San Diego 3700+ Overclock

H

HanZo

n00b
Joined
Dec 27, 2005
Messages
26
Sup Guys.

I just build this rig 2 weeks ago.

Fortron BlueStorm 500Watt
AMD Athlon64 3700+
OCZ EL Platinum Rev2 TCCD 2-2-2-5
DFI Lanparty UT NF4 Ultra-D
EVGA Geforce 7800GT w/copper heatsink
Pioneer DVD-Burner
Zalman CNPS7700-Cu



Currently running at: 1:1 ratio

CPU : 300 x 9 = 2700 mhz @ 1.425 Volt
Ram timing: 2.5-4-3-7 @ 2.9 Volt


Genie BIOS Settings:

FSB Bus Frequency - 300
LDT/FSB Frequency Ration - Auto
CPU/FSB Frequency Ratio - 9
PCI eXpress Frequency - 100Mhz

CPU VID StartUp Value - StartUp

CPU VID Control - 1.425v
CPU VID Special Control - auto
LDT Voltage Control - 1.20v
Chip Set Voltage Control - 1.50v
DRAM Voltage Control - 2.9v

Temperatue: idle 27 celcius, load 37-39 celcius.

I already tried 240 x 12 = 2600 mhz at timing of 2-3-3-5 and its stable. CPU volt is at default, ram is at 2.9 volt.

what do u think? should I go for 2700 or 2600 on tight timing. how can I achieve 2800mhz?
 
2700 1:1 is pretty good. You want 2.8 Ghz? run 280x10 1:1, a little more volts.
 
Absolute0 said:
2700 1:1 is pretty good. You want 2.8 Ghz? run 280x10 1:1, a little more volts.
Click to expand...


yeah I am satisfied with 2700mhz at 1:1. I tried 280 x 10 with 1.45 volt and the computer load into windows but froze when I do SuperPi benchmark.

so I guess thats the limit. I don't want to feed more volt to my cpu, cause it will decrease the life span.

Anyways I need some expert opinion. Should I just do 240 x 11 at 1:1 @ stock volts, timing of 2-3-3-5 @ 2.9v, or should I go for 300 x 9 @ 1.425 volt, timing of 2.5-4-3-7 @ 2.9 volts ?

Also y is CPU-Z showing I only have 1.39 volt to my cpu when I specified in BIOS 1.425 volt?
 
You did very well. I need 1.52V to get my sig system to 2750 (250x11)
 
DougLite said:
You did very well. I need 1.52V to get my sig system to 2750 (250x11)
Click to expand...

What is ur ram timing? is RAM timing really important? should I compromise ram timing for CPU power?
 
I had 1GB of Winbond BH-5, which ran 1.5-2-2-5 @ 5:6 divider for 196MHz on the RAM on the above mentioned system, and I went to running 2GB of Corsair Value @ the same divider/clock, but 3-3-3-8 timings...SuperPi 1M didn't budge, still 31s.
 
DougLite said:
I had 1GB of Winbond BH-5, which ran 1.5-2-2-5 @ 5:6 divider for 196MHz on the RAM on the above mentioned system, and I went to running 2GB of Corsair Value @ the same divider/clock, but 3-3-3-8 timings...SuperPi 1M didn't budge, still 31s.
Click to expand...

that's pretty good considering u have 2 gig. My score is 30.6s for 1MB @ 2.5-4-3-7
 
Also how is my Temperature?

CPU - idle 27-29 celcius, load or after playing quake 4 or Counter-strike 37-39 celcius
PWMIC - idle 37celcius, load 39-40 celcius
Chipset: 48 celcius, max I see is 49 celcius?

what is the safest temperature on these chip?
 
