Problem w/ 4890s in Crossfire

[BlueSkys]

Ok so I replaced my old gpu with a 4890, then a week later added another in crossfire. I didn't see any difference in fps in games so I started to wonder...

I'm playing at my monitors native resolution of 1920X1200 (except for 3DMark, which stays at 1280X1025)

Here's where I'm at right now, and Sooooo confused. I tried each 4890 separately and then together in crossfire, with the bridge in place and Crossfire enabled under catalyst, and here's what I've got...

3Mark06: Each individually scored 16012 exactly and together scored 18250, even though I'm watching the frame rates during the benchmarking and they are EXACTLY the same as they were individually.

World in conflict benchmark (Very High): Again, individually and separate they are getting an average of 36 frames per second. One of the singles actually got a 38 average. High and low fps are the same as well (variation of 1 or 2 fps.)

Crysis Warhead: Again the same... together or separate I'll play through the same part and watch the fps be virtually identical.

And why is it that when they test the 4890 for instance here... http://www.techpowerup.com/reviews/ATI/HD_4890_CrossFire/19.html it shows at LEAST 20fps increase among all resolutions in World in Conflict?

And why is it that on their test rig, which isn't so far off of mine, it gets 41fps increased to 78 in World in Conflict and I get 36 increased to well... 36.

It shows in GPUZ at the bottom under ATI Crossfire: Enabled (2GPUs)(Unsure on Vista64) Now I'm watching rivatuners low level hardware monitoring and when in crossfire, both graphs showing GPU usage fluctuate in the same manner although very slight differences which would indicate that in crossfire they are both working and working somewhat independantly. This in conjunction with the consistent though small bump in the 3DMark06 score would indicate that the crossfire is working, but NO INCREASE IN FRAMERATES ANYWHERE would indicate otherwise... and make me want to shoot myself in the face.

Please help!!! thx

My rig:
Cpu:Intel Q6600 oc'd to 3.3 ghz
GPU: 2 X XFX Radeon HD 4890 in Crossfire
OS: Vista 64
Mobo: MSI P45 Platinum
RAM: 4gb OCZ 800mhz
Power: Antec Truepower Quattro 850w

 

[Xazax]

Q6600 is holding you back abit there i would say hence the minor difference in 3dmark 06 which is heavily CPU bound, also the two games you mention, tend to hate ATI cards(Crossfire doesnt scale well either) and favor Nvidia more, in games like H.A.W.K.S Those HDF4890's would be producing some sweet FPS

 

[DaMulta]

Q6600 VS Intel Core i7 920 = 920 stomps the shit out of it.

You in what they say bottle neck. Your better using your 2ed card as a Eye-candy card most of the time.

Open up CCC and turn AA all the way up, if you have already done this then that's why you are not seeing a increase at all. Turning it all the way up uses one card for more Eye-Candy! Which is what ATi calls Super AA

 

[DaMulta]

Also install this little hot fix from Microsoft
http://support.microsoft.com/kb/936710

 

[BlueSkys]

It says the update does not apply in my system... my 3.3ghz Q6600 is really bottlenecking these cards?? And at 3.8 in the review they get a 30fps increase while I maybe get a negligible few??

 

[BlueSkys]

And I swear to god, I just ran the benchmark for hawks and with 1 card, high: 184, avg: 59

With 2 cards, crossfire enabled, high:134, avg: 59

???

 

[FelipeV]

Not only your clock is holding you back, but you will need more bandwidth, just overclocking your CPU to a higher clock wont do much of a help, you will need fsb at least higher then 470 to increase your bandwidth.

As you can see here
http://img90.imageshack.us/img90/2678/overclock.jpg
Increasing CPU Clock didnt help that much.

Crysis and WH, both love bandwidth and clock for MGPU setups, I and a friend was looking at this last week comparing his GTX 295 with my 4870X2, any clock and bandwidth is always wlecome.

You need bandwidth, after that you can use clock to get more fps.

A i7 setup offers good bandwidht compared to a Core 2 setup, probably a sotck i7 can do the same as a overclocked Core 2 in terms of bandwidth.

Another example of how MGPU setups are affected by overclock and increase in bandwidth and clock speeds. Its not a CF, but the effect its the same.

http://img134.imageshack.us/img134/2248/crysisislandhighcputest.jpg
http://img134.imageshack.us/img134/7056/crysisislandveryhighcpu.jpg

 

[3870x2]

Q6600 is holding you back abit there i would say hence the minor difference in 3dmark 06 which is heavily CPU bound, also the two games you mention, tend to hate ATI cards(Crossfire doesnt scale well either) and favor Nvidia more, in games like H.A.W.K.S Those HDF4890's would be producing some sweet FPS

It is well known that quads do great with xfire, and Q6600 OCed to 3.3 is a very good processor, the 6600 holds its own just fine when OCed 1.1ghz
It has to be something else. The only thing I can think of is immature drivers. Check your setup against other reviews, google 4890 xfire.

Just to clarify, a better CPU is always a great option, but this isnt a "bottleneck", it is CPU scaling.

 

[TRIPTEX_CAN]

@ 1920 his CPU bottleneck with that q6600 should be negligible. Crossfire Warhead scales really well in my experiences.

@ OP. Completely remove all your drivers and related software and reinstall the latest ATI catalyst driver for your system. Make sure CAtalyst AI is not set to "Disabled" in CCC and you should actually set it to "Advanced" when playing Crysis Warhead.

You should get more of a boost than you are. ESPECIALLY @ 1900x1200 your CPU is not the issue.

