I have always been interested in the use of cheap and efficient ARM based technology in trading. During the past few years – thanks to the pioneering work set out by the raspberry pi – there have been a great number of credit card sized platforms released with last year bringing the first Octa-core ARM offers to the market. As I wrote in previous posts it is possible to use these new and powerful platforms for back-testing and trading. This year we have seen the rise of several new platforms that can be used for this purpose and today I want to write a post about these octa-core platforms, the advantages/disadvantages of each one and how they may be effectively used for building efficient computer clusters for system mining, such as machine learning system mining. I will also talk about how they compare with the platform I am presently using and what factors anyone who wants to use these computers should consider.
–
–
The first octa-core credit card sized computer was the ODROID-XU4 which I bought and still use for machine learning based trading system mining. The ODROID-XU4 retails at 74 USD, comes with no charger, with no storage and with no wifi or bluetooth connectivity. This means that what you get for 74 USD is basically a “bare-bones” motherboard and you will practically need to spend about another 45 USD for shipping, an eMMC 16Gb module, a power adapter and an ethernet cable to practically use the ODROID-XU4. In the end we’re looking at a 120 USD computer that achieves about the same performance as a single logic core of an i7 4770 computer. To match the i7 you would need 8 of these platforms, meaning you would need to spend around 960 USD which is significantly more expensive than what it would take you to get an i7 6th generation processor that would out-perform your cluster of ODROID computers.
However there have been some recent developments in octa-core computers that may increase the efficiency of this type of setup. In particular FriendlyARM has created the NanoPC and NanoPi offers which are also octa-core processors that – unlike the ODROID-XU4 – come in a regular instead of a big-little configuration, meaning that all of the cores can be used to their full power. The main difference between the NanoPC (60 USD) and the NanoPi (35 USD) is that the NanoPC includes an eMMC 8Gb module for storage built into the board while the NanoPi does not contain storage so you need to buy an additional MicroSD card. The eMMC storage is much faster than the MicroSD storage so you will want to have the eMMC for the fastest performance. Considering the lack of storage of the NanoPi this means that you will need to spend at least an extra 7 USD to buy an 8Gb SD at a minimum but you might end up spending more depending on the amount of data storage you need for your trading needs.
–
–
Another important aspect to consider is heat dissipation, while the ODROID-XU4 includes a heat-sink and fan combo that does a decent job at cooling the ODROID, the NanoPC and NanoPi do not include any cooling and this cooling needs to be purchased separately. For the NanoPC this means spending an extra 1.99 USD on the heat-sink while for the NanoPi you need to spend an additional 5 USD on the heat-sink and fan combo. Cooling is extremely important with these processors because of the throttling that is applied if the upper limit temperature is exceeded, if this happens the processor clock speed will be lowered tremendously and you will get the same performance that you would get from a much slower ARM processor. People experimenting with these boards online have found that the NanoPi overheats much more – due to its smaller form factor – but both boards benefit greatly from using active cooling and from applying proper heat-paste to the heat-sink/processor interface.
If you factor in power-supply costs plus all the above the NanoPC will end up costing you around 70 USD while the NanoPi will end up costing around 55 USD. You can probably make substantial additional cost reductions by using a normal PC power supply to power the boards via the GPIO pins and using something like liquid cooling – immersing the boards in mineral oil and using a radiator for example? – to more effectively cool the boards but in the end you will probably not be able to get below the 70 USD mark when including shipping for the NanoPi and around 90 USD for the NanoPC. Although this is still cheaper than the ODROID-XU4 – showing the technology is getting more accessible – you’re still paying more than what you would pay for a 6th generation i7 that would have better overall performance. Although ARM technology is getting there we are yet to see a technology that is indeed more cost effective than traditional X86 intel based processor setups when it comes to initial costs. There is however a gain in energy terms when compared with intel processors. A fully loaded i7-6700 setup consumes a total of around 143W (see here) while a matching setup (9 NanoPi) would consume around 90W (if we consider they use the entire 10W power supply which they probably do as my ODROID-XU4 does).
–
–
Despite the lowering costs of ARM technology and the existence of a significant variety of new boards using octa-core technology there still aren’t any commercially available credit card sized computers that can really provide extremely lower initial costs for the same amount of TFlops as compared with intel processors. However the power savings can be significant so someone wanting to do a lot of number crunching might consider the longer term TFlop/watt savings that the ARM computers can provide. However the ARM platforms are also reasonably expected to be more likely to fail and more poorly supported than the intel processors so this need to potentially have to replace components more frequently should also be taken into account. I do hope someone comes up with a board that has a ton of ARM processors, sparing us the need to construct clusters. If you would like to learn more about how you too can use a computer like the ODROID-XU4 for back-testing and trading using software that can be compiled to function in ARM processors please consider joining Asirikuy.com, a website filled with educational videos, trading systems, development and a sound, honest and transparent approach towards automated trading.strategies.