Crossfire questions

[Candymancan21]

Hey guys i have a few questions for you. I got my 4890 the other day and its much better then my 4850 but i decited to try crossfire between them because i havent done crossfire or SLI before.

The thing is tho when i run 3dmark vantage i get terrible screen tearing and have to stop the benchmark. When i run the crysis benchmark i dont have screen tearing but i do notice stuttering. I guess this is the microstutter, or it might have to do with the fact that the 4850 is a 512mb card. What i noticed during the crysis benchmark was my fps was around 35ish the whole time but the Min/max was like 9/50 and fraps was telling me my fps was in the low 20's even tho the game was saying 35, and with crossfire disabled running the 4890 by itself the fps was about 24ish but min max was about 20/28 there was no stuttering it was pretty smooth. Do you guys think its because of the 512mb on the 4850 causing that stutter i was see'ing or is it crossfire doing this? I dont know what the benchmark settings were at i just hit the benchmark in the game folder and it started running.

 

[apoppin]

Have you looked in CCC to see if CrossFire-X is even enabled?
- make sure to install the drivers twice if you don't see a Primary and a Linked adapter in the information section of Catalyst Control center

Do not disable Catalyst AI - set it to standard

No, the 512MB of the 4850 won't affect Crysis - not even up to 19x12 resolution
- however, pairing a weaker card with a much stronger card will not give you too much extra FPS and may even mess with your minimums

You set the Crysis settings IN-GAME and then run the benchmark with THOSE settings

 

[Candymancan21]

Originally posted by: apoppin
Have you looked in CCC to see if CrossFire-X is even enabled?
- make sure to install the drivers twice if you don't see a Primary and a Linked adapter in the information section of Catalyst Control center

Do not disable Catalyst AI - set it to standard

No, the 512MB of the 4850 won't affect Crysis - not even up to 19x12 resolution
- however, pairing a weaker card with a much stronger card will not give you too much extra FPS and may even mess with your minimums

You set the Crysis settings IN-GAME and then run the benchmark with THOSE settings

Yea its enabled, and i see both GPU's. I triedp laying crysis just now with crossfire enabled and it just crash's everytime before i even jump outa the airplane. No idea what the problem is. I mean i have no intentions of running crossfire with this 4850 anyway but i still want to see what all the mumbo jumbo is about.

The game settings were using 1680x1050 8xAA, and the highest details in game. You sure the 512mb is not affected by that because i think it is. Fallout3 started to stutter with 8xAA on my 4850 ?


Edit: now i cant even run the benchmark the game just crash's, and now i had a driver failer. Wierd.

 

[apoppin]

i have tested Crysis at 16x10 with a 4870-X2 + 4870/512MB and there was no noticeable difference with a 4870-X2 + 4870-1GB

Have you looked in "diagnostics" in CCC ?
- you have to check the box to enable it

i would try installing the latest drivers - again - overtop the old ones without uninstalling them

 

[Candymancan21]

Yes crossfire is enabled for the 3rd time, im not that ignorant.... The crash i found out was due to my cpu, i guess my psu isnt enough to handle crossfire with my high cpu oc so i lowerd my cpu clock down by 300mhz just so i could test this and it works fine now

512mb does affect games, did you run 8xAA like i am. Its causing alot of stuttering. When i lower it down to 2x and 4x tho it gets better. I never played crysis through before because with my 4850 turning AA up i had to much stuttering, even my 8800GTS 640mb ran better.

Fallout3 does the same thing with my 4850 i couldnt use any higher then 4xAA or i would start see'ing microstuttering. This was long before the 4890's even existed.

Well i think i got it working now, and im gettign wierd results. With crossfire disabled i get about 34fps falling outa the airplane 8xAA, with crossfire enabled i get 23fps. Not sure why my fps is lower with crossfire enabled.

