![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
CleanSlate
Supreme [H]ardness
- Joined
- Mar 28, 2003
- Messages
- 5,263
I have been looking at the specs of the 7970 and gtx 680 just scratching my head wondering what allows the 680 to catch up to the 7970 with ease and I have concluded something important (assuming it isn't just bad drivers... A necessary assumption since I can't change driver quality independently).
Memory bandwidth doesn't mean a damn thing beyond a certain point for the 7970 in almost any resolution benchmarked, the gtx 680 seems to have proven that. The main fault in the 7970 architecture (in comparison to the gtx 680) is going to be the texture fill rate/gpu clock. One rather interesting note is that generally memory bandwidth is actually more costly in heat increase per clock terms in comparison to the texture fill rate heat increase per clock.
So, when I get my 7970 back from RMA this coming week I'm going to reduce the ram clock substantially while leaving the gpu clock at stock to make sure this theory holds up with this particular architecture and to test the heat dissipation increase differential. This will tell me whether or not my theory holds water. Once I do this and confirm my hypothesis I will begin finding the sweet spot for the ram clock. After this I will be free to overclock the hell out of the gpu clock.
I have essentially proven that ram bandwidth means nothing past a certain speed, for relevant resolutions that is, at this point.
My assumption here is that the reduction in ram clock will substantially reduce the heat creation from the excessive and unnecessary stock ram clock which will allow me much more head room for overclocking the gpu clock. This will then unlock the true power of the gpu in general and give me a 7970 that will not only beat a gtx680 but will also beat it by a cozy margin.
Of course this all rests on the assumption that the excess heat created by the excessive ram clock is substantial and in fact obtrusive thus leaving that extra substantial heat dissipation for more useful gpu clocks.
I'm interested in hearing thoughts and ideas here. If any one has a 7970 and would like to try this method, please do and post the results.
Memory bandwidth doesn't mean a damn thing beyond a certain point for the 7970 in almost any resolution benchmarked, the gtx 680 seems to have proven that. The main fault in the 7970 architecture (in comparison to the gtx 680) is going to be the texture fill rate/gpu clock. One rather interesting note is that generally memory bandwidth is actually more costly in heat increase per clock terms in comparison to the texture fill rate heat increase per clock.
So, when I get my 7970 back from RMA this coming week I'm going to reduce the ram clock substantially while leaving the gpu clock at stock to make sure this theory holds up with this particular architecture and to test the heat dissipation increase differential. This will tell me whether or not my theory holds water. Once I do this and confirm my hypothesis I will begin finding the sweet spot for the ram clock. After this I will be free to overclock the hell out of the gpu clock.
I have essentially proven that ram bandwidth means nothing past a certain speed, for relevant resolutions that is, at this point.
My assumption here is that the reduction in ram clock will substantially reduce the heat creation from the excessive and unnecessary stock ram clock which will allow me much more head room for overclocking the gpu clock. This will then unlock the true power of the gpu in general and give me a 7970 that will not only beat a gtx680 but will also beat it by a cozy margin.
Of course this all rests on the assumption that the excess heat created by the excessive ram clock is substantial and in fact obtrusive thus leaving that extra substantial heat dissipation for more useful gpu clocks.
I'm interested in hearing thoughts and ideas here. If any one has a 7970 and would like to try this method, please do and post the results.
Last edited:
