Intermediate conclusions
The PINE A64+ looks good as a host for a bitboard chess program, using either a single board or a cluster. But accurate performance numbers will have to wait until some empirical data can be had.
Because the Cortex A53 SoC and the RAM is targeted for use in battery powered devices, there are surely some significant compromises where throughput is traded for low power operation. I'll guess that the biggest of these is the effective speed of non-cached memory, all eight billion transistors of it. If the core portions of a chess program's code and data can fit into the L1 cache, then maybe slow RAM is not too much of a concern. But I'll bet that access to any transposition table will be very slow compared to a desktop machine.
I have some doubt about the claimed 1 GiB Ethernet link; in some other single board computers, the Ethernet link is implemented as a bridge form the board's USB 2.0 circuitry, and such may limit the top speed of the Ethernet port to 480 Mbps. Could the PINE board also have such a limitation? Perhaps so, with the port being able to talk with a 1 Gbps peer, but be limited to a 480 Mbps throughput. And there are some questions about latency and its effect upon cluster software design.
Can the project hold the price to US$19 and keep on schedule? I've got some real doubt about this given that the non-profit Raspberry Pi Foundation, with all of its support from Broadcom and a guarantee of a huge captive market from the British education establishment, charges nearly twice as much for the less capable Raspberry Pi B2.
I don't think that the US$16 price advantage vs the Raspberry Pi B2 is that significant since both boards require the addition of an SD card, a power supply, an enclosure, and an Ethernet cable (or WiFi USB adapter). After adding the cost of these necessities, I can see why many would prefer the slightly more expensive Raspberry Pi package when the Pi has so much support. This would apply even more so if the primary applications are happy with running in a 32 bit environment.
PINE A64+ 1.2 GHz quad core 64 bit: US$19
Moderators: hgm, Rebel, chrisw
-
- Posts: 12538
- Joined: Wed Mar 08, 2006 8:57 pm
- Location: Redmond, WA USA
Re: Intermediate conclusions
$190 would get you 40 cores at 1.2GHz.
$1900 would get you 400 cores.
You would need a system with a good ARM cross compiler to get any real use out of it (other than as cheap diskless terminals).
$1900 would get you 400 cores.
You would need a system with a good ARM cross compiler to get any real use out of it (other than as cheap diskless terminals).
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.
-
- Posts: 12538
- Joined: Wed Mar 08, 2006 8:57 pm
- Location: Redmond, WA USA
Re: Intermediate conclusions
The price also does not include the SD card.
I do not know if the units can boot over a network disk device, so the price may be rather misleading if they cannot.
The real price would be $19 + SD Card price if you need it to boot up.
I do not know if the units can boot over a network disk device, so the price may be rather misleading if they cannot.
The real price would be $19 + SD Card price if you need it to boot up.
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.
-
- Posts: 1334
- Joined: Sun Jul 17, 2011 11:14 am
Re: Intermediate conclusions
I think the PINE will have an advantage even if the memory bandwidth has been neutered (which I doubt, since smartphone users complain if the bandwidth is too small) because the PINE uses DDR3 compared to the Pi's LPDDR2.
As for clock cycles, Turbo Mode on the Pi 2 would give you extra memory bandwidth, and 100MHz extra clock speed, which slightly narrows the gap.
Matthew: out
As for clock cycles, Turbo Mode on the Pi 2 would give you extra memory bandwidth, and 100MHz extra clock speed, which slightly narrows the gap.
Matthew: out
Some believe in the almighty dollar.
I believe in the almighty printf statement.
I believe in the almighty printf statement.
-
- Posts: 4366
- Joined: Fri Mar 10, 2006 5:23 am
- Location: http://www.arasanchess.org
Re: Intermediate conclusions
+ power supply etc.
Still, it's a real computer and certainly low power.
I have an original Raspberry Pi lying around .. I really only used it to see if my program would run on ARM hardware (it does) but I've thought about using it for a low-end chess server bot.
--Jon
Still, it's a real computer and certainly low power.
I have an original Raspberry Pi lying around .. I really only used it to see if my program would run on ARM hardware (it does) but I've thought about using it for a low-end chess server bot.
