This……is the holy grail of all CPUs – The Threadripper 7995WX with 96 CORES and 192 Threads… But who would possibly spend over 10 LAKHS for just a single CPU?
Here’s the answer – and it’s called Algorithmic Trading.
Because this… is not how Trading works anymore – If you want to be the Harshad Mehta of 2024 *minus all the illegal part*-you don’t need bookies, in fact, you don’t even need to be in a trade center.
No, no – the modern-day traders are actually chilling in their rooms with a high-performing PC and 9 monitors.
And going one step further, some traders, or should I say developers actually build algorithms & AI models that do all the Trading for them.
These models have dug into tons of data since the stock market began, and that’s what we call Algorithmic or Quantitative Trading.
Now, this is not the same as High-Frequency Trading – HFT is a popular subset of Quantitative trading – it involves buying & selling 100s of options in less than a second, But Quantitative Trading can be fast, slow or somewhere in between.
It refers to the whole approach of using Computers & Data to find the best trading opportunities – instead of simply relying on your Gut Feeling.
and this Server is built for a company that builds Quantitative Trading Models.
Now, the reason why CPUs are used instead of GPU is because algorithms trained using CPUs are just more accurate, almost twice as much.
For everyday tasks – the lower accuracy of GPUs isn’t a big problem, but for financial & trading models, where even a small mistake can lead to huge losses – you’d want your AI to be as accurate as possible. Even if the training takes a much longer time.
So if you’re a Quant Developer or a Trading Firm, and you want to build your own system for Trading Algorithms, Go for a high-end CPU instead of a GPU.
And the Ryzen Threadripper Series is a great option to get started. this will make sure you have the highest accuracy for your trading models.
The second benefit of Choosing a CPU over a GPU is the RAM capacity.
And with the latest Thread Ripper 7000 series, you actually have 2 options for the Motherboard Chipset – the TRX50 and the WRX90, and the main difference between the 2 is the RAM capacity.
TRX50 only supports 4 channels of memory while WRX90 supports 8 channels. Now, your RAM should preferably be more than your dataset’s size – if not, you might face a lot of issues in training your model.
So this system has 8×64 – which is 512GB of DDR5 ECC Memory. But the maximum support on the WRX90 is 2 TERABYTES
That’s equal to getting 45 FLAGSHIP GPUs – without breaking the bank.
Also, ECC memory means almost no errors – and having 8 sticks of DDR5 memory gives you a combined memory bus of 512-bit. Which is much faster compared to the DDR6X memory of GPUs.
So If you want to increase the training speed on CPUs – go for the highest no. of RAM sticks.
So all of this housed in a server chassis with a 1200W redundant power supply makes sure – that our system gives the trader the best trading opportunities while also being reliable enough to make the split-second decisions required during the trade hours.
So, to sum it up – Hopefully, this blog guides you to pick the right hardware for building Trading Algorithms.
However, if you’re not into Trading – Now you know why some people choose CPUs for training AI models instead of GPUs – mainly because of the higher Accuracy and memory capacity.
So if you’re looking to build something like this of your own – Get in touch for a Free Consultation
