Overclocking Escapades: socket 939 Athlon 64 4000+ ClawHammer (CG)

[Tech Tweaker]

Well, when my 4000+ San Diego E6 turned out to be overly disappointing I figured I would see if the ClawHammer chip I had here could do any better.

Upon booting up I suddenly realized I still had it manually set up in the bios for running with the other CPU and the voltage was set at 1.35V as opposed to the default 1.5V of the ClawHammer. To my surprise though, it made it through POST just fine with the incorrect voltage setting, and even made it to the desktop.

So, out of curiosity of whether or not I could run it at this voltage and be stable I fired up Prime95 and CPUID HWMonitor so that I could monitor the temps during the test. To my amazement it seems to be completely stable, making it through 30 minutes on the blend test, during which it only inched its way past 40°C once and stayed below 40°C the rest of the time.

I have to ask, has anyone else ever ran a ClawHammer chip this far below its rated voltage, or is this uncommon? Were the AMD engineers just really generous when it came to the voltage on these and were they giving it more than necessary just in case it needed it?

It's actually still running as I type this, and everything seems perfectly normal.

Looks like a nice chip so far, with idle temps of 28-33°C and load temps of 34-41°. I must say I really had to try hard to get this chip to go over 39°C, it stayed under 40°C for over twenty minutes of Prime95, and never broke the 40°C barrier during multiple passes on SuperPi 1m, and several UCBench 2011 runs. I have to say from a temperature standpoint it is very nice (and encouraging) to find a chip that runs this cool at full load.

 

[RGone]

With that price at introduction, I doubt that many had one back then. That was a lot of money. You must be lucky in voltage and in having two 4000+ cpus. I am still running my FX-60 and it was a ton of coins back then too. Not a lot of folks ran one. Well the big guns but that was about it.

There were three of the 4000+ processors. 

1st: 
Microarchitecture	K8
Processor core  	ClawHammer
Core stepping   	CG
V core   	1.5V
Introduction date	Oct 19, 2004
Price at introduction     $729

2nd:
Microarchitecture	K8
Processor core   	San Diego
Core stepping  	E4
V core   	1.35V / 1.4V

3rd:
Microarchitecture	K8
Processor core   	San Diego
Core stepping   	E6
V core   	1.35V

 

[Woomack]

@Tech Tweaker
My Clawhammer FX-55 is overclocking about 100MHz on air , next 50MHz on water and up to 3.2GHz on SS.
What is killing oc is mainly overheating on these cpus. I didn't try to set lower voltage but I think it's pretty possible to make it run @1.35V.
I was able to pass UCB and some other benchies @3180MHz and 1.55V so not much higher than standard voltage - http://hwbot.org/submission/2235461_woomack_ucbench_2011_athlon_64_fx_55_clawhammer_68_mpt_score
My A8N-SLI is really bad and even if I wish I can't set higher voltage.
I'm not sure if your temps are real if you are using more standard air cooling. These cpus are generating lot of heat and I remember that on stock cooler ( this funny one with 2 heatpipes ) it was hitting 70*C+ under load.

 

[Tech Tweaker]

@Tech Tweaker
I'm not sure if your temps are real if you are using more standard air cooling. These cpus are generating lot of heat and I remember that on stock cooler ( this funny one with 2 heatpipes ) it was hitting 70*C+ under load.

I think the temperatures that are being reported are real. I can put my hand right at the base of the cooler at full load and it is just barely warm to the touch.

Some images for verification, this time at 1.3V.











I think the temperature has a lot to do with two things 1. the voltage being supplied to the CPU (when I was running this chip's brother an FX-53 with a stock cooler at the default 1.5V I was idling around 45°C and at load was getting up close to 60°C) 2. the cooler I'm using right now is probably overkill for this chip and could have something to do with my low temps.

There could be other factors contributing to the low temperatures as well though.

 

[Tech Tweaker]

Well, the absolute best I could muster with this one was 2.8GHz, any faster (even just to 2.85GHz) and it would blue screen no matter how much voltage I gave the CPU even with going to 1.85v (the absolute maximum my board supports).

