My benchmarks of 5770+5750!! Mix and match Crossfire

[busydude]

Test System:
CPU: AMD Phenom II X2 555 3.8Ghz
Memory: 4 GB DDR3 8-9-8-22-1T 1600Mhz
NB frequency: 2.4Ghz
GPU: XFX 5770(Reference) and XFX 5750 Both running at stock speeds
Drivers: Catalyst 10.4
Resolution: 1920X1080

AvP Benchmark Demo:
Texture quality: High
Shadow Quality: High
Anisotropic Filtering: x16
SSAO: On
Vertical Sync: off
DX11 Tessellation: On
DX11 Advanced Shadows: On
DX11 MSAA Samples: Off

Min:26/17
Max:77/50
Avg:44.854/29
Scaling efficiency: 54.5%

DiRT2:
All IQ settings: Ultrahigh/High
Dx11: On
Anisotropic Filtering: x16
MSAA: x4

Min:45/34
Max:76/49
Avg:63/41
Scaling efficiency: 53.66%

Furmark:
MSAA: x4

Min:34/15
Max:66/37
Avg:43.6/23
Scaling efficiency: 89.5%

JustCause 2:
Min:27/42
Max:53/73
Avg:41.7/62.7
Scaling efficiency: 50.36%

LostPlanet 2:
Dx 9c
All IQ settings: High
MSAA: x8
Scaling efficiency: 63.7%
5770+5750

5770

Mafia Benchmark:
All IQ settings: On/High
Anisotropic Filtering: x16
AA: On

Min:16/26
Max:47/62
Avg:30/46
Scaling efficiency: 53.33%

Uniengine Heaven:
Dx11
Shaders: High
Tessellation: Normal
Anisotropic Filtering: x4
AA: Off

Min:12/7
Max:83/51
Avg:33.7/20.8
Scaling efficiency: 62.0%

 

[magomago]

(a) Holy crap the size of those matlab plots are enormous and excessive; we aren't all running 1080p ya know.... (b) looks like quite a few of your trends are time shifted relative to the other trends; your data isn't aligned.

 

[happy medium]

Nice work, here is some benchmarks to compare those to. They are a bit old so 'm sure there a little faster now.

http://www.xbitlabs.com/articles/video/display/asus_eah5770-powercolor_pcs_hd5750.html

I was going to do the same thing, but my 5750 had a little more life left in it, and now I'm looking for a single card upgrade.

Edit: it looks like your cpu is holding you back. That scaling is not so good.

 

[3DVagabond]

For scaling efficiency are you using the 5770 or 5750 as the single reference card? If I missed where you said/showed it already, sorry.

 

[busydude]

Holy crap the size of those matlab plots are enormous and excessive; we aren't all running 1080p ya know.... (b) looks like quite a few of your trends are time shifted relative to the other trends; your data isn't aligned.

Yup, I am aware of that. I was too lazy to tinker with the data. Oh, and sorry for the huge pictures.

 

[busydude]

for scaling efficiency are you using the 5770 or 5750 as the single reference card? If i missed where you said/showed it already, sorry.

5770.

 

[happy medium]

Try overclocking the 5750? I wanna see if your cpu limited. Please?

 

[busydude]

Nice work, here is some benchmarks to compare those to. They are a bit old so 'm sure there a little faster now. http://www.xbitlabs.com/articles/vid...cs_hd5750.html I was going to do the same thing, but my 5750 had a little more life left in it, and now I'm looking for a single card upgrade. Edit: it looks like your cpu is holding you back. That scaling is not so good.

I have seen that article before. As you can see I am not using latest drivers, 10.8 do not support mix and match and I have not yet tried version 10.9 on this setup.

I will run these benches again after overclocking my CPU to 4Ghz.. and raising my NB to 2.8 Ghz. Will do it tomorrow, you will have to wait for them.

 

[happy medium]

I have seen that article before. As you can see I am not using latest drivers, 10.8 do not support mix and match and I have not yet tried version 10.9 on this setup.

I think they used the 10.3's, I can't remmber. I would consider that one of the best drivers.

 

[busydude]

I think they used the 10.3's, I can't remmber. I would consider that one of the best drivers.

