AMD Showcases Growing Momentum for AMD Powered AI Solutions from the Data Center to PCs

Micro짯soft, Dell Tech짯no짯lo짯gies, HPE, Leno짯vo, Meta, Ora짯cle, Super짯mi짯cro and others adopt new AMD Instinct MI300X and MI300A data cen짯ter AI acce짯le짯ra짯tors for trai짯ning and infe짯rence solutions 

AMD also laun짯ches ROCm 6 soft짯ware stack with signi짯fi짯cant per짯for짯mance opti짯miza짯ti짯ons and new fea짯tures for Lar짯ge Lan짯guage Models and Ryzen 8040 Series note짯book pro짯ces짯sors for AI PCs 

SAN JOSE, Calif., Dec. 06, 2023 (GLOBE NEWSWIRE) Today at the Advan짯cing AI event, AMD (NASDAQ: AMD) was joi짯n짯ed by indus짯try lea짯ders inclu짯ding Micro짯soft, Meta, Ora짯cle, Dell Tech짯no짯lo짯gies, HPE, Leno짯vo, Super짯mi짯cro, Aris짯ta, Broad짯com and Cis짯co to show짯ca짯se how the짯se com짯pa짯nies are working with AMD to deli짯ver advan짯ced AI solu짯ti짯ons span짯ning from cloud to enter짯pri짯se and PCs. AMD laun짯ched mul짯ti짯ple new pro짯ducts at the event, inclu짯ding the AMD Instinct MI300 Series data cen짯ter AI acce짯le짯ra짯tors, ROCm꽓 6 open soft짯ware stack with signi짯fi짯cant opti짯miza짯ti짯ons and new fea짯tures sup짯port짯ing Lar짯ge Lan짯guage Models (LLMs) and Ryzen꽓 8040 Series pro짯ces짯sors with Ryzen AI.

AI is the future of com짯pu짯ting and AMD is uni짯que짯ly posi짯tio짯ned to power the end-to-end infra짯struc짯tu짯re that will defi짯ne this AI era, from mas짯si짯ve cloud instal짯la짯ti짯ons to enter짯pri짯se clus짯ters and AI-enab짯led intel짯li짯gent embedded devices and PCs, said AMD Chair and CEO Dr. Lisa Su. 쏻e are see짯ing very strong demand for our new Instinct MI300 GPUs, which are the hig짯hest-per짯for짯mance acce짯le짯ra짯tors in the world for gene짯ra짯ti짯ve AI. We are also buil짯ding signi짯fi짯cant momen짯tum for our data cen짯ter AI solu짯ti짯ons with the lar짯gest cloud com짯pa짯nies, the industry셲 top ser짯ver pro짯vi짯ders, and the most inno짯va짯ti짯ve AI start짯ups 윿 who we are working clo짯se짯ly with to rapidly bring Instinct MI300 solu짯ti짯ons to mar짯ket that will dra짯ma짯ti짯cal짯ly acce짯le짯ra짯te the pace of inno짯va짯ti짯on across the enti짯re AI eco짯sys짯tem1.

Advan짯cing Data Cen짯ter AI from the Cloud to Enter짯pri짯se Data Cen짯ters and Supercomputers
AMD was joi짯n짯ed by mul짯ti짯ple part짯ners during the event to high짯light the strong adop짯ti짯on and gro짯wing momen짯tum for the AMD Instinct data cen짯ter AI accelerators.

