Saturday, January 6, 2007

Hitting the wall at 3.4 GHz

I ran Memtest86 v1.65 overnight for over 9 hours and 33 minutes (wall time) executing 19 passes without an error.

I restarted the system and allowed Nighthawk to settle in an idle state. The temperature readings at idle and 3.4 GHz overclock are: motherboard @ 30C/86F, CPU @ 49C/120F, Core #1 @ 55C/131F, Core #2 @ 56C/133F, and the GPU @ 53C/127F.

The first benchmark I ran was 3DMark06 and Nighthawk scored 8491 3DMarks, an improvement of 82 3DMarks over the 3.2 GHz benchmark. The graphics scores were similar but there was an improvement of 138 Marks in the CPU score.

The next benchmark I ran was the Everest Ultimate Cache & Memory Benchmark. There were improvements in all results, and the Memory Read result of 6856 MB/s now exceeds my highest score of 6781 MB/s achieved at no overclocking when memory was running at DDR2-1066.

I then ran 2 instances of SuperPi mod1.5 XS and took temperature readings at the end of the 20th loop of the 32M calculations. The temperatures were: motherboard @ 33C/91F, CPU @ 56C/133F, Cores #1 and #2 both @ 62C/144F, and GPU stayed at 53C/127F.

Both instances of SuperPi completed faster at 3.4 GHz versus 3.2 GHz overclock, completing 32M calculations 1 minute 14.031 seconds and 1 minute 14.312 seconds faster.



Everest Ultimate System Stability Test

My prior experience with the Everest Ultimate System Stability Test is that it really stresses the CPU more than other similar tools since I have observed it generate very high CPU temperature readings. This test keeps the CPU usage at 100% most of the time, with the usage dipping down to 50% when it switches tests.

The CPU temperature peaks at 72C/162F at 100% utilization, with both Cores #1 and #2 hitting 75C/167F.

During the periods when the Everest Ultimate System Stability Test is using the CPU at 50% while it is in transition between tests, the temperatures come down to more reasonable values.

The CPU temperature goes down to 57C/135F at 50% utilization, with both Cores #1 and #2 going down to 63C/145F.

At these very high temperature readings, I was not comfortable running this stability test longer than 10 minutes.

The good news is that all the temperatures are back at "idle" levels within 5 minutes of ending the stability test. The CPU temperature returns to 49C/120F, with Core #1 at 55C/131F and Core #2 at 56C/133F. The mother board and GPU temperatures stayed at 30C/86F and 53C/127F, respectively, throughout this stability test.

The Wall

I was feeling very good at this point so I fired up Johnny Lee's SP2004 ORTHOS as my final stability test with the intention of letting it run the rest of the day and overnight.

Nighthawk hit the wall just a little over 2 minutes into this stability test. Orthos stopped with the following messages: "FATAL ERROR: Rounding was 0.5, expected less than 0.4. Hardware failure detected, consult stress.txt file."

Windows XP MCE 2005 appeared to be running normally even after this error. I re-booted and tried a few more times before I gave up.

I lowered the overclock to 3.2 GHz and I have been running Orthos on Nighthawk since then, over 2 hours ago as I write this post. I can probably get over this hurdle by replacing the stock HSF and tweaking the voltages, but I am perfectly content at being able to overclock Nighthawk to 3.2 GHz, a respectable overclock.

Another one the rules I follow when overclocking is: Don't be greedy.

1 comment:

tomhole said...

Great update. You have hit a temperature wall, just like I did. A better HSF will help a little. In my case, the key to low temps was a better case cooling scheme.

I had temps just like yours with the stock HSF and stock case cooling. A Scythe Infinity HSF got me 8C. CHanging the cooling scheme in my case got me another 8C. I now idle at 35C. This is @ 3.6 GHz, 400 FSB. 1.4375 Vcore, 2.15 Vmem, 1.45 Vfsb and 1.4 Vnb. Max temp with Prime95 is 60C. Orthos is 57C. This is cooler than my system ran @ 2.66GHz, no overclocking and a stock HSF.

So, if you want to go further, you need to get the temps down.

Tom