Any plans for a 64 bit ODroid-MC1 with ARMv8?

Post Reply
Braden
Posts: 6
Joined: Fri May 29, 2020 1:44 pm
languages_spoken: english
Has thanked: 3 times
Been thanked: 1 time
Contact:

Any plans for a 64 bit ODroid-MC1 with ARMv8?

Post by Braden » Wed Jun 10, 2020 3:26 pm

Hi,

ODroid-MC1 has proven to be very good indeed. This paper https://www.sciencedirect.com/science/a ... 9X18322015 found that when using ODroid-MC1 for heavy duty Scientific Computing (Lattice_Boltzman flow calculations) it was only 4 times slower than $9000 server grade CPU's used for high performance computing.

They researchers concluded that it the would be likely that ODroid-MC1 would be significantly improved if it was 64 bit.

Is there any plans for a 64-bit ODroid-MC1 using ARMv8?

Thanks,

Braden

User avatar
odroid
Site Admin
Posts: 34642
Joined: Fri Feb 22, 2013 11:14 pm
languages_spoken: English, Korean, Japanese
ODROIDs: ODROID
Has thanked: 824 times
Been thanked: 712 times
Contact:

Re: Any plans for a 64 bit ODroid-MC1 with ARMv8?

Post by odroid » Wed Jun 10, 2020 3:30 pm

32bit XU4/HC1/MC1 have still much faster computing power than 64bit ODROID-C4.
Image

You can consider the ODROID-N2 if you need more computing horses. But the price is almost double or triple.
These users thanked the author odroid for the post:
Braden (Wed Jun 10, 2020 3:41 pm)

Braden
Posts: 6
Joined: Fri May 29, 2020 1:44 pm
languages_spoken: english
Has thanked: 3 times
Been thanked: 1 time
Contact:

Re: Any plans for a 64 bit ODroid-MC1 with ARMv8?

Post by Braden » Wed Jun 10, 2020 3:40 pm

Hi, thanks for that quick reply. I am a scientist, and I code, but I know nothing of hardware, so forgive this next part if it is ignorant.

I have downloaded and attached the paper since it is not free. They will explain their thoughts on ODroid-MC1 better than me, and you will understand them much better than me :)

I would be very interested on your thoughts. If they can run Lattice Boltzmann, then I can run molecular simulations, which is very impressive!

That is also a market HardKernell should look into - there are lots of us simulators who could use 32 CPU's (or more) at home.

Thanks,

Braden
Attachments
ODROID_MC1_Scientific_Flow_Problems.pdf
(1.43 MiB) Downloaded 13 times

User avatar
odroid
Site Admin
Posts: 34642
Joined: Fri Feb 22, 2013 11:14 pm
languages_spoken: English, Korean, Japanese
ODROIDs: ODROID
Has thanked: 824 times
Been thanked: 712 times
Contact:

Re: Any plans for a 64 bit ODroid-MC1 with ARMv8?

Post by odroid » Wed Jun 10, 2020 4:01 pm

That paper seems to be quite complicated for me.
Since I am not a scientist, it was not easy to understand what is Lattice Boltzmann exactly.
I hope other experts in this forum can give you any hint.

Braden
Posts: 6
Joined: Fri May 29, 2020 1:44 pm
languages_spoken: english
Has thanked: 3 times
Been thanked: 1 time
Contact:

Re: Any plans for a 64 bit ODroid-MC1 with ARMv8?

Post by Braden » Wed Jun 10, 2020 4:07 pm

Oh, the actual science is not very important. There is just alot of floating point arithmetic and quite a bit of memory used to hold arrays, and simulations can run for a long time, days even, although I don't think they ran it that long in this paper. I myself don't really know lattice-boltzmann either. I use different programs, but at the end of the day, they all store information in arrays and do matrix algebra.

It is just a really really demanding calculation. Most benchmarks are fairly simple and not useful tests for scientists. ODroid-MC1 did very well on a very demanding test!

Is there any way to test running scientific programs on ODroid-MC1 to see how it runs them? If it can run mine as well as it did the Lattice-Boltzmann calculations, I will be ordering one ASAP.

Thanks,

Braden

User avatar
mad_ady
Posts: 8155
Joined: Wed Jul 15, 2015 5:00 pm
languages_spoken: english
ODROIDs: XU4, C1+, C2, C4, N1, N2, H2, Go, Go Advance
Location: Bucharest, Romania
Has thanked: 568 times
Been thanked: 404 times
Contact:

Re: Any plans for a 64 bit ODroid-MC1 with ARMv8?

Post by mad_ady » Wed Jun 10, 2020 5:35 pm

There is at least an Odroid XU4 on the Odroid bench you can log into and play with remotely. It has the same cpu as mc1, but a different form factor. Search for Odroid Bench.
These users thanked the author mad_ady for the post:
Braden (Thu Jun 11, 2020 1:55 am)

Braden
Posts: 6
Joined: Fri May 29, 2020 1:44 pm
languages_spoken: english
Has thanked: 3 times
Been thanked: 1 time
Contact:

Re: Any plans for a 64 bit ODroid-MC1 with ARMv8?