It's ok to add a LITTLE voltage. Darn, you're not happy that you get >2.6GHz stable without raising voltage, and here I'm stuck somewhere in the area of 2.5 with raised voltage. I don't remember where the danger zone was, but, I think it was something to the effect of maybe +0.25V that you didn't want to go past. Well, look it up before really pushing things, but, I'd say don't be afraid to add just a little more juice to let it do what it's truly capable. And why 1:1? Pick up a calculator and run some figures. I'm sure you can find something that lets you get better if you aren't afraid to let the memory run differently. Athlon 64 is immune to the nforce 2 asyncronous memory timing bug by it's very nature, so it's not supposed to matter what you set the memory ratio to now. The idea of a proper overclock is to get everything as maxed as it safely can be without significantly decreasing lifetime, so max out CPU and memory both (FSB doesn't seem to matter as long as you keep that HyperTransport close to where it's supposed to be - 1GHz. At least, in benchmarks I saw practically no difference between 315x8x3 and my current 252x10x4. I guess AMD isn't lying about the new HyperTransport system. I think that it can only make a difference if absolutely saturated -- which is pretty darned hard to do -- at which point I suppose the faster FSB could still refresh faster. Games run the same for me at my current settings, so I gave my chipset a break and went down to this versus the 315 setup.) As for timings, well, to a certain extent, lower latencies will yield better performance increases than higher frequencies, but, the more you raise frequencies, the less higher latencies hurt. Considering that this is a fundamental nature of DDR2, that shouldn't be too shocking (ddr2 has higher latencies than ddr, but, gets such higher speeds with it's new layout that it's now caught up and maybe already surpassing ddr in performance.) Best way to tell, benchmarking since it gives you raw numbers (gaming won't be useful because you won't see a noticable enough difference to say for certain with modern or old games unless you set something pretty bad.)

Anyway, if you can't find a better mix, I suppose you'd probably see slightly better results with the memory running on tighter timings and a slight loss to the processor speed. I think you only need something like 2.2GHz, maybe these days now 2.5GHz to mostly remove cpu limitations, so I suspect memory will affect you more since it's still a pretty weak link in the chain (even on DDR2 systems really.)

Safest temperature, well, stay below 55C for about the best reasonable life span + most stability as a general rule of thumb for most chips. San Diego is no exception, though you'd probably have to unhook the CPU fan to get it to hit most san diegos to hit that (from what I hear, the ones that go really high can run hot though. Eg, the rare kind that can hit 3+GHz need water cooling usually to stay safe.) I used to have a hotter running mobile barton 2600+ which needed prime95 to push it as high as 55C, and it's not dead yet (now in my dad's HTPC and still chugging along nicely despite the dangerously warm case, though he doesn't get to oc to 2.5GHz like I did since it's so hot in there.) I can't tell you for certain how long I'd had the thing, but, I believe it was two years before I upgraded (I completely skipped the entire socket 754 generation) and I was keeping up pretty well regardless since it was clocking at 2.5GHz (and, honestly, the main upgrade I got was going from 512KiB L2 to 1MiB since in many of the main things I do the Athlon XP was just as powerful.) It's been in there quite a few months now and still running strong without errors until it hits 57C or so.

As for the chipset, well, I'd guess 55C there as well (just a guess.) Pretty high though, your system could use some more cooling I think. My case is running the system maybe 8-12C above ambient worst case (with the cpu 0-3C above that usually.) I'd say that if you're actually seeing numbers like 49C, then either you've got bad cooling on the motherboard, the case, or a crappy sensor that's way off. Don't know if mine might be a bit low -- I went overkill, with two front fans, two rear fans (excluding active PSU) and a 120mm side fan that I had to cut a big hole with a jgisaw to get in (super-generic case without even so much as a few vent holes in the side.) I think the most the system sensor has ever shown me was around 29, maybe as much as 32 when I visited my parents in their hot house.

I asked a while back (maybe not in here, possibly the AMD specific forum) to try to get a better idea once and for all of the overall effects of temperatures on chips. The general consensus seemed to me to be that in the area of 55C will still give you a good 10 years all but guaranteed on most cpus at least, with a decent possibility of longer still. I personally like to be 100% sure any chip I get WILL last more than 5 years because some (such as my current one) show good signs of being useful long after I upgrade. I figure at these temperatures, and advantages like that 1MiB L2, this will make a wonderful server/general purpose thing for everything from encoding to HTPC type stuff whenever I next upgrade (which will be a couple of years at least probably.) Considering that Prime95 pushes it to around 35C on a hot summer day, I honestly believe it will still be around for me to use.

