Distributed Endpoint Architecture

Single-Node Configuration: GPU + CXL Switch + 4 Type-3 Endpoints

HOST SYSTEM
🖥
CPU
Control Plane
âš¡
B200 GPU
80 GiB HBM3
8 TB/s
🔔
CXL Root
x16 Gen5
CXL 3.0 Switch
Non-Blocking Fabric • 2 TB/s Backplane
Endpoint 0
CXL Type-3
🧠
ARM Cores
8× Cortex-A78
💾
DDR5 DRAM
256 GiB @ 250ns
💿
NVMe Flash
4 TiB @ 15μs
Endpoint 1
CXL Type-3
🧠
ARM Cores
8× Cortex-A78
💾
DDR5 DRAM
256 GiB @ 250ns
💿
NVMe Flash
4 TiB @ 15μs
Endpoint 2
CXL Type-3
🧠
ARM Cores
8× Cortex-A78
💾
DDR5 DRAM
256 GiB @ 250ns
💿
NVMe Flash
4 TiB @ 15μs
Endpoint 3
CXL Type-3
🧠
ARM Cores
8× Cortex-A78
💾
DDR5 DRAM
256 GiB @ 250ns
💿
NVMe Flash
4 TiB @ 15μs
1 TiB
Total DRAM
16 TiB
Total Flash
240 GB/s
Aggregate BW
250 ns
DRAM Latency