I called Dell support and they replaced it with the exact same graphics card. Yes, you heard right. My geforce gt 220 was causing my computer to crash and then Dell replaced it with another geforce gt 220. And somehow that solved the problem; I have never had graphics card-related problems since.
Also, you should try installing the latest graphics card driver for it on the nvidia website.
The driver is not directly in cause, but definitely this is a severe compatiblity problem with the embedded firmware of this graphics card and the BIOS. Some PC manufacturers (notably for notebooks) designed their PC with this graphics card that unforunately does NOT have a flashable firmware (VGA BIOS) which incorrectly reports the "WMI" settings for memory contraints.
The GT220 card CANNOT decode memory addresses higher than 4GB. But as it has generally 500MB of VRAM (actually DDR2), it incorrectly declares it to the motherboard BIOS without this constraint for mapping this VRAM correctly below the 4GB barrier. Even if the BIOS does the correct thing, it may then remp your normal RAM (SDRAM on your motherboards) so that some of it will be placed above the 4GB barrier (and so usable by the OS and applications).
Unfortunately, the nVidia video driver forgets to check this constraint, and (notably when using CUDA or with memory intensive application or intensive use of the GPU) the driver incorrectly accepts to perform memory transfers (e.g. images, textures, or CUDA kernels) between the VRAM of the graphic card and some SDRAM on the motherboard (used inside applications for CUDA, or the Windows display compositor or games or OpenGL drivers) in memory blocks above the 4GB barrier: the OS should not be allowed to instruct the nVDIA hardware to perform accelerated DMA transfer between this graphics card with any block of memory above the 4GB threshold.
Normally the GT220M should be able to decode 37 bits of addresses on the bus... but the motherboard was also designed with a buggy north bridge (from nVidia) which allows transfering only 32 bits of addreses (the 5 higher bits, that should allow adressing up to are silently cleared by the nVidia North bridge).
So the fault is inside the design of the north bridge and the PC manufacturer forgot to include on the mother board the correct address filter for the bus connector going to the graphics card, or the manufacturer did not connect the pins: the motherboard was designed only for 32 bits, and could only allow up to 4GB (of memory, but forgot to count the amount needed for mapping the VRAM, or other PCI devices, which at that time of 32 bits was supposed to use a maximum of 4GB).
What is worse, is that these manufacturers designed these boards to support 64bit processors, but forgot to remvoe the hardware limits. Instead, they internally tweaked the Video BIOS and the BIOS in an incompatible way, with a trick that only worked with Windows 7 64 bit; this was made against the specifications of nVidia (which were not clear enough), and Intel, and Microsoft was also not informed of that.
So the only solution for that problem is to REMOVE some physical RAM if you intend to use this graphics card with its bogous Video BIOS, and its bogous north bridge, and a PC BIOS which was also buggy! As well you could not use any other OS version that the one that was shipped with the PC (Windows 7 64 bit, but, not the standard distribution by Microsoft). As well to work in Windows 7, you needed a specific patched version of the nvidia GT2xxx video drivers (and of its companion CUDA driver), which slowed down considerably the operations (internally the patch was for the tweaked nvideo driver to allocate a block of normal RAM below the 4GB threashold, and copy block with the CPU to that window if the OS or application ever passed a block of memory residing about the first 4GB of SDRAM memory, if ever then it attempted to use transfers by DMA. However this tweak made by these PC manufacturers were not correctly reported to Windows for the device manager.
The same hardware problem occured as well with any other OS: as soon as a DMA transfer is initiated with some RAM locatad above the 4GB barrier, in fact the graphics card or the faulty north bridge will start overwriting regions in the 1st gigabyte of memory, including thoese regions used by other devices for their window of buffers (like Ethernet adapters, or the legacy VGA adapter, or the bus controlers): this causes an instant crash of the system and instant shutdown.
You can use memory tests like MEMTEST86+ and you'll find that memory is not at all the issue. The CPU is not also the cause, and the PC BIOS is not the cause... as long as you don't attempt to use DMA with that graphics card, whose Video BIOS incorrectly says that it should decode 37 bits, i.e. up to 128GB, where in fact the combination of the north bridge on the motherboard can only perform safe DMA transfers only inside the 4GB barrier.
And today this severe bug (in the motherboard design, in the PC BIOS that contains no valid fix in the DMI/WMI memory tables, and in the VGA BIOS of the nvidia GT220M graphics card, which also does not detect properly the bug/limit in the north bridge, which was NEVER designed correctly to work with more than 32 bits, even if it was designed for 64 bit processors!) is causing the same crash when botting any other OS (e.g. you cannot safely boot any Linux distrib with a "live CD", as soon as it starts displaying graphics video: there's no fix as well for this undocumented bug of the nvidia north bridge used in some of these motherboards.
Note that replacing the grahic car with another model may not necessaily work, unless their VGA BIOS includes a detection for the fault bridges, to check the memory limits for using them with DMA transfers: som of these alternate VGA BIOS may also simply stop using any DMA (the effect of acceleration is almost lost, notably for transfers of textures, or for working directly with accelerate block moves.
The only solution is then to remove some RAM: if you have such a faulty motherboard (notably from Asus, HP, Lenovo on their notebook), you cannot simply remove the graphics card (as you have no other controler to generate the video on the motherboard or in the CPU with an embedded GPU). In other word, do not install more than 3GB of RAM, even if the PC was designed or preinstalled with 4GB of RAM (and a user guide or promotion saying that you could upgrade to 4GB or sometimes higher! This was a lie from the PC manufacturer and was NEVER tested correctly, except with their own old tweaked version of Windows 7n, which was also incompatible with normal windows updates because this was made against all Microsoft specifications for Windwos!). These manufacturer lied and some of them were not able to find any valid replacement for the GPU board (notably on notebooks with the faulty nVidia chipset). And they did not provide any way to fix the graphics card in the VGA BIOS, simply beause that VGA BIOS was simply NOT flashable (you could not even read the VGABIOS normally, it was designed to be hidden behind the VRAM using an internal "bank switch" hidden in the tweaked display driver)
So if your PC has 4GB or more, remove some of them: use a 2GB SDRAM in one slot, and a 1GB SDRAM in the 2nd slot if you have one (operating in dual channel will not longer work). Or simply don't use any accelerated graphics mode or CUDA: you can only safely use that PC with 4GB only in MSDOS, or old Windows 95/98, or in Linux with a text console only, or with just the standard VGA modes (without any acceleration). This nVidia GPU board (and the nVidia north and south bridges in the chipset of the motherboard) is completely useless if you have more than 3G of SDRAM installed (this configuration becomes compeltely incompatible, causing instant crashes of system at any time, most often with instant emergency shutdown, and the cause is not at all overheating, and not a bug in Windows or Linux).
nVidia did not provide a way to fix the VGA BIOS and did not even provide any workaround in the Windows driver, and did not document the bug so that it could be fixed in a Linux driver (remember that nVidia dose not publish any specification for its bridges and GPU, it is completely closed and opensourced drivers are developed only by reverse engineering). nVidia not only kept its secrets, but it lied to customers, and to PC manufacturers as well and even lied to Microsoft for Windows, by not respecting the industry standards for many published specifications! In fact nVidia instructed PC manufacturers to hide the truth and perform silent returns of products, with some PC manufacturers. Others did not have the support contract from nVidia and chose to be silent because they wanted to keep their contracts for new ranges of nVidia products, in order to have GPUs at low prices, they had volume licences paid or to honour and did not want to su^pport the cost.
In summary: NVIDIA IS DEFECT BY DESIGN, and LIES to every one !