PowerToTheUsers

After my previous Week of Overclock, about which I reported in http://www.madshrimps.be/vbulletin/f39/my-woc-43750/, it's time for another week of overclock. As this week is the week of the MSI-contest too, I will do some more SuperPi and 3DMark03-benchmarking, which wasn't planned before.

In short some information about my previous results:
I'm probably not the most extreme overclocker of our team, but as you can see in my HWBot-profile at http://www.hwbot.org/user.do?userId=7155, I benched with a lot of different hardware, especially Xeon- and Opteron-CPU's. Overclocking those server-oriented CPU's is quite different from overclocking other CPU's as there are very little enthusiast-boards for theses CPU's. For socket 1207 Opterons, there is the Asus L1N64-SLi-board, and for socket 771-Xeons you can use the Skulltrail-board.


Besides the 300 hardware-points, I scored a lot of global points with an IBM 16-core server. For some time I had the first place in wPrime 1024m and the third place in wPrime 32m, but now I dropped to the 2nd and 7th place. But still worth over 100 points.

Because of the HWBoints scored with different hardware, I still managed to get a top-3 position in our team, a 72nd place worldwide. But hopefully I can improve this during this Week of Overclock.

jmke

TOP 100 WorldWide is quite an achievement, will definitely earn you some points

__________________

PowerToTheUsers

The first result of this week:

Starring...

I already knew that 1 iRAM gave a nice boost in PCMark, but what with more iRAMs in RAID0? And why are most top-results with only 3 iRAMs, not with 4, 5, or even more?

The only way to find out which setup gave the best result, is to bench, bench and bench again.

The setup:

• Asus Striker II Formula mainboard with 780i-chipset (sorry MSI )
• Intel Core2Quad Extreme QX9650
• 2x1GB Crucial Ballistix Tracer PC2-8500
• OCZ 8800GTX (sorry MSI )
• 500GB Samsung sata2-disk

To test only the influence of the iRAM-setup, I ran the CPU and GPU on stock speed. The results:

• With the sataII disk: 9342 Marks
• 1 iRAM, no RAID: 13110 Marks
• 2 iRAMs in RAID0: 14237 Marks
• 3 iRAMs in RAID 0: 14492 Marks
• 4 iRAMs in RAID 0: 14581 Marks
• 3 iRAMs in RAID 0 and the windows swapfile on the 4th iRAM: 14507 Marks
• 4 iRAMs in RAID 5: 13670 Marks

Conclusions:
* Putting the swapfile on iRAM doesn't make a difference. I will check out if other benchmarks benefit more from the swapfile on iRAM.
* The "HDD - XP Startup" and "HDD - General usage" tests benefit most, "Multithreaded test 3: HDD Virus Scan" too, but to a lesser extent. I was surprised "File Decryption", "File Encryption" and "File Compression" weren't influenced by the use of iRAM(s), despite what the name suggests.
* Use RAID 0 (striping), as it's designed for maximum speed.
* Using 1 iRAM gives a very nice boost, and each extra iRAM gives another bonus. But the differences get smaller. Going from one iRAM to 2 gave 1127 more marks, but going from 3 to 4 gives another 89 Marks extra.

I compared my results with K|ngp|n's top-result, and for most tests my score was around half of his score, and the total Marks was also the half. Only the "Transparent Windows" was about 1/15th of his result! He got 15.567 windows/sec, I only managed to get about 1075 windows/s... What's the trick?

I tried to submit every result, but the ORB was down for some time, and when it is online, I get a message "Project submission failed. System information in the submitted project was corrupted." Any ideas? The only result I submitted succesfull was the 4iRAM-RAID5-result: http://service.futuremark.com/compare?pcm05=1525792 What's the trick?

Massman

Use Windows Vista with the Aero theme enabled for the PCMark05 benchmark, it's all in the transparent windows

PowerToTheUsers

Today, I worked further on PCMark05. First I tried installing the benchmark-suite on the 3 iRAM-setup, maybe shorter load-times would bring more Marks. This test turned out to make no difference. Every bench has a few Marks more or less, and this result fell in those ranges. Furthermore there was no test of the suite which had a remarkable gain.

The next thing I tested was an Ageia PhysX-card. This 128MB PPU (Physics Processing Unit) made no difference either. In the first run, the result of "Physics and 3D" was the highest I reached in the previous days, but in the second run it was one of the lowest scores. Just some general fluctuations, but in the end no extra advantage.

When trying to submit the results, I still get the same error message: Any help with that?

Next tests will be SuperPi, PiFast and wPrime with the program files and/or the swapfile on the sata-drive or on iRAM.

Pardons

Impressive contribution Johan

PowerToTheUsers

And indeed, there is a difference. I tested everything on stockspeed (QX9650@3GHz) and with only a slightly optimized OS.

In SuperPi (1M and 32M), the best results were reached where the program files and the windows swapfile were on the iRAM. In SuperPi 1M it made an average difference of 0.063 second: 15.547 with everything on the C-drive, 15.484 with everything on iRAM. In SuperPi 32M the result was 17m45.781 with everything on the C-drive and 17m43.843 with everything on iRAM: so an improvement of almost 2 seconds. I also noticed the halfway-results (one thing on C, the other on iRAM) had even worse results. It seems SuperPi likes to be run on the same drive as the swapfile, and best on iRAM.

In PiFast the results were different: on the C-drive the result was 32.43, on iRAM it was 32.44. Not a big difference, but here the halfway-results were slower too: 32.46 and 32.50 with the program on C and the swapfile on iRAM and the other way around.

The last benchmark was wPrime. Here the result was comparable to the results of PiFast: everything on the C-drive gave the fastest result, next was the full-iRAM result, and the mixed iRAM/C-drive results were the slowest.

Conclusion:
* Use a swapfile and the superpi-files on iRAM for SuperPi benches.
* Use a regular setup (program files and swapfile on C) for PiFast and wPrime

jmke

I like you analytical approach to OC

__________________

PowerToTheUsers

I installed Vista (Home Premium with SP1) and now the PCMark-score is 15628, so about 1000 Marks higher. Indeed the "Transparent Windows"-test scor is much higher now, but all other scores are about 5% lower

After I ran the februari08-patch, I could submit a result, but after the next run it failed again... I really need to get that problem sorted out

jmke

Rollback?

__________________