Yeah, 10.8 worked fine for single 5750, but when I started using Crossfire it crashed very frequently, even after applying the application profiles.

 

[busydude]

Edit: it looks like your cpu is holding you back. That scaling is not so good.

I am not so sure about my CPU being a bottle neck. The performance of 5770+5750 should fall right in between a single 5850 and 5870 and based on benches I say it is about right.

 

[MrK6]

I am not so sure about my CPU being a bottle neck. The performance of 5770+5750 should fall right in between a single 5850 and 5870 and based on benches I say it is about right.

It seems right on the money and I think it's darn impressive, especially considering they aren't even the same card. Thanks for taking the time to post this .

 

[railven]

What would you guys guess the performance gain would be with a 5870+5850 CrossFired?

It seems like a good 50% average scaling here. You think those two would have that kind of scaling?

 

[busydude]

What would you guys guess the performance gain would be with a 5870+5850 CrossFired?

5850 CF performs similarly to a 5970. 5870 CF is about ~20% faster. A ball park figure would be ~10% faster than a 5970.

 

[busydude]

Try overclocking the 5750? I wanna see if your cpu limited. Please?

I just did a quick benchmark run of DiRT 2. I found something interesting to report.

When I overclocked my 5750 to 850MHz(13.3% overclock), the averages did not change but the minimums improved substantially, from 45 to 53(17.7% improvement).

Next, when I overclocked the 5770 to 900MHz (6% overclock), the averages jumped from 63 to 68(8% improvement) and the minimum FPS was 55(22% improvement from stock).

 

[Skurge]

For all the scaling issues people complain about, the flexibility of Xfire is great.
The load balancing is excellent, especially with the different clocks that you tested with.

 

[busydude]

For all the scaling issues people complain about, the flexibility of Xfire is great. The load balancing is excellent, especially with the different clocks that you tested with.

Also, being an asymmetrical crossfire, I thought there would be a lot of micro stuttering, boy I was surprised. I do not say that there is no micro stuttering but going from 40 FPS to 65 FPS is really gratifying.

 

[Skurge]

Also, being an asymmetrical crossfire, I thought there would be a lot of micro stuttering, boy I was surprised. I do not say that there is no micro stuttering but going from 40 FPS to 65 FPS is really gratifying.

What rendering method are you using? AFR?

 

[happy medium]

I just did a quick benchmark run of DiRT 2. I found something interesting to report.

When I overclocked my 5750 to 850MHz(13.3% overclock), the averages did not change but the minimums improved substantially, from 45 to 53(17.7% improvement).

Next, when I overclocked the 5770 to 900MHz (6% overclock), the averages jumped from 63 to 68(8% improvement) and the minimum FPS was 55(22% improvement from stock).

Thanks alot, thats great, I kinda thought your minimums would rise.

 

[busydude]

What rendering method are you using? AFR?

To my knowledge, We cannot specify rendering method explicitly. Catalyst adapts the rendering method based on the game we are using.

Generally speaking:

OpenGL -> AFR
D3D -> supertiling

But AMD can release crossfire profiles which can change the rendering methods(I am sure you noticed a profile update for F1 2010 in 10.9 crossfire profiles). The user cannot change the way rendering is done.

 

[Skurge]

To my knowledge, We cannot specify rendering method explicitly. Catalyst adapts the rendering method based on the game we are using.

Generally speaking:

OpenGL -> AFR
D3D -> supertiling

But AMD can release crossfire profiles which can change the rendering methods(I am sure you noticed a profile update for F1 2010 in 10.9 crossfire profiles). The user cannot change the way rendering is done.

You can mess around with the rendering method with RadeonPRO. Try a few different methods and let us know. If you have the time that is.

 

[busydude]

You can mess around with the rendering method with RadeonPRO. Try a few different methods and let us know. If you have the time that is.

I personally never used RadeonPRO. Will give it a shot.
Will do it tomorrow night. Have an exam tomorrow. What a way to ruin the first day of F1 2010.. guh.

 

[Skurge]

I personally never used RadeonPRO. Will give it a shot.
Will do it tomorrow night. Have an exam tomorrow. What a way to ruin the first day of F1 2010.. guh.

Good, then you wont get more practice than me