problems for buying new gtx580 card

We are planning to buy GTX580 for scientific computing, and have several questions:
Our computer is Dell pricision T5500( should have PCI_Ex16 bus acording to dell’s website), the operating system is window 7.
1, can we use this new card (580) for scientific computing only, or say whether it is ok if we don’t connect this card to any monitor. ( I heard if the monitor is not connected to the card, the windows will disable the video card)
2, I heard that there is windows dog timer problem. The cuda code cannot take the resource for more than 5 seconds. My question is whether this times is for my whole the cuda code or for kernel only. We are going to run the code continuously for an hours, but each kernel will last than 1 second, will this cause any problem.
3, We currently have Quandro nvs 420 on own dell (precision T5500 computer). And we run the bandwith test code provided by nvidia, host to gpu bandwidth is only about 1GB/s. This seems to be too low, PCI_Ex16 should have 4GB/s bandwithd. For the new card, I guess we will have same host to device bandwith, that is very disappointed? any suggestion or comment?

If you want to run the GTX580 on a headless node, your best option probably is to run Linux. Under Windows, the card would still be subject to watchdog timer timeouts, as the TCC driver is restricted to Tesla cards. Follow the link in my signature for some more info (IIRC there are also some hints on how to set a longer timeout).

If you want a high-end card you might want to consider waiting a few weeks, that’s when the Kepler architecture is expected to be released. Tech websites have claimed that the release date should be near the end of march (not confirmed officially by NV). The card is rumoured to have 1536 SP:s ( up from 512 in GTX580 ) along with a so-so impressive bandwidth of 192 GB/s.

Take all of this with a grain of salt, but I can say that at least I wouldn’t go shopping unless I really needed to at this point.

thanks all.