What's the optimal RAM for algo trading?

[schizo]

Are you doing HFT? Because if not, the latency for VPS from say NY or MD to Chicago is insignificant.

I am in a way. But the question is does memory have much effect overall?

Anyway, I currently have 16 GB on my laptop and when I check the spec in the Task Manager, I hardly ever see the memory spike above 70% even when I open multiple charts. So I don't think it will be much of an issue with 8 gig when you only run 2 strats.

 

[schizo]

Hello, 8gb RAM should be plenty for running two strats in NinjaTrader, even if they are computationally intensive.

Send us an email to support@ninjamobiletrader.com and I can help you out with a free trial so that you can confirm everything works well. Make sure to mention that I offered the free trial in this thread.

Same offer available for everyone!

Thanks for the offer.

 

[Paroult80]

Ram is super cheap these days. I do intense algo stuff to calibrate option models but no automatic trading and 96gig has never failed me but I don't do windows only Linux

So it works only on Linux?

 

[d08]

So it works only on Linux?

Likely a deliberate choice as it was for me. Linux is much more efficient in every aspect and you have full control of your OS, unlike Windows, which is for gaming and simple browsing.

 

[NorgateData]

In my opinion, Linux app developers are more selective of the way they allocate RAM for their storage structures and I/O approaches versus the "all you can eat" approach of many Windows devs.

Prior to SSD (and NVMe) , you would need to think (and test) about the way I/O scheduling was performed with respect to hard drives to anticipate best performance for spinning drives. This is not to say that SSD becomes an "all you can eat" approach, but it certainly improves performance dramatically on larger data sets (i.e. a few GB and above).

Personally (skunkworks level) on the Windows realm I'm playing with DirectStorage - i.e. read from NVMe directly to a graphics card's RAM, completely bypassing the CPU/Motherboard RAM.

 

[A_Syed]

This may sound silly, I am using a Raspberry Pi 4b with 8GB to run the algo thats using 3 times frames and 2 symbols. i use wine and ubuntu desktop for Raspberry Pi as OS. I have never seen usage spike over 6GB. I do not use any GUI, It all comes down to how is your Algo optimized, OS used and how much you understand hardware requirements. Just do some experimentation.

 

[NorgateData]

Yes, it does sound silly. Your usage scenario is atypical (2 symbols, 3 time frames).

But, in general - why use cheap (sub?) consumer-grade hardware for a professional endeavour?

(I actually use multiple Pi-Zero, 2, 4 and 4B for my hobbyist projects)

 

[A_Syed]

Yes, it does sound silly. Your usage scenario is atypical (2 symbols, 3 time frames).

But, in general - why use cheap (sub?) consumer-grade hardware for a professional endeavour?

(I actually use multiple Pi-Zero, 2, 4 and 4B for my hobbyist projects)

I respect your opinion,

https://ubuntu.com/raspberry-pi

I read this link I am sharing and the white paper attached and was convinced its good enough for my use case scenario. Again this is just me sharing something I do, cheap doesn't always mean its not suitable......

 

[Jack1960]

Depends on your algorithm. RAM is cheap. Just get 64 GB. An elite trader can spend a few extra bucks.

 

[d08]

I am in a way. But the question is does memory have much effect overall?

Anyway, I currently have 16 GB on my laptop and when I check the spec in the Task Manager, I hardly ever see the memory spike above 70% even when I open multiple charts. So I don't think it will be much of an issue with 8 gig when you only run 2 strats.

Not really, memory won't be significant at all. If it runs without paging on 8gb and 64gb, there's no real difference speed wise. With paging, you'll be extending your ram via the SSD but even that is extremely fast these days with NVMe.