VIS­UAL COM­PUT­ING

NVIDIA GPU Tech­nol­ogy Work­shop South East Asia 2014

HWM (Malaysia) - - THINK -

Last month, NVIDIA held the GPU Tech­nol­ogy Work­shop South East Asia (GTW SEA) 2014 at the Sun­tec Sin­ga­pore Con­ven­tion & Ex­hi­bi­tion Cen­ter.

The one-day GPU con­fer­ence kicked off with key­note pre­sen­ta­tion by Marc Hamil­ton, NVIDIA’s Vice Pres­i­dent of So­lu­tion Ar­chi­tec­ture & Engi­neer­ing and John Tay­lor, CSIRO’s Direc­tor of eRe­search & Com­pu­ta­tional and Sim­u­la­tion Sciences, fol­lowed by tech­nol­ogy ses­sions and ex­hibits that show­cased prod­ucts and so­lu­tions from NVIDIA and its part­ners.

Dur­ing his key­note en­ti­tled ‘GPU Tech­nol­ogy: Past, Present, and Fu­ture’, Hamil­ton re­flected on a decade in GPU tech­nol­ogy, namely in the area of CUDA-en­abled GPUs, fol­lowed by the Pascal ar­chi­tec­ture, Jet­son TK1 devel­op­ment kit, as well as the GRID tech­nol­ogy port­fo­lio, which served as a re­cap of the com­pany’s GPU Tech Con­fer­ence that took place in March this year.

Hamil­ton fur­ther elab­o­rated on the three key fea­tures of the nextgen­er­a­tion Pascal GPU fam­ily: stacked DRAM chips, NVLink and uni­fied mem­ory. Stacked mem­ory in­te­grates sev­eral lay­ers of DRAM chips ver­ti­cally to de­liver sig­nif­i­cantly more band­width, en­ergy ef­fi­ciency and more than twice the ca­pac­ity of cur­rent GDDR5 chips. NVLink, on the other hand, pro­vides fast, ef­fi­cient data shar­ing be­tween GPU and CPU at five to 12 times the data rates of PCIe 3.0. Mean­while, uni­fied mem­ory al­lows the CPU to ac­cess the GPU’s mem­ory, en­sur­ing sim­pler pro­gram­ming and full band­width from each pro­ces­sor.

The sec­ond key­note, ‘Break­through Science on GPUs’ by John Tay­lor, was fo­cused on the su­per­com­put­ing ca­pa­bil­ity of one CSIRO GPU clus­ter named Bragg, a world-class com­pu­ta­tional and sim­u­la­tion science fa­cil­ity in Aus­tralia that har­nesses the power of 64 Tesla S1070s to ac­cel­er­ate anal­y­sis of large, com­plex re­search datasets in ar­eas, such as com­pu­ta­tional bi­ol­ogy, chem­istry and geo­sciences.

Apart from the key­notes, the at­ten­dees were given the op­por­tu­nity to learn the lat­est ad­vance­ments in vis­ual com­put­ing through a num­ber of hands-on demon­stra­tions and tech­ni­cal talks, which cov­ered topics like high-per­for­mance com­put­ing (HPC), big data, vir­tu­al­ized desk­top in­fra­struc­ture (VDI), fi­nance, and pro­fes­sional vir­tu­al­iza­tion by NVIDIA ex­perts, like-minded peers and in­dus­try fig­ures.

The Jet­son TK1 devel­op­ment kit is armed with a Ke­pler GPU with 192 CUDA cores, and a 4-Plus-1 quad-core ARM Cor­tex A15 CPU, for a to­tal per­for­mance of 326 GFLOPS.

NVIDIA demon­strated the GRID Test Drive at its booth, where a note­book PC was con­nected to the vir­tual desk­top. We were able to ma­nip­u­late Ira’s face as part of the FaceWorks in­ter­ac­tive tech demo, and stream YouTube videos in HD. The over­all ex­pe­ri­ence was smooth, with only the oc­ca­sional frame drops over a 4G con­nec­tion.

Newspapers in English

Newspapers from Malaysia

© PressReader. All rights reserved.