Dedicated Server for computations

I am thinking about hiring a dedicated server to run my SQ.  Few questions:

1. Is having 12 cores really twice as fast 6 cores in reality?  I have choice between 6, 8, and 12 cores.  Need to know if it is really worth the value.

2. Choice between Windows or Linux.  I know SQ requires Windows to run, but would it be slowed down at all if running LInux and using the Windows compatable software to run windows based applications?

3. Choice of RAM.  2 GB, 32, 64 GB.  What difference would this really make?

I appreciate your advice, guidance, experience to help me make an informed decisoin.

-Stearno

1) As others reported SQ3 cannot fully utilize multicore CPU in standard way of use but there are options available. See here https://strategyquant.com/forum/topic/3397-genetic-vs-random-cpu-usage/?hl=load

2) This would be better answered by some Linux guy here on the forum

3) More RAM would help you when using larger populations so it really depends how you use genetic research. I have 12GB and it’s enough for me

Hello Stearno!

I will tell You from my own expirience,

1. I already have 2 proccessors with 8 cores + HT = 16cores +HT = 32 threads . That is working fine for me

2. I tried to set up on Linux + VMware , but VMware is limited to virtual proccessor, so I cannot had more than 8 threads (instead of 32!) , Now I have Windows Server 2012 R2 and works good. But firstly You need to install some of software to make it work good.

3. I have 64GB of ram and believe me – I can tottaly fullful it! 🙂

Matusiak,
Thanks for that insight from your experience. I ended up getting 12 cores with 24 threads. It has 24 ram. I also went with Windows 2012,because I figured it would be faster. Maybe I will switch after new version comes out and will work on Linux.

I set it to use all threads and 12 ram. With watching the monitor I saw it was only using 33% of the processor. So installed a second SQ. Now it was up to 66% usage on the monitor.

My first night I was able to Create over 2 million strategies with 3 data sets and doing all robustness tests. So that’s not bad.

I think this is a good cost effective setup and will allow me to always have one generating and one doing WFM, etc. I will try adding another SQ and see what that does but just testing the waters so far.

Again, thanks for your experience. It helps to hear to make best decision for me.

Sent from my HUAWEI MT7-TL10 using Tapatalk

2mln strategies + robustness tests?

I set ranking prefencies and it generates around 2-3 strategies per night with 100% processor utilization.

Haha, no it does not spit out 2 million valid strategies into the databank. It goes through 2 million variations to compare against the criteria, as can tell by the counting strategy number.

I started to pay attention because I wanted to find a guideline of roughly how many variations tested does it take to spit out valid strategies.

One guy said it took “14 million tests to find a valid strategy” . On my old computer I had to let it run for 4 weeks to get two valid strategies. But unfortunately, I did not see how many variations that actually was (and it was doing maybe 100,000 per day).

-Stearno

Sent from my HUAWEI MT7-TL10 using Tapatalk