  • Micro짯soft detail짯ed how it is deploy짯ing AMD Instinct MI300X acce짯le짯ra짯tors to power the new Azu짯re ND MI300x v5 Vir짯tu짯al Machi짯ne (VM) series opti짯mi짯zed for AI workloads.
  • Meta shared that the com짯pa짯ny is adding AMD Instinct MI300X acce짯le짯ra짯tors to its data cen짯ters in com짯bi짯na짯ti짯on with ROCm 6 to power AI infe짯ren짯cing workloads and reco짯gni짯zed the ROCm 6 opti짯miza짯ti짯ons AMD has done on the Llama 2 fami짯ly of models.
  • Ora짯cle unvei짯led plans to offer OCI bare metal com짯pu짯te solu짯ti짯ons fea짯turing AMD Instinct MI300X acce짯le짯ra짯tors as well as plans to include AMD Instinct MI300X acce짯le짯ra짯tors in their upco짯ming gene짯ra짯ti짯ve AI service.
  • The lar짯gest data cen짯ter infra짯struc짯tu짯re pro짯vi짯ders announ짯ced plans to inte짯gra짯te AMD Instinct MI300 acce짯le짯ra짯tors across their pro짯duct port짯fo짯li짯os. Dell announ짯ced the inte짯gra짯ti짯on of AMD Instinct MI300X acce짯le짯ra짯tors with their PowerEdge XE9680 ser짯ver solu짯ti짯on to deli짯ver ground짯brea짯king per짯for짯mance for gene짯ra짯ti짯ve AI workloads in a modu짯lar and sca짯lable for짯mat for cus짯to짯mers. HPE announ짯ced plans to bring AMD Instinct MI300 acce짯le짯ra짯tors to its enter짯pri짯se and HPC offe짯rings. Leno짯vo shared plans to bring AMD Instinct MI300X acce짯le짯ra짯tors to the Leno짯vo Think짯Sys짯tem plat짯form to deli짯ver AI solu짯ti짯ons across indus짯tries inclu짯ding retail, manu짯fac짯tu짯ring, finan짯cial ser짯vices and health짯ca짯re. Super짯mi짯cro announ짯ced plans to offer AMD Instinct MI300 GPUs across their AI solu짯ti짯ons port짯fo짯lio. Asus, Giga짯byte, Ingra짯sys, Inven짯tec, QCT, Wist짯ron and Wiwynn also all plan to offer solu짯ti짯ons powered by AMD Instinct MI300 accelerators.
  • Spe짯cia짯li짯zed AI cloud pro짯vi짯ders inclu짯ding Ali짯gned, Arkon Ener짯gy, Cir짯ras짯ca짯le, Cru짯soe, Den짯vr Data짯works and Ten짯sor짯wa짯ves all plan to pro짯vi짯de offe짯rings that will expand access to AMD Instinct MI300X GPUs for deve짯lo짯pers and AI startups.

Brin짯ging an Open, Pro짯ven and Rea짯dy AI Soft짯ware Plat짯form to Market
AMD high짯ligh짯ted signi짯fi짯cant pro짯gress expan짯ding the soft짯ware eco짯sys짯tem sup짯port짯ing AMD Instinct data cen짯ter accelerators.

  • AMD unvei짯led the latest ver짯si짯on of the open-source soft짯ware stack for AMD Instinct GPUs, ROCm 6, which has been opti짯mi짯zed for gene짯ra짯ti짯ve AI, par짯ti짯cu짯lar짯ly lar짯ge lan짯guage models. ROCm 6 boasts sup짯port for new data types, advan짯ced graph and ker짯nel opti짯miza짯ti짯ons, opti짯mi짯zed libra짯ri짯es and sta짯te of the art atten짯ti짯on algo짯rith짯ms, which tog짯e짯ther with MI300X deli짯ver an ~8x per짯for짯mance increase for over짯all laten짯cy in text gene짯ra짯ti짯on on Llama 2 com짯pared to ROCm 5 run짯ning on the MI250.2
  • Dat짯ab짯ricks, Essen짯ti짯al AI and Lami짯ni, three AI start짯ups buil짯ding emer짯ging models and AI solu짯ti짯ons, joi짯n짯ed AMD on stage to dis짯cuss how they셱e lever짯aging AMD Instinct MI300X acce짯le짯ra짯tors and the open ROCm 6 soft짯ware stack to deli짯ver dif짯fe짯ren짯tia짯ted AI solu짯ti짯ons for enter짯pri짯se customers.
  • Ope짯nAI is adding sup짯port for AMD Instinct acce짯le짯ra짯tors to Tri짯ton 3.0, pro짯vi짯ding out-of-the-box sup짯port for AMD acce짯le짯ra짯tors that will allow deve짯lo짯pers to work at a hig짯her level of abs짯trac짯tion on AMD hardware.

