4870X2 running slow!

[Wolverine7]

Hello,

I just got a new Visiontek 4870X2 card yesterday, and was eager to get it up and running. My previous video card was a 8800GTX, but I was eager to see a graphics boost (especially since my motherboard only has 1 pci-e x16 slot).

However, after running some benchmarks, I am really disappointed with the result. I'm using the latest Catalyst 8.8 drivers. In DX9 mode, the 4870X2 is actually SLOWER than the 8800GTX. I find this hard to believe, but both 3DMark06 and playing Age of Conan seem slower than the 8800!

My system:
(Modified) Dell XPS 410
Q6700 Quad Core Core 2 processor (2.7GHz)
Corsair 750W power supply
8GB PC8000 GSkill memory, running at DDR2-800
Benchmarks run on Vista Ultimate (32 bit)

Benchmark results:

3DMark Vantage (Extreme Settings @ 1600x1200)

Test 8800GTX 4870X2
GPU Score 3435 4410
CPU Score 9486 9343
GPU Test 1 9.75fps 12.41fps
GPU Test 2 10.38fps 13.44fps
CPU Test 1 1283.36 plans/s 1283.19 plans/s
CPU Test 2 13.31 steps/s 12.63 steps/s
Feature Test 1 512.99 Gtexels/s 771.71 Gtexels/s
Feature Test 2 4.31 Gpixels/s 5.35 Gpixels/s
Feature Test 3 5.97 fps 20.12 fps
Feature Test 4 20.67 fps 13.27 fps
Feature Test 5 26.31 fps 22.37 fps
Feature Test 6 15.31 fps 51.27 fps

Peak Wattage 230W 295W
(Computer plugged into Kill-A-Watt sensor)


3DMark06

Basic Edition, run at the default freebie 1280x1024 settings on both:

Test 8800GTX 4870X2
3DMark Score 12197 3DMarks 10943 3DMarks
SM2.0 Score 4979 4387
SM3.0 Score 5275 4621
CPU Score 3828 3776


I'm confused. Everything I've read points to the 4870X2 being something that should kill these scores. But it LOSES in DX9 testing to the 8800GTX, and barely wins in the DX10 tests. Not only that, but I would have expected the wattage load to be well over 400.

I've tried overclocking the 4870X2 to 800/950, pretty stable (can get to 800/980 to be right on the 'edge') Gains are about 5% boost in 3DMark vantage, nothing much to write home about.

What am I missing? At first I thought perhaps I'm bottlenecking the PCI-e interface somewhere, but that doesn't explain how it is slower than the old 8800 card... I should at least have gotten to those marks!

Does anyone else see these kind of results?

 

[monsta]

Is it overheating?

 

[Wolverine7]

No. It was running 82-85C on automatic mode. I actually forced the fan to 50% speed, which gets the temperature down into the mid-50s. That didn't affect the scores.

I do notice that the card will change it's own clock speed based on load. It's gone from 550Mhz to 750Mhz while sitting here in GPU-Z. Is there some way to lock this thing at full speed (kind of like disabling speedstep)?

 

[JAYDEEJOHN]

Thought I read somewhere that profiles now work with the 8.8 cats. Youre saying the core speed changes while in 3D as well?

 

[Wolverine7]

I don't know what it was doing while benchmarking (3D Mode). However that is one theory as to why it got such poor results. It definitely drops speed while sitting idle at the desktop.

 

[JAYDEEJOHN]

Dropping to 2D speeds is normal, whats somewhat strange is you watching it do it, or actually being able to see it do it. How about drivers? Did you thoroughly uninstall your old nVidia drivers? Like using driver sweeper? Go into safe mode etc?

 

[Wolverine7]

Actually, I wondered if Driver Cleaner was necessary, since I did have some leftover nvidia garbage initially. However, tests before and after Driver Cleaner didn't make a difference on 3DMark scores.

Interesting observation, assuming I'm interpreting GPU-Z correctly.

GPU 1 is running at 750/500 according to GPU-Z
GPU 2 is running at 507/500 according to GPU-Z

ATI CCC has overdrive selected on both and set to 800/980. Does GPU-Z 0.2.7 correctly report GPU speed? I'm actually playing Age of Conan as I test these, so you'd think it would have the GPU at 800 like I set the drivers...

