A 4nm Peta-Scale AI SoC
with Unmatched TPS/W
Future-proofed for any AI workloads, REBEL-Quad combines high-throughput compute
with massive HBM3e bandwidth and scalable chiplet integration
—setting a new standard in performance per watt.
Compute
2,048 TFLOPS (Dense)
Memory
144GB HBM3e
4.8TB/s
One chip, enough for tens of billions of parameters
Chiplet
4TB/s
World’s First UCIe-Advanced Chiplet2.5D advanced packaging in production
Power
350 ~ 700W
Air-cooledDesigned for peak efficiency, through near-perfect utilization
Scalability
256GB/s, 2x PCIe Gen5
1.6TB/s ETH, Available SoonMassive scalability for inference at large scale
Ready to Support
Your Data Center
ATOM™-Max Now in Mass Production
Engineered for hyperscale AI workloads, ATOM™-Max delivers
up to 1,024 TOPS of performance with PCIe Gen5 x16 direct card-to-card communication.
Boosted performance. Minimal latency. Ready to deploy.
Rebellions 리벨리온
Delivering Industry-Leading
Performance-per-Watt
Rebellions, an AI accelerator startup, provides AI hardware for data centers—
cards, servers, and rack-scale solutions—alongside our proprietary
full-stack software that natively supports PyTorch and vLLM.
Text-to-Image
Text Generation
Multi-Modal
Object Detection
200+ Featured models
As Familiar as Ever,
With More to Explore
- Developer-Firendly Environment
- 200+ Open-Source Gen AI Models at Your Fingertips
Certified Hardware Partnerships
Offices
Korea
8F, 102, Jeongjail-ro 239, Seongnam-si, Gyeonggi-do
803-C, 700, Daewangpangyo-ro, Bundang-gu, Seongnam-si, Gyeonggi-do
Japan
Toranomon Hills Business Tower 15F, 1-17-1 Toranomon, Minato-ku, Tokyo, Japan
Saudi Arabia
SA Riyadh #4303, Ask Shaikh Ali Al Hamad As Salihi Street, Al Nakheel District