 

[apoppin]

if you crank up the AA you will exceed the 512MB bandwidth
- 8xAA is kinda silly for Crysis imo

 

[Candymancan21]

Originally posted by: apoppin
if you crank up the AA you will exceed the 512MB bandwidth
- 8xAA is kinda silly for Crysis imo

That was my entire point on the 512mb argument tho. Anyway im getting more error from the ati driver and the game is crashing again so i guess it wasnt a power issue at least i think.

Something else is wrong maybe, this is starting to not even be worth trying out nothing but hassels so far. I reistalled the driver 9.5 like you said and it didnt do anything. FPS is not increasing with crossfire enabled. In fact all it does it lower fps or cause the game to crash... uhg any idea's ?

 

[Candymancan21]

Ok i got it working for sure now, with 0xAA crossfire enabled i get about 42fps falling outa the plane, with it disabled i get about 39. So 10% improvement. Does that sound about right ?

Seems like it does, and no wonder i was see'ing lower performance with AA enabled. Look at this review the 4850x2 gets lower fps then a single 4890 at 1680x1050 and higher because of the 512mb of memory. Thats the same thing i was see'ing, but when i turned AA off completetly i saw a improvement over the 4890 but only 3fps and this review shows the same exact thing. One thing to note tho is that my 4850 alone would have gotten probably like 20fps so if im getting 42 with crossfire then its deff working. It just isnt worth it over my single 4890 because of the memory like i thought, thats why i never got a 2nd 4850 to start with.

It does seem however in other games like call of duty even with 512mb its alot faster. I wish i had call of duty to check for myself.



http://www.xbitlabs.com/articl...rossfirex_4.html#sect2

 

[apoppin]

Originally posted by: Candymancan21
Ok i got it working for sure now, with 0xAA crossfire enabled i get about 42fps falling outa the plane, with it disabled i get about 39. So 10% improvement. Does that sound about right ?

Seems like it does, and no wonder i was see'ing lower performance with AA enabled. Look at this review the 4850x2 gets lower fps then a single 4890 at 1680x1050 and higher because of the 512mb of memory. Thats the same thing i was see'ing, but when i turned AA off completetly i saw a improvement over the 4890 but only 3fps and this review shows the same exact thing. One thing to note tho is that my 4850 alone would have gotten probably like 20fps so if im getting 42 with crossfire then its deff working. It just isnt worth it over my single 4890 because of the memory like i thought, thats why i never got a 2nd 4850 to start with.

It does seem however in other games like call of duty even with 512mb its alot faster. I wish i had call of duty to check for myself.



http://www.xbitlabs.com/articl...rossfirex_4.html#sect2

that is about right with CrossFire-X. Crysis is pretty CPU limited anyway and the 4850 does not add much to a 4890 except in a few games, evidently.

Coincidentally, i finished benchmarking 15 games featuring HD 4890 and CrossFire-X in *11* configurations:

4870-X2 + 4890 (925/1050)
4870-X2 + 4870 (750/900)
4890-CF (925/1050)
4890-CF (850/975)
4890 (925/1050) + 4870 (800/1050)
4890 (925/1050) + 4870 (750/900)
4870 (750/900) + 4890 (750/900)
4870-X2 (750/900)
4890 (925/1050)
4890 (850/975)
4870 (750/900)

i will have a review up on it this weekend; but if you have questions, i can give you a preview

 

[Candymancan21]

Ok thanks.

I decited to try some real testing, and my results are this

Crysis Ultra high details/1680x1050/

ATI 4850

8xAA.... 7fps
4xAA.... 20fps
2xAA.... 24fps
0xAA.... 26fps



ATI 4890

8xAA.... 27fps
4xAA.... 33fps
2xAA.... 35fps
0xAA.... 39fps


ATI 4890/4850 CF

8xAA.... 9fps
4xAA.... 22fps
2xAA.... 39fps
0xAA.... 45FPS

...........................................................

3dmark Vantage Performance

ATI 4850

7600 GT1. 20FPS GT2. 20FPS



4890

9668 GT1. 30FPS GT2. 30FPS



4890/4850 CF

10955 GT1. 36fps GT2. 36fps

................................................................


