512mb 256-bit DDR5 or 868mb 448-bit DDR3

[Magusigne]

I know I'm indirectly comparing the GT2xx and 4xxx lines but seriously.

As a consumer I'd like to know how this matters when it comes down to gaming at high resolutions.. I was always told that high resolution=more VRAM

What about faster VRAM? I really don't know, so enlighten me.

 

[airhendrix13]

Think of the speed of the VRAM like the speed of a flowing river, the bit interface like the width of the river, and the capacity like the depth (best anology I could think of). GTX2xx has a slower flowing river, but is wider and deeper so more data (water) can flow through. The 4xxx has a fast flowing, more narrow river, which arguably gets the same amount of data through.

In terms of resolution, having more VRAM does help for the most part, but so does having faster VRAM.

Both matter is the simple answer. As of right now at super high resolutions like 2560 x 1600, the GTX280 comes out on top, but the 4870 is in a close second at that resolution.

 

[vgkarthik88]

great analogy there! if only reviewers *ahem* used that kinda language Everyone would understand! but in any case the end result you will see is the same. GDDR5 chips are pricier ( thats why the same core RV770 is $100 costlier in 4870 ) but u save by having a smaller bus. But as for your config, ur better of with a 4870

 

[Munky]

This indirect comparison is quite meaningless taken out of context with the video card as a whole unit. On the basic level, memory speed and bus width increase the available bandwidth, so 1800mhz gddr5 on a 256-bit bus would be roughly equivalent to 900mhz gddr3 on a 512-bit bus. As to what matters, the best answer is to just look at the benches. For example, the g92 8800gts is faster than the g80 gts in almost every test, despite the latter having more total memory and more bandwidth.

 

[Extelleron]

In terms of the bus width / speed of the memory, the only thing that matters is memory bandwidth. This is easy to calculate.

HD 4870 = 3600MHz GDDR5 * (256-bit/8000) = 115.2 GB/s
GTX 260 = 2000MHz GDDR3 * (448-bit/8000) = 112.0 GB/s

The HD 4870 has slightly more memory bandwidth than the GTX 260, but the two get their high memory bandwidth in different ways. The GTX 260 relies on a very wide bus but slow memory, meanwhile the HD 4870 relies on very fast memory connected to the GPU via a fairly narrow bus. Airhendrix13's explanation of this can help you visualize the difference.

As for the amount of VRAM, 512MB vs 896MB, they cannot be directly compared in this case. For nVidia GeForce cards, >512MB of memory seems to make a difference in certain situations. Theoretically more memory should give you better performance in high-res / AA+AF situations, but the GTX 260 doesn't seem to have any advantage in these scenarios over the HD 4870. AMD's cards use VRAM much more effectively, so they can do just as well with 512MB of memory as nVidia's cards can do with 896MB.

These benches can help show what I am talking about: http://www.pcgameshardware.de/...enchmark-Test/&page=10

Oblivion at high-res + Quarl's texture pack certainly consumes a lot of VRAM, yet the HD 4870 dominates here at high res + AA/AF.

 

[lopri]

My theory is that AMD's architecture is more efficient at moving/processing data in and out of RAM. AMD has been working in this side since the infamous 'ring-bus' of R600. On the other hand, NV started tying up processing units with x amount of memory since G80. I think this architectural decision affects the memory performance.

Said that, even if AMD's architecture is more efficient in handling memory, if an abslute amount of memory needed for rendering is beyond what a card has it will also hit a wall. Especially if such amount of RAM is required consistently.