Adventures in Overclocking with a Dual Core Athlon 64

Status
Not open for further replies.

fury

Posts: 13   +0
I just recently got through overclocking an Athlon 64 X2 3800+ system to 2.55ghz and thought I might share my experiences in case it might help someone not encounter all the trouble I went through :haha:

First off, here are the specs I started out with

Gigabyte K8N-SLI (apparently not the best board to OC with, but it worked in the end)
Athlon 64 X2 3800 (Week 33, '05. stock: 2ghz per core)
Antec Neo HE 500w power supply
BFG Geforce 7800 GT OC (stock: 425 / 1.05)
2x1024mb of OCZ Platinum PC3200 (stock: 400mhz CL2, 3-2-5)
Random Soundblaster Live that was laying around the house
Linksys Wireless card
Windows XP x64 edition

OK, so the board starts out at 200mhz, with a 5x multiplier for Hypertransport speed, for a total of 1,000mhz. This, as I found out after days of trial and error, is the key to any overclocking walls with this particular setup.

My initial attempts involved simply raising the 200mhz bit by bit. First to 205, then a quick round of benchmarks, then 210, and a quick round of benchmarks. So far so good. Wham. It hits the fan. Board won't even get past the POST screen at 215.

I tried raising the timings of the memory in case it was simply a bad set (as OCZ is particularly known for being able to go beyond stock at the tight timings it is rated at) - no go. I tried lowering the CPU multiplier, no go. I tried raising all the voltages, no go. On a somewhat related note, this board does not allow you to specify exact voltages for anything except the CPU. The chipset, DDR, and HT voltage settings are only expressed in values of +0.n - the maximum for which is +0.3, +0.2, and +0.3 respectively, and all of which I have set to maximum. The CPU can go anywhere from 0.8v to 1.75v in varying increments, so I attempted a massive 1.75v overvolt on the CPU to get it to emit any signs of life. Don't try this at home. In fact, I wouldn't even recommend that I try that anymore. Cause it still didn't work.

Then it dawned on me that the HT might be sensitive to overclocking, so I lower it down to 4x. Damn. Lower it down to 3x. 2.5x. 2x. 1.5x. Still nothing. Then I hit the bottom of the barrel, 1x, essentially bringing the HT bus to the same level as a regular old DDR FSB. Well, whaddya know. It posts. And boots. And benchmarks quite nicely.

I start getting a little crazy and pumping up the FSB. 220 goes by without incident. 230. I stop at 233 and check to make sure everything is stable - folding@home, Prime95, Sandra, 3dmark, etc. because I just like the number 233 (incidentally, that's faster than the first CPU that I recall being fast - the Pentium 133) and ironically enough, that's as high as the board will go. So at this point I'm still wondering if there is any possible way to squeeze more juice out of it now that I've hit the RAM's expected wall (I read on OCZ support forums that 230mhz is typically the highest that this particular model will go).

There is a feature on my board to run the RAM at a different clock speed vs. the FSB and I expect that any decent board not made to put in a Compaq computer will be able to use this feature... I made use of it to run the RAM divider at 2 / 1.83, and started on my FSB pumping journey again. 240. 245. 250. All whizzed by with no problems (outside of the occasional needing to bump the voltage on the CPU). At this point I was quite satisfied with the overclocking performance of my processor and board, as benchmarks went by without a hitch and the system was noticeably snappier - even moreso than when I initially installed it, having known previously only a single CPU system with generic RAM at crap timings.

I decided to push it to the very limit. 255 FSB, leaving the RAM at the max 466 (due to the divider) - and overclock the video card as well. Since I had already found out the video card's maximums before I started going nuts on everything else, this part was rather painless. A quick run of the NVTray application to enable Coolbits overclocking on my new Windows install, then I go in and crank it up to 495mhz core, 1.2ghz memory (apparently not the best overclock for the card, but it'll do for me).

What is it now?