GRID

4850

8xAA.... 44fps
4xAA.... 50fps
2xAA.... 54fps
0xAA.... 65FPS

4890

8xAA.... 83fps
4XAA.... 94fps
2xAA.... 102fps
0xAA.... 119fps

4890/4850

8xAA.... 75fps
4xAA.... 91fps
2xAA.... 104fps
0xAA.... 123fps

........................................

Fallout 3

4850

8xAA..... 38fps
4xAA..... 45fps
2xAA..... 52F[S
0xAA..... 64fps


4890

8xAA..... 64fps
4xAA..... 77FPS
2xAA..... 84fps
0xAA..... 98fps


4890/4850

8xAA.... 60FPS
4xAA.... 75FPS
2xAA.... 80fps
0xAA.... 91fps





I have come to the conclusion based off these results that a single 4890 is better then a crossfired 4850 with 512mb. Fps deff went up vs my 4850 by itself sometimes a 80% improvement, but when its compared to the 4890 by itself it just doesnt make sense to use for heat and power reasons. Another reason would be the fact that the 512mb of memory isnt enough. Its very aparent in crysis, but it doesnt show in Fallout3 fps wise but when you play the game you do notice alot of hitching when you turn and stuff.

So im just going to keep my 4850 out because it also lowers my cpu overclock due to power, im glad i didnt buy a 2nd one i knew the lack of memory would cause a problem. This is why i never recomended duel 4770's to cheat your way to a 4870 or 4890. There 512mb would just limit any gain. At least when you play it becomes noticable with more AA from the hitching

 

[apoppin]

This is why i never recomended duel 4770's to cheat your way to a 4870 or 4890. There 512mb would just limit any gain. At least when you play it becomes noticable with more AA from the hitching

You example is extreme - pairing 4890 with 4850.
-- it would not be such a problem for CFired 4770s - which would beat up on 4890

.. and then again, most gamers do not use 8xAA with Crysis when it runs so slow anyway

i think the min/average/max also paints a better picture of CrossFire than just using averages

 

[Candymancan21]

Originally posted by: apoppin

This is why i never recomended duel 4770's to cheat your way to a 4870 or 4890. There 512mb would just limit any gain. At least when you play it becomes noticable with more AA from the hitching

You example is extreme - pairing 4890 with 4850.
-- it would not be such a problem for CFired 4770s - which would beat up on 4890

.. and then again, most gamers do not use 8xAA with Crysis when it runs so slow anyway

i think the min/average/max also paints a better picture of CrossFire than just using averages

How is it extreme it runs down to 4850 speeds anyway, it doesnt matter regardless the 512mb single 4850 or crossfired with the 4890 it still stutters when 4xAA is applied to crysis or fallout3. Look at the crysis results, notice how in crossfire and single the 8x and 4x results are the same. Thats the memory. Thats why i wouldnt use duel 4770's unless they had 1GB of memory. Im glad i didnt buy a 2nd 4850 because of this also. In fallout3 the fps is fine but its has major shutters when using 8xAA, and with 4xAA its still noticable outside when you run around.

 

[Zstream]

Originally posted by: Candymancan21

Originally posted by: apoppin

This is why i never recomended duel 4770's to cheat your way to a 4870 or 4890. There 512mb would just limit any gain. At least when you play it becomes noticable with more AA from the hitching

You example is extreme - pairing 4890 with 4850.
-- it would not be such a problem for CFired 4770s - which would beat up on 4890

.. and then again, most gamers do not use 8xAA with Crysis when it runs so slow anyway

i think the min/average/max also paints a better picture of CrossFire than just using averages

How is it extreme it runs down to 4850 speeds anyway, it doesnt matter regardless the 512mb single 4850 or crossfired with the 4890 it still stutters when 4xAA is applied to crysis or fallout3. Look at the crysis results, notice how in crossfire and single the 8x and 4x results are the same. Thats the memory. Thats why i wouldnt use duel 4770's unless they had 1GB of memory. Im glad i didnt buy a 2nd 4850 because of this also. In fallout3 the fps is fine but its has major shutters when using 8xAA, and with 4xAA its still noticable outside when you run around.