Read here for more infor짯ma짯ti짯on about AMD Instinct MI300 Series acce짯le짯ra짯tors, ROCm 6 and the gro짯wing eco짯sys짯tem of AMD-powered AI solutions.

Con짯tin짯ued Lea짯der짯ship in Advan짯cing AI PCs
With mil짯li짯ons of AI PCs ship짯ped to date, AMD announ짯ced new lea짯der짯ship mobi짯le pro짯ces짯sors with the launch of the latest AMD Ryzen 8040 Series pro짯ces짯sors that deli짯ver robust AI com짯pu짯te capa짯bi짯li짯ty. AMD also laun짯ched Ryzen AI 1.0 Soft짯ware, a soft짯ware stack that enables deve짯lo짯pers to easi짯ly deploy apps that use pre짯trai짯ned models to add AI capa짯bi짯li짯ties for Win짯dows appli짯ca짯ti짯ons. AMD also dis짯c짯lo짯sed that the upco짯ming next-gen 쏶trix Point CPUs, plan짯ned to launch in 2024, will include the AMD XDNA꽓 2 archi짯tec짯tu짯re desi짯gned to deli짯ver more than a 3x increase in AI com짯pu짯te per짯for짯mance com짯pared to the pri짯or gene짯ra짯ti짯on3 that will enable new gene짯ra짯ti짯ve AI expe짯ri짯en짯ces. Micro짯soft also joi짯n짯ed to dis짯cuss how they are working clo짯se짯ly with AMD on future AI expe짯ri짯en짯ces for Win짯dows PCs.

Sup짯port짯ing Resources

  • Watch the full AMD Advan짯cing AI Keynote
  • Read more about the announce짯ments made during Advan짯cing AI here
  • Learn more about the AMD 30x25 ener짯gy effi짯ci짯en짯cy initia짯ti짯ve here
  • Fol짯low AMD on X
  • Con짯nect with AMD on Lin짯ke짯dIn

About AMD
For more than 50 years AMD has dri짯ven inno짯va짯ti짯on in high-per짯for짯mance com짯pu짯ting, gra짯phics and visua짯liza짯ti짯on tech짯no짯lo짯gies. Bil짯li짯ons of peo짯p짯le, lea짯ding For짯tu짯ne 500 busi짯nesses and cut짯ting-edge sci짯en짯ti짯fic rese짯arch insti짯tu짯ti짯ons around the world rely on AMD tech짯no짯lo짯gy dai짯ly to impro짯ve how they live, work and play. AMD employees are focu짯sed on buil짯ding lea짯der짯ship high-per짯for짯mance and adap짯ti짯ve pro짯ducts that push the boun짯da짯ries of what is pos짯si짯ble. For more infor짯ma짯ti짯on about how AMD is enab짯ling today and inspi짯ring tomor짯row, visit the AMD (NASDAQ: AMDweb짯siteblogLin짯ke짯dIn and X pages.