EDIT: BTW, stock voltage for a San Diego 3700+ is 1.35 on some, 1.4 on others (newegg isn't clear.) I suspect you see 1.39 because your board looses 0.01 (which is well within expected parameters, and perhaps even erring a little less than most.) I'm running mine at 1.45 or so and this is FAR below the danger zone. Raising my voltage doesn't seem to increase my overclock any further. Keep it in the 50s and never touch 60s and you should be ok with slightly raised voltages. Start hitting really high temperatures and that raised voltage is going to hurt even more though.
 
bump the core volatage to 1.5 and the DDR voltage can go up to 3.5 if needed. Might be a good idea to bump up the chipset voltage too.
 
UPDATE


RIGHT now I'm running 280 x 10 @ 1.45v on a divider 5/6, DRAM 2-3-3-7 @ 2.8v. Pass SuperPi 32M on 27m 266s. STable

is it better? or worst cause its on a divider of 5/6.

Please great ELDERS, should I keep this or go back to 240 x 11, 1:1, 2-3-3-5 2.8v

And is 2.9v on TCCD too much?
 
Cpu mhz is what you're after. Don't be afriad to use a divider, thats what theyre there for. I'd suggest dropping your ram speed down to 1:2 and see just how far you can push your cpu, then adjust your htt/multis/dividers accordingly to get the max out of everything. There's no real performance gain running the ram 1:1, especially if it means loosing 100, 200, or more mhz on the cpu.
 
bump the core volatage to 1.5 and the DDR voltage can go up to 3.5 if needed. Might be a good idea to bump up the chipset voltage too.
Click to expand...

Please don't listen to this guy. You have TCCD, and you should never go above 2.9V or so. Losen your timings to 2.5-3-3-7 and bump up the HTT. Frequency > timings. There is a pretty good chance your memory will be able to do 300 1:1 at 2.5-4-4-10 or 3-3-3-10 or something loose like that. TCCD/TCC5 isn't a huge fan of voltage, so I would stick with ~2.8V
 
2uantuM said:
Please don't listen to this guy. You have TCCD, and you should never go above 2.9V or so. Losen your timings to 2.5-3-3-7 and bump up the HTT. Frequency > timings. There is a pretty good chance your memory will be able to do 300 1:1 at 2.5-4-4-10 or 3-3-3-10 or something loose like that. TCCD isn't a huge fan of voltage, so I would stick with ~2.8V
Click to expand...


it did go 300 htt 1:1, 300 x 9 @ 1.45v, timing of 2.5-4-3-7 @ 2.9v, stable pass prime.

but should I keep that? I heard high htt doesn't gain much.
 
Ok I'm happy now. I have three option.

1. 300 x 9 @ 1.425v, 1:1 ratio, Ram timing: 2.5-4-3-7 @ 2.9 Volt

2. 240 x 11 @ default volt, 1:1 ratio, Ram timing: 2-3-3-5 @ 2.9 Volt

3. 280 x 10 @ 1.45v, 5:6 ratio, Ram timing: 2-3-3-7 @ 2.8v

Which one should I go for?
 
i would probably choose the last choice since i like the higher mhz...ur lucky u dont have to raise voltage much on ur cpu...i had to go up to 1.52v too to get 2750mhz (same as the guy above i think) but my ram only took 2.65v and it was stable up to 285mhz with 2.5-3-3-7 timings...i think my memory controller isnt very good htough cause my cpu failed prime at anything over 275mhz :( now i raised my voltage a good bit but not as high as others (into the 1.6v range...im not that brave ;)) heheheh nice overclock
 
2uantuM said:
do #3, but 1:1 with loose timings. You'll see higher performance.
Click to expand...


loose timing? Why? if I can get stable at 2-3-3-7, y bump it up?

Hey current I'm trying 288 x 10 with same timing as #3, but I'm curious about LDT multiplier, should I put auto, x3, x4?
 
