NVIDIA'S DEEP LEARNING ACCELERATOR MEETS SIFIVE'S FREEDOM PLATFORM - FRANS SIJSTERMANS (NVIDIA) & YUNSUP LEE (SIFIVE) - RISC-V
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
NVIDIA'S DEEP LEARNING ACCELERATOR MEETS SIFIVE'S FREEDOM PLATFORM Frans Sijstermans (NVIDIA) & Yunsup Lee (SiFive)
NVDLA — NVIDIA DEEP LEARNING ACCELERATOR
IP Core for deep learning – part of NVIDIA’s Xavier SOC
Optimized for Convolutional Neural Networks (CNNs), computer vision
Targeted towards edge devices, IoT
Industry standard formats and parameterized
Why open source NVDLA
Encourage Deep Learning applications
Invite contributions from the community
©2018 NVIDIA CORPORATION 2
NVDLA – TOP LEVEL ARCHITECTUE
Control Bus
SM SM
Configuration SMblock
and control SM
Convolutional Convolution core
Post-processing
Buffer (MAC array)
Memory interface
SDRAM On-Chip RAM
©2018 NVIDIA CORPORATION 3
AREA, PERFORMANCE, AND POWER
SMALL CONFIGURATION LARGE CONFIGURATION
INT8 data path INT8, INT16, FP16 data path
1 RAM interface 2 RAM interfaces
No advanced features Weight compression
EXAMPLE EXAMPLE
Config: 512 MACs, 256KB buffer Config: 1024 8b + 512 16b MACs, 256KB
Area: 1.4mm2 in 16nm Area: 2.4mm2 in 16nm (excl. TCM)
Perf: 93 fps ResNet50 Perf: 230 INT8/115 FP16 fps ResNet50
Power: 107mW Power: 348 / 475 mW for INT8 /FP16
©2018 NVIDIA CORPORATION 4
SW ARCHITECTURE
compiler Run time
Caffe model params
Application
Compile time User Mode
Driver
parser compiler
Kernel Mode
Driver
loadable
DLA hardware
©2018 NVIDIA CORPORATION 5
Inserting video: Insert/Video/Video from File.
Insert video by browsing your directory and selecting OK.
File types that works best in PowerPoint are mp4 or wmv
VIDEO FILE
©2018 NVIDIA CORPORATION 6
There are lots of custom chips you can build with NVDLA
+ + +
7 COPYRIGHT 2018 SIFIVE. ALL RIGHTS RESERVED.
SiFive Chip
Designer
Design
Infrastructure
Come to my Keynote talk “Opportunities and Challenges of
Building Silicon in the Cloud” tomorrow morning at 9:20am!
EDA
IP
Fab
Package/Test
COPYRIGHT 2018 SIFIVE. ALL RIGHTS RESERVED.
SiFive Freedom Chip Platforms
• SiFive Tapes Out Multiple Base Platforms
– Demonstrates silicon capability of each platform
SiFive Freedom Chip Platforms – Enables RISC-V software development
– Reduces risk for customer
– Proves our and matures Design flow for each
platform
Next talk in theHighsame
Low-cost, 32-bit
room will
performance, 64- •
introduce brand new
Customization Capabilities
microcontroller
platform:
Freedom Revolution
bit multi-core
platform:
Chip – Platform with
Add/remove DesignShare and SiFive IP
– Customization of SiFive CPU IP
• Edge Computing
• Industrial IOT
HBM2 and
• Storage system56-112Gb/s SerDes
controllers (SSD) – Customer can add own IP into Platform
• AI Smart Camera • Datacenter (accelerators, co-processors, other IP)
• Edge Inference (AI) Accelerators
• Wearables • Linux applications
• Networking / • From Prototype to Production
Baseband
– SiFive handles all logistics, incl. fab, package, test
– SiFive scales to production
– Final delivery is packaged, tested, qualified Silicon
9 COPYRIGHT 2018 SIFIVE. ALL RIGHTS RESERVED.
Freedom Unleashed 64-bit Multi-Core RISC-V Linux Platform
• 1.5+ GHz U54-MC SiFive CPU
• 1x E51: 16KB L1I$, 8KB DTIM with ECC support
• 4x U54: 32KB L1I$, 32KB L1D$ with ECC support
• Single- and Double-precision floating-point support
• 2MB Banked L2$ with directory-based cache-coherence & ECC support
• ChipLink
• Serialized Chip-to-Chip Coherent TileLink Interconnect
Freedom U540, FCBGA, manufactured in TSMC 28nm
• DDR3/4, GbE, Peripherals
10 COPYRIGHT 2018 SIFIVE. ALL RIGHTS RESERVED.HiFive Unleashed: World’s First Multi-Core RISC-V Linux Dev Board
• SiFive FU540-C000 (built in 28nm)
• 8 GB 64-bit DDR4 with ECC
• Gigabit Ethernet Port
• 32 MB Quad SPI Flash
• MicroSD card for removable storage
• MicroUSB for debug and serial
communication
• Digital GPIO pins
• FMC connector for future expansion with
add-in cards
11 COPYRIGHT 2018 SIFIVE. ALL RIGHTS RESERVED.HiFive Unleashed with Microsemi PolarFire Expansion Board 12 COPYRIGHT 2018 SIFIVE. ALL RIGHTS RESERVED.
HiFive Unleashed with Xilinx VCU118 Evaluation Kit 13 COPYRIGHT 2018 SIFIVE. ALL RIGHTS RESERVED.
Freedom Development Kit Comes with Linux BSP Based on Debian or
Fedora
65% April
0% March
https://wiki.debian.org/RISC-V
14 COPYRIGHT 2018 SIFIVE. ALL RIGHTS RESERVED.COPYRIGHT 2018 SIFIVE. ALL RIGHTS RESERVED.
15COPYRIGHT 2018 SIFIVE. ALL RIGHTS RESERVED.
16DEMO Setup: HiFive Unleashed + NVDLA
• NVDLA small config
HiFive Unleashed FPGA
– 2048 MACs, 512 KB
• NVDLA mapped onto Xilinx
VU118 Evaluation Kit
I/O interfaces
RISC-V
NVDLA • NVDLA running open-source
CPU FPGA
YOLOv3 object recognition
• Linux OS running on HiFive
Mem IF Mem IF Unleashed
– Easy to port over
umd/kmd from ARM
DRAM
• Demo setup built with
OpenCV thanks to Debian
17 COPYRIGHT 2018 SIFIVE. ALL RIGHTS RESERVED.Check out HiFive Unleashed + NVDLA demo at SiFive Booth!
COPYRIGHT 2018 SIFIVE. ALL RIGHTS RESERVED.
18Customize your Freedom Chip with NVDLA Today!
• Open-source IP cores further lower the bar to implement RISC-V-based products
• Freedom chip platform offers a complete template SoC with software support
• Freedom Unleashed + NVDLA is a great starting point for smart IoT SoCs and devices
• Everything is open-sourced, so check it out and contribute yourself!
– NVDLA
• https://github.com/nvdla/hw
• https://github.com/nvdla/sw
• http://nvdla.org
– Freedom Platform
• https://github.com/sifive/freedom
• https://github.com/sifive/nvidia-dla-blocks
• Once you’re ready, please come talk to us for your RISC-V AI chip needs!
19 COPYRIGHT 2018 SIFIVE. ALL RIGHTS RESERVED.End COPYRIGHT 2018 SIFIVE. ALL RIGHTS RESERVED.
You can also read
