Lean Machine 2006 Part 3
11.02.06 - 11:13pm
The long delayed results for the Lean Machine are in. I apologize for the lengthy delay, thanks for sticking around. I’ve built some fast computers in the past and I’ve managed to push them even further but this particular machine bested all my previous records at stock speeds and then just demolished everything when I began overclocking it.
The Build
The original specifications for this build didn’t change much over the course of the build. Asides from changing the memory out for some more lenient sticks and holding off on the optical drives due to one not being in stock and the other out of my current price range(yeah,say what you will), the Lean Machine has held true. The list of parts below shows a compromise between price and performance with the majority of the pieces having margins where I could expand upon their initial performance with a few tweaks.
- CPU: Intel Core 2 Duo E6400 $224
- Motherboard: Gigabyte GA-965P-DS3 $145
- Video: eVGA 7900GTO $250
- Memory: 2x Crucial 1GB DDR2-667 10th Anniversary $235/li>
- HDD Array: 2x Seagate Barracuda 300GB $160
- Optical Drive 1: LITE-ON SHC-52S7K-05 $30
- Optical Drive 2: Plextor PX-755SA $105
- Cooling: Custom Phase Cooling $220
- Power Supply: OCZ GameXStream 700w $130
- Sound: Integrated Sound
- Case: Antec SOHO Server $40 used
- Paint Job: “Custom” Laguna Seca Blue $80
- Screens: 2x Samsung 205BW $580
- Speakers: Logitech Z-680 $220
- Keyboard: Logitech G15 $70
- Mouse: Logitech MX Revolution $90
- Operating System: Windows XP Professional and Vista RC1 $200
During the initial build when I assembled the system for pictures and my aircooled benchmarking I utilized the iSight in two Macbooks to create some great time lapse videos. My video and Paul Stamatiou’s video offer great shots of the system coming together and they themselves are mildly entertaining. Once I had finished assembling the computer the sheer size of the video card struck me. Before when I had it sitting on my desk I didn’t think much of it but once the system was assembled, the monstrous heatsink and PCB seemed to dominate the computer. Utilizing the stock Intel box heatsink made for a very quiet machine, the 7900GTX heatsink on the GTO does a great job dissipating the heat while keeping the fan whine to a minimum.
Initial Impressions
Naturally the first thing I did when the system booted for the first time was jump into the bios and take a peek around. The motherboard supports vcore up to 2 volts, vdimm up to 2.4 volts, and I believe 2 volts on the mch. I’ve heard reports of the voltage regulation on this board not being up to snuff in regards to stability but the sheer option pumping 2 volts through the processor was tantalizing considering it’s stock voltage is around 1.25 volts or so. One thing I disliked about the bios was the fact that Gigabyte hides all the memory timings within the the bios. To enable manipulation of the memory timings and subtimings you need to hit ctrl+F1 in the main bios screen. The screen will flicker and then the memory timings will be available to you in the MB Intelligent Tweaker section of the bios. The lack of a CPU temp cutoff was a bit of a bummer also as that has saved me from cooking my processor before but it is something I can do with ought.
I swapped out the twin hard drives for a single 74 gig raptor as I didn’t relish in the thought of re-installing windows multiple times and dealing with loading up the RAID drivers. The computer seemed to tear through the windows installation process which was a small sign of its speed. Once windows was booted up and the necessary drivers, patches, and programs were installed I began testing the limits of the system. My original desire was to run every benchmark I knew of in the various configurations that I set the system up with however after running through the suite once at stock speeds I decided to cut the tests down to SuperPI 1M, 2M, 16M, 32M, 3DMark01, 05, 06, and AquaMark. I ran the AquaMark and 3DMark01 to show how the system scaled with the processor as they were heavily CPU limited and I ran the other 3D benchmarks to show how the system reacted overall as it was a more even match.
SuperPi
SuperPi is one of the longest running and most highly toted CPU benchmarks in the world. At its core SuperPi is a simple program used to calculate Pi to a specific degree of accuracy. This can range from 1 million decimal places to 32 million decimal places. The larger the degree of accuracy the greater the computation time. Having a standardized measurement to compete around (calculation times), overclockers have created a scene of sorts around breaking SuperPi world records. I believe the current record for SuperPi 1M is below 9 seconds utilzing a X6800 at 5.8 GHz or so, the times and speeds change so often that I have trouble keeping track of it. Either way, the SuperPi calculations go to show how well this system scales as clockspeed and memory bandwidth increases.
I know that is a lot of material to digest, now imagine running all these tests 3 times each to achieve the best result. I ran Super Pi 3 times per measurement as the best result tends to come out after the 3rd run and possibly longer than that. I managed to get results for 2133MHz(8*266), 2666MHz(8*333), 3200MHz(8*400), and 3600MHz(8*450) along with a few “suicide” shots above 3600MHz.
3DMark
FutureMark, the creators of the PCMark and 3DMark series, have done a great job pushing the limits of current and future hardware with the advances in their simulations. These tests determine how effective your hardware is at rendering a specific scene and then gives you a score based on how many frames were rendered and the speed at which they were rendered. Depending on the test, each 3DMark has evolved into testing a different part of the system, mainly due to advances in hardware. 3DMark01 barely offers a challenge to video cards today as the rendering is relatively simple however it is a great CPU test as it takes a lot of work for the processor to constantly feed the graphics card information to keep it busy. 3DMark05 and 06 have the opposite effect as my tests show increases in graphical power considerably increase the score. These benchmarks are great as they showcase some beautiful technologies however watching them run for 8+ hours straight can get to be a bit mind numbing.