Cau짯tio짯na짯ry Statement
This press release con짯ta짯ins for짯ward-loo짯king state짯ments con짯cer짯ning Advan짯ced Micro Devices, Inc. (AMD) such as the fea짯tures, func짯tion짯a짯li짯ty, per짯for짯mance, avai짯la짯bi짯li짯ty, timing, expec짯ted bene짯fits of, and expec짯ted demand for AMD pro짯ducts and tech짯no짯lo짯gy inclu짯ding the AMD Instinct꽓 MI300 Series data cen짯ter AI acce짯le짯ra짯tors, Ryzen꽓 8040 Series pro짯ces짯sors, and ROCm꽓 6 open soft짯ware stack, which are made pur짯su짯ant to the Safe Har짯bor pro짯vi짯si짯ons of the Pri짯va짯te Secu짯ri짯ties Liti짯ga짯ti짯on Reform Act of 1995. For짯ward-loo짯king state짯ments are com짯mon짯ly iden짯ti짯fied by words such as 쐗ould, 쐌ay, 쐃xpects, 쐀elie짯ves, 쐏lans, 쐇ntends, 쐏ro짯jects and other terms with simi짯lar mea짯ning. Inves짯tors are cau짯tio짯ned that the for짯ward-loo짯king state짯ments in this press release are based on cur짯rent beliefs, assump짯ti짯ons and expec짯ta짯ti짯ons, speak only as of the date of this press release and invol짯ve risks and uncer짯tain짯ties that could cau짯se actu짯al results to dif짯fer mate짯ri짯al짯ly from cur짯rent expec짯ta짯ti짯ons. Such state짯ments are sub짯ject to cer짯tain known and unknown risks and uncer짯tain짯ties, many of which are dif짯fi짯cult to pre짯dict and gene짯ral짯ly bey짯ond AMD셲 con짯trol, that could cau짯se actu짯al results and other future events to dif짯fer mate짯ri짯al짯ly from tho짯se expres짯sed in, or impli짯ed or pro짯jec짯ted by, the for짯ward-loo짯king infor짯ma짯ti짯on and state짯ments. Mate짯ri짯al fac짯tors that could cau짯se actu짯al results to dif짯fer mate짯ri짯al짯ly from cur짯rent expec짯ta짯ti짯ons include, wit짯hout limi짯ta짯ti짯on, the fol짯lo짯wing: Intel Corporation셲 domi짯nan짯ce of the micro짯pro짯ces짯sor mar짯ket and its aggres짯si짯ve busi짯ness prac짯ti짯ces; eco짯no짯mic uncer짯tain짯ty; cycli짯cal natu짯re of the semi짯con짯duc짯tor indus짯try; mar짯ket con짯di짯ti짯ons of the indus짯tries in which AMD pro짯ducts are sold; loss of a signi짯fi짯cant cus짯to짯mer; impact of the COVID-19 pan짯de짯mic on AMD셲 busi짯ness, finan짯cial con짯di짯ti짯on and results of ope짯ra짯ti짯ons; com짯pe짯ti짯ti짯ve mar짯kets in which AMD셲 pro짯ducts are sold; quar짯ter짯ly and sea짯so짯nal sales pat짯terns; AMD셲 abili짯ty to ade짯qua짯te짯ly pro짯tect its tech짯no짯lo짯gy or other intellec짯tu짯al pro짯per짯ty; unfa짯vorable cur짯ren짯cy exch짯an짯ge rate fluc짯tua짯tions; abili짯ty of third par짯ty manu짯fac짯tu짯r짯ers to manu짯fac짯tu짯re AMD셲 pro짯ducts on a time짯ly basis in suf짯fi짯ci짯ent quan짯ti짯ties and using com짯pe짯ti짯ti짯ve tech짯no짯lo짯gies; avai짯la짯bi짯li짯ty of essen짯ti짯al equip짯ment, mate짯ri짯als, sub짯stra짯tes or manu짯fac짯tu짯ring pro짯ces짯ses; abili짯ty to achie짯ve expec짯ted manu짯fac짯tu짯ring yields for AMD셲 pro짯ducts; AMD셲 abili짯ty to intro짯du짯ce pro짯ducts on a time짯ly basis with expec짯ted fea짯tures and per짯for짯mance levels; AMD셲 