I do not understand, the 4890 + 4850 is the same as two 4850's in crossfire. If you noticed in your benchmarks the 2x and 0x generate nearly double the FPS. Anything beyond that you will be frame buffer limited thus not allowing the performance to scale. Also you should notice the min. FPS is generally higher with crossfire, at least for me it was.

Two 4770's 1gb is probably the best bang for the buck.

 

[apoppin]

Originally posted by: Zstream

Originally posted by: Candymancan21

Originally posted by: apoppin

This is why i never recomended duel 4770's to cheat your way to a 4870 or 4890. There 512mb would just limit any gain. At least when you play it becomes noticable with more AA from the hitching

You example is extreme - pairing 4890 with 4850.
-- it would not be such a problem for CFired 4770s - which would beat up on 4890

.. and then again, most gamers do not use 8xAA with Crysis when it runs so slow anyway

i think the min/average/max also paints a better picture of CrossFire than just using averages

How is it extreme it runs down to 4850 speeds anyway, it doesnt matter regardless the 512mb single 4850 or crossfired with the 4890 it still stutters when 4xAA is applied to crysis or fallout3. Look at the crysis results, notice how in crossfire and single the 8x and 4x results are the same. Thats the memory. Thats why i wouldnt use duel 4770's unless they had 1GB of memory. Im glad i didnt buy a 2nd 4850 because of this also. In fallout3 the fps is fine but its has major shutters when using 8xAA, and with 4xAA its still noticable outside when you run around.

I do not understand, the 4890 + 4850 is the same as two 4850's in crossfire. If you noticed in your benchmarks the 2x and 0x generate nearly double the FPS. Anything beyond that you will be frame buffer limited thus not allowing the performance to scale. Also you should notice the min. FPS is generally higher with crossfire, at least for me it was.

Two 4770's 1gb is probably the best bang for the buck.

No .. it is not the same

at least 4890+4870 is faster than 4870+4870

and 4870-X2 + 4890 is consistently faster than 4870-X2 + 4870

in EACH of these configurations - almost without exception - pairing a faster 2nd card in CrossFire-X is faster than pairing the same card with a slower card:

4870-X2 + 4890 (925/1050) [faster]
4870-X2 + 4870 (750/900) [slower]

4890-CF(925/1050)[faster]
4890-CF(850/975) [slower]

4890 (925/1050) + 4870 (800/1050) [faster]
4890 (925/1050) + 4870 (750/900) [slower]

4870 (750/900) + 4890 (750/900) [variable; slightly faster]
4870-X2 (750/900) [variable; slightly slower]


These individual cards were just used to show the scaling with my system:

4890 (925/1050)
4890 (850/975)
4870 (750/900 & 800/1050)
4870-X2 (750/900)

i am certain

edited for clarity

 

[Candymancan21]

Thats interesting, cant wait for you post your results with those config's so i can see the diff between say a 4870/4870 CF and a 4890/4870 cf. I wish i had a 1GB 4850 so i could actually see the benefits but i couldnt due to the frame buffer limit i had. I was always told that if you have a 4870 and crossfire with a 4850 the 4870 essentally turns into another 4850 so its just like 2x4850's. I always thought that was a little fishy tho because even if the 4870 downclocked to the 4850's speeds it does have GDDR5 memory which runs at lower timmings then the 4850. So i always thought it would actually be slower, because a 4870 single downclocked to a 4850 is actually slower.

Of course that 6fps the 4850 had over my card on 0AA the card was overclocked to 700/1050, my 4890 i managed to get to 975/1100 last nite and the fps went from 39 to 43 average. So the gap went from 6fps to only 3.

