0% found this document useful (0 votes)
464 views6 pages

Klimax 210

The document describes a computing system architecture with multiple compute nodes connected via a PCIe fabric subsystem. Each compute node contains two CPU modules connected via QPI links and supporting up to 16 GPUs, with high-speed memory and fabric links. Additional nodes include a storage node for SSDs and a DBMS server node for database applications to access GPU resources.

Uploaded by

JhonCastro
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
464 views6 pages

Klimax 210

The document describes a computing system architecture with multiple compute nodes connected via a PCIe fabric subsystem. Each compute node contains two CPU modules connected via QPI links and supporting up to 16 GPUs, with high-speed memory and fabric links. Additional nodes include a storage node for SSDs and a DBMS server node for database applications to access GPU resources.

Uploaded by

JhonCastro
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 6

PCIe Device # 1 128Gbps MEM MEM

GPU, MIC, FPGA, or HCA/SSD

PCIe Device # 2 128Gbps


GPU, MIC, FPGA, or HCA/SSD

128Gbps

CPU Module
PCIe Device # 3 128Gbps

CPU #0
CPU #1
GPU, MIC, FPGA, or HCA/SSD

PCIe Device # 4 128Gbps


GPU, MIC, FPGA, or HCA/SSD
MEM MEM

PCIe Device # 5 128Gbps


GPU, MIC, FPGA, or HCA/SSD

PCIe Device # 6 128Gbps


GPU, MIC, FPGA, or HCA/SSD

PCIe Device # 7
128Gbps
GPU, MIC, FPGA, or HCA/SSD
MEM MEM

PCIe Fabric Subsystem


PCIe Device # 8
128Gbps
GPU, MIC, FPGA, or HCA/SSD

PCIe Device # 9

(22 port of PCIe gen.3/x16, 960Gbps switching, 1:5 blocking)


128Gbps 128Gbps
GPU, MIC, FPGA, or HCA/SSD

CPU #0
CPU #1

..
CPU Module/ Reserved
PCIe Device # 20 128Gbps MEM MEM
GPU, MIC, FPGA, or HCA/SSD
-
GPU # 1 MEM
GPU # 2

#0
CPU
CPU Module
GPU # 3

GPU # 4

Compute Node

QPI
GPU # 5

#1
CPU
GPU # 6

GPU # 7 MEM

GPU # 8


MEM

GPU # 16

#0
CPU
QPI
IB HCA # 1

CPU Module / Reserved

#1
CPU
PCIe SSD # 1
MEM

PCIe Fabric Subsystem


PCIe SSD # 2
PCIe SSD # 1 GPU # 1
MEM MEM
PCIe SSD # 2 GPU # 2

#0
#0

CPU
CPU

CPU Module
CPU Module
PCIe SSD # 3 GPU # 3

DBMS Server

Storage Node
PCIe SSD # 4 GPU # 4

QPI
QPI
PCIe SSD # 5 GPU # 5

GPU-DBMS Application

#1
#1

CPU
CPU
PCIe SSD # 6
MEM GPU # 8 MEM
PCIe SSD # 7

PCIe SSD # 8 IB HCA # 1

PCIe SSD # 9 MEM MEM


RAID Adapter # 1


#0

#0
CPU

CPU
PCIe SSD # 16 RAID Adapter # 2

QPI

QPI
IB HCA # 1 PCIe SSD # 1

CPU Module / Reserved

CPU Module / Reserved


#1

#1
PCIe SSD # 2

CPU

CPU
RAID Adapter # 1


MEM MEM

PCIe Fabric Subsystem

PCIe Fabric Subsystem


RAID Adapter # 2 PCIe SSD # 6
Host Memory Space
S

Load data

Initialize data

Memcp H2D GPU Memory Space


GPU Memory Space
Loop

Call kernel #1 Kernel on GPU #1

Call kernel #2 Kernel on GPU #2


Call kernel #n Kernel on GPU #n

Memcp D2D

Memcp D2H

Restore data

You might also like