--Jon
-
- Posts: 373
- Joined: Thu Aug 14, 2008 3:21 am
- Location: Albuquerque, NM
Re: PINE A64+ 1.2 GHz quad core 64 bit: US$19
I have the OrangePi Plus which is very similar-- the main difference is having also EMMC storage which is a lot faster than using a MicroSD card.flok wrote:Yeah or this one: [link to $15 OrangePi PC removed]
1GB ram, quadcore, 1.6GHz
It's hard to recommend without caveats. If you expect 1.6 Ghz, or accelerated video or hardware codec decompression, you will be disappointed.
If you want a headless server to run chess and are willing to add a heatsink + fan, then you can get exactly what you want.
There is a push to get working video/decompression, which somewhat works. But don't expect it to work out of the box unless running android.
The boards (all of the H3 OrangePi boards) are overclocked by default and have a poorly chosen dvfs table. (dynamic voltage scaling) The result is you will not get 1.6 Ghz without adequate cooling and since it is overvolted to begin with, you risk longevity. There is also concern that the memory is not entirely stable at its default overclocked settings. I haven't experienced this, but I haven't run any serious testing of that yet.
Don't read this is a purely negative-- I actually like the device. Just be aware what you are getting into. This not a beginner friendly device because of the default overclock, the closed source MALI drivers, and the fact that the kernel isn't mainline supported yet. This may eventually change, but it's likely that by the time it does there will be better low cost options.
I plan on running a chess engine on the device soon. I'm waiting for thermal adhesive to arrive. But from my tests using regular grease, I can reliably run 1 core for a chess engine at 1.6 without a fan. I had planned on doing that, but I think instead I will switch to 1.2 Ghz and run 2 cores. This will reduce the voltage a lot and actually run much cooler. FYI, 1.2 is still an overclock. The spec is 1.0 Ghz, and if you run that with a proper dvfs table, you do not even need a heatsink. That would be the case even with all 4 cores.
-
- Posts: 4675
- Joined: Mon Mar 13, 2006 7:43 pm
Lessons learned
One lesson learned is that an original Raspberry Pi B doesn't always take CPU overclocking very well. I managed to brick mine after a few days of overclocking experiments.
-
- Posts: 793
- Joined: Sun Aug 03, 2014 4:48 am
- Location: London, UK
Re: Lessons learned
How did you brick a Raspberry Pi? Isn't all the code on the SD card?sje wrote:One lesson learned is that an original Raspberry Pi B doesn't always take CPU overclocking very well. I managed to brick mine after a few days of overclocking experiments.
Did you manage to actually destroy the hardware? I thought it has hard-coded temperature and voltage limits to make it pretty much impossible.
Disclosure: I work for DeepMind on the AlphaZero project, but everything I say here is personal opinion and does not reflect the views of DeepMind / Alphabet.
-
- Posts: 1334
- Joined: Sun Jul 17, 2011 11:14 am
Re: Lessons learned
This could be before Turbo Mode was added, when it still voided the warranty and ran at that clock constantly, but I agree that it should not have bricked the Pi unless you used the current_override command.matthewlai wrote:How did you brick a Raspberry Pi? Isn't all the code on the SD card?sje wrote:One lesson learned is that an original Raspberry Pi B doesn't always take CPU overclocking very well. I managed to brick mine after a few days of overclocking experiments.
Did you manage to actually destroy the hardware? I thought it has hard-coded temperature and voltage limits to make it pretty much impossible.
Matthew:out
Some believe in the almighty dollar.
I believe in the almighty printf statement.
I believe in the almighty printf statement.
-
- Posts: 4675
- Joined: Mon Mar 13, 2006 7:43 pm
Re: Lessons learned
Oh, it was broke for sure. I tried two different SD cards, then reformatted the cards with fresh boot images (verified) and tried again. No go; the Pi was busted with only the power LED lit and no activity on any of the others.
The overclocking was done via the standard raspi-config utility; no other intervention.
However, the board had lasted for a very long time with many hours of full-load operation, both at the standard 700 MHz rate and at various overclocked rates. So I had already gotten my money's worth from my first Pi. I now have two B2 boards, and no problems yet.
The overclocking was done via the standard raspi-config utility; no other intervention.
However, the board had lasted for a very long time with many hours of full-load operation, both at the standard 700 MHz rate and at various overclocked rates. So I had already gotten my money's worth from my first Pi. I now have two B2 boards, and no problems yet.