06-18-2014, 01:52 AM
Is there any planned support for ARMHF as a host architecture? The Parallella (OpenCL) and Nvidia Jetson TK1 (CUDA) both provide reasonable performance in low power, low cost mobile boards.
06-18-2014, 01:52 AM
06-26-2014, 12:56 PM
does anyone know which speeds we could expect? is it worth it?
07-11-2014, 08:18 AM
Apologies for the late reply, I've done some testing and wanted to share the results.
First off, in the interest of full disclosure, I own both of these boards.
My approach to benchmarking has been to use an open-source platform like JtR to gather results and compare them to public benchmarks. JtR has proven to be fairly easy to jump in and make adjustments, as I've been able to separate out host architecture and acceleration without too many code changes.
Testing on the Jetson has shown it to be about 5-10 times slower (sometimes more for recent boards) than desktop GPUs, certainly nothing to write home about. The upside is that it achieves this for ~$200 a unit in a 5x5 inch board that consumes less than 5 watts, compared to the cost of host + GPU and a minimum recommendation of 300 watts.
I'm still working on getting the OpenCL libs working on the Parallella, but the native example they have for JtR against bcrypt is testing at >2K c/s, more than half of what an i7-2600K can do. Not bad for a $99 board a bit larger than a credit card!
I would expect there to be continued investments in ARM architectures supporting GPGPU processing, as it allows greater density at lower cost and power consumption. I don't know how many people would be looking to build a cracking cluster using these boards, but I would consider the mobility to be worth the time investment.
First off, in the interest of full disclosure, I own both of these boards.
My approach to benchmarking has been to use an open-source platform like JtR to gather results and compare them to public benchmarks. JtR has proven to be fairly easy to jump in and make adjustments, as I've been able to separate out host architecture and acceleration without too many code changes.
Testing on the Jetson has shown it to be about 5-10 times slower (sometimes more for recent boards) than desktop GPUs, certainly nothing to write home about. The upside is that it achieves this for ~$200 a unit in a 5x5 inch board that consumes less than 5 watts, compared to the cost of host + GPU and a minimum recommendation of 300 watts.
I'm still working on getting the OpenCL libs working on the Parallella, but the native example they have for JtR against bcrypt is testing at >2K c/s, more than half of what an i7-2600K can do. Not bad for a $99 board a bit larger than a credit card!
I would expect there to be continued investments in ARM architectures supporting GPGPU processing, as it allows greater density at lower cost and power consumption. I don't know how many people would be looking to build a cracking cluster using these boards, but I would consider the mobility to be worth the time investment.
07-28-2014, 07:30 PM
I have already started building such a cluster....
![[Image: dgOOk1e.jpg]](https://i.imgur.com/dgOOk1e.jpg)
I have another 6 coming in shortly with a planned goal of reaching aprox 100 in the cluster.
I have another 6 coming in shortly with a planned goal of reaching aprox 100 in the cluster.
03-26-2015, 05:51 PM
Hi,
Just inquiring on any progress made with OpenCL on parallella ? I understand COPRTHR is now in final ?
Pyrex, what kind of performance do you get from such a setup. What ratio performance ( results/# boards) Are these Jetson TK dev boards ?
Just inquiring on any progress made with OpenCL on parallella ? I understand COPRTHR is now in final ?
Pyrex, what kind of performance do you get from such a setup. What ratio performance ( results/# boards) Are these Jetson TK dev boards ?
12-20-2015, 12:19 PM
Hi,
how to compile successfully on Nvidia Tx1 or TK1 platform? any base requirement or Makefile suggestion?
how to compile successfully on Nvidia Tx1 or TK1 platform? any base requirement or Makefile suggestion?
12-20-2015, 01:54 PM
(12-20-2015, 12:19 PM)spider82 Wrote: [ -> ]Hi,
how to compile successfully on Nvidia Tx1 or TK1 platform? any base requirement or Makefile suggestion?
CUDA 7 won't work on a TK1. Source: https://devtalk.nvidia.com/default/topic...tk1-board/
12-20-2015, 05:49 PM
(12-20-2015, 01:54 PM)_NSAKEY Wrote: [ -> ](12-20-2015, 12:19 PM)spider82 Wrote: [ -> ]Hi,
how to compile successfully on Nvidia Tx1 or TK1 platform? any base requirement or Makefile suggestion?
CUDA 7 won't work on a TK1. Source: https://devtalk.nvidia.com/default/topic...tk1-board/
How about TX1? I got both.