Search
× Search
Thursday, March 28, 2024

Archived Discussions

Recent member discussions

The Algorithmic Traders' Association prides itself on providing a forum for the publication and dissemination of its members' white papers, research, reflections, works in progress, and other contributions. Please Note that archive searches and some of our members' publications are reserved for members only, so please log in or sign up to gain the most from our members' contributions.

One survey for GPU vs FPGA in the financial trading algorithm based on the background of FPGA OpenCL SDK.

photo

 Chuan Du, Experienced Ultra-low Latency Trading Platform Engineer

 Friday, June 28, 2013

Hi, I want to see whether most trading companies are using GPU platform to accelerate algorithm.And what are their opinions about FPGA platform. Particularly, after altera delivered OpenCL SDK which allows traders to use c algorithm on FPGA platform. In my opinion, FPGA will show its potential in the HFT algorithm as the high level hdl synthesis compiler develops. Could you share your ideas? Thank you.


Print

5 comments on article "One survey for GPU vs FPGA in the financial trading algorithm based on the background of FPGA OpenCL SDK."

photo

 Mark Brown mark@markbrown.com, Global Quantitative Financial Research, International Institutional Trading, Algorithmic Modeling.

 Sunday, July 7, 2013



its my experience fpga far out performs anything we have tried - so much so we have stopped trying to improve it nor anything speed related and just work on the algos now.


photo

 Chuan Du, Experienced Ultra-low Latency Trading Platform Engineer

 Sunday, July 7, 2013



@Mark These days I also made some research about the the utility of fpga in financial industry. GPU has better perfromance in float algos than FPGA. And the critical latency for trading is the traffic latency from client to fiancial server which many people tried to use Microwave product to resolve. So generally FPGA can only be suitable to reduce the low level hardware latency such as hardware TCP/IP.

1. Is this your means of "fpga far out performs anything we have tried"? Or can you share one or two bad performance of FPGA in your experience?

2. Algos should use one platform to run such as GPU,DSP,Manycore,FPGA. Which platform is your first choice? Thank you.


photo

 Christopher Reeves, Programmer at A2X Capital

 Wednesday, July 10, 2013



I think that using a GPU would be good for a task where the time saved to do the computation outweighs the time you take to prepare the data for a GPU calculation. FPGA is good for doing simple math fast.

As far as HFT it would be hard to do real time simulations on the GPU while trading, so I would prefer to do that beforehand and like optimize some variables on historic data.

As far as microwave networks they seem overpriced and I think Tradeworx uses one to the chicago exchange. Thesys technologies has microwave technology but I do not know which trading venues they route to but I am also interested in this technology.

do you know anyone who routes to the exchange over a microwave network?


photo

 Chuan Du, Experienced Ultra-low Latency Trading Platform Engineer

 Wednesday, July 10, 2013



I do not know people who focus on the microwave technology in financial industry. However I had 4 years work experience in NSN and Dragonwave which are leading microwave product providers. So I am very clear about this technology in telecommunication field.


photo

 Sheng Li, Senior FICC Trader at Guotai Junan Securities Co., Ltd

 Friday, August 2, 2013



Preparing the data is often one of the most time consuming part, esp if one is trading tens of thousands of products with all market micriostructure features.

Please login or register to post comments.

TRADING FUTURES AND OPTIONS INVOLVES SUBSTANTIAL RISK OF LOSS AND IS NOT SUITABLE FOR ALL INVESTORS
Terms Of UsePrivacy StatementCopyright 2018 Algorithmic Traders Association