GPU brand compatibility question?

[ale555666]

Let say I buy a XFX 4850 but later on decide to do crossfire, but instead of getting another XFX video card I buy a HIS 4850 iceq4.
Will that work or do they have to be the same brand to work in crossfire?

 

[xc0mmiex]

they will work together just fine... if they have different clocks, the higher one will clock down, same goes for memory

 

[ale555666]

Thanks

 

[daedalus685]

they will work together just fine... if they have different clocks, the higher one will clock down, same goes for memory

That is incorrect.

While the memory amount will scale to the lowest card the clocks will not.

The cards work asynchronously in crossfire, it is not like SLI. A faster card will remain faster, and will improve performance slightly. While there seem to be driver issues with how much return you get from clocking the cards out of sync, neither will adjust its core speed from what you (or the bios) have it set to.

 

[xc0mmiex]

That is incorrect.

While the memory amount will scale to the lowest card the clocks will not.

The cards work asynchronously in crossfire, it is not like SLI. A faster card will remain faster, and will improve performance slightly. While there seem to be driver issues with how much return you get from clocking the cards out of sync, neither will adjust its core speed from what you (or the bios) have it set to.

http://ati.amd.com/technology/crossfire/faq.html


5. What happens when graphics cards with varying memory speeds and engine clocks are paired together?

A. In this scenario, overall performance will be adjusted to match the memory speeds and engine clocks of the lower clocked card. To obtain ideal performance, graphics cards should be paired correctly.


uhhhh, did i miss something?

 

[daedalus685]

The performance is roughly matched, the cards do not change their clocks though. The tasks assigned to each card changes. You can still clock one higher than the other and get a very marginal performance increase.

In first generation crossfire and in SLI the clocks are actually changed to run in sync, which is what you said in your first post, this is not how current crossfire works.

 

[xc0mmiex]

The performance is roughly matched, the cards do not change their clocks though. The tasks assigned to each card changes. You can still clock one higher than the other and get a very marginal performance increase.

In first generation crossfire and in SLI the clocks are actually changed to run in sync, which is what you said in your first post, this is not how current crossfire works.

can i get a link?

 

[daedalus685]

I had always assumed that crossfire worked like SLI until I was yelled at by the mod for the ATI forums for "spreading the rumor that crossfire downclocks the card when it actually runs asynchronously."

When I asked for a document to explain exactly how it works I got some "trade secret" mumbo jumbo.

At any rate, it is easy enough to show on your own.

Here is a link to the 8.1 driver news from fudzilla. Mixed crossfire was added for the first time with these drivers.

http://www.fudzilla.com/index.php?option=com_content&task=view&id=5641&Itemid=1

You can see on the first page them demonstrating clocking the cards independantly.

Google 4850+4870 crossfire, and 3850+3870 (the first combo to support mixed crossfire) for some reviews on the performance. The performance is almost identical to crossfire of the slowest cards. Hardware canucks has a thorough look.

If you have crossfire yourself, you can show youself that they don't underclock really easily, just use overdrive to clock them differently and observe the GPUz report of them both. The clocks will not change from what is user specified, and they can be changed independant of each other.

I have asked here several times for a ati white paper on the subject but none seem to exist at all. But it is easy enough to show that the cards are not changing thier clock rate. Thus I do not know if the drivers simply feed a different amount of informatino to each card proportional to its speed (not likely as this would improve performance) or if the drivers task each card with half of the work, add it in the buffer, and the fastr card simply waits for the slower to finish before the frame is displayed (How i believe it works).

I'd like to get mroe info on this but the topic always seems to provide dead silence.

 

[daedalus685]

Check wiki too, while the sentence doesnt itself have a reference this does appear:

"Unlike older series of Radeon cards, different HD 3800 series cards can be combined in CrossFire, each with separate clock control." http://en.wikipedia.org/wiki/ATI_CrossFire

Besides the addition of crossfireX it has not changed since the 3000 series.

 

[xc0mmiex]

lol yeh i saw the wiki post too... but failed to find anything in detail searching google for over 15 minutes, thinking that ati website may have some solid answers and i found the link posted earlier... i also knew that you could also select the clocks but i just thought they were automatically disregarded... plus since i've been here on the forums i herd they underclocked from many many people. I guess this is just another case of you have to believe in what they say without any proof of it.

 

[daedalus685]

Aye, I know for sure how crossfire used to work was what everyone says, as you used to be able to show the gpu downclock (old school crossfire with the dvi dongle). However, since that is how SLI works, everyone just assumes that is what happens now. I'm rather certain it isnt, but noone can tell me for sure either way.

It doesnt really matter though, the performance is still almost identical to the same crossfire with the slower card. It only comes into play when you use a game that doesnt support a multi GPU configuration even with crossfire turned on, as in these situations the faster clocks are actually important. I more interested from a curiosity and technical standpoint.