Read this also. --> http://forums.techpowerup.com/showthread.php?t=51219

 

[BlueSkys]

It is well known that quads do great with xfire, and Q6600 OCed to 3.3 is a very good processor, the 6600 holds its own just fine when OCed 1.1ghz
It has to be something else. The only thing I can think of is immature drivers. Check your setup against other reviews, google 4890 xfire.

Just to clarify, a better CPU is always a great option, but this isnt a "bottleneck", it is CPU scaling.

I'm totally comparing my gains to the techpowerup review I linked above... where they gain like 30 fps in world in conflict.

@ 1920 his CPU bottleneck with that q6600 should be negligible. Crossfire Warhead scales really well in my experiences.

@ OP. Completely remove all your drivers and related software and reinstall the latest ATI catalyst driver for your system. Make sure CAtalyst AI is not set to "Disabled" in CCC and you should actually set it to "Advanced" when playing Crysis Warhead.

You should get more of a boost than you are. ESPECIALLY @ 1900x1200 your CPU is not the issue.

Read this also. --> http://forums.techpowerup.com/showthread.php?t=51219

And I saw this earlier and followed the steps exactly to see if it helped... and no gains

 

[TRIPTEX_CAN]

What was the "old GPU" you replaced with the 4890?

You might want to try a registry sweet to remove residuals form the old drivers.

If that doesnt work... Reinstall Windows.

 

[BlueSkys]

An 8800GTS 512 (G92)

 

[mastrdrver]

So what cat drivers are you running?

Are you running 2 bridges or 1?

I've seen reports where people have said that they see limitations at higher resolutions with just 1 bridge compared to 2. I can't comment though as I've never had a physical 2 card setup.

 

[BlueSkys]

Running Catalyst 9.4 and driver version 8.60

Tried w/ 1 or 2 bridges...

Just ran a crysis benchmark program that actually showed me a near 100% increase in framerates... in the benchmark it went from an avg of 16 with Xfire disabled to an average of 30.

Tried the exact same thing with the Hawx and World in conflict benchmark programs, and again the Xfire showed NO increase in fps. Maybe I screwed something up when looking at crysis before.

I can't imagine this is right... but I'm really starting to think that it's completely hit or miss with crossfire whether or not it helps. I don't understand why in the reviews they get increases across the board, but I'm not seeing any possible problems on my end.

At this point, I would certainly advise against paying for 2 cards...

 

[TRIPTEX_CAN]

When you have your system setup right Crossfire works really well. I dont think I would say hit or miss as I see high % gains more often then not but there are issues that arise when you install Crossfire over old Nvidia drivers.

 

[iandh]

This isn't just bottleneck, there is definitely a software or hardware issue in your system... I ran 4870 CF on a Q6600 and saw a HUGE difference.

 

[BlueSkys]

I reinstalled windows completely when I switched my mobo and added the first 4890... then reinstalled the drivers when I added the second as instructed by the optimization thread posted above.

So confused...

 

[DaMulta]

Can you give us CCC Screen shots with GPU-Z Screen shots?

You can host pics here
www.techpowerup.org

Just hit print screen, open paint, ctrl-v, save as and go to that link.

 

[purecain]

i'd reinstall windows....

going from green to red always causes problems....good luck with it...

 

[BlueSkys]

i'd reinstall windows....

going from green to red always causes problems....good luck with it...

I did.

 

[thraxed]

You don't really have a problem, just a crappy motherboard. P45 boards, when in crossfire use 8x on each slot, so 2 cards in crossfire the equivalent too 1 card @ 16x, atleast on MSI motherboards. This is why you don't see a gain. For the test setup, it runs two slots at 16x in crossfire. So its actually doubling the pci bandwidth when two cards are in use. Your system robs half of the bandwith from the first slot so ya can use crossfire.

 

[BlueSkys]

Can you give us CCC Screen shots with GPU-Z Screen shots?

You can host pics here
www.techpowerup.org

Just hit print screen, open paint, ctrl-v, save as and go to that link.

Here ya go

http://img.techpowerup.org/090501/1168.jpg
http://img.techpowerup.org/090501/2824.jpg
http://img.techpowerup.org/090501/3.jpg
http://img.techpowerup.org/090501/4.jpg

 

[BlueSkys]

You don't really have a problem, just a crappy motherboard. P45 boards, when in crossfire use 8x on each slot, so 2 cards in crossfire the equivalent too 1 card @ 16x, atleast on MSI motherboards. This is why you don't see a gain. For the test setup, it runs two slots at 16x in crossfire. So its actually doubling the pci bandwidth when two cards are in use. Your system robs half of the bandwith from the first slot so ya can use crossfire.

This is the case with the MSI P45 Platinum, but I've read that running with 2 cards in x8 doesn't rob bandwidth and should be fine for running xfire...

Can I get someone else to verify this?

 

[TRIPTEX_CAN]

This is the case with the MSI P45 Platinum, but I've read that running with 2 cards in x8 doesn't rob bandwidth and should be fine for running xfire...

Can I get someone else to verify this?

From all the tests I've seen the loss of bandwidth between x8 and x16 isnt really noticeable until you hit 2560x1600 and at the the difference can be close to 15% or more depending on the game. Although @ 1900x1200 there is still a loss. With that said even accounting for the x8 "limitation" crossfire gains should be more than noticeable in almost ever D3D application.

http://www.tweaktown.com/articles/1472/intel_p45_vs_x48_crossfire_performance/index.html

 

[grunt_408]

well you have pcie2 so when they go to 8x they still work well. when i had my 4850's in crossfire they got around 17k in 3d mark 06