6870 crossfire with PCI-E X4

[Skurge]

So following the trend of people benching their hardware, I'm going to do the same.

I've got a motherboard with 1 PCI-E 2.0 X16 and a PCI-E 2.0 X4 slot. I'll be getting a second 6870 in the next few days. So I'm going to benchmark single and dual card configs.

I'm going to bench at 1920x1080 for all games.
Everything will be at stock clocks.
The cards are clocked at 915/4200 from factory.

3DMark11
Crysis
Crysis 2
BF3 (I'll need help on how best to bench this)
Dirt 3
F1 2011
AvP
Skyrim (need help with this too)
Stalker COP

System Specs

Intel Core i3-2100 3.1Ghz
Gigabyte H67A-UD3H
8GB DDR3-1333
Gigabyte Radeon 6870 1GB WindForce 3X
2TB WD Caviar
Windows 7 Home Premuim x64

Anything else you guys wanna see me bench, just say so.
For the games that don't have a built in benchmark tool. I'm going to use manual runs and use fraps to get an average over 3 runs.

 

[Skurge]

3DMark11
3DMark Score - P3895
Graphics Score - 4074
http://3dmark.com/3dm11/2604104

Dirt 3
Ultra Settings, 4xMSAA
Average - 52.67FPS
Minimum - 44.85FPS

F1 2011
Ultra settings, 4xMSAA
Average - 53FPS
Minimum - 44FPS

Crysis 2

Metro 2033
Average Framerate: 37.00
Max. Framerate: 72.93
Min. Framerate: 9.70

STALKER: COP
MSAA Off
SSAO Mode: Default
SSAO Quality: Low
Tessellation: On
Contact Hardening Shadows: On

I wanted to use settings that were playable on a single card in all areas including the sun shafts test.

 

[Skurge]

Reserved

 

[tweakboy]

A degration in performance will come, however nothing noticable.. gl

 

[Skurge]

A degration in performance will come, however nothing noticable.. gl

Yeah, but I'm getting the card real cheap so, I don't mind that.

 

[Smartazz]

Thanks for starting this thread. Sorry I couldn't take the plunge first with my 16x/4x motherboard. I hope your CPU doesn't bottleneck such a powerful gpu setup.

 

[Smartazz]

If you have Metro 2033 and the Witched 2, could you bench those? If not, I can benchmark them when I get the second card.

 

[Skurge]

Don't have witcher 2, but I do have metro2033. I was also worried about the CPU, but we will see when you get a second card.

I'll start single card benchmarks later today.

 

[Necc]

Mass effect 2, Left 4 Dead 2, StarCraft 2 if you can.

 

[Termie]

You're going to be ok in Metro, Stalker, and AvP on the CPU side, but you'll get seriously CPU bottlenecked in BF3 if that PCIe x4 slot holds up ok. Starcraft will likely give you CPU troubles as well. And in my testing of Crysis, it's incredibly inconsistent. I'm going to bench it today to see if I can get any good numbers. Crysis 2 scaled a bit better in my experience (but the DX11/High-res pack borked my install and I could never test it with those add-ons). Dirt3/F1 (same engine?) require more than 2 cores, but the HT on your 2100 might be able to compensate. Skyrim I've never tried, but obviously everyone says it's massively CPU-bottlenecked, but more based on clock speed than cores (unfortunately you have a problem either way).

FYI, Mass Effect 2 is capped at 60 unless you can find a command to get around it, and one 6870 will max that out.

3DMark11 should be able to tell you right away how your x4 slot is scaling. Get those numbers first (one card and then two), and then decide if you want to test all the games.

 

[Smoblikat]

Metro 2033 with DX11 turned on, thatll be cool to see.

 

[Smartazz]

Do you know if your pci-e 4x slot share bandwidth with the sata devices?

 

[Skurge]

Do you know if your pci-e 4x slot share bandwidth with the sata devices?

I checked, Can't see anything about that on the gigabyte site.

 

[Termie]

3DMark11
3DMark Score - P3895
Graphics Score - 4074
http://3dmark.com/3dm11/2604104

Dirt 3
Ultra Settings, 4xMSAA
Average - 52.67FPS
Minimum - 44.85FPS

F1 2011
Ultra settings, 4xMSAA
Average - 53FPS
Minimum - 44FPS

As I suspected, Dirt3 and F1 are basically the same game. The performance with a single card is identical in both, so no need to bench both of them in crossfire unless it's really easy.

By the way, to bench BF3, I recommend doing both single-player and multi-player. Single-player should be pretty easy. Just choose one of the on-foot missions and bench it for a couple of minutes in a walkthrough, using the same starting points each time. For multi-player, one approach you can take is to bench during a battle where you are unlikely to get killed. Caspian is considered one of the most demanding maps. I have benched over by the antenna, running around the hill a few times while viewing the battle at other flags. That at least gets some strain on the computer, without messing up your benchmark 20 seconds in by getting you killed. Also, for consistency, just use the "high" setting. Your system should be able to run it reasonably well with one card, which is your starting point, and you'll obviously get excellent performance with dual cards. Ultra would absolutely kill a single card, making it unplayable, and the MSAA would overrun the VRAM limit on two cards, making it fairly unplayable there too.