Athlon 64 X2 3800 @ 2.55ghz, 1.65v (faster than an X2 4800 and more able to leap tall buildings in a single bound)
BFG Geforce 7800 GT OC @ 495 / 1.2
2x1024mb of OCZ Platinum PC3200 @ PC3700, CL2.5, 3-3-6, +0.2v

All on air cooling. The Athlon 64 X2 is the only item that contains aftermarket cooling hardware - the Zalman CNPS7700Cu with the fan controller set to high.

At this point I start up the benchmarks and prepare to **** my pants (Sorry for the expletive but it's justified, damnit :D)

SiSoftware Sandra

Benchmark Results for CPU Arithmetic
Dhrystone ALU : 24948 MIPS (beats out all of the reference scores)
Whetstone iSSE3 : 10834 MFLOPS (bested only by the Pentium 4 reference score)

Benchmark Results for Memory Bandwidth
RAM Bandwidth Int Buff'd iSSE2 : 6425 MB/s
RAM Bandwidth Float Buff'd iSSE2 : 6372 MB/s
Both scores mopped the floor with even the best reference score

3dmark05 Pro
Complete benchmark (Default settings, except ALL tests checked in the selection dialog), with the video card settings at highest performance
8,417 3dmarks (7,177 CPU marks) - compare: http://service.futuremark.com/compare?3dm05=1696074

Observations
  • The HT bus speed multiplier did not seem to have any discernible effect on the speed of the system, either perceived or benchmarkable. It seems that nothing quite saturates a DDR 255mhz FSB quite yet, so a HT multiplier of 1x doesn't bother me a bit since it allows me such a great overclock.
  • Overclocking the FSB to 266 (keeping the processor at the 10x multiplier still) and lowering the RAM divider to make it run at 441mhz lowered the RAM bandwidth to 6.3gb but was a higher percentage of the bus's potential speed than 6.4gb at 466 - 90% vs 85%. Thus, it seems that memory bandwidth efficiency is positively affected by a faster processor which makes sense since the controller is on the chip instead of the board. I imagine people with elegant liquid cooling setups can manage an even higher overclock and thus enjoy an even higher percentage of use of the available memory bandwidth. Unfortunately the system was not stable enough at this speed to finish 3dmark
  • AMD processors rock
  • OCZ ram rocks
  • Rocks don't feel particularly well when bashed against one's head
  • I overclocked as many MHz out of these two cores as I had on my previous system altogether: 1100 MHz

Hope this mess of text happens to help someone in the same situation, and if not, I hope it at least entertained you.
 
Thanks :D

I should also note that I am running a Prime 95 torture test (blend) to maximize the chance of finding instability. Two instances to ensure both cores are blazing.

Current temp is 58°C, the Zalman fan is still at full speed on the fan controller (I hear you can plug it directly to the mobo to get a few hundred more RPMs out of it though)

Tomorrow if the Prime 95 stress test goes well I may try and lower the voltage a bit and plug the fan directly to the motherboard to see if I can get that down to a more reasonable temperature.

Prime 95 benchies said:
Best time for 512K FFT length: 18.370 ms.
Best time for 640K FFT length: 23.787 ms.
Best time for 768K FFT length: 28.830 ms.
Best time for 896K FFT length: 34.649 ms.
Best time for 1024K FFT length: 38.472 ms.
Best time for 1280K FFT length: 48.963 ms.
Best time for 1536K FFT length: 59.887 ms.
Best time for 1792K FFT length: 72.554 ms.
Best time for 2048K FFT length: 81.090 ms.
Best time for 2560K FFT length: 107.759 ms.
Best time for 3072K FFT length: 131.566 ms.
Best time for 3584K FFT length: 158.925 ms.
Best time for 4096K FFT length: 177.190 ms.
Best time for 58 bit trial factors: 2.977 ms.
Best time for 59 bit trial factors: 2.979 ms.
Best time for 60 bit trial factors: 3.160 ms.
Best time for 61 bit trial factors: 3.260 ms.
Best time for 62 bit trial factors: 3.872 ms.
Best time for 63 bit trial factors: 4.165 ms.
Best time for 64 bit trial factors: 5.367 ms.
Best time for 65 bit trial factors: 6.369 ms.
Best time for 66 bit trial factors: 6.330 ms.
Best time for 67 bit trial factors: 6.301 ms.
 