abili짯ty to gene짯ra짯te reve짯nue from its semi-cus짯tom SoC pro짯ducts; poten짯ti짯al secu짯ri짯ty vul짯nerabi짯li짯ties; poten짯ti짯al secu짯ri짯ty inci짯dents inclu짯ding IT outa짯ges, data loss, data brea짯ches and cyber-attacks; poten짯ti짯al dif짯fi짯cul짯ties in ope짯ra짯ting AMD셲 new짯ly upgraded enter짯pri짯se resour짯ce plan짯ning sys짯tem; uncer짯tain짯ties invol짯ving the orde짯ring and ship짯ment of AMD셲 pro짯ducts; AMD셲 reli짯ance on third-par짯ty intellec짯tu짯al pro짯per짯ty to design and intro짯du짯ce new pro짯ducts in a time짯ly man짯ner; AMD셲 reli짯ance on third-par짯ty com짯pa짯nies for design, manu짯fac짯tu짯re and sup짯p짯ly of mother짯boards, soft짯ware, memo짯ry and other com짯pu짯ter plat짯form com짯pon짯ents; AMD셲 reli짯ance on Micro짯soft and other soft짯ware ven짯dors sup짯port to design and deve짯lop soft짯ware to run on AMD셲 pro짯ducts; AMD셲 reli짯ance on third-par짯ty dis짯tri짯bu짯tors and add-in-board part짯ners; impact of modi짯fi짯ca짯ti짯on or inter짯rup짯ti짯on of AMD셲 inter짯nal busi짯ness pro짯ces짯ses and infor짯ma짯ti짯on sys짯tems; com짯pa짯ti짯bi짯li짯ty of AMD셲 pro짯ducts with some or all indus짯try-stan짯dard soft짯ware and hard짯ware; cos짯ts rela짯ted to defec짯ti짯ve pro짯ducts; effi짯ci짯en짯cy of AMD셲 sup짯p짯ly chain; AMD셲 abili짯ty to rely on third par짯ty sup짯p짯ly-chain logi짯stics func짯tions; AMD셲 abili짯ty to effec짯tively con짯trol sales of its pro짯ducts on the gray mar짯ket; impact of govern짯ment actions and regu짯la짯ti짯ons such as export regu짯la짯ti짯ons, tariffs and trade pro짯tec짯tion mea짯su짯res; AMD셲 abili짯ty to rea짯li짯ze its defer짯red tax assets; poten짯ti짯al tax lia짯bi짯li짯ties; cur짯rent and future claims and liti짯ga짯ti짯on; impact of envi짯ron짯men짯tal laws, con짯flict mine짯rals-rela짯ted pro짯vi짯si짯ons and other laws or regu짯la짯ti짯ons; impact of acqui짯si짯ti짯ons, joint ven짯tures and/or invest짯ments on AMD셲 busi짯ness and AMD셲 abili짯ty to inte짯gra짯te acqui짯red busi짯nesses;  impact of any impair짯ment of the com짯bi짯ned company셲 assets; rest짯ric짯tions impo짯sed by agree짯ments gover짯ning AMD셲 notes, the gua짯ran짯tees of Xilinx셲 notes and the revol짯ving cre짯dit faci짯li짯ty; AMD셲 indeb짯ted짯ness; AMD셲 abili짯ty to gene짯ra짯te suf짯fi짯ci짯ent cash to meet its working capi짯tal requi짯re짯ments or gene짯ra짯te suf짯fi짯ci짯ent reve짯nue and ope짯ra짯ting cash flow to make all of its plan짯ned R&D or stra짯te짯gic invest짯ments; poli짯ti짯cal, legal and eco짯no짯mic risks and natu짯ral dis짯as짯ters; future impairm짯ents of tech짯no짯lo짯gy licen짯se purcha짯ses; AMD셲 abili짯ty to attract and retain qua짯li짯fied per짯son짯nel; and AMD셲 stock pri짯ce vola짯ti짯li짯ty. Inves짯tors are urged to review in detail the risks and uncer짯tain짯ties in AMD셲 Secu짯ri짯ties and Exch짯an짯ge Com짯mis짯si짯on filings, inclu짯ding but not limi짯t짯ed to AMD셲 most recent reports on Forms 10멚 and 10멡.

