AMD Launches First 5nm ASIC-based Media Accelerator Card to Power New Era of Interactive Media Services at Scale

Purpose-built video processing architecture featuring an AV1 accelerated pipeline delivers 32x 1080p streams per card with AI-enabled video quality optimization

SANTA CLARA, Calif., April 06, 2023 (GLOBE NEWSWIRE)  AMD (NASDAQ: AMD) today announ짯ced the AMD Alveo꽓 MA35D media acce짯le짯ra짯tor fea짯turing two 5nm, ASIC-based video pro짯ces짯sing units (VPUs) sup짯port짯ing the AV1 com짯pres짯si짯on stan짯dard and pur짯po짯se-built to power a new era of live inter짯ac짯ti짯ve strea짯ming ser짯vices at sca짯le. With over 70% of the glo짯bal video mar짯ket being domi짯na짯ted by live con짯tent1, a new class of low-laten짯cy, high-volu짯me inter짯ac짯ti짯ve strea짯ming appli짯ca짯ti짯ons are emer짯ging such as watch par짯ties, live shop짯ping, online auc짯tions, and social streaming.

The Alveo MA35D media acce짯le짯ra짯tor deli짯vers the high chan짯nel den짯si짯ty, with up to 32x 1080p60 streams per card, power effi짯ci짯en짯cy and ultra-low-laten짯cy per짯for짯mance cri짯ti짯cal to redu짯cing the sky짯ro짯cke짯ting infra짯struc짯tu짯re cos짯ts now requi짯red for sca짯ling such com짯pu짯te inten짯si짯ve con짯tent deli짯very. Com짯pared to the pre짯vious gene짯ra짯ti짯on Alveo U30 media acce짯le짯ra짯tor, the Alveo MA35D deli짯vers up to 4x hig짯her chan짯nel den짯si짯ty2, 4x max lower laten짯cy in 4K3 and 1.8x grea짯ter com짯pres짯si짯on effi짯ci짯en짯cy4 to achie짯ve the same VMAF score a com짯mon video qua짯li짯ty metric. 

We work짯ed clo짯se짯ly with our cus짯to짯mers and part짯ners to under짯stand not just their tech짯ni짯cal requi짯re짯ments, but their infra짯struc짯tu짯re chal짯lenges in deploy짯ing high-volu짯me, inter짯ac짯ti짯ve strea짯ming ser짯vices pro짯fi짯ta짯b짯ly, said Dan Gib짯bons, gene짯ral mana짯ger of AECG Data Cen짯ter Group, AMD. 쏻e deve짯lo짯ped the Alveo MA35D with an ASIC archi짯tec짯tu짯re tail짯o짯red to meet the bespo짯ke needs of the짯se pro짯vi짯ders to redu짯ce both capi짯tal and ope짯ra짯ting expen짯ses for deli짯ve짯ring immersi짯ve expe짯ri짯en짯ces to their users and con짯tent crea짯tors at scale.

Pur짯po짯se-Built Video Pro짯ces짯sing Unit
The Alveo MA35D uti짯li짯zes a pur짯po짯se-built VPU to acce짯le짯ra짯te the enti짯re video pipe짯line. By per짯forming all video pro짯ces짯sing func짯tions on the VPU, data move짯ment bet짯ween the CPU and acce짯le짯ra짯tor is mini짯mi짯zed, redu짯cing over짯all laten짯cy and maxi짯mi짯zing chan짯nel den짯si짯ty with up to 32x 1080p60, 8x 4Kp60, or 4x 8Kp30 streams per card. The plat짯form pro짯vi짯des ultra-low laten짯cy sup짯port for the main짯stream H.264 and H.265 codecs and fea짯tures next-gene짯ra짯ti짯on AV1 trans짯co짯der engi짯nes deli짯ve짯ring up to a 52% reduc짯tion in bit짯ra짯te for band짯width savings ver짯sus a com짯pa짯ra짯ble soft짯ware imple짯men짯ta짯ti짯on5.

AMD셲 announce짯ment of the new Alveo MA35D add-in card is an exci짯ting advance짯ment of video acce짯le짯ra짯ti짯on for data cen짯ters and is an important step in buil짯ding out a ful짯ly-fled짯ged eco짯sys짯tem to sup짯port royal짯ty-free, high-defi짯ni짯ti짯on video devices, pro짯ducts, and ser짯vices, said Matt Frost, Alli짯ance for Open Media Chair. 쏬ive strea짯ming pro짯vi짯ders are loo짯king for hig짯her den짯si짯ty, lower power, lower laten짯cy AV1 solu짯ti짯ons and by addres짯sing the짯se, Alli짯ance mem짯bers such as AMD are hel짯ping faci짯li짯ta짯te AV1 deploy짯ment and over짯all adoption. 

