[TR] Graphcore AI chip competes with Nvidia A100 - Overclock.net - An Overclocking Community

Forum Jump: 

[TR] Graphcore AI chip competes with Nvidia A100

Reply
 
Thread Tools
post #1 of 15 (permalink) Old 07-16-2020, 10:30 AM - Thread Starter
Vandelay Industries
 
maltamonk's Avatar
 
Join Date: May 2015
Location: US/UK
Posts: 1,782
Rep: 85 (Unique: 63)
[TR] Graphcore AI chip competes with Nvidia A100

Quote:
British-made chip offers 250 TeraFLOPS of FP16.16 performance that beats Nvidia A100 in complexity
Quote:
Graphcore, a U.K.-based AI computing startup, has unveiled its new Colossus MK2 GC200 intelligence processing unit (IPU) featuring 59.4 billion transistors, making it currently the most complex chip in the world.

The IPU was designed specifically for machine intelligence tasks and can scale out to offer 1 PetaFLOPS of FP16.16 compute horsepower in a 1U machine or up to 16 ExaFLOPS in a datacenter.
Quote:
“J.P. Morgan is evaluating Graphcore’s technology to see if our solutions can accelerate their advances in AI, specifically, in the NLP and speech recognition arenas,” a statement by the company reads.
Source: https://www.techradar.com/news/this-...monster-ai-cpu
maltamonk is offline  
Sponsored Links
Advertisement
 
post #2 of 15 (permalink) Old 07-16-2020, 10:52 AM
sudo apt install sl
 
Join Date: Dec 2009
Posts: 7,038
Rep: 197 (Unique: 133)
Exciting, I read about this exciting IPU's performance this morning at hpcwire. Finally someone who can compete with nVidia at large-scale training. Heard many good things about their software stack Poplar.
Attached Thumbnails
Click image for larger version

Name:	Graphcore-Colossus-MK2-GC200_IPU-M2000-Server_Chip_1-1030x496.jpg
Views:	19
Size:	63.9 KB
ID:	360726  


Silent
(20 items)
CPU
Core i9 9900K... CoffeeTime! @ 5.1Ghz w/ 1.36v
Motherboard
Maximus VIII Formula
GPU
Titan RTX @ 2100Mhz/2075Mhz
RAM
TeamGroup Xtreem 32GB 3200Mhz CL15
Hard Drive
HP EX950 2TB
Hard Drive
Samsung 850 Evo 1TB
Hard Drive
Samsung 850 Evo 1TB
Power Supply
EVGA SuperNova 1200w P2
Cooling
EK Supremacy Full Copper Clean
Cooling
XSPC D5 Photon v2
Cooling
Black Ice Gen 2 GTX360 x2
Cooling
EK-Vector RTX RE Ti - Copper + Plexi
Case
Thermaltake Core X5 Tempered Glass Edition
Operating System
Solus Linux
Monitor
Acer XF270HUA
Keyboard
Cherry MX Board 6.0
Mouse
Logitech G600
Mouse
Alugraphics GamerArt
Audio
Definitive Technology Incline
Audio
SMSL M8A
▲ hide details ▲

Last edited by WannaBeOCer; 07-16-2020 at 11:02 AM.
WannaBeOCer is online now  
post #3 of 15 (permalink) Old 07-16-2020, 12:07 PM
Original 16-bit Genesis®
 
Omega X's Avatar
 
Join Date: Mar 2013
Location: That gap between the couch cushion.
Posts: 1,732
Rep: 68 (Unique: 45)
Uhoh, That is going to be a ding on the bottom line for Nvidia.
Omega X is offline  
Sponsored Links
Advertisement
 
post #4 of 15 (permalink) Old 07-16-2020, 01:24 PM
mfw
 
ToTheSun!'s Avatar
 
Join Date: Jul 2011
Location: Terra
Posts: 8,499
Rep: 521 (Unique: 240)
How did everyone let a startup make such a stupidly superior chip? It's kind of weird in this industry.

