Crossfire and SLI performance question

rvjones10

Distinguished
Jun 18, 2009
33
0
18,530
I am trying to understand the benefit or performance equivalent of Crossfire/SLI setup as compared to a single graphics card.

Example: I have two ATI RX3850 OC cards that are each 512meg / 256bit bit which i am thinking of setting up in crossfire. Simple math tells me that I would get the performance of a 1gb / 512 bit card which i know is not true.

So my questions are what single card performs similar to these two cards? and how much of difference in performance if the PCIEx slots on the mobo are 16x/8x as opposed to 16x/16x?

Also regarding the two ATI RX3850 OC cards, I see lots of cards that are 512mb / 256 bit with a wide range of price points. Is there another spec on the cards that I should be looking at to judge performance?
 
Solution
What you get with crossfire is actually the equivalent of a faster core, but still with only 512 MB of graphics memory (since the textures have to be mirrored in both video cards' memory). Off the top of my head, I'd guess that a pair of 3850s would perform on par with a 4850, but that's more of a guess than any actual data.

As for specs to look at? Look at the memory clock, memory bus width, and memory amount as far as memory is concerned (in most cases, clock and bus width are more important than amount), and look at the number of shaders as well as the core clock for the core. If I remember right, your 3850 has 320 shaders, and they're clocked somewhere around 600MHz. Now, the number of shaders (sometimes called cores) is only...

rvjones10

Distinguished
Jun 18, 2009
33
0
18,530
Just wanted to add that i just read the sticky on Crossfire/ SLI and realize that there are many factors so a definitive answer is probably impossible with out more specific info.

If anyone has anything to add or a comment on my last question, it would be appreciated
 
What you get with crossfire is actually the equivalent of a faster core, but still with only 512 MB of graphics memory (since the textures have to be mirrored in both video cards' memory). Off the top of my head, I'd guess that a pair of 3850s would perform on par with a 4850, but that's more of a guess than any actual data.

As for specs to look at? Look at the memory clock, memory bus width, and memory amount as far as memory is concerned (in most cases, clock and bus width are more important than amount), and look at the number of shaders as well as the core clock for the core. If I remember right, your 3850 has 320 shaders, and they're clocked somewhere around 600MHz. Now, the number of shaders (sometimes called cores) is only directly comparable when the cards are from the same manufacturer. For example, you can tell the 4850, with 800 shaders, will be quite a bit faster than the 3850, with 320 shaders. However, the GTX 285, with 240 shaders, beats both because the Nvidia shaders are different than the ATI ones.
 
Solution