![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
SquadBumrush
n00b
- Joined
- Dec 14, 2009
- Messages
- 14
Hey gang!
I'm hoping that Kyle & Co. can offer some clarification on the specific methods they use to determine optimal stable overclocks for Radeon 5xxx VGA cards. I know from reading numerous [H] reviews that they prefer to OC the core first, then nudge the memory clocks to the point where performance starts to degrade (since the Radeon 5xxx GDDR has error correction and can compensate by adversely underclocking). However, I'm afraid that I'm not knowledgeable enough yet to recognize situations when the memory overclocking is having the adverse effect.
I'm in the process of OCing my Asus EAH5850 using a Thermalright Shaman + VRM-R4 combo, setting voltage/clocks with MSI Afterburner and testing stability with MSI Kombustor. I also ran some memory benchmarks with SiSoft Sandra 2010 SP3 and noticed what appears to be some significant memory performance degradation any time I OC the core. To clarify, the pertinent memory benchmarks demonstrate the expected improvements as I OC memory from the stock 1000MHz up to 1245Mhz (with stock core clock). However, anytime I increase the core clock from stock 725MHz (using the previous memory OCs), those same memory benchmarks show adverse results.
As an example, running the Video Memory Bandwidth benchmark...
(Value #1 is Aggregate Memory Performance, Value #2 is Internal Memory Bandwidth, Value #3 is Data Transfer Bandwidth; higher scores are better in all cases)
725/1000: 20.58, 98.65, 4.3
725/1100: 21.28, 105.8, 4.28
725/1200: 22.2, 115.32, 4.27
725/1245: 22.56, 119.47, 4.26
All values increase as the memory clock increases (with the exception of Data Transfer Bandwidth, though the variance appears to be trivial). Next I raise the core clock with the expectation that the memory benchmarks will be mostly unaffected...
800/1245: 21.4, 115.36, 4
900/1245: 20, 119, 3.37
1010/1245: 18.4, 119.54, 2.83
As you can see, Aggregate Memory Performance and Data Transfer Bandwidth are negatively affected while Internal Memory Bandwidth improves. Can anyone shed some light on these results? Is there a better method for benchmarking these memory overclocks? I'm really trying to gain a better understanding of the effect that the OCing has on the memory and how to recognize when I've crossed into "bad" timings.
Thanks for any help!
I'm hoping that Kyle & Co. can offer some clarification on the specific methods they use to determine optimal stable overclocks for Radeon 5xxx VGA cards. I know from reading numerous [H] reviews that they prefer to OC the core first, then nudge the memory clocks to the point where performance starts to degrade (since the Radeon 5xxx GDDR has error correction and can compensate by adversely underclocking). However, I'm afraid that I'm not knowledgeable enough yet to recognize situations when the memory overclocking is having the adverse effect.
I'm in the process of OCing my Asus EAH5850 using a Thermalright Shaman + VRM-R4 combo, setting voltage/clocks with MSI Afterburner and testing stability with MSI Kombustor. I also ran some memory benchmarks with SiSoft Sandra 2010 SP3 and noticed what appears to be some significant memory performance degradation any time I OC the core. To clarify, the pertinent memory benchmarks demonstrate the expected improvements as I OC memory from the stock 1000MHz up to 1245Mhz (with stock core clock). However, anytime I increase the core clock from stock 725MHz (using the previous memory OCs), those same memory benchmarks show adverse results.
As an example, running the Video Memory Bandwidth benchmark...
(Value #1 is Aggregate Memory Performance, Value #2 is Internal Memory Bandwidth, Value #3 is Data Transfer Bandwidth; higher scores are better in all cases)
725/1000: 20.58, 98.65, 4.3
725/1100: 21.28, 105.8, 4.28
725/1200: 22.2, 115.32, 4.27
725/1245: 22.56, 119.47, 4.26
All values increase as the memory clock increases (with the exception of Data Transfer Bandwidth, though the variance appears to be trivial). Next I raise the core clock with the expectation that the memory benchmarks will be mostly unaffected...
800/1245: 21.4, 115.36, 4
900/1245: 20, 119, 3.37
1010/1245: 18.4, 119.54, 2.83
As you can see, Aggregate Memory Performance and Data Transfer Bandwidth are negatively affected while Internal Memory Bandwidth improves. Can anyone shed some light on these results? Is there a better method for benchmarking these memory overclocks? I'm really trying to gain a better understanding of the effect that the OCing has on the memory and how to recognize when I've crossed into "bad" timings.
Thanks for any help!