Problem w/ 4890s in Crossfire

[Mussels]

you uhh, wouldnt happen to have Vsync enabled would you?

 

[BlueSkys]

Not that I know of... I don't see it in CCC

 

[Mussels]

Not that I know of... I don't see it in CCC

its there under 3D options. try forcing it to off.

 

[mastrdrver]

Under all settings.

 

[BlueSkys]

Wait for vertical refresh?

 

[Mussels]

Wait for vertical refresh?

yeah thats the one.

force it off and see what happens.

 

[BlueSkys]

Well, it was put to off unless application specifies, so I forced it off and went to run hawx. I was getting an average of 59fps with 1 4890, and with 2 in crossfire. That should have been a tip off to me that vsync was enabled. So I disabled it in the hawx options and re ran the benchmarks. This time 99fps average (high of 198) with crossfire! I'm thinking, sweet... maybe that was the problem. So I disable the second card and see what a single card does.... Avg. of 97fps. (High of 218)

WTF!!!!
I'm getting pretty convinced crossfire is a sham...

 

[erocker]

It's no sham considering it works for everyone else.

Follow this please:
After CrossFiring several cards this has never failed me.

http://forums.techpowerup.com/showthread.php?t=52502

^^Did you do everything in that link?

Also, have you tried installing motherboard chipset drivers with both cards in? It's definitely worth a shot.

 

[BlueSkys]

It's no sham considering it works for everyone else.



^^Did you do everything in that link?

Also, have you tried installing motherboard chipset drivers with both cards in? It's definitely worth a shot.

I did do everything in that link...

On the forums in response to my problems, I'm getting people (including moderators) saying that it just goes to show what they've been saying for a while. That sli/crossfire is just a gimmick. This is my only experience with multi-gpu, and it's not a good one.

My system specs are all posted, and I seem to have everyone stumped as to why I'm getting no real increase in performance even though I have doubled my investment in gpus.

I'll try to reinstall the chipset drivers and will update.

 

[DaMulta]

I have had good experience with dual cards. In fact I say run two cards or noting. Running the extra things inside CCC, and inside Nvidia control panel really takes two cards to take advantage of them.

Also they are pushing CF more and more because they are working on dual core GPUs which will force them to have near perfect drivers.


I will have two 4890 cards in on Thursday. That I will be running on the new AM3FX platform.
---
I will post the difference in performance I see between the cards.
-----
I'm starting to wonder if you have a bad card, or a bad CF cable.

 

[DaMulta]

Now I did run two hd2900xt cards on a FX-62 system. It was bottle necked HARD!

I would get 10 grand in 06 with one card and 12 grand with two cards.

So in gaming what I did was use the 2ed card as I candy. I would just turn on every setting in CCC BUT v-sync and enjoy my games clearer. This is something that you may end up doing also.....

 

[BlueSkys]

---
I will post the difference in performance I see between the cards.
-----
I'm starting to wonder if you have a bad card, or a bad CF cable.

I have tried each card individually and benchmarked them both... they are identical. And I have tried with both bridges installed, as well as only one (on either connection).

And I would really like to see the difference in performance you see...

 

[BlueSkys]

Now I did run two hd2900xt cards on a FX-62 system. It was bottle necked HARD!

I would get 10 grand in 06 with one card and 12 grand with two cards.

So in gaming what I did was use the 2ed card as I candy. I would just turn on every setting in CCC BUT v-sync and enjoy my games clearer. This is something that you may end up doing also.....

This sounds like what I'm experiencing... I can't imagine I'm getting bottlenecked with my Q6600 @ 3.3Ghz though...

 

[DaMulta]

I can!

That is an older chip! You can not compare chips with Mhz

Lets put it this way two QX9650(the Xeon chips) bench as fast as ONE i7965
Both are one thousand dollar chips both are clocked at 3ghz, but the other one will run circles around it.


Try your games with maxed out CCC I bet you like it a lot. I do and in fact run my games like that when I do play. Benchmarking is the only time I turn it down to performance mode like your wanting.


Other than that look into ways of getting your CPU to run at 4Ghz 24/7.

 

[mastrdrver]

Are your clocks on both cards jumping to 3d speeds in games?
Do you have the molex connected on the board that is to give power to the PCI-e slots?
What kind of loads did you see with the Riva Tuner hardware monitor?

 

[SonDa5]

I think something is not configured right or maby the cards or MB are bad.