 

[trents]

Is that cooler a Thermalright Ultra 120 extreme?

 

[Tech Tweaker]

Is that cooler a Thermalright Ultra 120 extreme?

Good guess, you were close.

It's a Thermalright Ultima 90.

Actually I'm not sure I could fit a TRUE on this board with the top PCIe slot being so close to the CPU socket, as it is with the Ultima 90 the heatsink comes within about a half-inch of my video card. Though I had been hoping to mount a TRUE on this one at some point if it were possible.

 

[keny]

Wow $730 on release and now there on eBay for £26 ($40)that's got to hurt if you bought on of these things on release. I was toying with the idea of getting into the socket 939 benching as there is loads of cheap cpu's out there to bench but the decent 939 boards still cost a small fortune.

 

[Tech Tweaker]

Well, I've upgraded the CPU heatsink to a Thermalright Ultima 90, upgraded the mosfet cooling from individual small copper heat sinks to a large Thermalright heatpipe cooler, upgraded the case and fans, and still can't get this chip any higher than 2.8GHz. It's strange, I'm still hitting the same exact barrier I was hitting with the stock heatsink and my old case. Would have thought the upgrades would have yielded some kind of improvement in pushing back the barrier on my overclock, but it doesn't seem to have yielded any improvements there.

I can get it to the desktop at 2.9GHz sometimes, but the system quickly locks up once it gets that far. Takes 1.75V just to get the system to attempt to load the OS at any speed above 2.8GHz, which is strange considering it only needs 1.6-1.625V to be relatively stable at 2.8.

Think I may have hit the so called "Thermal wall" of this CPU. Strange thing is it's not running all that hot as far as I can see, at 2.8GHz my idle temps are in the mid to upper thirties Celsius. At stock it idles in the low to mid thirties. My load temps are still well within check too, as I believe they were still at or below 50°C. So, I'm not entirely sure it is a thermal wall, maybe just a design limitation of this CPU (maybe just my particular batch/date code was a lemon).

 

[Tech Tweaker]

Well, good news is it's more-or-less rock-solid stable at 2.8GHz, unfortunately though it requires 1.65V to achieve that level of stability.

I tested it with Super Pi 32m and wPrime 1024M last night for some boints and it completed both, though Super Pi required me to increase the voltage higher than what wPrime needed. wPrime would run fine at 1.625V, but Super Pi needed at least 1.650V in order to actually complete the benchmark without encountering an error.

Have yet to find any settings at which it is stable enough to run Prime95 though, as it seems to run into errors right around 3-4 minutes into testing every time.

 

[Tech Tweaker]

Well, after killing my old A64 4000+ Clawhammer, I finally got to testing out a new one today.

New one is utter crap. That or the old one was golden, I'm not sure which.

New one won't run faster that 2.6GHz without crashing, locking up, or blue screening. Higher voltages don't help.

It seems that this particular stepping is just mediocre at best and is pretty near to it's max at the stock 2.4GHz.

It's either the CPU, my board, or the OS is borked because I've tried multiple sticks of memory and it still won't run any faster.

Old one took 1.52V to run at 2.6GHz stable, this one needs 1.55V.

Old one took 1.6V to run at 2.8GHz stable, this one needs 1.65V to even want to boot into the OS at 2.8GHz and is not even remotely stable.

This one is actually deteriorating as time goes by, last night I was running benches in UCBench 2011 and wPrime at 2.8GHz, today though it barely wants to run at 2.8GHz at all.

Wonder if I can Frankenstein the old one back to life...

 

[Tech Tweaker]

It lives!

Apparently my old 4000+ ClawHammer wasn't dead, it just doesn't seem to get along very well with my board. Another of DFI's weird quirks on their high-end boards I guess.

For some reason I can't get it to post with any memory in the primary memory slots (orange slots) I got it to run once I swapped the memory to the yellow slots though.

I hope there isn't something wrong with my memory slots though, that would suck big time.

I might have to pull another board out of storage to bench this chip again, since it doesn't seem to get along too well with my DFI SLI-DR Expert.