2uantuM said:
enforcer17 said:
bump the core volatage to 1.5 and the DDR voltage can go up to 3.5 if needed. Might be a good idea to bump up the chipset voltage too.
Click to expand...
Please don't listen to this guy. You have TCCD, and you should never go above 2.9V or so. Losen your timings to 2.5-3-3-7 and bump up the HTT. Frequency > timings. There is a pretty good chance your memory will be able to do 300 1:1 at 2.5-4-4-10 or 3-3-3-10 or something loose like that. TCCD/TCC5 isn't a huge fan of voltage, so I would stick with ~2.8V
Click to expand...
Well, you're right that he should be very very careful when raising anywhere beyond 2.9. For one, it definitely needs good cooling from what I read. Still, some TCCD chips are going to manage to tighten timings even more with higher voltages -- IF you take care of them with better cooling and such. That said, I would definitely recommend not jumping straight to 3.5V even if you had some darned good active cooling on those things. If you do improve cooling, step up slowly and keep that memory as cool as you can so it will last. Anyway, 2.8, 2.9V are definitely much safer and still get darned good overclocks on TCCD, so it definitely makes sense to use that if you need to. Besides, enforcer seemed to assume that the system can even SET 3.5V... Not many can... Few boards even do 2.9V, fewer still go above that at all. I bet 3.5 would be just about only on DFI boards or a few with a ddr booster. Oh, and you ONLY raise chipset/cpu voltage when it increases overclock. Never do this just for the fun of it like enforcer implied. Just bump things up just enough to be 100% stable. It is ok to raise voltages a bit, but, if it's not needed, don't do it. I used to have to raise mine to get 315 and it was fine. Still, don't be afraid to go up a little as long as you're staying within a safe range for your setup. Your CPU won't die noticably faster if you raise voltages slightly, I promise. Now, if you raise them a lot, THEN it will. San Diegos are more sensitive than Venices I was told, so don't go too far. I think it was around 1.55 that you want to switch to water cooling before you go any higher due to the potential risk.

2uantuM said:
do #3, but 1:1 with loose timings. You'll see higher performance.
Click to expand...
Why would you want to loosen timings? If it's stable and it's not about to tear up due to heat, nothing's wrong, right? Heck, it's running on a lower voltage in setup number 3.

Anyway, I don't like how people switch FSB and HTT. Somehow HTT means what used to be called FSB and HyperTransport now is called FSB. That said, you will see FAR more difference from your HyperTransport changing than from your "HTT" (aka what used to be FSB) changing. I've personally tested this with benchmarks on my own system which could set 315MHz fsb and give me the same cpu/memory settings and saw very little difference, so went with this since it's a little less stressful to the chipset (not that it makes much difference since suddenly it stopped needing a raised voltage to set 315.) In other words, you heard correctly. Keep that HyperTransport a reasonable number (don't set it to 500MHz or something dumb like that) and beyond that total speed matters far more. Don't be afraid to use different memory ratios to max things out. 1:1 has no performance advantage anymore (remember, the CPU itself controls the memory now, not the chipset, so there's no synchronization problem since the cpu just uses a divider of it's own clock to sync to the memory.)

EDIT: Oh yeah, auto on the hypertransport multiplier means it defaults to what the CPU says is stock. In other words, it will be x5. This will prevent your system from booting, so don't do that. You'll need to manually specify the setting to make sure it never goes over 1GHz (well, as you see on my setup, you might get away with a few mhz over, but, not much at all. HyperTransport just can't overclock right now. I wonder if some future boards/cpus will be able to overclock that too though.)
 
HanZo said:
loose timing? Why? if I can get stable at 2-3-3-7, y bump it up?

Hey current I'm trying 288 x 10 with same timing as #3, but I'm curious about LDT multiplier, should I put auto, x3, x4?
Click to expand...

Use ldt x 3. You won't notice any difference if ht bus is 1000, 864, or 750mhz.
 
Why would you want to loosen timings? If it's stable and it's not about to tear up due to heat, nothing's wrong, right? Heck, it's running on a lower voltage in setup number 3.
Click to expand...

Higher Ram Frequency = Higher bandwidth. In option 3, hes running a 5/6 divider for no reason. Running 1:1 would be very advantageous. You're talking DDR 466 vs DDR 560. 2-3-3-6@233 isn't as fast as 2.5-4-3-7@285.

Running a higher ram frequency dosn't have much to do with the HTT frequency, besides the divider. You wont see a difference between DDR400 with an HTT of 1000, 800, 600 etc. But you WILL notice a difference if you're running at DDR600 at a HTT of 1000, 800 and 600. Just because you wont notice any peformance gains from running a higher HTT, you will notice performance gains from running higher ram frequencies.

