AMD releases 'world's fastest' graphics card
One board, two GPUs.
AMD has launched its two-GPU Radeon HD 4870 X2 graphic card, as expected, with a host of suppliers offering boards based on the design.
AMD claimed it's the "world's fastest" graphics card.
The first in our in-tray: Sapphire's HD 4870 X2. Both GPUs pack in 800 unified shaders in cores clocked at 750MHz. Each connects to 1GB of 900MHz GDDR 5 graphics memory, linked over a 256-bit bus.
Sapphire's HD 4870 X2: world's fastest?
The two-slot, actively cooled card sports the usual TV-out port and a pair of DVI slots - Sapphire includes adaptors for VGA and HDMI.
The two GPUs operate co-operatively using AMD's CrossFire X technology. Put two of the new cards in a system with a suitable motherboard, and all four GPUs can work together to render scenes.
The HD 4870 X2 is available from today at the $549 (£288/€369) price point. Gamers on a budget may prefer to wait until September when AMD will roll out the $399 (£209/€268) HD 4850 X2
The 4850 X2 uses the same GPU as the 4870 X2, but clocked to 625MHz. Boards will contain 2GB of GDDR 3 memory.
That's right... but wait until they roll out the blazing 6MHz clock... I hear you'll actually have a 24 BIT address space with that sucker! Woohoo!
...640k will be all the RAM you will ever need.
Because the limit only kicks in at 4gb using a 32-bit operating system. Explained another way:
512MB system ram + 512MB video ram = 1gb total ram. Under the 4gb limit, so perfectly acceptable.
4gb system ram + 2gb video ram = 6gb total ram. You'll see 2gb system ram and 2gb video ram. 2gb system ram will sit there doing nothing.
2gb system ram + 4gb video ram = 6gb total ram. You won't have any system ram to play with, which is a Bad Thing (tm).
Where we've produced this 4gb ram limitation comes from 2^32 which is 4294967296 bytes or 4194304 kbytes or 4096mb or 4gb.
Hope that clears things up for you.
As was previously explained, a 32-bit system addresses 4Gb of ram. The issue with the cards, if all of the onboard card ram is mapped into the main memory address space, is that two of them contain enough ram to use *all* of the addressable space. Your system (graphics + system) uses a total of 1Gb, well within the limit, so your graphics card and motherboard memory can both be mapped into the 4Gb limit address space.
A 64 bit OS can *potentially* address much more memory (up to a maximum of 2^64 bytes), but practical hardware configurations will likely address much less.
Just a thought: your system might run a little faster if you have at least 1Gb of system ram. 512Mb would have it paging like a bad thing under many scenarios. For example, running Unreal Tournament 3 under XP with this much ram means that the game is largely unusable.
It's not using RAM, it's using address space.
If you have filled the address space with RAM then the graphics cards _memory_mapped_ space will overlap and you either lose RAM or you lose graphics card.
In your case you have space for both so everyone is happy and gay.
BTW Doesn't the kernel immediately claim 1GB of address space in XP?