 

[Skurge]

As I suspected, Dirt3 and F1 are basically the same game. The performance with a single card is identical in both, so no need to bench both of them in crossfire unless it's really easy.

By the way, to bench BF3, I recommend doing both single-player and multi-player. Single-player should be pretty easy. Just choose one of the on-foot missions and bench it for a couple of minutes in a walkthrough, using the same starting points each time. For multi-player, one approach you can take is to bench during a battle where you are unlikely to get killed. Caspian is considered one of the most demanding maps. I have benched over by the antenna, running around the hill a few times while viewing the battle at other flags. That at least gets some strain on the computer, without messing up your benchmark 20 seconds in by getting you killed. Also, for consistency, just use the "high" setting. Your system should be able to run it reasonably well with one card, which is your starting point, and you'll obviously get excellent performance with dual cards. Ultra would absolutely kill a single card, making it unplayable, and the MSAA would overrun the VRAM limit on two cards, making it fairly unplayable there too.

Thanks for the tips, it takes about 3mins to bench F1 and Dirt. So I'll just bench both. My internet is having issues here, so it's taking me ages to download the Stalker and AvP benches. I already have the games, but they don't have built in benchmarks.

 

[Skurge]

I've only really played Crysis 2 for a few minutes and haven't gone back to it, but why does the game look so blurry? It's as if it's got MLAA on or something.

 

[Skurge]

Got my second 6870, still waiting on the PSU. It will still be a few days. So I'll have time to bench all the games, including the ones you suggested. Gonna have a few more up tonight.

 

[Bman123]

I'm looking forward to seeing your results

 

[Skurge]

I'm looking forward to seeing your results

Unfortunately I don't have MW3, Fallout 3 or LA Noir to test for you. Wish I did.

 

[Termie]

OP - I think I'd drop Crysis from your list. I did some testing on my setup, and found that I'd hit 45fps average (1920x1200, 4xAA, Very High details), but that my GPUs were only being used at 50-75%. Not having benched Crysis since the early days (on an 8800gt), I wasn't aware that it could be CPU-bottlenecked. Sure enough, with today's graphics horsepower, it absolutely is. Take a look at these graphs:

http://www.tomshardware.com/reviews/radeon-hd-6990m-geforce-gtx-580m-deception,3118-3.html

While it won't be bottlenecked on a single 6870 (equivalent to the 6990m they tested above), dual cards are enough to blow right past the GPU bottleneck and put you into CPU-bottleneck territory. If you include this game in your test, you'll only skew the crossfire scaling analysis.

 

[Skurge]

OP - I think I'd drop Crysis from your list. I did some testing on my setup, and found that I'd hit 45fps average (1920x1200, 4xAA, Very High details), but that my GPUs were only being used at 50-75%. Not having benched Crysis since the early days (on an 8800gt), I wasn't aware that it could be CPU-bottlenecked. Sure enough, with today's graphics horsepower, it absolutely is. Take a look at these graphs:

http://www.tomshardware.com/reviews/radeon-hd-6990m-geforce-gtx-580m-deception,3118-3.html

While it won't be bottlenecked on a single 6870 (equivalent to the 6990m they tested above), dual cards are enough to blow right past the GPU bottleneck and put you into CPU-bottleneck territory. If you include this game in your test, you'll only skew the crossfire scaling analysis.

Yeah, I'll drop it. Thanks for the heads up. Going to test BF3 as soon as my connection improves and I can get playable latency online.

 

[RavenSEAL]

I made an 8x/8x thread for comparison

 

[Skurge]

I made an 8x/8x thread for comparison

Nice, I'm just waiting on the PSU and I should have crossfire benches up. Also, whats up with the AvP benchmark tool. I run it and I can't change settings and or get FPS results. What kind of benchmark is that? I don't even know what folder it's installed in.

 

[Termie]

Yeah, I'll drop it. Thanks for the heads up. Going to test BF3 as soon as my connection improves and I can get playable latency online.

Looking forward to the benchmarks. As a heads up, be very careful about your fan speed when enabling and disabling crossfire between benchmarks. I found that without a restart after disabling crossfire, Afterburner couldn't control my single-card, and my GPU got up to 90C at 22% fan in the test below. Don't want to ruin your brand new card!

For those who are interested, here are my single-card and multi-card Crysis benchmarks at stock 5850 speeds:

DX10 1900x1200, AA=4x, Vsync=Disabled, 64 bit test, FullScreen
Custom Quality Values: ALL VERY HIGH

Single-card: Run #1- DX10 1900x1200 AA=4x, 64 bit test, Quality: ~~ Overall Average FPS: 27.69

Dual-card: Run #1- DX10 1900x1200 AA=4x, 64 bit test, Quality: ~~ Overall Average FPS: 41.675

That's exactly 50% scaling, and the GPU was pegged at 99% in the single-card run, but my cards were at about 75% in dual-card mode. From the Tom's benchmarks I posted earlier, it seems there's a point where the game becomes CPU-bottlenecked, as my benches are very similar to both a 580 and a 6990 (Tom's got 48fps, but was on a faster i7-990x).

 

[Smartazz]

Termie, how has your experience with microstuttering been? That's my chief concern with going multi-GPU.