Hmm, apparently I was overclocking the video card too hard and it was causing the system to lock up. :confused: I didn't know the video card could lock the system up.

One of the cores failed the Prime 95 torture test. The other torture test was still going when I woke up. I have lowered the CPU voltage in case it was a heat issue, raised memory timings, and I am going to try again tonight.
 
fury said:
I didn't know the video card could lock the system up.

you bet. are you just ocing the clock speeds on the card? or the memory timings and such

a very nice post btw
 
Just the clock speeds on the card. I don't mess around with that advanced memory timing stuff on the card - too easy to permanently damage the card, from what I hear.

It's hard for me to mess up system RAM on this board...I just reset it twice in quick succession before it POSTs and it reverts to safe speed settings and such...
 
Lasted through the night on a Prime95 FPU stress test with the CPU at 2.5 ghz (250 FSB), 1.55v, memory at 458, 2.5-3-3-5 1T.

I tried to get the memory to run at 500, but no matter how I set the timings (even tried 3-4-4-8 2T) it wouldn't budge

CPU was at 49 at the hottest; I connected the Zalman fan directly to the motherboard and it runs a couple hundred RPM faster than on the fan controller's highest speed setting. Oddly enough, it seems quieter to me with it directly connected to the motherboard than with it connected to the speed controller and set to high.
 
3800 X2 @2.53 Ghz

I also got my 3800 X2 running at 2.53Ghz

Here's my Story

I used to have 3000+(1.8Ghz) venice but I got it running at 2.53Ghz.. I saw this Fry's Electronics ad --> 3800 X2 (retail-PIB) plus a mobo for only $279.00 so I went to the store rushing early in the morning to grab one..

So excited that I change my old 3000+ venice to this new dual core CPU when I got home without changing/resetting my mobo bios setup.. I was shocked when I saw the clock speed when I turned it on @ 2.53Ghz.. my new X2 adapted the motherboard setting of my old 3000+ venice..

I tried stressing it for stability using sandra but fortunately its average temp is 45C.. case temp is around 30C.. Room temp is around 25-30C.. BTW,, im using stock AMD air cooling on my X2..

I wasnt expecting this result--> in the sandra benchmark it clearly defeated the AMD X2 4800+ ..

just wondering where to get (if there are) drivers for AMD X2



my setup..
AMD 3800+ X2 @ 2.53Ghz
ASUS A8n-SLI premium
2 x 512mb kingston value RAM :D
ATI Radeon X600 256mb :unch:
2 120Gb WD (7200rpm) --raid 0
Audigy 2zs platinum.. 6.1 logitech speaker
19" samsung 940b LCD
Plus 1 newbie user--> and thats me
 
jcX4Ever, howd you get the RAM to OC?

I've had a problem OC'ing my X2 3800+ past 2200Mhz... I believe the problem to be my cheap memory. I have 2GB Corsair Value RAM (400Mhz). I see that you have 1GB Kingston Value RAM and you were able to get to 2530Mhz. What mobo are you using? What are your memory settings?

BTW, Im using an ASUS A8R-MVP (ATI X-fire)
 
Oops, RTFS (Read the freekin signature)

Ahh, answered one of my questions. Youve got an ASUS A8n-SLI premium.
I absolutely love ASUS mobo's, and I traded an ECS in for this one. I bought a package deal from Fry's with the ECS and X2 3800 and promptly returned the ECS for a refund. ECS is pure crap.
 
Status
Not open for further replies.
Back