis­cuss how they are working clo­se­ly with AMD on future AI expe­ri­en­ces for Win­dows PCs.

Sup­port­ing Resources

• Watch the full AMD Advan­cing AI Keynote
• Read more about the announce­ments made during Advan­cing AI here
• Learn more about the AMD 30x25 ener­gy effi­ci­en­cy initia­ti­ve here
• Fol­low AMD on X
• Con­nect with AMD on Lin­ke­dIn

About AMD
For more than 50 years AMD has dri­ven inno­va­ti­on in high-per­for­mance com­pu­ting, gra­phics and visua­liza­ti­on tech­no­lo­gies. Bil­li­ons of peo­p­le, lea­ding For­tu­ne 500 busi­nesses and cut­ting-edge sci­en­ti­fic rese­arch insti­tu­ti­ons around the world rely on AMD tech­no­lo­gy dai­ly to impro­ve how they live, work and play. AMD employees are focu­sed on buil­ding lea­der­ship high-per­for­mance and adap­ti­ve pro­ducts that push the boun­da­ries of what is pos­si­ble. For more infor­ma­ti­on about how AMD is enab­ling today and inspi­ring tomor­row, visit the AMD (NASDAQ: AMD) web­site, blog, Lin­ke­dIn and X pages.

Cau­tio­na­ry Statement
This press release con­ta­ins for­ward-loo­king state­ments con­cer­ning Advan­ced Micro Devices, Inc. (AMD) such as the fea­tures, func­tion­a­li­ty, per­for­mance, avai­la­bi­li­ty, timing, expec­ted bene­fits of, and expec­ted demand for AMD pro­ducts and tech­no­lo­gy inclu­ding the AMD Instinct™ MI300 Series data cen­ter AI acce­le­ra­tors, Ryzen™ 8040 Series pro­ces­sors, and ROCm™ 6 open soft­ware stack, which are made pur­su­ant to the Safe Har­bor pro­vi­si­ons of the Pri­va­te Secu­ri­ties Liti­ga­ti­on Reform Act of 1995. For­ward-loo­king state­ments are com­mon­ly iden­ti­fied by words such as “would,” “may,” “expects,” “belie­ves,” “plans,” “intends,” “pro­jects” and other terms with simi­lar mea­ning. Inves­tors are cau­tio­ned that the for­ward-loo­king state­ments in this press release are based on cur­rent beliefs, assump­ti­ons and expec­ta­ti­ons, speak only as of the date of this press release and invol­ve risks and uncer­tain­ties that could cau­se actu­al results to dif­fer mate­ri­al­ly from cur­rent expec­ta­ti­ons. Such state­ments are sub­ject to cer­tain known and unknown risks and uncer­tain­ties, many of which are dif­fi­cult to pre­dict and gene­ral­ly bey­ond AMD’s con­trol, that could cau­se actu­al results and other future events to dif­fer mate­ri­al­ly from tho­se expres­sed in, or impli­ed or pro­jec­ted by, the for­ward-loo­king infor­ma­ti­on and state­ments. Mate­ri­al fac­tors that could cau­se actu­al results to dif­fer mate­ri­al­ly from cur­rent expec­ta­ti­ons include, wit­hout limi­ta­ti­on, the fol­lo­wing: Intel Corporation’s domi­nan­ce of the micro­pro­ces­sor mar­ket and its aggres­si­ve busi­ness prac­ti­ces; eco­no­mic uncer­tain­ty; cycli­cal natu­re of the semi­con­duc­tor indus­try; mar­ket con­di­ti­ons of the indus­tries in which AMD pro­ducts are sold; loss of a signi­fi­cant cus­to­mer; impact of the COVID-19 pan­de­mic on AMD’s busi­ness, finan­cial con­di­ti­on and results of ope­ra­ti­ons; com­pe­ti­ti­ve mar­kets in which AMD’s pro­ducts are sold; quar­ter­ly and sea­so­nal sales pat­terns; AMD’s abili­ty to ade­qua­te­ly pro­tect its tech­no­lo­gy or other intellec­tu­al pro­per­ty; unfa­vorable cur­ren­cy exch­an­ge rate fluc­tua­tions; abili­ty of third par­ty manu­fac­tu­r­ers to manu­fac­tu­re AMD’s pro­ducts