My AMD Overclock

R

runLoganrun

Gawd
Joined
Jul 4, 2005
Messages
601
I went from 2200 to 2596 (236*11). That's about an 18% increase. Is that a good overclock? I have low latency xms memory, 2-2-2-5, that I had to loosen the timings on to accomplish this. My question is, how does the slower timings on the memory affect performance? Like for me, if I, hypothetically, was able to get a 2486 (226*11) clock and keep the 2-2-2-5 timings, would that be better? Is there a line where the speed increase surpasses the loss of having looser timings?
 
Need to know a whole lot more information about what you're running before anyone can tell you if that's 'good' (very subjective term by the way) or not.

What OS?
What Motherboard?
What CPU?
What socket type for CPU?
What precise make/model/timing ram?

How have you benchmarked your overclock? Prime95 for 12hrs (per core if in a dual-core rig)? Or just repeating loops of 3dmark or something (ie; not a great idea, you need true RAM and CPU stress from Prime or SuperPi or equivilant to truly establish a base-line for stability).

edit - also, if you're running an A64 - bear in mind that your memory latency has a VERY minimal impact on overall performance. The A64's just don't need tight timings to obtain the same level core performance - so tighter timings aren't all they used to be. That's just IF you're running an A64 S754 or S939 though.
 
There's a sticky on the anandtech.com forums comparing timing and mem speeds with OC on the CPU end. Best way for you to tell is to bench your own system with several configs and compare.

I run with a 2:3 divider because fast ram isn't worth the performance/price hit I'd take with my rig. If you look at my specs, I'm all about getting something for nothing and having a decent rig on a budget.
 
I hope the OP comes back because I have almost the exact same overclock. I am just starting and learning but I was kinda curious how "good" mine was too. But in my case I can list what I have.

Im at 235x11 with my memory set to 470mhz

AMD 64 Newcastle 130nm +3500 939 (x11 multiplier, voltage set to 1.55)
Asus A8n-SLI deluxe (bios revision 1016)
G.SKill PC4000 2x1gb DDR500 3.4.4.8 timings (voltage set to 2.8)
stock cooling (though I have everything in a windtunnel case and I think that is helping to keep temps down because under Prime 95 load I hit 57 degrees)

What do you think? Should I try for higher? If so what should I try to tweak? I can play BF2 with everything turned on except 4xAA (have a 6800gt)
 
Lukano said:
Need to know a whole lot more information about what you're running before anyone can tell you if that's 'good' (very subjective term by the way) or not.
Click to expand...

Asus A8N-SLI Deluxe, Socket 939
Corsair TWINX1024-3200XL memory (2.5-3-3-7)
AMD Athlon 3500+ Winchester (@2596, 236*11, 1:1)

I ran this for 18 hrs on Small FFT torture test, Prime 95, with no errors, and for 23 hours on the Blend. (initially I used Memtest to find approximate limits). Yeah, I thought that the memory timings were pretty crucial still on this, that's what I read somewhere anyways. With keeping the timings at 2-2-2-5, I could only reach about 218 HTT, and that was at 2.90v DRAM. I was told also that the memory divider was a bad idea, so I've been keeping it at 1:1. I tested some benchmarks in SiSoft Sandra, CPU Arithmetic and Mixed Multi-Media. Both had approx. 18% increase when compared to stock settings results. So it seems pretty right-on with what I'd expect. I didn't know what other test to run or how to get a better idea of how the memory was performing.
 
Memory dividers are actually a very good idea

And if you're going to overclock your ram 1:1, loosen the timings a bit so it can go somewhere.
 
runLoganrun said:
Asus A8N-SLI Deluxe, Socket 939
Corsair TWINX1024-3200XL memory (2.5-3-3-7)
AMD Athlon 3500+ Winchester (@2596, 236*11, 1:1)

I ran this for 18 hrs on Small FFT torture test, Prime 95, with no errors, and for 23 hours on the Blend. (initially I used Memtest to find approximate limits). Yeah, I thought that the memory timings were pretty crucial still on this, that's what I read somewhere anyways. With keeping the timings at 2-2-2-5, I could only reach about 218 HTT, and that was at 2.90v DRAM. I was told also that the memory divider was a bad idea, so I've been keeping it at 1:1. I tested some benchmarks in SiSoft Sandra, CPU Arithmetic and Mixed Multi-Media. Both had approx. 18% increase when compared to stock settings results. So it seems pretty right-on with what I'd expect. I didn't know what other test to run or how to get a better idea of how the memory was performing.
Click to expand...

I would dare say that whomever told you 1:1 was a bad idea, was misinformed or working with older/different architechtures and hardware.

The A64 memory controller is within the CPU itself - therefore any changes to the HTT (FSB) directly affects the speed at which it communicates to the memory. 1:1 just means the CPU talks to the memory at a 1:1 ratio - which is effectively running the ram at the exact same speed as the HTT.

Using a divider in no way, shape, or form - will degrade your performance. All it does is allow you to keep your ram running at it's most optimal speeds while gaining more ground on the HTT. A faster CPU will make miles more difference on performance sensitive tasks than slightly speedier ram.

So try using a divider on that RAM, and using a lower multiplier for the CPU. Try 250x10 - and if that works, 260x10. 270x10, 290x9, 300x9, 310x9... etc. Just see how high you can push your HTT (note : you man need to loosen the timing on the ram if you intend to try to keep running it at 236mhz - or just do 2-2-2-5 but use the divider to keep it around 200-210mhz at least until you've gotten your maximum CPU OC based off of HTT).

Once you've maxed your HTT (FSB), then you can trim back voltage and try to compensate for what are likely going to be higher temps. If I recall correctly those Winny's will likely peak at about 2500-2600mhz but I also jumped directly into Venice cores from AXP - so I'm not 100% sure.

Then and only then try to push your ram high but loosening the divider constraints.
 
Is having a higher HTT better? I've tested the max clock of the cpu and it is approx 2630 (no long duration tests at that speed). The HTT I was able to get up to 328. I can get 250 with the ram at 2.5-3-3-7. So now I've got an overall clock of 2596 with 236*11. Is it better to re-work that and have not too much higher total clock, because it doesn't have much more room, but have higher HTT, using a ram divider?
 
Id have to agree. Memory divider is like a magic wand. I love it. Can buy slower cheaper memory and use tighter timings and not lose any performance while still maintaining a great overclock. I love it.
 
runLoganrun said:
Is having a higher HTT better? I've tested the max clock of the cpu and it is approx 2630 (no long duration tests at that speed). The HTT I was able to get up to 328. I can get 250 with the ram at 2.5-3-3-7. So now I've got an overall clock of 2596 with 236*11. Is it better to re-work that and have not too much higher total clock, because it doesn't have much more room, but have higher HTT, using a ram divider?
Click to expand...

Your HTT is equivlant to what was the FSB on older architechtures. Your FSB / HTT is what directly affects your CPU clock speed.

10mhz on the CPU will add more performance to your machine than 1mhz on RAM.

Therefore getting the best of both worlds - absolute BEST CPU overclock then going back and pushing the ram to it's fastest/tightest timings.

So your 236x11 may actually be pretty darned close to your maximum overclock. All in all it's not a bad speed gain - but there's a chance you could get more. Thus the suggestion for a x10 multi with a higher HTT to see if I can keep the ram running around the same level, but squeeze a bit more out of your CPU.
 
You must log in or register to reply here.
Back
Top