 

[JAYDEEJOHN]

Hmmm, I may have seen this in another thread. If I find it, Ill post

 

[No1sFanboy]

I just installed my 4870x2 tonight so I can offer very little experience but some comparisons. I also just swapped out with an 8800gtx.

If I select the tab at the bottom of GPU-Z I see the same difference in clock speed. I guess that is normal.

My load power has gone up from a peak of 311 to a peak of 430 using a watt meter. The 430 was for one second but I'm typically sitting in the 350's while running graphics.

I haven't had time to try much yet but 3dmark06 is up from 12288 with the 8800gtx (overclocked to 621/1053) to 14920 with the stock 4870x2. This score seems in line with my dual core e6600 at 3.2. I also just did a quick test with Crysis and also see unquantified improvement there.

Edit to add: It looks like only your CPU scores are down. Do you have any other past CPU tests to compare to like a superpi or something? Your benchmark runs should return almost identical cpu scores.

Maybe try removing some of that unused RAM for another test. Your probably only seeing 2.5 gig of that 8 gig you have installed.

 

[Wolverine7]

Nos1sFanboy, the thing I'm watching is the "GPU Memory Clock" in GPU-Z. Does it ever come up off 500.0 MHz for you? I'm in a game and it's still showing 500, which seems wrong.

The tab at the bottom switches between processors, but neither seems to get to full memory speed.

 

[No1sFanboy]

OK I just loaded Rainbow six vegas 2 to try this. I had both CC and GPU-z on the desktop. CC showed the clocks going up to 750/900 from the 2d settings of 507/500. GPUZ did not show the new clock speed until I swapped between the Graphics card and Sensors tabs at the top of the window.

Now that I've done that both show the new clock speed and are staying clocked up with the game minimized.

 

[roadrunner197069]

I think you have some driver conflicts. If all else fails, a reformat sould get you on your way.

I just built a e7300 3.6
4850
2g ram
12053 3dmark06.

 

[swifty_morgan]

I'm curious.......... why 8gigs of memory on the 32bit opsys ?

 

[Wolverine7]

I actually run 64 bit Vista partition as my "stable" setup. The 32 bit was my original partition that has turned more into a testbed.

FYI - For those with 32 bit OS, the 2 gigs on the video card = 2 gigs remain for Windows! Down from 3 gigs with my 8800GTX.

Update on some more test observations:

1) Cloning/Expanding to a second monitor appears to hog one of the GPUs for that task. I assume that to get full speed, the other display needs to be disabled.

2) Playing Conan in Windowed mode had my 1st GPU at 750/500. Once I kicked Conan into full screen mode, CPU-Z did show the game going to 800/950. However, this still only happened on only 1 of the GPUs.. the 2nd one stays at 507/500. I'm guessing this is the root of my problem, my benchmarks probably reflect a 4870 (x1) performance. This happens in both 32 and 64 Vista.

 

[Wolverine7]

For those with more successful 4870x2 installs, if you leave GPU-Z running, background updates checked, and tab set to the 2nd GPU, does your 2nd GPU also show full speed frequencies while in 3D gaming?

 

[goonting]

Running DX9? why?
HD 4870 X2 @ 1600x1200 only? why? Its purposely built for higher resolutions with eye candies on.

its really unusual that we compare most of the time 3dmark scores... I always go for the card that suites my needs at specific resolutions.

check this benchies... as you can see Nvidia products are quite powerful at lower resolutions. As you turn on more eye candy and higher resolutions, the true power of HD4870X2 will show...it was built for 1920x1080 or more

http://www.hardwarecanucks.com/forum/hardware-canucks-reviews/9225-palit-radeon-hd-4870-x2-2gb-video-card-review-16.html

 

[JAYDEEJOHN]

Aha, its the windowed mode thats causing it, just found that old thread http://www.xtremesystems.org/forums/showthread.php?t=198126&page=6 Try going full window, it should fix it

 

[Wolverine7]

goonting - 1600x1200 is all my 20" FP monitor will do.

