Disclaimer: This is an ES CPU, retail may or may not have changes from this sample.
Note: Check out Part I article to know the test methodology.
Goals and Objectives: The primary goal of this overclocking experiment is to reach 50% overclock without hard modding and costly cooling requirements.
Success Indicator: There will be four(4) success indicators:
- Reaching 50% on all four cores
- Stable multi-threaded Prime95 with aircool heatsink for at least an hour
- Completion of all benchmarks
- Zero Physical Changes on the hardware (no hard mod, no lapping, etc)
Test Set Up
Processor Intel® Xeon® X3210(2.13GHz) Quad Core
Motherboard: Asus® P5B Deluxe BIOS 1004
Memory: Apacer Value DDR2-800MHz 2x1024MB(CL5-5-5-18 SPD)
Hard Drive: Seagate* Barracuda 120GB SATA RAID 0
Video Card: Asus® 7900GTX 512MB v93.74
Power Supply: Delta 670W EPS12V (server grade)
Ambient Temperature: 32c
Heatsink/Fan: Do not expect water cooling, or phase change from this Blog article. The best I could muster is an aging Thermalright® XP-90 with a Delta® brushless fan model FFC0912DE (1.5a) and an LGA775 retention mechanism. This combo has served me well, and while they are not the best air cooling out there, I have no other heatsink available except the stock reference cooler. Thermaltake® Big Typhoon, Sunbeam® Tuniq Tower, Sycthe® Ninja Plus, Zalman® CNPS9700 or even Coolermaster® GeminII are all great heatsink, and arguably, the best right now in the market but availability and most importantly $$$ is scarce right now. But hey, I have used it in my Prescott days, so it should perform equally well (fingers crossed).
Motherboard: The motherboard Asus® P5B Deluxe is selected as the motherboard of choice, since this seems to be the most stable I have tested when it comes to high FSB overclocking. Performance, on the other hand, is not as best as their i975 brethren and manual selection of bootstrap is not possible which is a major drawback of this board, specially when hitting FSB beyond 400MHz. Another major drawback is the severe voltage drop of the vCore. Asus® almost always have this "feature" in every boards they released in the market. While I don't find the vdroop a major concern with dual core, it's proven to be a major pain with quad core. Both the performance and vdroop issue will be dealt with in order to level set the expectations from this board.
Memory: I have no plans of bringing this baby thru 500FSB, and target is a meager 400FSB. Some might think that the target is too low, but hey, this is quad core and early indication of quad core overclocking are all capped somewhere at 340MHz to 360MHz, with some hitting 390MHz. Anyway, newer BIOS and motherboard revisions has shown that they can push Kentsfield, some (*cough* eVGA) even recalled their board because of poor Kentsfield overclocking experience. Anyway, so for this round, my value RAM DDR2-800MHz 2x1GB dual-sided from Apacer® will be enough for the time being (cough *sour grape*).
Hard Drive: Two Seagate® 160GB SATA/3Gbps/8Mb will be used and configured as RAID 0 using Intel® Matrix RAID configuration with their ICH8R chipset. Not the fastest, but it's the most reliable and compatible that I have used so far. The hard drive will be partitioned in two, one will be for the OS installation, and the other partition will just be for miscellaneous data.
Video Card: I am thinking of using lowly PCI video card, but since I am not going to do magnificent score anyway, I decided that an Asus® 7900GTX/512MB is good enough. The driver is the latest (dated November 2006) downloadable for this video card at the time of testing. No tweak is done on this video card, and operated on stock.
Power Supply: A Delta® 670w model DPS-670DSB REV: S2. It's a full featured power supply and actually stable enough to run even server processors. It's an industrial grade power supply, and would be more than enough for my specifications.
Chassis: The testing will be done on an open test bed so there will be no chassis.
With all the preparations done, the issue about performance with the seemingly changing and uncontrollable bootstrap on the Asus® P5B Deluxe will be ignored, and numbers coming out of them will be taken as is, without regard if there's any "linear" performance improvement or degradation. The other issue about the horrible voltage droop will be taken care of if and when needed.
I set my memory and FSB ratio to 1:1, and set the memory timing manually to CL5-5-5-15. vDIMM is set to Auto just to be sure I am on the relatively safe zone. PCIe is also set to Auto, Spread Spectrum is disabled, PCI Frequency is manually set to 33MHz. An initial test run has been done, to record the stock idle and load temperature.