 

[apoppin]

Originally posted by: Candymancan21
Thats interesting, cant wait for you post your results with those config's so i can see the diff between say a 4870/4870 CF and a 4890/4870 cf. I wish i had a 1GB 4850 so i could actually see the benefits but i couldnt due to the frame buffer limit i had. I was always told that if you have a 4870 and crossfire with a 4850 the 4870 essentally turns into another 4850 so its just like 2x4850's. I always thought that was a little fishy tho because even if the 4870 downclocked to the 4850's speeds it does have GDDR5 memory which runs at lower timmings then the 4850. So i always thought it would actually be slower, because a 4870 single downclocked to a 4850 is actually slower.

Of course that 6fps the 4850 had over my card on 0AA the card was overclocked to 700/1050, my 4890 i managed to get to 975/1100 last nite and the fps went from 39 to 43 average. So the gap went from 6fps to only 3.

there is Zero doubt that 4890+4870 CF-X is faster than 4870-X2
- and 4890 (925/1050) + 4870 (800/1050) is much faster than 4890 (925/1050) + 4870(750/900)
- and 4890 (925/1050) + 4870 (750/900) is much faster than 4890 (750/900) + 4870 (750/900)
i have all the results now - i even finished the graphs/charts/images and have uploaded them for publication tonight. i will send you a link when it is up.

What you saw are my raw conclusions. Drivers will sometimes make Tri-Fire [4870-X2+4890] slower than 4890 (950/1050) CrossFire. But generally, CrossFire-X compensates for the slower card and it "adds" to the frame rates.

if you put two cards into CrossFire-X and overclock *one*, the frame rates increase - so clearly one card does not "clock down" to the slower card.

What was interesting was running 4870-X2 {essentially 4870 CrossFire} against 4890 [750/900] + 4870 [750/900] - essentially 4870 CF - BUT the the 4890+4870 was generally slightly faster then 4870-X2 :Q

 

[Candymancan21]

Originally posted by: apoppin

Originally posted by: Candymancan21
Thats interesting, cant wait for you post your results with those config's so i can see the diff between say a 4870/4870 CF and a 4890/4870 cf. I wish i had a 1GB 4850 so i could actually see the benefits but i couldnt due to the frame buffer limit i had. I was always told that if you have a 4870 and crossfire with a 4850 the 4870 essentally turns into another 4850 so its just like 2x4850's. I always thought that was a little fishy tho because even if the 4870 downclocked to the 4850's speeds it does have GDDR5 memory which runs at lower timmings then the 4850. So i always thought it would actually be slower, because a 4870 single downclocked to a 4850 is actually slower.

Of course that 6fps the 4850 had over my card on 0AA the card was overclocked to 700/1050, my 4890 i managed to get to 975/1100 last nite and the fps went from 39 to 43 average. So the gap went from 6fps to only 3.

there is Zero doubt that 4890+4870 CF-X is faster than 4870-X2
- and 4890[925/1050]+4870[800/1050] is much faster than 4890[925/1050]+4870[750/900]

i have all the results now - i even finished the graphs/charts/images and have uploaded them for publication tonight. i will send you a link when it is up.

What you saw are my raw conclusions. Drivers will sometimes make Tri-Fire [4870-X2+4890] slower than 4890 [950/1050] CroissFire. But generally, CrossFire-X compensates for the slower card and it "adds" to the frame rates.

if you put two cards into CrossFire-X and overclock *one*, the frame rates increase - so clearly one card does not "clock down" to the slower card.

What was interesting was running 4870-X2 {essentially 4870 CrossFire} against 4890 [750/900] + 4870 [750/900] - essentially 4870 CF - BUT the the 4890+4870 was generally slightly faster then 4870-X2 :Q

Yea that makes sense now that you mention it, when i was looking at GPU-Z and the CCC during the tests my 4890 didnt downclock to the 4850 speeds. It did however run cooler then it did by itself even tho load said 80% so i dunno if it did downclock and just didnt show it, or if something else was going on. GPU-z said the 4890 went to its 900/1000mhz clock during crossfire. Another thing i noticed was the 4850 had a lower load then the 4890 did. During Fallout 3 my 4850 was like at 40% load, the 4890 was at 80%

 

[apoppin]

i edited my post to make it clearer

at any rate, it appears there is some load balancing with CrossFire-X .. in your case, it seems that some of the load is just shifted to the 4850; and sometimes when the drivers are imperfect, it chokes. otoh 4870 and 4890 appear to be a better "match' in CF-X; especially if you OC the 4870 you get quite close to 4890 CF.