Even still, the link you posted proves a point in their 3DMark06 benchmark. My 8800GTX score is relatively in line with the 8800 in their table for default settings. The 4870x2 is running about half the score of the Palit card in that review. This is my dilemma.

jaydeejohn - full window fixed one GPU. I still don't have any signs of any frequency movement on the other GPU. This may explain the half score, although my half score seems lower than a single 4870 in that review table...

 

[JAYDEEJOHN]

But if theyre trying to CF, the one could be disrupting the other. It wouldnt be like just running one. So full screen doesnt elimanate it?

 

[Wolverine7]

My observation is using Alt-Tab, since I'm playing in full screen mode. However, if I leave a GPU-Z window running, actively plotting frequency on the 2nd GPU pulldown, it never moves. I can Alt-tab into the game, play around, etc., but coming back to see the plotted trend shows nothing.

On the other hand, GPU #1 plots just fine at 800/950..

Whether it's internal crossfire or not, I'd suspect that the 2nd GPU would be reporting what it was doing.

 

[tri-sli]

I recently upgraded to the 4870x2 from 3x8800GTX’s. I was hoping to see some improvements when playing games in high resolutions HOWEVER, upon installing the card, the performance overall was pretty poor, I had tried all 3 drivers available at the time 8.7, 8.8B & 8.8 official release but nothing helped. In COD, I went from ~180FPS to 64FPS with the new card. Crysis took a plummet immediately.

The 4870x2 works slower than 2x8800GTX’s too.

I am wondering whether you guys are using Nvidia motherboards? I have heard in other forums that the chipset could be causing issues with the onboard xFire?

 

[Scotteq]

I do know there are issues with a 4870(XX) on some P38/48 mobos which requires a Bios update to clear. Sometimes pretty bad, as in 'No Workee At All'. I don't know if you have potential issues there, since the mobo type hasn't been mentioned yet. But mention it as a possibility.

Have you tried cycling to/from Crossfire mode? If the information I read elsewhere is correct, you should be able to do that and it may make a difference.


Tri - My understanding is that Crossfire does not work on nVidia motherboards at all. Though since a 4870X2 is a single card, in theory you shouldn't be affected. However... I wouldn't be surprised to hear about an ATi card not being able to run properly in an nVidia board.

 

[tri-sli]

I do know there are issues with a 4870(XX) on some P38/48 mobos which requires a Bios update to clear. Sometimes pretty bad, as in 'No Workee At All'. I don't know if you have potential issues there, since the mobo type hasn't been mentioned yet. But mention it as a possibility.

Have you tried cycling to/from Crossfire mode? If the information I read elsewhere is correct, you should be able to do that and it may make a difference.


Tri - My understanding is that Crossfire does not work on nVidia motherboards at all. Though since a 4870X2 is a single card, in theory you shouldn't be affected. However... I wouldn't be surprised to hear about an ATi card not being able to run properly in an nVidia board.

Hi Scotteg,

Apologise, I should have mentioned my specs.

Asus Striker II Formula 780i
Q6600 @ 3.4Ghz
3x8800GTX (and a 4870x2 which I have taken out)

I have heard the NF200 chip on the motherboard may be causing the issue but it just feels like both cores are not working at the same time. Hard to explain.

I have tried the card and drivers on a fresh installation of Vista and XP to which neither were able to let the card perform.

I have Intel's i7 in the pipeline for November if it does get released in time and an x58 board. I am hoping this will be able to let the card perform as intended.

 

[Scotteq]

I have heard the NF200 chip on the motherboard may be causing the issue but it just feels like both cores are not working at the same time. Hard to explain.

Yah - Like I mentioned, it wouldn't surprise me... I know that chip is there to enable SLI. I do not know if it somehow actively interferes with Crossfire setups, save that I know Crossfire doesn't work in nVidia boards. In Theory, since a 4870x2 is a 'single' card and only takes the one slot, it should work. But there is also a siggy on the forums here basically saying ...if a theory doesn't work in real life, then it's a sh*tty theory...

But as I haven't had any direct experience with an nVidia motherboard during this past generation, I'd be a poor choice to tap for troubleshooting assistance.

 

[Shantei]

Motherboard