Use ldt x 3. You won't notice any difference if ht bus is 1000, 864, or 750mhz.
Click to expand...

I agree, ldtx3 is the way to go. As long as your LDT*HTT<1000, you're golden.

HyperTransport just can't overclock right now. I wonder if some future boards/cpus will be able to overclock that too though.
Click to expand...

It isn't that you can't overclock it, its just that there is no reason to. When you go over stock, you start to see some instability with no added performance gain.
 
2uantuM said:
Higher Ram Frequency = Higher bandwidth. In option 3, hes running a 5/6 divider for no reason. Running 1:1 would be very advantageous. You're talking DDR 466 vs DDR 560. 2-3-3-6@233 isn't as fast as 2.5-4-3-7@285.

Running a higher ram frequency dosn't have much to do with the HTT frequency, besides the divider. You wont see a difference between DDR400 with an HTT of 1000, 800, 600 etc. But you WILL notice a difference if you're running at DDR600 at a HTT of 1000, 800 and 600. Just because you wont notice any peformance gains from running a higher HTT, you will notice performance gains from running higher ram frequencies.



I agree, ldtx3 is the way to go. As long as your LDT*HTT<1000, you're golden.



It isn't that you can't overclock it, its just that there is no reason to. When you go over stock, you start to see some instability with no added performance gain.
Click to expand...






are u saying I should use 300 x 9 option?
 
No, option 3, 1:1

Higher CPU frequencies are more beneficial than higher ram frequencies when the CPU gain is 100MHz.
 
2uantuM said:
Running a higher ram frequency dosn't have much to do with the HTT frequency, besides the divider. You wont see a difference between DDR400 with an HTT of 1000, 800, 600 etc. But you WILL notice a difference if you're running at DDR600 at a HTT of 1000, 800 and 600. Just because you wont notice any peformance gains from running a higher HTT, you will notice performance gains from running higher ram frequencies.
Click to expand...
Ok, oops on my part. I didn't do the math on all that so didn't realize that you were saying to loosen the timings for higher frequency. Yeah, well, I do agree with that much. Lower timings help a lot, but, not as much as a considerably higher frequency (which DDR2 now proves.)



It isn't that you can't overclock it, its just that there is no reason to. When you go over stock, you start to see some instability with no added performance gain.
Click to expand...
No reason to? I imagine you do loose noticable performance if a perfect overclock is ruined by having to set the hypertransport really low since it won't oc worth anything. That's mainly what I meant. However, I was talking about future hardware. Who can say that a 1.2GHz HyperTransport won't make a giant leap in performance on some potential setup a few years from now? That said, I'm not disagreeing entirely, just saying that there is a certain point where it wiill hurt to set it too low.


And yes, you usually see a bigger jump from running your CPU faster. Well, you can benchmark things to get an idea where the line between the CPU and memory benefits more with one over the other, but, generally, 100MHz on an AMD cpu makes a pretty noticable difference.
 
No reason to? I imagine you do loose noticable performance if a perfect overclock is ruined by having to set the hypertransport really low since it won't oc worth anything. That's mainly what I meant. However, I was talking about future hardware. Who can say that a 1.2GHz HyperTransport won't make a giant leap in performance on some potential setup a few years from now?
Click to expand...

The hypertransport link isn't very saturated, which is why lower the frequency down to 800 or 600 doesn't hurt performance. If you have water trickling down a huge pipe, making the pipe bigger won't make the water move any faster. But like you said, if you set the hypertransport really low, you will probably see a performance hit. But there is no situation where that would happen with todays hardware and probably not tomorrows hardware, either. I don't think DDR2 will be able to saturate the hypertransport link.
 
2uantuM said:
No, option 3, 1:1

Higher CPU frequencies are more beneficial than higher ram frequencies when the CPU gain is 100MHz.
Click to expand...


hey, I guess u didn't read my post earlier, option 3, I have to run it on a divider 5/6 or it cannot be stable.
 
He was saying to raise latencies a bit to get the higher memory frequency because you should see better performance that way than with the lower frequencies. Well, of course, test both ways to be sure what works best for your situations (some things benefit more from low latencies, but, most benefit more from speed and low latencies just help.)
 
You must log in or register to reply here.
Back
Top