Disappointing 6970CF performance

[OVerLoRDI]

My rig is in my sig, processor is running at 3.5ghz currently.

My question revolves around Metro 2033. Looking at the 6970 review on anandtech, the performance at 1920x1200 is 66fps
http://images.anandtech.com/graphs/graph4061/34635.png

Using the same benchmark settings at 1920x1080 (11% less pixels), I'm getting 52fps, a ~27% difference, which is not insignificant.

Similar performance differences are seen with Crysis. My CCC settings are all default.

So my question is what is holding my machine back? Is a Phenom II X4 @3.5ghz really cpu limiting me at these high settings? Or should I look elsewhere

 

[Skurge]

Yeah, I would say your phenom is holding you back.

 

[busydude]

I don't know how much difference a DDR2 Vs DDR3 memory makes, but that might be an issue too.

 

[RavenSEAL]

6970CF with a phenom X4 940? Yea...not gonna happen. Giant bottleneck, might as well have gotten 1 6950.

 

[cusideabelincoln]

Are your 6950s also overclocked to 6970 speeds? That will also account for the performance difference. Also the DDR2 and your motherboard may also chip away at the performance, if you're running 16x + 4x or maybe even 8x + 8x.

If you account for those little things along with the your processor being a bit slower, that would probably add up.

If you try to turn up the settings a tiny bit does your performance stay the same or drop?

http://alienbabeltech.com/main/?p=22167&page=12

In Crysis with 5870 Crossfire the Phenom II @ 3.2 GHz is about 20% slower than the Core i7 @ 3.2 Ghz, so that falls in line with what you're experiencing. You have an insanely fast GPU setup, and multi-GPU setups seem to be more affected more by CPU power than single card setups. Probably has to do with how the driver utilizes the CPU to drive the cards, along with the fact dual cards are just plain faster and will exhibit CPU bottlenecks more easily.

Also I suspect Crysis is also influenced by system memory bandwidth and speed.

 

[toyota]

6970CF with a phenom X4 940? Yea...not gonna happen. Giant bottleneck, might as well have gotten 1 6950.

I would not go that far but yeah a Phenom 2 could easily bottleneck a 6970 CF setup by 20% or more.

 

[Daedalus685]

The thing about most comptuer bottlenecks is that they aren't simply going to let you reach a point where the performance stops improving regardless of GPU increases in any realistic situation (There are examples of this of course, but most are at stupidly low resolutions or in the case of Civ 5 corrected via Nvidia's new drivers).

What you will notice is a decrease in scaling (not as much benefit going to a faster GPU) and simply less performance compared to a faster CPU. You're likely noticing the latter. You simply can't compare results you get to a review conducted on an over clocked 6 core i7.

What kind of results does a single card get you? I'd imagine you are losing some CF scaling but I bet your system is noticeably slower than Anand's with a single card as well.

Of course if the scaling is in the dumps this likely means you can get relatively "free" increase to resolution/eye candy.

 

[Makaveli]

Processor bottleneck for sure!

 

[OVerLoRDI]

6970CF with a phenom X4 940? Yea...not gonna happen. Giant bottleneck, might as well have gotten 1 6950.

That was the original plan, then the unlocking of the 6950 proved reliable so I grabbed 2.

Consensus is the CPU is the problem, and I am not surprised. I figured it might be, but I was surprised about the disparity in performance especially at such high detail levels. I figured the CPU wouldn't be THAT much of a bottleneck, apparently it was a foolish assumption.

A platform upgrade is in my future, SB looked tempting, but I'm glad I waited, I need all the functional SATA ports I can get.

Thanks guys

 

[3DVagabond]

Can you squeeze any more of an O/C out of your CPU? If it is the bottleneck (which I believe is likely) then it will show it by improving the performance. I've seen tests where the CPU bottleneck disappeared @ 3.8GHz when comparing AMD to Intel in games.

 

[Castiel]

Been well known that an X4 cannot properly push a multi GPU setup

 

[toyota]

Can you squeeze any more of an O/C out of your CPU? If it is the bottleneck (which I believe is likely) then it will show it by improving the performance. I've seen tests where the CPU bottleneck disappeared @ 3.8GHz when comparing AMD to Intel in games.

there is nothing magical about 3.8. its just going to come down the specific game and setup he has that will determine how much he is losing compared to having an i7/i5.

