Get ready y'all, I have a long one for you today.
This has taken most of the day, plus a few hours last night, to set up and gather... I hope you enjoy.
As many people know, LGA775 has so many CPUs that are very overclockable, inexpensive, and they largely don't draw a lot of power, unless you're looking at Netburst. But the 45nm Yorkfield CPUs especially seem to run cool, and are super easy to OC on pretty much any board. So, I purchased 6x Core 2 Duo E8500 processors to satiate my own curiosity, and get some form of binning done for the first time in my life.
Components:
- ASUS P5Q Turbo v1.02g (P45 Chipset),
with modded BIOS ("5th anniversary edition Final")
- 1x2GB OCZ Reaper DDR2, rated for 1066MHz
- XFX GTX 285 (stock)
- Super Flower Leadex III Bronze Pro 650W
- Sandisk X400 SSD (128GB)
- DeepCool Gammaxx 400 w/ 2x Arctic P12 in push-pull config locked at 100% PWM, MX-2 thermal paste
Constant settings:
- vCore: 1.550v*
- FSB Term. Voltage / CPU VTT: 1.40v
- NB Voltage: 1.40v
- SB Voltage: 1.10v
- CPU GTL Reference: 0.65x
- vDIMM: 2.20v
- PCIe SATA Voltage: 1.50v
- CPU PLL Voltage: 1.50v
- LLC: Enabled
- Spread Spectrum (CPU & PCIe): Disabled
- PCIe Frequency: 100
- FSB Strap to NB: 333MHz, DRAM frequency set to slowest available ratio
- Memory Timings: CAS 5-5-5-18
- CPU Margin Enhancement: Performance Mode
- DRAM Static Read Control: Disabled
- DRAM Read Training: Disabled
- Mem. OC Charger: Enabled
- AI Clock Twister: Strong
- C1E: Disabled
- Hardware Virtualization: Disabled
*was conditionally altered for 1 CPU, which exceeded thermal capacity.
All other OC settings, not relating to FSB frequency or CPU multiplier for OC, are auto.
Miscellaneous:
- Thermal paste application: Always approx. pea-sized dot in center of IHS. Cooler mounted with pins pushed in diagonal order. All MX-2 used is from the same 65G syringe, itself being less than 1 month old.
- Ambient temperature 22-23⁰C (~71-74⁰F), as measured by cheap IR thermal gun pointed at the table the test bench is on.
- OS was kept disconnected from the internet at all times.
Thermal paste application photos, these are from SLB9K_#4, which incidentally was the hottest CPU. This is only an example, I did not photograph every CPU's mount. For SLB9K_#4, at least a bad mount can be ruled out:
This is the order in which the CPU push-pins were depressed. This was consistent for every single CPU. Additionally, the CPU cooler was always mounted in the same direction (DeepCool logo same orientation as Yageo inductors in next picture).
Stability measurement:
- 30 min Prime95 blend torture test, HWiNFO as only other active program
- CPU-Z validation for reference
- Max Bootable Frequency is defined as getting into Windows 10 64-bit desktop and the mouse can be moved, amount of time pre-BSOD is inconsequential.
~~~
Ordered and tested by production date.
