I am looking for advice on building a NinjaTrader 8 PC.

I need a machine built for algorithmic trading that has the power to scan/trade all of the stocks on the NASDAQ and NYSE during regular trading hours.

(I am using an 8-core 2008 MacPro (2.8 GHz Xeon E5462 x2) with Windows 7 (64-bit). It functions just fine for coding, but I am scaling up my trading in the near future. My internet connection is 60 Mbps.)

Advice on all the standard bits: CPU, RAM, SSD, operating system, etc. would be greatly appreciated.

Thanks!

You might look at a Ram Disk to store your needed(recent) historical data (/db folder). Your heaviest cost will be in loading up all that data at startup and the file I/O overhead for that many symbols. Other than that the usual advice applies: the more cores/threads the better, use SSD's, ect. Based on my brief 1 year of using NT8, my advice is max out your RAM and Cores/Threads. Focus on the data bandwidth through the system/software, if NT8 runs out of RAM it goes to the drive and your performance degrades.

Here is a thread on the subject...

Speeding up NinjaTrader with a RAM drive (ramdisk) I have been able to make a large performance gain in NT 7 by storing the NT database files on a RAM drive and I thought that I would share what I have done. The time taken by NT to change the time frame of a chart from 3 to 5 minutes was 46 seconds and …

Not sure i perfectly understood your target setting.

Having all Nasdaq and Nyse symbols open at once, in NT 8, that doesn't sound like a good plan to me.
Did you actually test that already in dev ?

I would recommend to make a module and take in the necessary feed for those symbols outside of NT8 and then inject the orders into NT 8 and use NT8 for routing to your broker (if you are on IB, you may as well inject
directly in TWS).

Then you use NT 8 to look and follow and kill the trade if needed.

I am pretty sure that opening 3100 nasdaq charts with a strategy attached to it, is a receipt for disaster..

How fare are you with your concept ?
I would recommend doing a small POC (proof of concept)

I run an algorithmic robot on 21 futures in parallel, it runs on various machines,
- intel 20 core (i7-6950 / 128 RAM / 1TB SSD
- intel 6 core (i7-4960X / 32 RAM / 250 GB SSD
- AMD vishera 8 core (FX 8350 (very hot !!) / 64 GB RAM / 1 TB SSD

all of the machines perform equally (using NT7) , no lag noticed

Thanks for your reply. You have brought up some points that I would like to explore further.

My algorithm is 85-90% complete. I have been learning C# over the past 6 months, having just switched from StrategyDesk (TD Ameritrade). NinjaTrader is very robust, comparatively.

Being that I am new to NT8, I am still trying to figure out the best way to trade. My plan, up to this point, has been to follow the NinjaTrader instructions. As an example, I was going to assign the full NASDAQ list to my strategy in the “Strategies” tab. I would enable all of the symbols, and proceed with trading throughout the day.

I am currently in the backtesting phase. My computer runs out of RAM (16 GB) when I load more than 800 symbols. This bottleneck brought me here seeking PC spec advice.

I am new to scanning a large number of stocks. I, also, do not know anything about how to “make a module and take in the necessary feed for those symbols outside of NT8 and then inject the orders into NT 8 and use NT8 for routing to your broker.”

I am VERY interested in the module you are describing. Any info that you could provide would be a huge help! (It is so new to me, I do not know the proper terminology.)

I am currently with TD Ameritrade because of the free data feed. I will be switching to IB, when I start to trade.

I am still trying to learn the specifics about which data feed to get when I switch to IB (IQFeed vs. Nanex). And, if Nanex is the correct one, how do I get that feed into NinjaTrader 8. So much still to learn…

The issue you are describing is exactly what i was afraid of. I also think that a bigger machine will only move the issue to the next level, i simply think it may not be the best approach.

I think that IQfeed has even a memory issue if you subscribe 3000 symbols, @sam028 would be able to comment on that.

How complex is your algorithm ? (do you need dozens of exotic indicators from NT ?)
Does it require real-time feed and generate real-time signals ? (as compared to EOD)

It is important to design your system correctly, so that when you extend it, it doesn't fall over.

What i built is also in c#, it takes in IQfeed tick data real-time, calculates all kinds of things, generates the orders and manages the orders. Orders are injected into NT and i can then watch and (if i want) change or kill trades.

I was going to mention that concern as well, but felt it best to let the discussion progress a bit. I like the proof of concept suggestion, grow the concept organically, start with 10 symbols and see if you can get that working. You will learn a lot just doing that.

We are running roughly 50 symbols from IQ Feed through NT8 at a 1 tick interval, startup (and restarts) is the challenge (ie. the Ram Disk suggestion). Once up and stable resource usage is minimal, not a problem at all. That didn't happen by magic, it took a lot of time and hard work to get to this point. You really have to take the time up front and design all this correctly to avoid throughput bottle necks. I think somewhere between 50 and 100 symbols you are going to hit a wall, my initial thought was every N-symbols you will have to add another server.

I also think it's not realistic.
As far as I know IQFeed limit is 500 symbols in real time for example, for more you may use something like NxCore (which won't work with Ninja without creating a custom data feed).
Assuming receiving 3000 symbols data in parallel most trading platform won't be able to deal with the huge amount of data involved.

@rleplae: the problem I had is with parallel historical data download where the IQFeed client is crashing with more than 1000
download in parallel (now I limit my queries so there's never more than 512 download in parallel).

rleplae 

I would recommend to make a module and take in the necessary feed for those symbols outside of NT8 and then inject the orders into NT 8.

I am trying to understand this sentence from some months ago. How to make a module to be calculated outside of NT8, only to inject the results of the calculation back into NT8 once trade decisions have been made?

Is the module in a different, more advanced language? If so, what language? A different app maybe. I am keenly interested in understanding what you meant by a "module".

Thanks in advance.

ArmKnuckle 

I am trying to understand this sentence from some months ago. How to make a module to be calculated outside of NT8, only to inject the results of the calculation back into NT8 once trade decisions have been made? Is the module in a different, more advanced language? If so, what language? A different app maybe. I am keenly interested in understanding what you meant by a "module". Thanks in advance.

You can do this in c# (or any other language you feel comfortable with)