 

[RavenSEAL]

there is nothing magical about 3.8. its just going to come down the specific game and setup he has that will determine how much he is losing compared to having an i7/i5.

It can't hurt though...

 

[thilanliyan]

Even at 1920 in METRO 2033 (which is very GPU intensive) he is being bottlenecked by a 3.5GHz PH2?

OP, under/over clock your CPU to other speeds and see if performance goes down/up in a linear fashion. That way you can actually determine if it's the CPU. If performance doesn't change then the bottleneck is somewhere else.

 

[nyker96]

Processor bottleneck for sure!

+1, your x4 will definitely hold you back in most cases.

 

[OVerLoRDI]

Even at 1920 in METRO 2033 (which is very GPU intensive) he is being bottlenecked by a 3.5GHz PH2?

OP, under/over clock your CPU to other speeds and see if performance goes down/up in a linear fashion. That way you can actually determine if it's the CPU. If performance doesn't change then the bottleneck is somewhere else.

Performance in Metro seems to cap at 3.5. I saw framerate go from 40 to 52 from 3.0ghz to 3.5ghz. 3.6ghz added no benefit in metro 2033.

 

[thilanliyan]

Can you go a bit higher to 3.8GHz maybe? If you're not getting any improvement at all going from 3.5 to 3.6GHz then I wouldn't be so quick to conclude it's a CPU bottleneck.

 

[littlezipp]

My rig is in my sig, processor is running at 3.5ghz currently.

My question revolves around Metro 2033. Looking at the 6970 review on anandtech, the performance at 1920x1200 is 66fps
http://images.anandtech.com/graphs/graph4061/34635.png

Using the same benchmark settings at 1920x1080 (11% less pixels), I'm getting 52fps, a ~27% difference, which is not insignificant.

Similar performance differences are seen with Crysis. My CCC settings are all default.

So my question is what is holding my machine back? Is a Phenom II X4 @3.5ghz really cpu limiting me at these high settings? Or should I look elsewhere

Just ran the Metro benchmark for comparison purposes for you...
•Average Framerate: 58.00
•Max. Framerate: 227.79
•Min. Framerate: 10.17

Options: Resolution: 1920 x 1080; DirectX: DirectX 11; Quality: Very High; Antialiasing: MSAA 4X; Texture filtering: AF 16X; Advanced PhysX: Disabled; Tesselation: Enabled; DOF: Disabled

i7 970 at 4.0GHZ
2x XFX Unlocked 6950's at 880/1375

 

[cusideabelincoln]

Performance in Metro seems to cap at 3.5. I saw framerate go from 40 to 52 from 3.0ghz to 3.5ghz. 3.6ghz added no benefit in metro 2033.

Is your northbridge overclocked and by how much?

The thing about most comptuer bottlenecks is that they aren't simply going to let you reach a point where the performance stops improving regardless of GPU increases in any realistic situation (There are examples of this of course, but most are at stupidly low resolutions or in the case of Civ 5 corrected via Nvidia's new drivers).

What you will notice is a decrease in scaling (not as much benefit going to a faster GPU) and simply less performance compared to a faster CPU. You're likely noticing the latter. You simply can't compare results you get to a review conducted on an over clocked 6 core i7.

What kind of results does a single card get you? I'd imagine you are losing some CF scaling but I bet your system is noticeably slower than Anand's with a single card as well.

Of course if the scaling is in the dumps this likely means you can get relatively "free" increase to resolution/eye candy.

Yeah, I don't think it's quite as simple as a "hard" CPU clockspeed-performance bottleneck. I think there are several factors, which can include the clockspeed, which are playing a role. Memory bandwidth and possibly PCI-E bandwidth can also take away a few (significant few) performance points. This is why we can't simply blame everything on the CPU since his system and Anandtech's systems are vastly different.

 

[Powermoloch]

yeah..your CPU is holding you back.

Time for an upgrade !