Overclocking Escapades: socket 939 Athlon 64 3800+

[Tech Tweaker]

Overclocking Escapades: socket 939 Athlon 64 3800+ Venice (E6)

When I started out to see what a socket 939 Athlon 64 3800+ Venice chip's limits were, I was encouraged, but then my hopes were quickly dashed.

I got it to do 2.6GHz on stock volts (1.4V), anything beyond that though required a massive voltage increase to 1.47-1.5V+. I hit a major wall at 2.7GHz and couldn't get stability past that, I got it up to 2.75-2.77GHz once or twice, but it wasn't very stable and upon reboot I couldn't get it to POST. Voltage increases had little to no effect, I went all the way to 1.55V and even upped the NB voltage, but it didn't help as I still couldn't get it stable past 2.7GHz.

Reducing the multiplier to 11.5x down from the stock 12x helped a little, as I could then get it to boot at 2.75GHz, but it locked up soon after entering the OS and would have likely required more voltage than my board could supply to remain stable.

Temps were good, at 2.7GHz I was idling in the high 30's/low 40's°C, and at load I still wasn't hitting 50°C.

I don't know if I've just got a stubborn CPU or a CPU that's a lemon when it comes to overclocking. I'd heard stories of folks achieving 2.7GHz+ on stock volts and was intrigued, my own results weren't nearly as impressive to me.

 

[TsunamiJuan]

Chances are if the voltage isn't doing anything to push the speed up. Than your probably at the thermal wall. Which is common for amd chips. Chances are if you ran that same voltage on say phase change system or dry ice, you would probably be able to push the multis more without having to up the voltage for a bit.

I have also found that sometimes I have to drive ram at higher voltages on AMD systems .5-.10, sometimes by a small amount, others by a larger amount .15-.40v.

 

[Tech Tweaker]

Chances are if the voltage isn't doing anything to push the speed up. Than your probably at the thermal wall. Which is common for amd chips. Chances are if you ran that same voltage on say phase change system or dry ice, you would probably be able to push the multis more without having to up the voltage for a bit.

I have also found that sometimes I have to drive ram at higher voltages on AMD systems .5-.10, sometimes by a small amount, others by a larger amount .15-.40v.

Hmm... Maybe, but I didn't think that was possible since my temps are at the normal range (idling at 37-40°C at 2.7GHz, which as I've seen is normal for most of these Venice cores) (this particular one is not a normal CPU temp-wise though, and at stock speeds idles at 30-33°C).

I think I may have to let it sit at 2.6GHz max on this particular CPU, since it seems most stable at that overclock and oddly the temps are nearly identical to stock speeds, surprisingly at the stock voltage.

Man, this CPU is underwhelming overclock-wise. An 8.5% overclock is nothing for me, I could get a 17% overclock out of my S754 Sempron with hardly any effort or voltage increases at all and my S939 Athlon 64 3200+ Venice (2GHz stock) would do 2.4GHz (20% overclock) without too much effort (would get into the OS at 2.6GHz, but wasn't very stable) (wish I hadn't sold that chip).

I wouldn't think I'd have to increase my ram voltage much with the ram I'm running, since the stuff I've got can run at up to DDR480-500 at 2.7-2.8 volts, and I was running it under-clocked with the 166MHz divider. Recommended specs for these sticks are DDR400/PC3200 at 3-3-3-8 with 2.5V, I was running them at DDR386-395/PC3088-3160 at 2.5-3-3-7 with 2.6V. I was kind of hoping under-clocking the memory would free up some extra potential for the overclock, but that didn't happen.

I'm testing another setup now on Prime to see if it's stable, running it at 2.6GHz on stock volts with the memory at 1:1 (DDR434/PC3500) at 2.5-3-3-7 with 2.6V. I tried 2.65GHz later on yesterday after my initial post, but couldn't seem to get it stable at that speed even with the memory under-clocked.

If I had a better board I might be able to drop the multi and up the FSB, but I'm limited to around 260MHz on this board

 

[TsunamiJuan]

Yeah i never tried overclocking my amd 64 3200 till I had it cold. But that only took me about 400-500mhz higher. Then I hit the thermal wall. I had to hardmod my board and run ln2 to get it up to 2800.

Have you increased the north bridge voltages at all?

 

[Tech Tweaker]

Yeah i never tried overclocking my amd 64 3200 till I had it cold. But that only took me about 400-500mhz higher. Then I hit the thermal wall. I had to hardmod my board and run ln2 to get it up to 2800.

Have you increased the north bridge voltages at all?

Yes, I tried that. Made no difference though.

I'm testing another setup now on Prime to see if it's stable, running it at 2.6GHz on stock volts with the memory at 1:1 (DDR434/PC3500) at 2.5-3-3-7 with 2.6V. I tried 2.65GHz later on yesterday after my initial post, but couldn't seem to get it stable at that speed even with the memory under-clocked.

If I had a better board I might be able to drop the multi and up the FSB, but I'm limited to around 260MHz on this board

Tested this setup on Prime95, it was stable for 33 mins, then I ended the test because I some chores I needed to go do around the house.

Max temp reached was 50°C.

 

[TsunamiJuan]

Yeah your probably at a thermal wall than. Btw is that 50c core, or external?

 

[Tech Tweaker]

Yeah your probably at a thermal wall than. Btw is that 50c core, or external?

Shame. I wonder if my other 3800 would do any better...

Yep, that's the core temp.

 

[TsunamiJuan]

Wouldn't hurt to try, if theres a large age differance between the two in manufacturing time there will be a higher chance of the newer clocking higher.

 

[Tech Tweaker]

Well, I swapped the other 3800 in to test it out.

