systemml-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Niketan Pansare" <npan...@us.ibm.com>
Subject Re: GPU hardware support for SystemML bleeding edge.
Date Wed, 28 Mar 2018 18:06:05 GMT

Hi Janardhan,

We support Kepler and higher. Our kernels are compiled with sm_30 and we
depend on CUDA 8 and CuDNN 5. Both support kepler and higher. Please see
https://en.wikipedia.org/wiki/CUDA#GPUs_supported

>>  1. we have tuned a kernel only for a specific hw configuration(eg.
Maxwell)
with `asm()` code, can we integrate it into our project
>> 2. And can we ignore this hw specific optimized kernel for other
incompatible hardware and simply use the present kernels.

I would recommend that you first attempt to implement the kernel in .cu and
move to specific hardware optimization only if necessary. If you plan to do
latter because its absolutely necessary, I recommend that
- you optimize newer generations (i.e. Pascal) rather than older
generations
- and ensure that you optimize common cases (in a typical DL networks)
first.

That will make a better case to integrate hardware specific kernels into
SystemML.

For the second question, I refer you to
http://www.jcuda.org/jcuda/doc/index.html.

Thanks,

Niketan Pansare
IBM Almaden Research Center
E-mail: npansar At us.ibm.com
http://researcher.watson.ibm.com/researcher/view.php?person=us-npansar



From:	Janardhan Pulivarthi <janardhan.pulivarthi@gmail.com>
To:	dev@systemml.apache.org, Niketan Pansare <npansar@us.ibm.com>,
            Nakul Jindal <nakul02@gmail.com>, reinwald@us.ibm.com, Matthias
            Boehm <mboehm7@googlemail.com>
Date:	03/28/2018 08:57 AM
Subject:	GPU hardware support for SystemML bleeding edge.



Greetings,

What are the hardware(hw) configuration that SystemML (an opensource
project) needs to support?

i.e., do we have support for all of these?
 - Kepler
 - Maxwell
 - Pascal
 - Volta

Lets say,
1. we have tuned a kernel only for a specific hw configuration(eg. Maxwell)
with `asm()` code, can we integrate it into our project
2. And can we ignore this hw specific optimized kernel for other
incompatible hardware and simply use the present kernels.

@Niketan - If the answer for the 1 & 2 is yes, then we first tune for
maxwell and then we support others, case by case as they (openai) have
done.

Thanks,
Janardhan



Mime
  • Unnamed multipart/related (inline, None, 0 bytes)
View raw message