on a time­ly basis in suf­fi­ci­ent quan­ti­ties and using com­pe­ti­ti­ve tech­no­lo­gies; avai­la­bi­li­ty of essen­ti­al equip­ment, mate­ri­als, sub­stra­tes or manu­fac­tu­ring pro­ces­ses; abili­ty to achie­ve expec­ted manu­fac­tu­ring yields for AMD’s pro­ducts; AMD’s abili­ty to intro­du­ce pro­ducts on a time­ly basis with expec­ted fea­tures and per­for­mance levels; AMD’s abili­ty to gene­ra­te reve­nue from its semi-cus­tom SoC pro­ducts; poten­ti­al secu­ri­ty vul­nerabi­li­ties; poten­ti­al secu­ri­ty inci­dents inclu­ding IT outa­ges, data loss, data brea­ches and cyber-attacks; poten­ti­al dif­fi­cul­ties in ope­ra­ting AMD’s new­ly upgraded enter­pri­se resour­ce plan­ning sys­tem; uncer­tain­ties invol­ving the orde­ring and ship­ment of AMD’s pro­ducts; AMD’s reli­ance on third-par­ty intellec­tu­al pro­per­ty to design and intro­du­ce new pro­ducts in a time­ly man­ner; AMD’s reli­ance on third-par­ty com­pa­nies for design, manu­fac­tu­re and sup­p­ly of mother­boards, soft­ware, memo­ry and other com­pu­ter plat­form com­pon­ents; AMD’s reli­ance on Micro­soft and other soft­ware ven­dors’ sup­port to design and deve­lop soft­ware to run on AMD’s pro­ducts; AMD’s reli­ance on third-par­ty dis­tri­bu­tors and add-in-board part­ners; impact of modi­fi­ca­ti­on or inter­rup­ti­on of AMD’s inter­nal busi­ness pro­ces­ses and infor­ma­ti­on sys­tems; com­pa­ti­bi­li­ty of AMD’s pro­ducts with some or all indus­try-stan­dard soft­ware and hard­ware; cos­ts rela­ted to defec­ti­ve pro­ducts; effi­ci­en­cy of AMD’s sup­p­ly chain; AMD’s abili­ty to rely on third par­ty sup­p­ly-chain logi­stics func­tions; AMD’s abili­ty to effec­tively con­trol sales of its pro­ducts on the gray mar­ket; impact of govern­ment actions and regu­la­ti­ons such as export regu­la­ti­ons, tariffs and trade pro­tec­tion mea­su­res; AMD’s abili­ty to rea­li­ze its defer­red tax assets; poten­ti­al tax lia­bi­li­ties; cur­rent and future claims and liti­ga­ti­on; impact of envi­ron­men­tal laws, con­flict mine­rals-rela­ted pro­vi­si­ons and other laws or regu­la­ti­ons; impact of acqui­si­ti­ons, joint ven­tures and/or invest­ments on AMD’s busi­ness and AMD’s abili­ty to inte­gra­te acqui­red busi­nesses;  impact of any impair­ment of the com­bi­ned company’s assets; rest­ric­tions impo­sed by agree­ments gover­ning AMD’s notes, the gua­ran­tees of Xilinx’s notes and the revol­ving cre­dit faci­li­ty; AMD’s indeb­ted­ness; AMD’s abili­ty to gene­ra­te suf­fi­ci­ent cash to meet its working capi­tal requi­re­ments or gene­ra­te suf­fi­ci­ent reve­nue and ope­ra­ting cash flow to make all of its plan­ned R&D or stra­te­gic invest­ments; poli­ti­cal, legal and eco­no­mic risks and natu­ral dis­as­ters; future impairm­ents of tech­no­lo­gy licen­se purcha­ses; AMD’s abili­ty to attract and retain qua­li­fied per­son­nel; and AMD’s stock pri­ce vola­ti­li­ty. Inves­tors are urged to review in detail the risks and uncer­tain­ties in AMD’s Secu­ri­ties and Exch­an­ge Com­mis­si­on filings, inclu­ding but not limi­t­ed to AMD’s most recent reports on Forms 10‑K and 10‑Q.

AMD, the AMD Arrow logo, AMD Instinct, Ryzen and com­bi­na­ti­ons the­reof, are trade­marks of Advan­ced Micro Devices, Inc. Other names are for infor­ma­tio­nal pur­po­ses only and may be trade­marks of their respec­ti­ve owners.

