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 announ­ced the AMD Alveo™ MA35D media acce­le­ra­tor fea­turing two 5nm, ASIC-based video pro­ces­sing units (VPUs) sup­port­ing the AV1 com­pres­si­on stan­dard and pur­po­se-built to power a new era of live inter­ac­ti­ve strea­ming ser­vices at sca­le. With over 70% of the glo­bal video mar­ket being domi­na­ted by live con­tent¹, a new class of low-laten­cy, high-volu­me inter­ac­ti­ve strea­ming appli­ca­ti­ons are emer­ging such as watch par­ties, live shop­ping, online auc­tions, and social streaming.

The Alveo MA35D media acce­le­ra­tor deli­vers the high chan­nel den­si­ty, with up to 32x 1080p60 streams per card, power effi­ci­en­cy and ultra-low-laten­cy per­for­mance cri­ti­cal to redu­cing the sky­ro­cke­ting infra­struc­tu­re cos­ts now requi­red for sca­ling such com­pu­te inten­si­ve con­tent deli­very. Com­pared to the pre­vious gene­ra­ti­on Alveo U30 media acce­le­ra­tor, the Alveo MA35D deli­vers up to 4x hig­her chan­nel den­si­ty², 4x max lower laten­cy in 4K³ and 1.8x grea­ter com­pres­si­on effi­ci­en­cy⁴ to achie­ve the same VMAF score – a com­mon video qua­li­ty metric. 

“We work­ed clo­se­ly with our cus­to­mers and part­ners to under­stand not just their tech­ni­cal requi­re­ments, but their infra­struc­tu­re chal­lenges in deploy­ing high-volu­me, inter­ac­ti­ve strea­ming ser­vices pro­fi­ta­b­ly,” said Dan Gib­bons, gene­ral mana­ger of AECG Data Cen­ter Group, AMD. “We deve­lo­ped the Alveo MA35D with an ASIC archi­tec­tu­re tail­o­red to meet the bespo­ke needs of the­se pro­vi­ders to redu­ce both capi­tal and ope­ra­ting expen­ses for deli­ve­ring immersi­ve expe­ri­en­ces to their users and con­tent crea­tors at scale.”

Pur­po­se-Built Video Pro­ces­sing Unit
The Alveo MA35D uti­li­zes a pur­po­se-built VPU to acce­le­ra­te the enti­re video pipe­line. By per­forming all video pro­ces­sing func­tions on the VPU, data move­ment bet­ween the CPU and acce­le­ra­tor is mini­mi­zed, redu­cing over­all laten­cy and maxi­mi­zing chan­nel den­si­ty with up to 32x 1080p60, 8x 4Kp60, or 4x 8Kp30 streams per card. The plat­form pro­vi­des ultra-low laten­cy sup­port for the main­stream H.264 and H.265 codecs and fea­tures next-gene­ra­ti­on AV1 trans­co­der engi­nes deli­ve­ring up to a 52% reduc­tion in bit­ra­te for band­width savings ver­sus a com­pa­ra­ble soft­ware imple­men­ta­ti­on⁵.

“AMD’s announce­ment of the new Alveo MA35D add-in card is an exci­ting advance­ment of video acce­le­ra­ti­on for data cen­ters and is an important step in buil­ding out a ful­ly-fled­ged eco­sys­tem to sup­port royal­ty-free, high-defi­ni­ti­on video devices, pro­ducts, and ser­vices,” said Matt Frost, Alli­ance for Open Media Chair. “Live strea­ming pro­vi­ders are loo­king for hig­her den­si­ty, lower power, lower laten­cy AV1 solu­ti­ons and by addres­sing the­se, Alli­ance mem­bers such as AMD are hel­ping faci­li­ta­te AV1 deploy­ment and over­all adoption.” 