AI-Enab짯led, Intel짯li짯gent Video Pipeline
The acce짯le짯ra짯tor fea짯tures an inte짯gra짯ted AI pro짯ces짯sor and dedi짯ca짯ted video qua짯li짯ty engi짯nes desi짯gned to impro짯ve the qua짯li짯ty of expe짯ri짯ence at redu짯ced band짯width. The AI pro짯ces짯sor eva짯lua짯tes con짯tent, frame-by-frame, and dyna짯mi짯cal짯ly adjus짯ts enco짯der set짯tings to impro짯ve per짯cei짯ved visu짯al qua짯li짯ty while mini짯mi짯zing bit짯ra짯te. Opti짯miza짯ti짯on tech짯ni짯ques include regi짯on-of-inte짯rest (ROI) enco짯ding for text and face reso짯lu짯ti짯on, arti짯fact detec짯tion to cor짯rect sce짯nes with high levels of moti짯on and com짯ple짯xi짯ty, and con짯tent-awa짯re enco짯ding for pre짯dic짯ti짯ve insights for bit짯ra짯te optimization.

 Cost-Effec짯tively Sca짯le Inter짯ac짯ti짯ve Media
Sca짯ling high-volu짯me strea짯ming ser짯vices requi짯res maxi짯mi짯zing the num짯ber of chan짯nels per ser짯ver while mini짯mi짯zing power and band짯width-per-stream. By deli짯ve짯ring up to 32x 1080p60 streams per card at 1 watt per stream6, a 1U rack ser짯ver equip짯ped with 8 cards deli짯vers up to 256 chan짯nels to maxi짯mi짯ze the num짯ber of streams per ser짯ver, rack or data center.

Soft짯ware Dev Kit and Pro짯duct Availability
The plat짯form is acces짯si짯ble with the AMD Media Acce짯le짯ra짯ti짯on soft짯ware deve짯lo짯p짯ment kit (SDK), sup짯port짯ing the wide짯ly used FFmpeg and Gstrea짯mer video frame짯works for ease of development.

Alveo MA35D media acce짯le짯ra짯tors are sam짯pling now with pro짯duc짯tion ship짯ments expec짯ted in Q3. To acce짯le짯ra짯te deve짯lo짯p짯ment, an Ear짯ly Access Pro짯gram is available to qua짯li짯fied cus짯to짯mers with com짯pre짯hen짯si짯ve docu짯men짯ta짯ti짯on and soft짯ware tools for archi짯tec짯tu짯ral exploration.

Sup짯port짯ing Resources

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 Twit짯ter pages.

Mike San짯chez
AMD Communications
(209) 2627458

1 Source: Blue짯wa짯ve Con짯sul짯ting and Rese짯arch, March 2022 
2  In published spe짯ci짯fi짯ca짯ti짯ons, the Alveo MA35D sup짯ports up to 32 1080p60 streams, while the Alveo U30 sup짯ports up to 8. Chan짯nel den짯si짯ty rati짯os remain the same regard짯less of reso짯lu짯ti짯on. ALV-002
3 In published spe짯ci짯fi짯ca짯ti짯ons, the Alveo MA35D deli짯vers 4X lower laten짯cy at 8ms vs. Alveo U30 deli짯ve짯ring 4K H.264 at 32ms, based on lowest laten짯cy capa짯bi짯li짯ty of each plat짯form. ALV-005
4 Based on test짯ing by AMD Labs in April 2023, using the VMAF scores of a Alveo MA35D AV1 encode com짯pared to Alveo U30 H.264 encode across (13) publicly available video files at various reso짯lu짯ti짯ons and bit짯ra짯tes. Actu짯al results may vary. ALV-009
5 Based on test짯ing by AMD Labs in March 2023, using the VMAF scores of Alveo MA35D H.264 encode, H.265 encode, and AV1 encode com짯pared to the VMAF score of an open source x264 ver짯y짯fast SW model across (13) publicly available video files at various reso짯lu짯ti짯ons and bit짯ra짯tes. Actu짯al results may vary. ALV-006
6 Typi짯cal power for 8 4K streams or 32 1080p60 streams esti짯ma짯ted at 35W, based on preli짯mi짯na짯ry test짯ing and sub짯ject to chan짯ge. 50W Total Ther짯mal Design Power (TDP