Post by Braden » Fri Jun 12, 2020 3:33 am

I ran GROMACS on the bench.

For 300 water molecules it ran quite well, however at a more realistic size of 2000 water molecules it was notably slower. I believe this likely points to the fact that GROMACS is not optimized for ARM, and memory starts to become an issue as system size grows larger. A lot more information is stored. 2000 water molecules is 6000 atoms, each atom needs to store x,y,z coords, the charge, the velocity, and the x,y, z force components acting on the atom. All of these are floats.

Would a 64 bit ARMv8 overcome memory issues?

User avatar
mad_ady
Posts: 8155
Joined: Wed Jul 15, 2015 5:00 pm
languages_spoken: english
ODROIDs: XU4, C1+, C2, C4, N1, N2, H2, Go, Go Advance
Location: Bucharest, Romania
Has thanked: 568 times
Been thanked: 404 times
Contact:

Re: Any plans for a 64 bit ODroid-MC1 with ARMv8?

Post by mad_ady » Fri Jun 12, 2020 3:41 am

You may need more and faster ram. Try your test on the N2 on the bench, see how it goes.
These users thanked the author mad_ady for the post:
Braden (Sun Jun 14, 2020 8:16 am)

Braden
Posts: 6
Joined: Fri May 29, 2020 1:44 pm
languages_spoken: english
Has thanked: 3 times
Been thanked: 1 time
Contact:

Re: Any plans for a 64 bit ODroid-MC1 with ARMv8?

Post by Braden » Fri Jun 12, 2020 10:49 am

Thanks, I ran GROMACS on the N2 for 2000 water molecules (same as MC1)

Here is what I found:

single thread ---> 2.4 ns/day (nanoseconds per day larger is better)
6 threads -------> 3.6 ns/day (nanoseconds per day larger is better)

In comparison, using the Graham compute cluster in Canada, which was using 2 x Intel E5-2683 v4 Broadwell @ 2.1GHz

single thread --> 13.4 ns/day
6 threads ------> 68.8 ns/day

So the N2 is ~5.6 slower than a very expensive CPU which is not too bad. Also, I forget what the MC1 did, I think, it was about 1.3 ns/day, so the N2 is faster. The other thing is that Gromacs is not optimzied for ARM. There are other molecular dynamics programs, such as openMM, that have versions designed for ARM. However, installing openMM is not as easy as "sudo apt-get install gromacs"" :)
These users thanked the author Braden for the post:
Luke.go (Fri Jun 12, 2020 10:55 am)

Luke.go
Posts: 505
Joined: Mon May 30, 2016 1:55 pm
languages_spoken: english
Has thanked: 52 times
Been thanked: 53 times
Contact:

Re: Any plans for a 64 bit ODroid-MC1 with ARMv8?

Post by Luke.go » Fri Jun 12, 2020 10:57 am

Braden wrote:
Fri Jun 12, 2020 10:49 am
Thanks, I ran GROMACS on the N2 for 2000 water molecules (same as MC1)

Here is what I found:

single thread ---> 2.4 ns/day (nanoseconds per day larger is better)
6 threads -------> 3.6 ns/day (nanoseconds per day larger is better)

In comparison, using the Graham compute cluster in Canada, which was using 2 x Intel E5-2683 v4 Broadwell @ 2.1GHz

single thread --> 13.4 ns/day
6 threads ------> 68.8 ns/day

So the N2 is ~5.6 slower than a very expensive CPU which is not too bad. Also, I forget what the MC1 did, I think, it was about 1.3 ns/day, so the N2 is faster. The other thing is that Gromacs is not optimzied for ARM. There are other molecular dynamics programs, such as openMM, that have versions designed for ARM. However, installing openMM is not as easy as "sudo apt-get install gromacs"" :)
And, ONE Intel E5-2683 v4 Broadwell was $1846.00. :)

User avatar
mad_ady
Posts: 8155
Joined: Wed Jul 15, 2015 5:00 pm
languages_spoken: english
ODROIDs: XU4, C1+, C2, C4, N1, N2, H2, Go, Go Advance
Location: Bucharest, Romania
Has thanked: 568 times
Been thanked: 404 times
Contact:

Re: Any plans for a 64 bit ODroid-MC1 with ARMv8?

Post by mad_ady » Fri Jun 12, 2020 3:25 pm

You know, if you get your friends to buy N2s for themselves and their families and set them up as media players you could have all of them connect back to a central VPN and use them as a quiet distributed compute cluster with awesome media playback capabilities! The more friends you have, the bigger the bigger your cluster. Media playback is done on dedicated hardware anyway.

Post Reply

Return to “General Topics”

Who is online

Users browsing this forum: No registered users and 0 guests