Stock Screenshot (Temperature at Load)
The vCore is set to 1.25v, since when setting it at Auto, the motherboard is giving the processor a whooping 1.6v and it droops to 1.5v during load. Hitting the next bootstrap level of 333MHz (2.66GHz) is fairly easy with this processor, and it is done with just 1.25v.
Reaching 366MHz FSB (2.93GHz) shows the vDroop problem is impacting multi-threaded Prime95, but it is benchstable. Adjusting the vCore to 1.28v helped alleviate the issue with the vDroop and the system is rock stable.
The 375MHz barrier is proving to be a difficult proposition. Changing several voltage options from vCore, to vChipset is not producing any significant stability. While all the benchmark are stable, mPrime95 is producing an error after 30mins and pushing vCore only worsens the scenario. The second issue of the motherboard I mentioned earlier needs to be explored. While the loaded temperature might be an issue, a vDroop mod will be checked first if it can help boost stability. After a few googling here and there, the droop mod by hicookie will be tried. But this is a hard mod, so instead of soldering a VR in there, a pencil mod will be done. After a 3min job, it proved to be succesful and stability is reached with a meager 1.31vCore.
After a few more minutes of pushing, 399MHz which is the 50% target and goal for this experiment has been reached. The ending vCore required at this speed of 3.2GHz is 1.36v. However, a lot of fans has been used to achieve this and all fan headers of the motherboard has proved useful.
Overclock Screenshot (Temperature at Load)
Considering the clock speed, performance, and temperature at load, this is an impressive overclock. With just a minor modification needed, which is the pencil mod which is a very quick job, a stable overclock with a steady vCore holding during load produces an excellent results. To recap, I have reached the goals and objective of this experiement thru:
- Reaching 50% on all four cores - checked
- Stable multi-threaded Prime95 with aircool heatsink for at least an hour - checked
- Completion of all benchmarks - checked
- Zero Physical Changes on the hardware (no hard mod, no lapping, etc) - crossed
The pencil mod is not a full blown hard mod, but in a sense, it is still a mod. I have two choices, get a better heatsink and bump the vCore higher to make up for the vDroop or do a pencil mod. I chose the pencil mod, failing one of my criteria, than spend hard earned money for a better heatsink. But hey, you can pitch in your idea whether my experiment is a success or failure in this regard.
Since the overclock is done relatively easy, I decided to try and push forward with the overclock. Note that going beyond 400MHz means I am not just overclocking the processor and chipset, but also the memory as well. Due to the time limitation, and my determination (or lack of), I am able to push the system up to 436MHz with 1.44v, resulting to 3.48GHz. I did not try to find the max FSB via lowering multi, since I always doubt my memory can go beyond 900MHz anyway and doing such might result to changing the northbridge heatsink, or worst, replacing my RAM.
Overclock Screenshot (Temperature at Load)
At 3.48GHz with 1.44vCore, Core Temp is showing 80c tops at full load for all cores using mPrime95. While it is almost impossible to load all four cores at 100% for a prolonged time, it's still important to see how stable the system is. All the fans are working at full speed to keep the system cool, and I have to add FET cooling all over the board. At this stage, I have achieved a 63% overclock, with mere air coolers and a measly 1.44vCore, and 3.48GHz is not too bad for a quad core too with 32c ambient room temperature.
CPU-Z Validation (Click on the image)
The following benchmark will be based on this final overclock. Enjoy...
Doom3: Low Quality-640x480
Cinebench: Render Score
Cinebench: Render Time
SANDRA: RAM Bandwidth
::Closing Thoughts:: I can not stress hard enough how this processor is truly a wonder in all aspects. It can overclock well, with a good cooler and measly 1.35v, and it can also work well with all the four cores working at full speed with voltage as low as 1.175v and silent fan. If you can not afford the Intel® Core™ 2 Extreme QX6700 or Intel® Core™ 2 Quad Q6600, then try to consider this one and give a good, solid and hard look.
Since this processor has a low clock multiplier, it is possible to really push that FSB high up in the air, and still achieve 50% overclock with good air cooling without a need for fire extinguisher. And for those who have better RAM and cooler, I am sure that they can achieve more than my max overclock of 63%. I am looking forward to extreme overclockers to achive a first ever 100% OC on quad core (if there's someone who hit it already, just let me know.
That is all for the day, it's been a busy day, FanBoy out...