 

[Candymancan21]

Yea im not using the 4850 i already took it out, i just wanted to see what crossfire was about but its to slow to really show any gains over just using the 4890 by itself. That and my PSU was under heavy load. Had to lower my cpu overclock to keep the system stable. I dont think i'll ever do crossfire anyway, my board is doing crossfire at 8xPCI-E. I also use a PCI sound card my Audigy so if had another 4870 or 4890 the duel slot cooler would litteraly be up against it and probably raise temps. It wouldnt be able to get any air flow to the fan i bet.

 

[apoppin]

Originally posted by: Candymancan21
Yea im not using the 4850 i already took it out, i just wanted to see what crossfire was about but its to slow to really show any gains over just using the 4890 by itself. That and my PSU was under heavy load. Had to lower my cpu overclock to keep the system stable. I dont think i'll ever do crossfire anyway, my board is doing crossfire at 8xPCI-E. I also use a PCI sound card my Audigy so if had another 4870 or 4890 the duel slot cooler would litteraly be up against it and probably raise temps. It wouldnt be able to get any air flow to the fan i bet.

i totally agree with you. Frankenfire is not always an advantage; especially if the cards are mismatched and there are any issues with thermals or power supply. Mine is marginal for TriFire with a 4890-XOC at 850w

However, i got hooked on CF when i bought a cheap 2900p and OCed it to XT speeds and paired it with my 2900xt

Things have not been the same ever since
:Q

 

[Candymancan21]

Hey off topic here, but im getting ready to mail in my rebate for my 4890. Usually i just cut out the UPC label and put in the envolope. Do you actually have to do that ? Or can i scan the image on my scanner and mail it to them that way ? I hate cutting my box's what if the card is defective and i have to RMA it ?

 

[apoppin]

Follow the instructions to the T - if you want the rebate[period!]
- if it says "Original UPC", that is what it means - just keep a copy for yourself

But wait till the ultimate last postmark day to mail it if you are worried about the card's failure; because NewEgg won't take it back with the sticker cut off - most etailers won't and you have to go through RMA should it fail instead of returning it to the seller

 

[Candymancan21]

ah i see says original guess i have to chop it off

 

[Tempered81]

Originally posted by: apoppin

Originally posted by: Candymancan21
Thats interesting, cant wait for you post your results with those config's so i can see the diff between say a 4870/4870 CF and a 4890/4870 cf. I wish i had a 1GB 4850 so i could actually see the benefits but i couldnt due to the frame buffer limit i had. I was always told that if you have a 4870 and crossfire with a 4850 the 4870 essentally turns into another 4850 so its just like 2x4850's. I always thought that was a little fishy tho because even if the 4870 downclocked to the 4850's speeds it does have GDDR5 memory which runs at lower timmings then the 4850. So i always thought it would actually be slower, because a 4870 single downclocked to a 4850 is actually slower.

Of course that 6fps the 4850 had over my card on 0AA the card was overclocked to 700/1050, my 4890 i managed to get to 975/1100 last nite and the fps went from 39 to 43 average. So the gap went from 6fps to only 3.

there is Zero doubt that 4890+4870 CF-X is faster than 4870-X2
- and 4890 (925/1050) + 4870 (800/1050) is much faster than 4890 (925/1050) + 4870(750/900)
- and 4890 (925/1050) + 4870 (750/900) is much faster than 4890 (750/900) + 4870 (750/900)
i have all the results now - i even finished the graphs/charts/images and have uploaded them for publication tonight. i will send you a link when it is up.

