Twin Frozr iii 7950 peculiar OC results

Lynx

Lynx

New member
Got my "MSI HD 7950 Twin Frozr III Boost Edition 3072MB GDDR5 PCI-Express Graphics Card **NEW REVISION**" a few days ago and have been toying around with it. I've encountered a few obstacles, such as the fact the Core volts wont change unless I have "Force Constant Voltage" enabled in Trixx control panel (although this is not neccassary as the below OC's are stable at stock volts)

These are my results from running the Crysis GPU benchmark:

Stock (960/1250) - Min = 41.25
OC (1100/1450) - Min = 41.16
OC (1100/1250) - Min = 42.40
OC (960/1450) - Min = 40.94
OC (1050/1400) Min = 40.77
And weirdest of all:
OC (1100/1450) with +20% power - Min 39.75

This seems completly counter intuitive to me..
If anyone could shed some light on this I would be grateful.

*EDIT - Forgot to mention that it is my SIG RIG, so I though maybe the CPU was bottle-necking?*
 
I think I read somewhere that AMD cards implement some kind of memory error correction, which kicks in when you go above the limit of your setup, and has the effect of slowing down the rendering. That could explain it. Easy way to find out, fix core mhz to 1100 (or any other value that is stable), start with mem. at 1250 mhz and bench it at 25 mhz increments (again, leave core mhz fixed). To minimize the effect of the CPU, in case is holding you back, I'd run the bench in the highest resolution your display supports.
 
Last edited:
Bottleneck sucks. Me and a friend have the same card as each other XFX DD 7950 yet his benchmark results are half mine! I have i5 4670k, he has Phenom II 965.
 
Xavi C said:
I think I read somewhere that AMD cards implent some kind of memory error correction, which kicks in when you go above the limit of your setup, and has the effect of slowing down the rendering. That could explain it. Easy way to find out, fix core mhz to 1100 (or any other value that is stable), start with mem. at 1250 mhz and bench it at 25 mhz increments (again, leave core mhz fixed). To minimize the effect of the CPU, in case is holding you back, I'd run the bench in the highest resolution your display supports.
Click to expand...

i'm pretty sure GDDR5 after you push it past its stable limit just start "misfiring" so rather than crashing the program like what happens with core speed it just doesn't work as effective.

@op i should OC your core to get your highest clock then set back to stock and then OC memory then try them both together.

this way you will know whether it's core or memory that is making the OC fail.

I personally use a spread sheet and type the scores of a benchmark or average FPS then you can see at what speeds the memory speed is decreasing performance.
 
vorticalbox said:
i'm pretty sure GDDR5 after you push it past its stable limit just start "misfiring" so rather than crashing the program like what happens with core speed it just doesn't work as effective.

@op i should OC your core to get your highest clock then set back to stock and then OC memory then try them both together.

this way you will know whether it's core or memory that is making the OC fail.

I personally use a spread sheet and type the scores of a benchmark or average FPS then you can see at what speeds the memory speed is decreasing performance.
Click to expand...

I'll do that when I get home tonight, I doubt I would get an extra 5 fps out of this card without a CPU upgrade, Oh well time to go Haswell!
 
Lynx said:
I'll do that when I get home tonight, I doubt I would get an extra 5 fps out of this card without a CPU upgrade, Oh well time to go Haswell!
Click to expand...

Mine as well wait for ivy-e at this rate... that will last you longer if you feel you won't upgrade in years.
 
You must log in or register to reply here.
Back
Top