Altera or Xilinx

I am wondering if anyone is using an FPGA for any part of their trading, from just offloading the network stack all the way to running the entire trading system inside an FPGA card. If so I would enjoy hearing your experience and why you chose one solution over another.

To get this started I thought I would title the thread on something relatively easily comparable, the choice of FPGA chip itself, but feel free to discuss any of it.

 

I'd look at using a GPU with OpenCL before going with a hardware implementation.

 

I would say look at Xilinx. Beecube and Mercury...

Please don't ask 8^(

Cordially,
-DD-

 

Thanks for all the replies. This was particularly interesting, as I had never heard of BeeCube or Mercury. Also led me to the FPGA site.

Thanks!

 

the whole point is to offload things before they hit the kernel (or to make sure they never even get there). can't do that with a gpu. not yet anyway.

 

i just read the nasdaq will use fpga to publish market data. And that tibco is putting there ftl middleware on a switch besting arista

 

Hi Nitro,

Your welcome.

If you don't have a deep understanding of computering, hardware and software and algorithmic first principles you will have a long road to hoe here. I'm not trying to disway, just have your eyes wide open 8^)

Here is some light reading to get you started at the International Journal of Reconfigurable Computing http://www.hindawi.com/journals/ijrc/

Cordially,
-DD-

"It's like they just gave you a new horse." -Dan Ingalls-

 

To just offload the network stack will generally get your poorer performance than a decent SolarFlare NIC install. Especially if you're running at reasonable speed (ie, 10GbE).

Altera has a slight technical lead over Xilinx at the moment - IMO - but the two companies constantly leap frog each other in little increments and over the course of an entire development cycle the difference between the two rarely amounts to anything meaningful.

GPU with OpenCL is terrible for anything network/latency sensitive, but has potential uses elsewhere in the computational chain. Classic YMMV.

Nasdaq's FPGA market data feed has been an embarrassing disaster. The street is littered with the corpses of failed FPGA development teams. Marketing literature from the likes of Fixnetix or Novasparks is embarrassing misleading shite, to be honest.

 

I've developed a ton of stuff in both. Altera is my current choice; the tool set is a huge plus. Never upgrade development environments unless necessary.

And dear god, never ever let a software engineer design HDL, or architect a FPGA solution, in either A or X. Its not software !!!

 

Yeah, I briefly considered the 7124fx

http://www.aristanetworks.com/en/products/7100series/7124fx/7124fx-development

I am just not sure I want to be stuck with a FPGA chip inside the switch. Solarflare has a card that attempts the same:

http://www.solarflare.com/Content/UserFiles/Documents/Solarflare_AOE_Overview.pdf

This is probably more along the right lines I think. Spending one time cost of $40,000 for one of these things is ok, but the recurring costs are what makes me balk. Implement an off the shelf feedhandler can run into a recurring cost of tens of thousands a month, or build it yourself which can take a year+.