Poker-AI.org

Re: has anyone built a workstation with a lot of Ram?

2015-09-03T20:37:15+00:00

I'm using an old refurbished rack server with 72GB of RAM and dual Xeon L5420 CPUs. It's needed for my nash-eq development which currently eats up ~60GB by itself. Bought the server for about $1K off of eBay about 1 year ago.

Statistics: Posted by epo — Thu Sep 03, 2015 8:37 pm

Re: has anyone built a workstation with a lot of Ram?

2015-07-28T04:40:47+00:00

Nasher is trying to do a lot of things with it, one of them was finding large NEs. I consider it an investment. It's quite nice not to be limited by RAM. Having 24 effective Xeon cores doesn't hurt either, or a RAID array.

Statistics: Posted by cantina — Tue Jul 28, 2015 4:40 am

Re: has anyone built a workstation with a lot of Ram?

2015-07-26T14:52:26+00:00

Quote:

Hmm so what exactly is it in the code that needs 200gb of ram?
Some kind of massive multithreaded loops or something?

After the bot is built do you still need all that ram or is it just during development?

This just makes me feel I don't have any idea how tough it is going to be to build a bot..

Nasher was trying to find a NE solution which requires a super computer (actually a whole warehouse full of them). If your not trying to do that then just about any computer would work. If your running like 100 tables then you would need some kind of dual processor server or at least something with 8-16 cores. It just depends on what type of number crunching your doing in real time. Some botters pre-compute all the equities using a lookup table so there is not much real time processing and some don't (like me).

I do all my development on a 2008 2 core laptop which works just fine. The computer you will need just depends on what your trying to accomplish.

Statistics: Posted by shalako — Sun Jul 26, 2015 2:52 pm

Re: has anyone built a workstation with a lot of Ram?

2015-07-26T09:28:33+00:00

spears wrote:

Getrdyforthe wrote:

Is it really so necassary to have so much RAM for botting?
Is it because of the openCV that uses a lot of ram? Or are you botting 100+ tables or something?
Don't you think I could get by if 2 tabling heads up tables with just 8gb ram?

Nasher used a load of RAM to build his Nash Equilibrium model using an approach similar to that used by the leading academic researchers. If you know of a way to build a bot without doing that you don't need tons of RAM. It's nothing to do with openCV, number of tables etc.

Statistics: Posted by Getrdyforthe — Sun Jul 26, 2015 9:28 am

Re: has anyone built a workstation with a lot of Ram?

2015-07-26T09:16:41+00:00

Getrdyforthe wrote:

Statistics: Posted by spears — Sun Jul 26, 2015 9:16 am

Re: has anyone built a workstation with a lot of Ram?

2015-07-25T20:30:11+00:00

Statistics: Posted by Getrdyforthe — Sat Jul 25, 2015 8:30 pm

Re: has anyone built a workstation with a lot of Ram?

2015-07-23T04:00:02+00:00

You can get refurbished Dell workstations with server mobos if you want to save a few $$$ and buy the RAM separately in bulk. Expect to pay at least $3k+ if you want that much RAM, and do plenty of research to see how the RAM factors in on the price. Make sure you get something that doesn't require ECC RAM too, otherwise you'll be paying a whole lot more.

If there's something you wanted to run on my workstation, PM me.

Statistics: Posted by cantina — Thu Jul 23, 2015 4:00 am

Re: has anyone built a workstation with a lot of Ram?

2015-07-21T22:11:39+00:00

It's not unusual for servers to be able to take 256GB ram.

Statistics: Posted by spears — Tue Jul 21, 2015 10:11 pm

has anyone built a workstation with a lot of Ram?

2015-07-21T17:18:52+00:00

I think I remember, that Nasher said he had a workstation with about 200GB RAM.

A lot of RAm is of course crucial for building a big abstraction and a good bot, but most consumer PCs only have up to 32GB. How can I build a PC with lets say 256GB RAM? Most mainboards support way less RAM. Are there any special manufacturers? Are there good tutorials, how to do it? Is it worth it? I was never that happy with renting a cloud and it's very expensive after a while.

Statistics: Posted by HontoNiBaka — Tue Jul 21, 2015 5:18 pm