3DMark01 scaled very well with increases in processor speed, stock processor speeds just weren’t up to snuff compared to the 7900GTO. 3DMark05 and 06 do a great job hammering all aspects of the system and I feel they do a great job showing how the system handles large rendering loads similar to those found in modern video games. Because I immediately started benchmarking the system when I had it up and running I hadn’t had a chance to actually play a video game but the scores that were coming out of it were very promising.
AquaMark
AquaMark is another test similar to 3DMark01 in that it tends to be heavily CPU limited. This proved to be very true as the scores below show. I wish I had the results from one of the runs I did at 3.8 GHz but the machine crashed when I took the screenshot to save my results. 3.8GHz hasn’t been stable since either, no surprise there.
Final Impression
After spending a few days tweaking and overclocking this computer I have learned a few things. My research paid off great dividends in regards to the quality of the parts that I chose however my choice in motherboard is greatly holding me back. Initially I heard that the DS3 was very picky in regards to memory overclocking. Due to the nature of the motherboard I was unable to push the memory past ~967 MHz. While this is a 300MHz overclock from stock speeds for the memory, I was really looking for 1066MHz out of this memory so i could attain a 100% overclock on the processor. No matter how many volts and how loose I set the memory the system would not budge past the upper 960’s. These same sticks have pushed past 1100 MHz in other boards so there is still hope left. Gigabyte is reportedly trying to solve this problem as it seems to be an issue within the bios, if there is a fix then I will have even faster results to post. As it stands, 3868MHz was the highest I could push this little processor, coming from a lowly 2133 MHz. I suppose an 81% overclock is nothing to laugh at and I commend Intel on a beautiful processor however it has so much more potential to be unleashed given the correct memory setup.
The video card was an awkward little package to deal with in itself. When I received the box it was plastered with stickers labeling it as a 7900GTO. Upon further investigation I found the box to be a 7900GTX box featuring essentially a 7900GTX with underclocked memory. In the early days of the 79xx series a large number of cards were returned for failing memory and voltage regulators. Well.. I managed to push my GTO from stock clocks of 650core/660memory to 720core/820memory, just a shade over the 800MHz on a stock GTX’s memory. However at these speeds the card began to generate a distinct whine from the voltage regulators on the card. This whine is similar to the noise you may here from televisions and other such devices, high pitched and piercing. If you can handle the whine then the card overclocks like a dream however if you are looking for silence, stick to stock clocks.
One last note about overclocking this particular system, the DS3 likes to undervolt the CPU by about .05 volts when under load. This isn’t a whole lot to deal with but keep this in mind as this could potentially be the difference between a stable and unstable overclock. The Core 2 Duos do a great job of keeping cool so feel free to be a bit more relaxed in regards to pushing volts through them. I managed my highest overclock at 1.35 volts +/- .05 volts which was the exact number that the processor is warranted to. If I had not been restricted by the ram them I might have pushed harder, perhaps I will be able to explore this later if Gigabyte gets their act straight.
Conclusion
I hope you enjoyed the results of this little experiment of mine, if anything I hope you enjoyed the screenshots. This particular setup is by far one of the best combined deals this year, coupling a top of the line graphics card with a low-end processor, mid-range board, and receiving bleeding edge results. That’s about it, now it is time for me to go frag some friends in a devastating game of BF2142.
Just realized that my graphs look like crap, I hope you guys can look past my terrible Excel skills and indulge in the information.
that’s a lean machine. i’m on a very fast connection, but for someone who is not, the first image might take a while for someone else, as it is 1500kb.
Photoshop Save for Web, 60-70% FTW.
Took a while to load. ;) None the less, nice. :-)
Tomorrow I’ll have to use photochop and save the images in a better format and link them, make it easier on the slower connections. Just another thing to add to the queue, such as the redesign of this website that is 9 days overdue. Thanks for the comments though, it is a blast to have this machine running :)
Nice machine and review. thanks for sharing. What phase cooling are you using? Can you also share how you are connecting two power supplies for 1200 watts total? Regarding MB’s,I have both a gigabyte 965 s3 and the asus p5b deluxe (was an oem version from newegg for $162, retail was too pricey for me at the time), and it seems the asus mb seems to be a little more stable and allow for slightly higher bus speeds, in case that helps, although reaching over 3.8 Ghz is already very impressive.
Claus, I built a 1/4 horsepower phasecooler to keep the system cool and I currently have it configured to maintain a -35 celsius evap temperature around 225 watts. I am hopefully going to have a stronger cascaded system built when I upgrade my motherboard to help push my processor a bit further. The p5b is a considerably better board than the S3 and DS3 motherboards mainly due to the fact that there are very little memory compatibility problems. Once you shoot past a 450MHz fsb your stability turns to crap, my maximum overclock was attained through pushing maximum volts through the motherboard, 2.4 through the memory, 5-5-5-18 timings on the memory, and ambient temperatures were around 0 celsius. In regards to the dual powersupply setup, I had plans for using 2 7900GTOs and a 975x chipset and hacked drivers for SLI and I had concerns over stressing a single powersupply. Since I didn’t pursue this route the excess power isn’t needed but may be needed for future upgrades. If I were to go this route then I would hook up the weaker powersupply to power the external connectors on the video cards and then use the stronger powersupply to power the motherboard and hard drives. All peripheral stuff would all get lumped onto the weaker powersupply as the motherboard will be drawing immense amounts of power due to overclocking the processor and graphics cards. let me know if you have any other questions.