MINDTECH ANNOUNCES ENHANCED SCALABILITY SUPPORT FOR CHAMELEON

LONDON, Nov. 15, 2019 /PRNewswire/ – Mind­tech Glo­bal Ltd, a UK based start-up, has announ­ced con­tai­ner sup­port to allow run­ning mul­ti­ple ins­tances of their Cha­me­le­on syn­the­tic data gene­ra­ti­on tool on clus­te­red plat­forms.  This approach enables full use of available com­pu­te resour­ces lea­ding to a signi­fi­cant reduc­tion in trai­ning cycle times for neu­ral net­works using Syn­the­tic data.

Sca­la­bi­li­ty by design

Mindtech’s Cha­me­le­on was built from the out­set with the capa­bi­li­ty to sca­le for gene­ra­ting lar­ge amounts of data to meet the demands of trai­ning visu­al neu­ral net­works for AI sys­tems.  The simu­la­tor incor­po­ra­tes a scrip­ting engi­ne, enab­ling easy capa­bi­li­ties for repea­ted simu­la­ti­on runs, chan­ging sin­gle varia­bles at a time.  The sce­na­rio edi­tor is inten­ded to allow the quick and easy crea­ti­on of mul­ti­ple dif­fe­rent sequen­ces, to crea­te the data requi­red for the inten­ded use case.

Exploi­ting scalability

The Sce­na­rio edi­tor and simu­la­ti­on scrip­ting engi­ne bet­ween them ensu­re that the Cha­me­le­on tools can easi­ly be used to gene­ra­te huge amounts of data.  To allow the user to effec­tively exploit this requi­res that simu­la­ti­on and ren­de­ring hard­ware be ful­ly uti­li­sed.  To achie­ve this the tools were desi­gned to ful­ly employ all resour­ces available, taking advan­ta­ge of mul­ti-thread pro­ces­sors and GPU acce­le­ra­tors available on the sys­tem hos­ting the simu­la­tor.  To fur­ther enhan­ce this capa­bi­li­ty, Mind­tech has updated the simu­la­tor and encap­su­la­ted it in a con­tai­ne­ri­sed form, to allow use of such tech­no­lo­gies as Sin­gu­la­ri­ty and Docker.  This in turns allows deploy­ment within a clus­te­red envi­ron­ment like Kuber­netes.  This new func­tion­a­li­ty allows for deploy­ment at ser­ver and cloud sca­le, enhan­cing the abili­ty to uti­li­se com­pu­te resour­ces available. 

The need for scalability

The crea­ti­on of the vir­tu­al worlds, used to crea­te Mindtech’s high qua­li­ty syn­the­tic data, requi­res the use of sophisti­ca­ted 3D ren­de­ring tech­ni­ques, as well as signi­fi­cant pro­ces­sor power, deli­ver­ed by GPU acce­le­ra­tors, to help ensu­re the vir­tu­al world beha­ves in a rea­li­stic, “life-like” man­ner.  Each trai­ning task for a neu­ral net­work will requi­re hundreds of thou­sands or even mil­li­ons of images, each indi­vi­du­al­ly ren­de­red, and anno­ta­ted for the trai­ning task.  The use of con­tai­ners allows the user to run mul­ti­ple simu­la­ti­on runs in par­al­lel, vast­ly redu­cing the time to crea­te data, and shor­tening the over­all trai­ning cycle for neu­ral networks.

Says Ogi Brkic, Cor­po­ra­te Vice Pre­si­dent and Gene­ral Mana­ger of the Data Cen­ter GPU Busi­ness Unit at AMD, Rade­on Tech­no­lo­gy Group: “The com­bi­na­ti­on of the latest AMD EPYC™ 7002 Series Pro­ces­sor and Rade­on Instinct™MI50 GPU acce­le­ra­tor repres­ents the best of what’s pos­si­ble for modern AI workloads. Mindtech’s Cha­me­le­on lever­a­ges this powerful com­pu­te plat­form to acce­le­ra­te syn­the­tic data gene­ra­ti­on and to help redu­ce trai­ning cycle times with a con­tai­ne­ri­zed approach.”

Avai­la­bi­li­ty

Mindtech’s Cha­me­le­on AI Tools sca­la­bi­li­ty using con­tai­ners will be shown on the AMD stand at Super­com­pu­ting 2019, Novem­ber 18–21, Den­ver, Colorado. 

Cont­act

For fur­ther infor­ma­ti­on, plea­se cont­act chris.longstaff@mindtech.global, or https://mindtech.global

AMD, the AMD logo, EPYC, Rade­on Instinct, and com­bi­na­ti­ons the­reof are trade­marks of Advan­ced Micro Devices, Inc.

SOURCE Mind­tech Glo­bal Ltd

RELATED LINKS

https://mindtech.global