AI-Enab­led, Intel­li­gent Video Pipeline
The acce­le­ra­tor fea­tures an inte­gra­ted AI pro­ces­sor and dedi­ca­ted video qua­li­ty engi­nes desi­gned to impro­ve the qua­li­ty of expe­ri­ence at redu­ced band­width. The AI pro­ces­sor eva­lua­tes con­tent, frame-by-frame, and dyna­mi­cal­ly adjus­ts enco­der set­tings to impro­ve per­cei­ved visu­al qua­li­ty while mini­mi­zing bit­ra­te. Opti­miza­ti­on tech­ni­ques include regi­on-of-inte­rest (ROI) enco­ding for text and face reso­lu­ti­on, arti­fact detec­tion to cor­rect sce­nes with high levels of moti­on and com­ple­xi­ty, and con­tent-awa­re enco­ding for pre­dic­ti­ve insights for bit­ra­te optimization.

 Cost-Effec­tively Sca­le Inter­ac­ti­ve Media
Sca­ling high-volu­me strea­ming ser­vices requi­res maxi­mi­zing the num­ber of chan­nels per ser­ver while mini­mi­zing power and band­width-per-stream. By deli­ve­ring up to 32x 1080p60 streams per card at 1 watt per stream⁶, a 1U rack ser­ver equip­ped with 8 cards deli­vers up to 256 chan­nels to maxi­mi­ze the num­ber of streams per ser­ver, rack or data center.

Soft­ware Dev Kit and Pro­duct Availability
The plat­form is acces­si­ble with the AMD Media Acce­le­ra­ti­on soft­ware deve­lo­p­ment kit (SDK), sup­port­ing the wide­ly used FFmpeg and Gstrea­mer video frame­works for ease of development.

Alveo MA35D media acce­le­ra­tors are sam­pling now with pro­duc­tion ship­ments expec­ted in Q3. To acce­le­ra­te deve­lo­p­ment, an Ear­ly Access Pro­gram is available to qua­li­fied cus­to­mers with com­pre­hen­si­ve docu­men­ta­ti­on and soft­ware tools for archi­tec­tu­ral exploration.

Sup­port­ing Resources

• Learn more about the Alveo MA35D Media Accelerator
• Check out the Alveo MA35D in the AMD booth at NAB, North Hall N2158
• Learn more about the Alli­ance for Open Media and its mem­ber companies
• Beco­me a fan of AMD on Face­book
• Fol­low AMD on Twit­ter

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 Twit­ter pages.

Cont­act:
Mike San­chez
AMD Communications
(209) 262‑7458
m.sanchez@amd.com

¹ Source: Blue­wa­ve Con­sul­ting and Rese­arch, March 2022 
²  In published spe­ci­fi­ca­ti­ons, the Alveo MA35D sup­ports up to 32 1080p60 streams, while the Alveo U30 sup­ports up to 8. Chan­nel den­si­ty rati­os remain the same regard­less of reso­lu­ti­on. ALV-002
³ In published spe­ci­fi­ca­ti­ons, the Alveo MA35D deli­vers 4X lower laten­cy at 8ms vs. Alveo U30 deli­ve­ring 4K H.264 at 32ms, based on lowest laten­cy capa­bi­li­ty of each plat­form. ALV-005
⁴ Based on test­ing by AMD Labs in April 2023, using the VMAF scores of a Alveo MA35D AV1 encode com­pared to Alveo U30 H.264 encode across (13) publicly available video files at various reso­lu­ti­ons and bit­ra­tes. Actu­al results may vary. ALV-009
⁵ Based on test­ing by AMD Labs in March 2023, using the VMAF scores of Alveo MA35D H.264 encode, H.265 encode, and AV1 encode com­pared to the VMAF score of an open source x264 ver­y­fast SW model across (13) publicly available video files at various reso­lu­ti­ons and bit­ra­tes. Actu­al results may vary. ALV-006
⁶ Typi­cal power for 8 4K streams or 32 1080p60 streams esti­ma­ted at 35W, based on preli­mi­na­ry test­ing and sub­ject to chan­ge. 50W Total Ther­mal Design Power (TDP)