![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
createcoms
Weaksauce
- Joined
- Mar 14, 2004
- Messages
- 90
Hi all 
Now it occurred to me the possibility this post is a waste, but I can't ascertain this since I'm failing to think creatively in my search parameters. I have spent some time digging through overclocking articles that deal with the Athlon 64 without reading about what I've discovered.
As in my signature, I've got a 939 board + Opteron 170 + 4x sticks of ram. Now I couldn't overclock this chip or the 165 chip before it to any serious degree. Example - this 170 was failing Prime95 before it even touched 2.2GHz!!!!! A similar wall existed with the previous 165, but interestingly enough not the 4400+ I started with. Now I had added an extra pair of ram between the first and second cpu, so it was suggested to me that the Athlon 64 wouldn't overclock well with 4 sticks of ram even with 2T timing.
This fascinated me since I understand memory timings and the relationship with the memory MHz, I appreciated the fact that 1-1 ratio was going to limit my FSB to 250 as per the ram spec but failed to see the connection with the actual core clock (not the FSB driving it or the multiplier or anything else).
Finding myself staring at the possibility of a system that wouldn't overclock (after all the effort with this hardcore watercooling) I decided to get creative and entertain the loony idea that memory settings might somehow appease the core. To get those of you who lack patience to the endpoint fast - as a result of this I'm now sitting stable at 2.8 and still going up.
Like I said, I haven't read of this rather peculiar relationship in A64 guides so the only way I previously was aware of the memory holding back the core was a direct correlation by way of 1-1 ratio and FSB-driven overclocking where you max out your memory's clock as you increase the FSB to push the core higher. It's obvious what happens there, and using dividers in that scenario allows the FSB to climb further still. But here, we've got a situation where the memory clock is well within spec (ie running at less than rated speed), the fsb is low (or high if we drop a multiplier) yet there's this undefined burden being placed on the core by the memory controller's specific parameters. Adjusting those (max async latency, read premable, data and dram drive strengths etc etc etc), and suddenly the core can climb higher - not the ram but the core. I adjust memory settings and I see a result outside of the memory context with no obvious (other than the blaringly obvious fact the memory controller is sitting on-die) relationship.
I hope I've made sense since it's hard to define what's still a bit murky in specifics. If I've missed all the threads and articles that have already been here I apologise, but this has only just been made known to me by personal experimentation!
Now it occurred to me the possibility this post is a waste, but I can't ascertain this since I'm failing to think creatively in my search parameters. I have spent some time digging through overclocking articles that deal with the Athlon 64 without reading about what I've discovered.
As in my signature, I've got a 939 board + Opteron 170 + 4x sticks of ram. Now I couldn't overclock this chip or the 165 chip before it to any serious degree. Example - this 170 was failing Prime95 before it even touched 2.2GHz!!!!! A similar wall existed with the previous 165, but interestingly enough not the 4400+ I started with. Now I had added an extra pair of ram between the first and second cpu, so it was suggested to me that the Athlon 64 wouldn't overclock well with 4 sticks of ram even with 2T timing.
This fascinated me since I understand memory timings and the relationship with the memory MHz, I appreciated the fact that 1-1 ratio was going to limit my FSB to 250 as per the ram spec but failed to see the connection with the actual core clock (not the FSB driving it or the multiplier or anything else).
Finding myself staring at the possibility of a system that wouldn't overclock (after all the effort with this hardcore watercooling) I decided to get creative and entertain the loony idea that memory settings might somehow appease the core. To get those of you who lack patience to the endpoint fast - as a result of this I'm now sitting stable at 2.8 and still going up.
Like I said, I haven't read of this rather peculiar relationship in A64 guides so the only way I previously was aware of the memory holding back the core was a direct correlation by way of 1-1 ratio and FSB-driven overclocking where you max out your memory's clock as you increase the FSB to push the core higher. It's obvious what happens there, and using dividers in that scenario allows the FSB to climb further still. But here, we've got a situation where the memory clock is well within spec (ie running at less than rated speed), the fsb is low (or high if we drop a multiplier) yet there's this undefined burden being placed on the core by the memory controller's specific parameters. Adjusting those (max async latency, read premable, data and dram drive strengths etc etc etc), and suddenly the core can climb higher - not the ram but the core. I adjust memory settings and I see a result outside of the memory context with no obvious (other than the blaringly obvious fact the memory controller is sitting on-die) relationship.
I hope I've made sense since it's hard to define what's still a bit murky in specifics. If I've missed all the threads and articles that have already been here I apologise, but this has only just been made known to me by personal experimentation!