SUNNYVALE, Calif. — Jan. 5, 2017 — Today AMD (NASDAQ: AMD) unvei­led preli­mi­na­ry details of its forth­co­ming GPU archi­tec­tu­re, Vega. Con­cei­ved and exe­cu­ted over 5 years, Vega archi­tec­tu­re enables new pos­si­bi­li­ties in PC gam­ing, pro­fes­sio­nal design and machi­ne intel­li­gence that tra­di­tio­nal GPU archi­tec­tures have not been able to address effec­tively. Data-inten­si­ve workloads are beco­ming the new nor­mal, and the par­al­lel natu­re of the GPU lends its­elf ide­al­ly to tack­ling them. Howe­ver, pro­ces­sing the­se huge new data­sets requi­res fast access to mas­si­ve amounts of memo­ry. The Vega architecture’s revo­lu­tio­na­ry memo­ry sub­sys­tem enables GPUs to address very lar­ge data sets spread across a mix of memo­ry types. The high-band­width cache con­trol­ler in Vega-based GPUs can access on-packa­ge cache and off-packa­ge memo­ries in a fle­xi­ble, pro­gramma­ble fashion using fine-grai­ned data movement.

“It is incre­di­ble to see GPUs being used to sol­ve giga­byte-sca­le data pro­blems in gam­ing to exa­byte-sca­le data pro­blems in machi­ne intel­li­gence. We desi­gned the Vega archi­tec­tu­re to build on this abili­ty, with the fle­xi­bi­li­ty to address the extra­or­di­na­ry breadth of pro­blems GPUs will be sol­ving not only today but also five years from now. Our high-band­width cache is a pivo­tal dis­rup­ti­on that has the poten­ti­al to impact the who­le GPU mar­ket,” said Raja Kodu­ri, seni­or vice pre­si­dent and chief archi­tect, Rade­on Tech­no­lo­gies Group, AMD.

High­lights of the Vega GPU architecture’s advance­ments include:

• The world’s most advan­ced GPU memo­ry archi­tec­tu­re: The Vega archi­tec­tu­re enables a new memo­ry hier­ar­chy for GPUs. This radi­cal new approach comes in the form of a new high-band­width cache and its con­trol­ler. The cache fea­tures lea­ding-edge HBM2 tech­no­lo­gy which is capa­ble of trans­fer­ring tera­bytes of data every second, doubling the band­width-per-pin over the pre­vious gene­ra­ti­on HBM tech­no­lo­gy. HBM2 also enables much grea­ter capa­ci­ty at less than half the foot­print of GDDR5 memo­ry. Vega archi­tec­tu­re is opti­mi­zed for strea­ming very lar­ge data­sets and can work with a varie­ty of memo­ry types with up to 512TB of vir­tu­al address space.

• Next-gene­ra­ti­on geo­me­try pipe­line: Today’s games and pro­fes­sio­nal appli­ca­ti­ons make use of incre­di­bly com­plex geo­me­try enab­led by the extra­or­di­na­ry increase in the reso­lu­ti­ons of data acqui­si­ti­on devices.  The hundreds of mil­li­ons of poly­gons in any given frame have mes­hes so den­se that the­re are often many poly­gons being ren­de­red per pixel. Vega’s next-gene­ra­ti­on geo­me­try pipe­line enables the pro­gramm­er to extra­ct incre­di­ble effi­ci­en­cy in pro­ces­sing this com­plex geo­me­try, while also deli­ve­ring more than 200% of the through­put-per-clock over pre­vious Rade­on archi­tec­tures.[i] It also fea­tures impro­ved load-balan­cing with an intel­li­gent workload dis­tri­bu­tor to deli­ver con­sis­tent performance.

• Next-gene­ra­ti­on com­pu­te engi­ne: At the core of the Vega archi­tec­tu­re is a new, next-gene­ra­ti­on com­pu­te engi­ne built on fle­xi­ble com­pu­te units that can natively pro­cess 8‑bit, 16-bit, 32-bit or 64-bit ope­ra­ti­ons in each clock cycle.[ii] The­se com­pu­te units are opti­mi­zed to attain signi­fi­cant­ly hig­her fre­quen­ci­es than pre­vious gene­ra­ti­ons and their sup­port of varia­ble data­ty­pes makes the archi­tec­tu­re high­ly ver­sa­ti­le across workloads.

• Advan­ced pixel engi­ne: The new Vega pixel engi­ne employs a Draw Stream Bin­ning Ras­te­ri­zer, desi­gned to impro­ve per­for­mance and power effi­ci­en­cy. It allows for “fetch once, sha­de once” of pixels through the use of a smart on-chip bin cache and ear­ly cul­ling of pixels invi­si­ble in a final sce­ne. Vega’s pixel engi­ne is now a cli­ent of the onboard L2 cache, enab­ling con­sidera­ble over­head reduc­tion for gra­phics workloads which per­form fre­quent read-after-wri­te operations.

GPU pro­ducts based on the Vega archi­tec­tu­re are expec­ted to ship in the first half of 2017.

Sup­port­ing Resources

• Check out the Vega Pre­view Cine­ma­tic Trailer
• Check out http://ve.ga, the Vega archi­tec­tu­re pre­view portal
• Beco­me a fan of Rade­on on Facebook
• Fol­low @Radeon on Twitter
• Fol­low @RadeonPro on Twitter
• Fol­low @RadeonInstinct on Twitter 

About AMD

For more than 45 years AMD has dri­ven inno­va­ti­on in high-per­for­mance com­pu­ting, gra­phics, and visua­liza­ti­on tech­no­lo­gies ― the buil­ding blocks for gam­ing, immersi­ve plat­forms, and the dat­a­cen­ter. Hundreds of mil­li­ons of con­su­mers, lea­ding For­tu­ne 500 busi­nesses, and cut­ting-edge sci­en­ti­fic rese­arch faci­li­ties around the world rely on AMD tech­no­lo­gy dai­ly to impro­ve how they live, work, and play. AMD employees around the world are focu­sed on buil­ding gre­at 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, Face­book and Twit­ter pages.