CPU
Intel 6700K
Motherboard
Asus Z170i Pro Gaming
GPU
Gigabyte 2070 Super Windforce OC
RAM
G.skill Trident Z 3200CL14 8+8
Hard Drive
Samsung 850 EVO 1TB
Hard Drive
Crucial M4 256GB
Power Supply
Corsair SF600
Cooling
Noctua NH C14S
Case
Fractal Design Core 500
Operating System
Windows 10 Education
Monitor
ViewSonic XG2703-GS
Keyboard
Vortex Race 3 RGB
Mouse
Glorious Model O Glossy White v2.0
Mousepad
SteelSeries QcK+
Audio
Fiio E17K v1.0 + Beyerdynamic DT 1990 PRO (Dekoni pads)
▲ hide details ▲
ToTheSun! is offline  
post #5 of 15 (permalink) Old 07-16-2020, 01:55 PM
New to Overclock.net
 
doom26464's Avatar
 
Join Date: Jun 2014
Posts: 930
Rep: 14 (Unique: 12)
Wait what? There a company that de throned the mighty king Nvidia on there top tier chip?

This for real??
doom26464 is offline  
post #6 of 15 (permalink) Old 07-16-2020, 02:04 PM
New to Overclock.net
 
EniGma1987's Avatar
 
Join Date: Sep 2011
Posts: 6,551
Rep: 348 (Unique: 257)
Id like to point out that this does not beat Nvidia A100, the title is just for clickbait. This chip does 250 TFlop of FP16, the A100 does 312TFlop (or 614TF with whatever "sparsity" is). The A100 also does a bunch of other things and can process higher precision floating point for the later parts of AI training. According to the article, this chip "beats the A100 in complexity", not performance. And by complexity they mean because it uses more transistors. This chip "competes with" the A100 in the same way a Radeon 7 Pro "competes" with an A100.

EniGma1987 is offline  
post #7 of 15 (permalink) Old 07-16-2020, 02:11 PM
sudo apt install sl
 
Join Date: Dec 2009
Posts: 7,038
Rep: 197 (Unique: 133)
Quote: Originally Posted by ToTheSun! View Post
How did everyone let a startup make such a stupidly superior chip? It's kind of weird in this industry.
It's what happens when comparing a GPU against a core designed specially for deep learning.

1,472 IPU-Cores vs 432 Tensor cores, it's about time nVidia just releases a TPU.

https://blogs.nvidia.com/blog/2020/0...cision-format/

Quote:
TensorFloat-32 is the new math mode in NVIDIA A100 GPUs for handling the matrix math also called tensor operations used at the heart of AI and certain HPC applications. TF32 running on Tensor Cores in A100 GPUs can provide up to 10x speedups compared to single-precision floating-point math (FP32) on Volta GPUs. Combining TF32 with structured sparsity on the A100 enables performance gains over Volta of up to 20x.
Quote: Originally Posted by EniGma1987 View Post
Id like to point out that this does not beat Nvidia A100, the title is just for clickbait. This chip does 250 TFlop of FP16, the A100 does 312TFlop (or 614TF with whatever "sparsity" is). The A100 also does a bunch of other things and can process higher precision floating point for the later parts of AI training. According to the article, this chip "beats the A100 in complexity", not performance. And by complexity they mean because it uses more transistors. This chip "competes with" the A100 in the same way a Radeon 7 Pro "competes" with an A100.
You are correct, they're only comparing the CUDA core performance not the Tensor core performance.

Silent
(20 items)
CPU
Core i9 9900K... CoffeeTime! @ 5.1Ghz w/ 1.36v
Motherboard
Maximus VIII Formula
GPU
Titan RTX @ 2100Mhz/2075Mhz
RAM
TeamGroup Xtreem 32GB 3200Mhz CL15
Hard Drive
HP EX950 2TB
Hard Drive
Samsung 850 Evo 1TB
Hard Drive
Samsung 850 Evo 1TB
Power Supply
EVGA SuperNova 1200w P2
Cooling
EK Supremacy Full Copper Clean
Cooling
XSPC D5 Photon v2
Cooling
Black Ice Gen 2 GTX360 x2
Cooling
EK-Vector RTX RE Ti - Copper + Plexi
Case
Thermaltake Core X5 Tempered Glass Edition
Operating System
Solus Linux
Monitor
Acer XF270HUA
Keyboard
Cherry MX Board 6.0
Mouse
Logitech G600
Mouse
Alugraphics GamerArt
Audio
Definitive Technology Incline
Audio
SMSL M8A
▲ hide details ▲
WannaBeOCer is online now  
post #8 of 15 (permalink) Old 07-16-2020, 03:18 PM
mfw
 
ToTheSun!'s Avatar
 
Join Date: Jul 2011
Location: Terra
Posts: 8,499
Rep: 521 (Unique: 240)
That makes more sense. So, basically, it's more expensive and less performant. What a great product!

