i am thinking about to using a Amazon EC2 cloud, because my hardware is a crap at home and i dont want to investing much money for this to become more speed, first i want to ask some of programmers if you have tried this and how can benefit from the speed.
if this may good and works fine, i plan to entering with this cloud to play on CCT13 or playing other tournaments with access this cloud over internet, so i need only to bring my laptop to connect into this cloud
i see that Jon Dart will be using a Amazon EC2 with his Arasan engine on CCT13, on 8 threads, but i see on Amazon a 2,5 x 8 core units, that makes 20 cores ?! , may i can using 20 cores of this?!
When they say 20 cores that's just a measure of compute capacity. It doesn't equate to 20 physical cores.
Most of the Amazon instance types are not suitable for programs like chess engines that want to have 100% CPU usage all the time. They aren't intended for that, and they are less powerful than a cheap quad.
But the Cluster Compute nodes are different. They have 8 physical cores and you get performance comparable to a dedicated 2x4 core Xeon.
jdart wrote:When they say 20 cores that's just a measure of compute capacity. It doesn't equate to 20 physical cores.
Most of the Amazon instance types are not suitable for programs like chess engines that want to have 100% CPU usage all the time. They aren't intended for that, and they are less powerful than a cheap quad.
But the Cluster Compute nodes are different. They have 8 physical cores and you get performance comparable to a dedicated 2x4 core Xeon.
--Jon
mmm...that sounds not good, but anyway if i can using 8 cores its enough too for me.
i didnt try it since yet, how its works and how much does it will be cost ,say 2 days for the CCT event only ?
expensive ?
seems difficult to become a sign without creditcard and if i show on youtube some videos tutorials its really complex to connect on the hardware, and how i can upload my engine and use it there ?
Engin wrote:
i didnt try it since yet, how its works and how much does it will be cost ,say 2 days for the CCT event only ?
expensive ?
seems difficult to become a sign without creditcard and if i show on youtube some videos tutorials its really complex to connect on the hardware, and how i can upload my engine and use it there ?
Last I checked it is $1.60 an hour. That is not exactly cheap but you are only paying for the hours you use - if you want to lease a comparable machine or part of one from most hosting providers, you have to pay a monthly fee ($300 - $400 a month typically).
Re setup, yes you do need a credit card I think. This type of instance only supports Linux. So you get a Linux root login and you have to set it up and manage it. If you're not comfortable doing that then this is not for you. Re upload, you can create a durable storage device (EC2 block storage) and mount it to your machine. File transfer is typically done using scp.
Ironically I just cancelled my free account today. It's an interesting concept but I don't really like how they word/perhaps do it for crunching.
They sell you time per core (as well as bandwidth and storage).
But say you wanted to rent 16 cores. Ideally I would want it on a machine with 16 cores, but you might be sharing 1 core on this machine, 2 on that, 5 on another, 1 on another and who knows literally in the world where those different "cores" are.
Web and Saas it seems great, but for my needs didn't like it for intensive crunching.
The cluster compute instance type is designed for number crunching. You are running a virtual image on a specific type of physical hardware (2 x Intel Xeon X5570, quad-core “Nehalem” architecture). However, this is really intended to be used by programs that have a distributed architecture and can make use of multiple compute nodes - if you can't do this you are limited by the power of one host and that is still limited to 8 cores. (You can rent a dedicated 16-core or higher machine (not from Amazon) but it's very expensive).
This is having some trouble staying connected to FICS (on US West Coast) from Amazon's Virginia data center. It worked better with ICC. (Amazon only provisions cluster instances in their East Coast zone).