What you saw are my raw conclusions. Drivers will sometimes make Tri-Fire [4870-X2+4890] slower than 4890 (950/1050) CrossFire. But generally, CrossFire-X compensates for the slower card and it "adds" to the frame rates.

if you put two cards into CrossFire-X and overclock *one*, the frame rates increase - so clearly one card does not "clock down" to the slower card.

What was interesting was running 4870-X2 {essentially 4870 CrossFire} against 4890 [750/900] + 4870 [750/900] - essentially 4870 CF - BUT the the 4890+4870 was generally slightly faster then 4870-X2 :Q

Hi Mark. You've used all the crossfire and tri-fire combinations through a number of games for your review site. What would you say is your favorite rig to play on. The one that produces the least hitching, the highest FPS and most fluid playback in games?

Is it 4890 OC Crossfire, or something else?

 

[apoppin]

Originally posted by: jaredpace

Originally posted by: apoppin

Originally posted by: Candymancan21
Thats interesting, cant wait for you post your results with those config's so i can see the diff between say a 4870/4870 CF and a 4890/4870 cf. I wish i had a 1GB 4850 so i could actually see the benefits but i couldnt due to the frame buffer limit i had. I was always told that if you have a 4870 and crossfire with a 4850 the 4870 essentally turns into another 4850 so its just like 2x4850's. I always thought that was a little fishy tho because even if the 4870 downclocked to the 4850's speeds it does have GDDR5 memory which runs at lower timmings then the 4850. So i always thought it would actually be slower, because a 4870 single downclocked to a 4850 is actually slower.

Of course that 6fps the 4850 had over my card on 0AA the card was overclocked to 700/1050, my 4890 i managed to get to 975/1100 last nite and the fps went from 39 to 43 average. So the gap went from 6fps to only 3.

there is Zero doubt that 4890+4870 CF-X is faster than 4870-X2
- and 4890 (925/1050) + 4870 (800/1050) is much faster than 4890 (925/1050) + 4870(750/900)
- and 4890 (925/1050) + 4870 (750/900) is much faster than 4890 (750/900) + 4870 (750/900)
i have all the results now - i even finished the graphs/charts/images and have uploaded them for publication tonight. i will send you a link when it is up.

What you saw are my raw conclusions. Drivers will sometimes make Tri-Fire [4870-X2+4890] slower than 4890 (950/1050) CrossFire. But generally, CrossFire-X compensates for the slower card and it "adds" to the frame rates.

if you put two cards into CrossFire-X and overclock *one*, the frame rates increase - so clearly one card does not "clock down" to the slower card.

What was interesting was running 4870-X2 {essentially 4870 CrossFire} against 4890 [750/900] + 4870 [750/900] - essentially 4870 CF - BUT the the 4890+4870 was generally slightly faster then 4870-X2 :Q

Hi Mark. You've used all the crossfire and tri-fire combinations through a number of games for your review site. What would you say is your favorite rig to play on. The one that produces the least hitching, the highest FPS and most fluid playback in games?

Is it 4890 OC Crossfire, or something else?

GTX-280 or HD4890 for single GPU

4890 CF clocked as high as you can get it for Multi-GPU (925/1050)

CrossFire-X3 is a novelty that in some cases can give you [much] higher maximums but often it suffers from lower minimums than CF.

My very next review is CrossFire-X eXplored and it is published tomorrow. You can see 8 CF/CF-X and 3 single GPU configurations compared over 15 benchmarks. Here is the very *rough* conclusion:

4870-X2 + 4890 (925/1050) [faster (and overall fastest)]
4870-X2 + 4870 (750/900) [slower]

4890-CF(925/1050)[faster]
4890-CF(850/975) [slower]

4890 (925/1050) + 4870 (800/1050) [faster]
4890 (925/1050) + 4870 (750/900) [slower]

4870 (750/900) + 4890 (750/900) [variable; slightly faster]
4870-X2 (750/900) [variable; slightly slower]


These individual cards were just used to show the scaling with my system:

4890 (925/1050 & 850/975)
4870 (750/900 & 800/1050)
4870-X2 (750/900)