AMD, the AMD Arrow logo, Rade­on 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.

Cau­tio­na­ry Statement

This press release con­ta­ins for­ward-loo­king state­ments con­cer­ning Advan­ced Micro Devices, Inc. (AMD) inclu­ding the fea­tures, func­tion­a­li­ty and expec­ta­ti­ons regar­ding AMD’s Vega GPU archi­tec­tu­re and the timing and avai­la­bi­li­ty of GPU pro­ducts based on AMD’s Vega archi­tec­tu­re, 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 docu­ment 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 may limit AMD’s abili­ty to com­pe­te effec­tively; AMD is par­ty to a wafer sup­p­ly agree­ment with GLOBALFOUNDRIES Inc. (GF) with obli­ga­ti­ons to manu­fac­tu­re pro­ducts at GF with cer­tain excep­ti­ons. If GF is not able to satis­fy AMD’s manu­fac­tu­ring requi­re­ments, its busi­ness could be adver­se­ly impac­ted; AMD reli­es on third par­ties to manu­fac­tu­re its pro­ducts, and if they are unable to do so on a time­ly basis in suf­fi­ci­ent quan­ti­ties and using com­pe­ti­ti­ve tech­no­lo­gies, AMD’s busi­ness could be mate­ri­al­ly adver­se­ly affec­ted; fail­ure to achie­ve expec­ted manu­fac­tu­ring yields for AMD’s pro­ducts could nega­tively impact its finan­cial results; the suc­cess of AMD’s busi­ness is depen­dent upon its abili­ty to intro­du­ce pro­ducts on a time­ly basis with fea­tures and per­for­mance levels that pro­vi­de value to its cus­to­mers while sup­port­ing and coin­ci­ding with signi­fi­cant indus­try tran­si­ti­ons; if AMD can­not gene­ra­te suf­fi­ci­ent reve­nue and ope­ra­ting cash flow or obtain exter­nal finan­cing, it may face a cash short­fall and be unable to make all of its plan­ned invest­ments in rese­arch and deve­lo­p­ment or other stra­te­gic invest­ments; the loss of a signi­fi­cant cus­to­mer may have a mate­ri­al adver­se effect on AMD; AMD’s receipt of reve­nue from its semi-cus­tom SoC pro­ducts is depen­dent upon its tech­no­lo­gy being desi­gned into third-par­ty pro­ducts and the suc­cess of tho­se pro­ducts; glo­bal eco­no­mic uncer­tain­ty may adver­se­ly impact AMD’s busi­ness and ope­ra­ting results; the mar­kets in which AMD’s pro­ducts are sold are high­ly com­pe­ti­ti­ve; AMD may not be able to gene­ra­te suf­fi­ci­ent cash to ser­vice its debt obli­ga­ti­ons or meet its working capi­tal requi­re­ments; AMD has a sub­stan­ti­al amount of indeb­ted­ness which could adver­se­ly affect its finan­cial posi­ti­on and pre­vent it from imple­men­ting its stra­tegy or ful­fil­ling its con­trac­tu­al obli­ga­ti­ons; the agree­ments gover­ning AMD’s notes and the secu­red revol­ving line of cre­dit (Secu­red Revol­ving Line of Cre­dit) impo­se rest­ric­tions on AMD that may adver­se­ly affect its abili­ty to ope­ra­te its busi­ness; uncer­tain­ties invol­ving the orde­ring and ship­ment of AMD’s pro­ducts could mate­ri­al­ly adver­se­ly affect it; the demand for AMD’s pro­ducts depends in part on the mar­ket con­di­ti­ons in the indus­tries into which they are sold. Fluc­tua­tions in demand for AMD’s pro­ducts or a mar­ket decli­ne in any of the­se indus­tries could have a mate­ri­al adver­se effect on its results of ope­ra­ti­ons; AMD’s abili­ty to design and intro­du­ce new pro­ducts in a time­ly man­ner is depen­dent upon third-par­ty intellec­tu­al pro­per­ty; AMD depends on third-par­ty com­pa­nies for the design, manu­fac­tu­re and sup­p­ly of mother­boards, soft­ware and other com­pu­ter plat­form com­pon­ents to sup­port its busi­ness; if AMD loses Micro­soft Corporation’s sup­port for its pro­ducts or other soft­ware ven­dors do not design and deve­lop soft­ware to run on AMD’s pro­ducts, its abili­ty to sell its pro­ducts could be mate­ri­al­ly adver­se­ly affec­ted; and AMD’s reli­ance on third-par­ty dis­tri­bu­tors and Add-in-Board part­ners sub­jects it to cer­tain risks. 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 Quar­ter­ly Report on Form 10‑Q for the quar­ter ended Sep­tem­ber 24, 2016.

[i] Data based on AMD Engi­nee­ring design of Vega.  Rade­on R9 Fury X has 4 geo­me­try engi­nes and a peak of 4 poly­gons per clock. Vega is desi­gned to hand­le up to 11 poly­gons per clock with 4 geo­me­try engi­nes.  This repres­ents an increase of 2.6x.  VG‑3

[ii] Dis­crete AMD Rade­on™ and Fire­Pro™ GPUs based on the Gra­phics Core Next archi­tec­tu­re con­sist of mul­ti­ple dis­crete exe­cu­ti­on engi­nes known as a Com­pu­te Unit (“CU”). Each CU con­ta­ins 64 shaders (“Stream Pro­ces­sors”) working tog­e­ther. GD-78