BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/Los_Angeles
X-LIC-LOCATION:America/Los_Angeles
BEGIN:DAYLIGHT
TZOFFSETFROM:-0800
TZOFFSETTO:-0700
TZNAME:PDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0700
TZOFFSETTO:-0800
TZNAME:PST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260402T024534Z
LOCATION:3000\, Level 3
DTSTART;TZID=America/Los_Angeles:20250625T140000
DTEND;TZID=America/Los_Angeles:20250625T141500
UID:dac_DAC 2025_sess110_RESEARCH203@linklings.com
SUMMARY:StreamingGS: Voxel-Based Streaming 3D Gaussian Splatting with Memo
 ry Optimization and Architectural Support
DESCRIPTION:Chenqi Zhang, Yu Feng, Jieru Zhao, and Guangda Liu (Shanghai J
 iao Tong University); Wenchao Ding (Fudan University); and Chentao Wu and 
 Minyi Guo (Shanghai Jiao Tong University)\n\n3D Gaussian splatting (3DGS) 
 has gained popularity for its efficiency and sparse Gaussian-based represe
 ntation. However, 3DGS struggles to meet the real-time requirement of 90 f
 rames per second (FPS) on resource-constrained mobile devices, achieving o
 nly 2 to 9 FPS. Existing accelerators focus on compute efficiency but over
 look memory efficiency, leading to redundant DRAM traffic. We introduce St
 reamingGS, a fully streaming 3DGS algorithm-architecture co-design that ac
 hieves fine-grained pipelining and reduces DRAM traffic by transforming fr
 om a tile-centric rendering to a memory-centric rendering. Results show th
 at our design achieves up to 45.7× speedup and 62.9× energy savings over m
 obile Ampere GPUs.\n\nTopics: AI\n\nTracks: AI3: AI/ML Architecture Design
 \n\nSession Chairs: Yonggan Fu (Georgia Institute of Technology, Nvidia) a
 nd Bokyung Kim (Rutgers University)\n\n
END:VEVENT
END:VCALENDAR