____________________________

¹ Mea­su­re­ments con­duc­ted by AMD Per­for­mance Labs as of Novem­ber 11th, 2023 on the AMD Instinct™ MI300X (750W) GPU desi­gned with AMD CDNA™ 3 5nm | 6nm Fin­FET pro­cess tech­no­lo­gy at 2,100 MHz peak boost engi­ne clock resul­ted in 163.4 TFLOPs peak theo­re­ti­cal dou­ble pre­cis­i­on Matrix (FP64 Matrix), 81.7 TFLOPs peak theo­re­ti­cal dou­ble pre­cis­i­on (FP64), 163.4 TFLOPs peak theo­re­ti­cal sin­gle pre­cis­i­on Matrix (FP32 Matrix), 163.4 TFLOPs peak theo­re­ti­cal sin­gle pre­cis­i­on (FP32), 653.7 TFLOPs peak theo­re­ti­cal Ten­sor­Float-32 (TF32), 1307.4 TFLOPs peak theo­re­ti­cal half pre­cis­i­on (FP16), 1307.4 TFLOPs peak theo­re­ti­cal Bfloat16 for­mat pre­cis­i­on (BF16), 2614.9 TFLOPs peak theo­re­ti­cal 8‑bit pre­cis­i­on (FP8), 2614.9 TOPs INT8 floa­ting-point performance.
Published results on Nvi­dia H100 SXM (80GB) GPU resul­ted in 66.9 TFLOPs peak theo­re­ti­cal dou­ble pre­cis­i­on ten­sor (FP64 Ten­sor), 33.5 TFLOPs peak theo­re­ti­cal dou­ble pre­cis­i­on (FP64), 66.9 TFLOPs peak theo­re­ti­cal sin­gle pre­cis­i­on (FP32), 494.7 TFLOPs peak Ten­sor­Float-32 (TF32)*, 989.4 TFLOPs peak theo­re­ti­cal half pre­cis­i­on ten­sor (FP16 Ten­sor), 133.8 TFLOPs peak theo­re­ti­cal half pre­cis­i­on (FP16), 989.4 TFLOPs peak theo­re­ti­cal Bfloat16 ten­sor for­mat pre­cis­i­on (BF16 Ten­sor), 133.8 TFLOPs peak theo­re­ti­cal Bfloat16 for­mat pre­cis­i­on (BF16), 1,978.9 TFLOPs peak theo­re­ti­cal 8‑bit pre­cis­i­on (FP8), 1,978.9 TOPs peak theo­re­ti­cal INT8 floa­ting-point performance.
Nvi­dia H100 source:
https://resources.nvidia.com/en-us-tensor-core/

* Nvi­dia H100 GPUs don’t sup­port FP32 Tensor.
MI300-18
² Text gene­ra­ted with Llama2-70b chat using input sequence length of 4096 and 32 out­put token com­pa­ri­son using cus­tom docker con­tai­ner for each sys­tem based on AMD inter­nal test­ing as of 11/17/2023. Con­fi­gu­ra­ti­ons: 2P Intel Xeon Pla­ti­num CPU ser­ver using 4x AMD Instinct™ MI300X (192GB, 750W) GPUs, ROCm® 6.0 pre-release, PyTorch 2.2.0, vLLM for ROCm, Ubun­tu® 22.04.2. Vs. 2P AMD EPYC 7763 CPU ser­ver using 4x AMD Instinct™ MI250 (128 GB HBM2e, 560W) GPUs, ROCm® 5.4.3, PyTorch 2.0.0., Hug­ging­Face Trans­for­mers 4.35.0, Ubun­tu 22.04.6.
4 GPUs on each sys­tem was used in this test. Ser­ver manu­fac­tu­r­ers may vary con­fi­gu­ra­ti­ons, yiel­ding dif­fe­rent results. Per­for­mance may vary based on use of latest dri­vers and opti­miza­ti­ons. MI300-33

³ An AMD Ryzen “Strix point” pro­ces­sor is pro­jec­ted to offer 3x fas­ter NPU per­for­mance for AI workloads when com­pared to an AMD Ryzen 7040 series pro­ces­sor. Per­for­mance pro­jec­tion by AMD engi­nee­ring staff. Engi­nee­ring pro­jec­tions are not a gua­ran­tee of final per­for­mance. Spe­ci­fic pro­jec­tions are based on refe­rence design plat­forms and are sub­ject to chan­ge when final pro­ducts are released in mar­ket. STX-01.