 

[Frizz]

This sounds like what I'm experiencing... I can't imagine I'm getting bottlenecked with my Q6600 @ 3.3Ghz though...

I think you really are being bottlenecked, 4890 is faster than gtx295 in crossfire and I couldn't get rid of bottlenecks for shit on my gtx295 at 3.6ghz with my CPU.

4850 crossfire on the other hand took me until 3.4ghz on my Q6600 to get rid of their bottlenecks, one 4890 is fine under 3.3ghz since it performs only a tiny bit slower than 4850 crossfire (4850 crossfire equivalent of 1x gtx285 or 1x 4850x2)

I'm not sure how to do the maths on this but you will hardly see a difference in your gaming benchies with 2 cards, I'm assuming with 4890's your overclocks become handicapped in crossfire which is why there are slower frames running 2. But then again I'm not even sure weather you still get the performance boost or not with your CPU only running 3.3ghz.

So I may be completely wrong about my assumption.

Other culprits could be the MOBO PCIE slots or your config as the above poster mentioned.

 

[Mussels]

i have some pics in regards to bottlenecking.

3.5Ghz Q6600, stock 4870's

3.5Ghz Q6600, 4870's OC'd (800/1000)

Notice the score difference... or lack thereof. I'm confident that its CPU limited.

 

[TRIPTEX_CAN]

i have some pics in regards to bottlenecking.

3.5Ghz Q6600, stock 4870's
SNIP

3.5Ghz Q6600, 4870's OC'd (800/1000)
SNIP

Notice the score difference... or lack thereof. I'm confident that its CPU limited.

Can you run that exact same test @ a higher resolution.... 1680x1050 or higher please?

 

[Mussels]

Can you run that exact same test @ a higher resolution.... 1680x1050 or higher please?

i've just moved to win 7 and the xeon, but yeah i can re-test at 1080P if you want.


new CPU, 4Ghz
Stock GPU's
1280x1024 test (compared to Q6600, for bottlenecking purposes)

OC'd GPU's

1920x1080 tests

Since it gave an invalid score due to the changed res, i only did the GPU tests for the stock cards. the GPU score was all thats relevant here anyway.

stock GPU

OC'd GPU

More of a difference, but still negligible.

 

[BlueSkys]

Are your clocks on both cards jumping to 3d speeds in games?
Do you have the molex connected on the board that is to give power to the PCI-e slots?
What kind of loads did you see with the Riva Tuner hardware monitor?

A- Yeah, the clocks are jumping to full speed during games
B- Yeah, I remembered to connect the molex lol
C- In RivaTuner it shows that both cards are reaching 100% load during games.

 

[mastrdrver]

I did a quick look and didn't see anyone ask, but are you sure your cpu is not overheating and triggering the throttling? Even with it disabled in the bios, it will still kick in. It just doesn't do is so quick.

I ran across a post on [H] where someone was having a similar non scaling issue with crossfire and ended up being that the cpu was throttling because of temps.

 

[BlueSkys]

I did a quick look and didn't see anyone ask, but are you sure your cpu is not overheating and triggering the throttling? Even with it disabled in the bios, it will still kick in. It just doesn't do is so quick.

I ran across a post on [H] where someone was having a similar non scaling issue with crossfire and ended up being that the cpu was throttling because of temps.

Good question bro... but unfortunately thats not it either. I've been monitoring my cpu temps and they never get above 60 on full load. My giant Zalman heatsink has been working awesome lol

 

[human_error]

Just to check you're not playing in windowed mode are you? Crossfire doesn't like windowed mode.

Also to see if your system thinks crossfire is running right click on you ati icon next to your clock on the bottom right of your screen then select one of your cards>crossfirex settings> show crossfire x icon.

With that enabled you get a massive ATI badge in the top right of your screen when playing games where crossfire is running - if you see it then it's not that your system isn't enabling crossfire for some reason, if you don't see it then crossfire isn't getting enabled and we need to figure out why.

 

[SonDa5]

4850 crossfire on the other hand took me until 3.4ghz on my Q6600 to get rid of their bottlenecks, one 4890 is fine under 3.3ghz since it performs only a tiny bit slower than 4850 crossfire (4850 crossfire equivalent of 1x gtx285 or 1x 4850x2)

I didn't see any "bottle neck" issues with the HD4890. Just solid performance even at slower CPU speeds. I know what you are refering to about the HD4850. They open up to fast CPUs.