#1: E8500, SLB9K (E0)
- Serial: Q939A475
- Made: 2009, wk 39
- Max stable frequency: 4.370GHz
- Max bootable frequency: 4.560GHz (480 FSB)
- Multiplier/FSB: 9.5x / 460MHz
- End DRAM Frequency: 920MHz
-
CPU-Z Validation
#2: E8500, SLB9K (E0)
- Serial: Q945A839
- Made: 2009, wk 45
- Max stable frequency: 4.512GHz
- Max bootable frequency: 4.607GHz (485 FSB)
- Multiplier/FSB: 9.5x / 475MHz
- End DRAM Frequency: 950MHz
-
CPU-Z Validation
#3: E8500, SLB9K (E0)
- Serial: Q035B863
- Made: 2010, wk 35
- Max stable frequency: 4.560GHz
- Max bootable frequency: 4.750GHz (500FSB)
- Multiplier/FSB: 9.5x / 480MHz
- End DRAM Frequency: 960MHz
-
CPU-Z Validation
#4: E8500, SLB9K (E0)
- Serial: Q038C609
- Made: 2010, wk 38
- Max stable frequency: 4.560GHz
- Max bootable frequency: 4.655GHz (490FSB)
- Multiplier/FSB: 9.5x / 480MHz
- End DRAM Frequency: 960MHz
-
CPU-Z Validation
*Note: Oddly drew 10-20W more than previous CPUs @ identical settings + frequency (measured by OnSemi ASP0800, = EPU2). Also peaked about 5-7⁰C higher than all previous CPUs, again same settings. 80⁰C vs 72-75⁰C previous.
First chip that was thermally limited (none before were @ 80⁰C), so to test I lowered voltage to 1.500V. Consumed similar power, and was at similar temps as others @ 1.550V. Not an issue of thermal paste application. End OC achieved @ 1.550V. Max 91W, previous 70-80W.
#5: E8500, SLB9K (E0)
- Serial: Q040D896
- Made: 2010, wk 40
- Max stable frequency: 4.607GHz
- Max bootable frequency: 4.892GHz (!) (515FSB)
- Multiplier/FSB: 9.5x / 485MHz
- End DRAM Frequency: 970MHz
-
CPU-Z Validation
*Note: This was even worse than the last one on power, consuming 115W and instantly hitting 96⁰C on both cores @ 1.550V, 450FSB for initial sanity check. Dropping to 1.500V for final clocks.
Consumes as much @1.500V as E8500#4 does @ 1.550v (@ same frequency). Yet, it clocked the highest thus far for raw frequency booting. Seems to have good potential for high clocks with actual tuning, just uses stupidly more power for some reason. Final OC still hit peak 95⁰C, 115W @ OnSemi ASP0800 (ASUS EPU2).
#6: E8500, SLB9K (E0)
- Serial: Q041B237
- Made: 2010, wk 41
- Max stable frequency: 4.322GHz
- Max bootable frequency: 4.560GHz (480FSB, insta-crashed after 10s of desktop)
- Multiplier/FSB: 9.5x / 455MHz
- End DRAM Frequency: 910MHz
-
CPU-Z Validation
* Note: Finally, a return to thermal normalcy... didn't instantly hit 95⁰C. Unfortunately, this chip is still not very good. Should have been a mobile CPU
Something interesting that I didn't realize until after testing: apparently today (February 10th, 2022) is the
exact 10-year anniversary of the last date of Intel shipping out SLB9K processors out to OEMs. The longevity of this platform is insane; Ivy Bridge came out that same year... I was very surprised to learn how long they had truly been making these.
Comments on this testing: I know that 1.55v or even 1.50v vCore is quite high for these CPUs. My objective was not to measure how well the CPUs would perform once tweaked (lowest voltage for given frequency, adjusting VTT, etc.), but rather to see what differences exist between different samples when operating within otherwise identical conditions. This is only a sample size of 6, and I purchased them all from the same eBay seller, so I don't know the history of any of these chips either. Therefore I don't feel comfortable extrapolating this data to Yorkfield overall, nor even to SLB9K CPUs as a whole. Also, I didn't track temperatures formally throughout testing; I merely kept an eye on them to see if there were notable differences. Again, max frequency within identical parameters - CPU excepted - was the goal, not tracking power or thermal differences, so I can't really comment on the differences between my samples scientifically.
This is what happens when you complete buildzoid's pentagRAM... I'm surprised OCZ ever went out of business, their modules clearly bend time and space.
I may do this again in the future with 6x Xeon X5675's...
I keep most of my tech in a coat/utility closet. I have three desktops in the room, but still plenty of space.