CPU
Intel 6700K
Motherboard
Asus Z170i Pro Gaming
GPU
Gigabyte 2070 Super Windforce OC
RAM
G.skill Trident Z 3200CL14 8+8
Hard Drive
Samsung 850 EVO 1TB
Hard Drive
Crucial M4 256GB
Power Supply
Corsair SF600
Cooling
Noctua NH C14S
Case
Fractal Design Core 500
Operating System
Windows 10 Education
Monitor
ViewSonic XG2703-GS
Keyboard
Vortex Race 3 RGB
Mouse
Glorious Model O Glossy White v2.0
Mousepad
SteelSeries QcK+
Audio
Fiio E17K v1.0 + Beyerdynamic DT 1990 PRO (Dekoni pads)
▲ hide details ▲
ToTheSun! is offline  
post #9 of 15 (permalink) Old 07-16-2020, 03:54 PM
sudo apt install sl
 
Join Date: Dec 2009
Posts: 7,038
Rep: 197 (Unique: 133)
Quote: Originally Posted by ToTheSun! View Post
That makes more sense. So, basically, it's more expensive and less performant. What a great product!
They're only comparing CUDA core performance against their IPU. Nvidia's 432 Tensor cores crush their IPUs on paper. We'll have to wait for MLPerf results.

https://mlperf.org/training-results-0-6/

https://www.nvidia.com/en-us/data-center/a100/
Attached Thumbnails
Click image for larger version

Name:	Screenshot_20200716_155155.png
Views:	21
Size:	91.9 KB
ID:	360760  


Silent
(20 items)
CPU
Core i9 9900K... CoffeeTime! @ 5.1Ghz w/ 1.36v
Motherboard
Maximus VIII Formula
GPU
Titan RTX @ 2100Mhz/2075Mhz
RAM
TeamGroup Xtreem 32GB 3200Mhz CL15
Hard Drive
HP EX950 2TB
Hard Drive
Samsung 850 Evo 1TB
Hard Drive
Samsung 850 Evo 1TB
Power Supply
EVGA SuperNova 1200w P2
Cooling
EK Supremacy Full Copper Clean
Cooling
XSPC D5 Photon v2
Cooling
Black Ice Gen 2 GTX360 x2
Cooling
EK-Vector RTX RE Ti - Copper + Plexi
Case
Thermaltake Core X5 Tempered Glass Edition
Operating System
Solus Linux
Monitor
Acer XF270HUA
Keyboard
Cherry MX Board 6.0
Mouse
Logitech G600
Mouse
Alugraphics GamerArt
Audio
Definitive Technology Incline
Audio
SMSL M8A
▲ hide details ▲
WannaBeOCer is online now  
post #10 of 15 (permalink) Old 07-19-2020, 12:51 AM
Performance is the bible
 
Join Date: Apr 2009
Posts: 6,996
Rep: 441 (Unique: 304)
That depends on software.

Nvidia's biggest advantage beside the raw power, is software. They had many years of advantage in that regard and know how to route and offload the processing required to the tensor cores.
So if the software can run well on tensor core processing power, it will be able to take the big advantage of the A100 and that chip will not really be a threat to nvidia.
If the software can't or someone write a software that is not compatible with the tensor cores, than it will make sense to run on it on more "conventional" processing power, and for that the A100 might not be the best.

But also note that nvidia can do int4, int8 and have 2:1 FP32 which that chip doesn't. So I don't think they are going to really bring a threat to nvidia in systems that tensor really goes into work.


Defoler is online now  
Reply

Quick Reply
Message:
Options

Register Now

In order to be able to post messages on the Overclock.net - An Overclocking Community forums, you must first register.
Please enter your desired user name, your email address and other required details in the form below.
User Name:
If you do not want to register, fill this field only and the name will be used as user name for your post.
Password
Please enter a password for your user account. Note that passwords are case-sensitive.
Password:
Confirm Password:
Email Address
Please enter a valid email address for yourself.
Email Address:

Log-in



Currently Active Users Viewing This Thread: 1 (0 members and 1 guests)
 
Thread Tools
Show Printable Version Show Printable Version
Email this Page Email this Page


Forum Jump: 

Posting Rules  
You may post new threads
You may post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are Off
Pingbacks are Off
Refbacks are Off