AMD, the AMD Arrow logo, AMD Instinct, Ryzen and com짯bi짯na짯ti짯ons the짯reof, are trade짯marks of Advan짯ced Micro Devices, Inc. Other names are for infor짯ma짯tio짯nal pur짯po짯ses only and may be trade짯marks of their respec짯ti짯ve owners.


1 Mea짯su짯re짯ments con짯duc짯ted by AMD Per짯for짯mance Labs as of Novem짯ber 11th, 2023 on the AMD Instinct꽓 MI300X (750W) GPU desi짯gned with AMD CDNA꽓 3 5nm | 6nm Fin짯FET pro짯cess tech짯no짯lo짯gy at 2,100 MHz peak boost engi짯ne clock resul짯ted in 163.4 TFLOPs peak theo짯re짯ti짯cal dou짯ble pre짯cis짯i짯on Matrix (FP64 Matrix), 81.7 TFLOPs peak theo짯re짯ti짯cal dou짯ble pre짯cis짯i짯on (FP64), 163.4 TFLOPs peak theo짯re짯ti짯cal sin짯gle pre짯cis짯i짯on Matrix (FP32 Matrix), 163.4 TFLOPs peak theo짯re짯ti짯cal sin짯gle pre짯cis짯i짯on (FP32), 653.7 TFLOPs peak theo짯re짯ti짯cal Ten짯sor짯Float-32 (TF32), 1307.4 TFLOPs peak theo짯re짯ti짯cal half pre짯cis짯i짯on (FP16), 1307.4 TFLOPs peak theo짯re짯ti짯cal Bfloat16 for짯mat pre짯cis짯i짯on (BF16), 2614.9 TFLOPs peak theo짯re짯ti짯cal 8멳it pre짯cis짯i짯on (FP8), 2614.9 TOPs INT8 floa짯ting-point performance.
Published results on Nvi짯dia H100 SXM (80GB) GPU resul짯ted in 66.9 TFLOPs peak theo짯re짯ti짯cal dou짯ble pre짯cis짯i짯on ten짯sor (FP64 Ten짯sor), 33.5 TFLOPs peak theo짯re짯ti짯cal dou짯ble pre짯cis짯i짯on (FP64), 66.9 TFLOPs peak theo짯re짯ti짯cal sin짯gle pre짯cis짯i짯on (FP32), 494.7 TFLOPs peak Ten짯sor짯Float-32 (TF32)*, 989.4 TFLOPs peak theo짯re짯ti짯cal half pre짯cis짯i짯on ten짯sor (FP16 Ten짯sor), 133.8 TFLOPs peak theo짯re짯ti짯cal half pre짯cis짯i짯on (FP16), 989.4 TFLOPs peak theo짯re짯ti짯cal Bfloat16 ten짯sor for짯mat pre짯cis짯i짯on (BF16 Ten짯sor), 133.8 TFLOPs peak theo짯re짯ti짯cal Bfloat16 for짯mat pre짯cis짯i짯on (BF16), 1,978.9 TFLOPs peak theo짯re짯ti짯cal 8멳it pre짯cis짯i짯on (FP8), 1,978.9 TOPs peak theo짯re짯ti짯cal INT8 floa짯ting-point performance.
Nvi짯dia H100 source:

* Nvi짯dia H100 GPUs don셳 sup짯port FP32 Tensor.
2 Text gene짯ra짯ted with Llama2-70b chat using input sequence length of 4096 and 32 out짯put token com짯pa짯ri짯son using cus짯tom docker con짯tai짯ner for each sys짯tem based on AMD inter짯nal test짯ing as of 11/17/2023. Con짯fi짯gu짯ra짯ti짯ons: 2P Intel Xeon Pla짯ti짯num CPU ser짯ver using 4x AMD Instinct꽓 MI300X (192GB, 750W) GPUs, ROCm짰 6.0 pre-release, PyTorch 2.2.0, vLLM for ROCm, Ubun짯tu짰 22.04.2. Vs. 2P AMD EPYC 7763 CPU ser짯ver using 4x AMD Instinct꽓 MI250 (128 GB HBM2e, 560W) GPUs, ROCm짰 5.4.3, PyTorch 2.0.0., Hug짯ging짯Face Trans짯for짯mers 4.35.0, Ubun짯tu 22.04.6.
4 GPUs on each sys짯tem was used in this test. Ser짯ver manu짯fac짯tu짯r짯ers may vary con짯fi짯gu짯ra짯ti짯ons, yiel짯ding dif짯fe짯rent results. Per짯for짯mance may vary based on use of latest dri짯vers and opti짯miza짯ti짯ons. MI300-33

3 An AMD Ryzen 쏶trix point pro짯ces짯sor is pro짯jec짯ted to offer 3x fas짯ter NPU per짯for짯mance for AI workloads when com짯pared to an AMD Ryzen 7040 series pro짯ces짯sor. Per짯for짯mance pro짯jec짯tion by AMD engi짯nee짯ring staff. Engi짯nee짯ring pro짯jec짯tions are not a gua짯ran짯tee of final per짯for짯mance. Spe짯ci짯fic pro짯jec짯tions are based on refe짯rence design plat짯forms and are sub짯ject to chan짯ge when final pro짯ducts are released in mar짯ket. STX-01.