It achieved 2.7GHz with slightly lower voltage of 1.42-1.44V, temps remain about the same as the other though, idling at 35-39°C.

I did manage to get this one to boot at 2.8GHz, but it was not at all stable and locked up as soon as I started a 1m run on SuperPi. It also required me to set it to the highest setting I'm comfortable going to in bios of 1.55V, which is actually 1.52-1.54V in the OS since there's a bit of vdroop. If there were a 1.575V setting in bios I would try that, but oddly it jumps from 1.55V to 1.65V, despite going in .025V increments from 1.4 to 1.55V.

It can complete a run of SuperPi 1m at 2.75GHz, but requires setting voltage to 1.55V to do it, and idles a good 5-7°C above normal idling in the 37-40°C range.

Somehow I didn't expect miracles from this second chip though, since when I was removing the other one I checked the date codes and they are identical, they're also the same stepping (NBBWE).

 

[TsunamiJuan]

Yeah I figured it was a thermal wall for that chip, It's really interesting how chip quality for overclocking ranges sometimes. Some of the older chips there are extremely wide differance between chips performance and thermal walls and voltage loads. There is all ways some degree of difference even if you get chips cut out of the same wafer.

 

[Tech Tweaker]

I finally broke the 2.8GHz barrier today.

It seems the cheapo power supply I had been using was holding me back a bit because when I went for 2.7GHz it would quickly become unstable almost immediately after I would start up prime95. So, on a hunch, I swapped it out for my Corsair TX750W. Voila, instability is gone, ran prime 95 for twenty minutes without issue at the same settings. Just goes to show you shouldn't scrimp on the power supply.

So, I decided to make a run for 2.8GHz, and to my surprise it worked and I was able to get to the desktop and snap a screen shot of CPU-Z's CPU tab.

Now there is the next thing that is holding me back, it seems my northbridge cooling isn't sufficient as it is running rather warm (and may in fact be overheating and causing my lockups when I'm going for high overclocks). I guess that'll be my next upgrade on that system.

 

[OptyTrooper]

Chances are you have a mediocre stepping Venice. It's not uncommon for then to reach their limit around where your at.

 

[Tech Tweaker]

Put in another 3800+ to try out tonight, this time an LBBWE stepping rather than NBBWE.

Results seem encouraging so far, it idles at 23-27°C at stock and overclocked to 2.8GHz it idles at 29-33°C in a room that is around 21-22°C.

This one requires less voltage to run 2.8 than the others did, taking only 1.475V to make it to the desktop and it appears to be stable at 1.5V whereas the others I've tried needed 1.55V to even be marginally stable, and even then they ran very hot.

 

[Tech Tweaker]

Found what seems to be the absolute limit for this LBBWE stepping chip, 2994.5MHz.

 

[\dev\goat]

That is sick. I had a 3000+ and a 3200+, both of them maxed out slightly above 2.6. can't remember my voltages. Had one of those huge thermalright coolers though. can't remember what its called but its sitting beside me

 

[TsunamiJuan]

If you took that chip cold i am guessing you could probably get another 500mhz out of it, at around 1.8v

 

[keny]

I've been messing with the 939 stuff lately, managed to get this old 3500 to over 2.8ghz but it took a lot of vcore to get it there(1.75v) , temps never seem to be a problem and it will bench under 40c, I'm not familiar with benching this skt so it's a learning curve for me. What sort of voltage will the nforce ultra chipset take ?

http://valid.canardpc.com/show_oc.php?id=2331850

 

[OptyTrooper]

I've been messing with the 939 stuff lately, managed to get this old 3500 to over 2.8ghz but it took a lot of vcore to get it there(1.75v) , temps never seem to be a problem and it will bench under 40c, I'm not familiar with benching this skt so it's a learning curve for me. What sort of voltage will the nforce ultra chipset take ?

http://valid.canardpc.com/show_oc.php?id=2331850

I assume you mean nForce4 Ultra. That being the case it can take just about anything your board can throw @ it assuming temps are acceptable (I like to keep them under 40C if possible). Even then it likely won't kill it (unless you run it hot for a long time) just will get wonky.

 

[ssjwizard]

I managed to kill one nForce 4 ultra out of about 5 dead socket 939 boards that I have(opty I might want to see if you can ressurect one of these from parts off another) and that was from a complete fan failure(pre existing condition, I didnt notice) on the chipset while pumping 1.9v through it with ~2.0v going through the CPU.

 

[Tech Tweaker]

That is sick. I had a 3000+ and a 3200+, both of them maxed out slightly above 2.6. can't remember my voltages. Had one of those huge thermalright coolers though. can't remember what its called but its sitting beside me

View attachment 108651

Never had a 3000+ myself, but the last 3200+ I overclocked maxed out around 2.5-2.6GHz without increasing the voltage much.

I have (or have had) an A64 3200+ Venice, an A64 3700+ San Diego (never have overclocked it), three or four A64 3800+ Venice chips that I've overclocked (2.8Ghz max on all but one of them so far), an A64 4000+ Clawhammer (2.8GHz max) and an A64 4000+ San Diego (2.96GHz max), an A64 X2 4200+ Manchester (never overclocked yet), two A64 X2 4200+ Toledo chips (never overclocked yet, traded one of them), and an A64 X2 4400+ Toledo (never overclocked).

Actually, the 3200+ Venice is the next chip I'm planning to bench. I kind of regretted selling my last one without ever making a submission to the bot with it, so I bought a new one to play with.

I think that heat sink is a Thermalright SI-128, probably about the biggest heat sink I've ever seen, other than the Thermalright Ultra 120 Extreme.