I agree,a single cpu is more then fast enough for only trading, but what do you then with your real time TA and backtesting,what if you do trading in stocks and futures at the same time. For me the risk that something sooner or later shall go wrong is real,i don't take that risk with my money,1 missing trade is enough for the value of a complete system. When i built-up a new system or do an importend upgrade this must always run 24/7 no less than 14 days before i use it for trading. http://www.tradingcomputers.com/
If I were in the market for a new trading computer, I'd try to hold off a while. We've just seen our 1st round of "multi-core"... dual core, so far. However, we've been alerted to "quad core" being in the works. Multi-core, multi-threading is the wave of the future. We'll see more, better, cheaper a bit later on. Sometime in the next generations of chipsets I'll bet they will have a set-up where the user indicates which processes he wants to have priority for each core. Maybe a primary and secondary for each ... and the rest will be accommodated as needed but as lower priority.
gnome, You are right to consider this as a whole new ball game. The true revolution will be in software. Although multiprocessors have been used for at least 20 years in different forms, writing software for these has been a discipline by itself. Processes don't simply have to be assigned to a processor, but concurrent processes have to communicate with eachother (IPC) and have to be kept synchronous. Environments like servers, being often reproduced 1000's of times can be developed economically for SMP. Doing this in case of any garden variety coded application is altogether a different matter. No simple tricks with a compiler will do this for the average programmer. In fact, the future is far from clear. Will we go SMP or will we go cluster or something different still? Intel for example, eager to push on with its 2 or n processes in the same shrink-wrap package, is frantically trying to develop even a new language to cope with these problems. A period of at least 5 years or so will probably be required in order to get up to maturity with such initiative. Of course, multicores will perhaps enable you to run some unrelated tasks concurrently. But which serious trader is really looking to play some games on his same rig while running active realtime trades? We may want to do more sophisticated things in our trading though. That's possible, but you will have to jump into MP software engineering. This adds a considerable degree of complexity to design and ... debugging. No easy free rides today.
Fair enough. But what if waiting was not an option? If you had to buy a computer now, what type of processor would it have?
P4 or Athlon... I don't think it makes a significant difference. But as some reviews have said "dual core performs significantly better in multi-tasking environments", taking a chance on a new dual core would also be reasonable.
The Athlon X2s have had consistently better performance than the P4 X2s in reviewers' tests.... best "bang for the buck" is said to be in the low end... the X2 3800.
I don't believe that esignal currently supports multithreading. The esignal development team have stated that it doesn't on the esignalcentral.com website. They have said that they will start to convert the application once the CPUs are in widespread usage - which has yet to occur. Runningbear
This topic has been taken up several times before. Multithreading IS NOT Multiprocessing. You can run multiple threads in ONE process. You need distinct processes to run on distinct CPU's. One process has at minimum one thread. How to do this? Not so simple, depends on the problem to be solved and the programmer solving it. No simple or easy gimmicks exist like today's vendors want to make you believe. No 'automatic' conversion solutions of old software exist today and perhaps never will. Remember: 2 * 3Ghz = 3GHz? = 3.5GHz? = 4+GHz? (depends on a lot of factors) Never 6GHz. If you want to know: MEASURE (not easy either) Those rosy easy days of doubling speed every 18 months are finished. The manufacturers ran into the wall of Physics.