S_A_V Posted May 2, 2013 Posted May 2, 2013 (edited) I had some experience of PSC overclocking under subzero temps two years ago, but it was done the ghetto way - without any RAM pot, just using aluminum foil: Temps was not good at all, I only got around -30C...-35C (measured from thermocouple attached near RAM IC). Anyway, my Lynnfield was BCLK-limited, so I stuck to 2908 MHz no matter what cooling. I was surprised so much teams used LN2 cooling for RAM on AOOC 2012, so I decided it's time to move forward and ordered RAM pot as well as my new CPU pot: Now I only have my old Hyper kit (very good on any platform, but no go with Ivy Bridge) and one samsung kit for daily usage and some 3D. Both is not the best choice for subzero, I guess. That's why I took G.Skill 2000C6 PIS kit (F3-16000CL6D-4GBPIS based on X-Series PSC ICs) from my teammate to test with KPC Ney pro memory cooler. I quickly test this RAM kit on air with 3770K and it was nothing special - 2600 9-12-9 for 32M or so. Max clock test bed: - AMD FX-8120 - ASUS Crosshair V Formula - G.Skill F3-16000CL6D-4GBPIS - Palit GeForce 7300GT Sonic - Corsair Neutron GTX - Corsair AX1200 SuperPi 32M test bed: - Intel Core i7-3770K - ASUS Maximus V Extreme - G.Skill F3-16000CL6D-4GBPIS - ASUS MATRIX-HD7970-P-3GD5 - Corsair Neutron GTX - Corsair AX1200 Preparation: I remove the heatspreaders from ram kit using hair dryer and some old plastic card. Then cleaned ICs from glue with acetone and cover both sides with rubber eraser to protect from condensation and for better contact with pot plates and better heat transfer. ASUS motherboards was covered with rubber eraser: KPC Venom and Ney Pro was insulated with thin sticky dielectric tape and Armacell Armaflex self-adhering insulating tape: Arctic Cooling Ceramique thermal paste used between memory ICs, plates and RAM pot: Voltage / temp scaling: PSC scales quite well with voltage and temperature. This G.Skill 2000C6 PIS kit scales up to 1.80V-1.85V on air cooling. For subzero temps it runs fine at 2.10V-2.15V. No CB or CBB, so it was easy to bench without needs to control temperature. Temps - T1 for CPU and T2 for memory: Testing results: FX-8120 was tested before for CPU_NB overclocking and it is good for up to 3600 MHz. Bus speeds around 300 MHz was also not a problem. I expected some high memory frequency validation, but I got only 3100 for dual channel and 3130 MHz for single channel. 3130 MHz 11-14-14-35 2T 2.10V -170C Single Channel: http://valid.canardpc.com/show_oc.php?id=2788041 May be this CPU can't do high CPU_NB, bus and memory clock at the same time, IDK. Ok, what about 3+ GHz with tight timings? 3058 MHz 7-11-7-27 1T 2.10V -170C Dual Channel: http://valid.canardpc.com/show_oc.php?id=2789955 3010 MHz 6-10-6-27 1T 2.10V -170C Dual Channel: http://valid.canardpc.com/show_oc.php?id=2788195 Not sure what is current WRs with CL6 and CL7. I did quick search the web and found some old results was done with Kingston memory and Lynnfield CPU: http://ocaholic.ch/modules/smartsection/item.php?itemid=609&sel_lang=english The difference that it was done in single channel and high tRCD/tRP. Semi-stable clock for booting windows and run some benchmarks with 6-10-6-27 timings was around 2800 MHz: SuperPi 1M @ 2817 MHz 6-10-6-27 1T 2.10V -170C Dual Channel: Cinebench R11.5 @ 2804 MHz 6-10-6-27 1T 2.10V -170C Dual Channel: FX-8120 is too slow for SuperPi 32M and I changed testbed to 3770K for this bench. SuperPi 32M - 6.17.219 @ 2501 MHz 6-9-5-22 1T 2.15V -170C Dual Channel: http://HWBOT.org/submission/2377941 SuperPi 32M @ 2500 MHz 5-9-5-20 passed 15 loops and failed (no waza): Booting at 2600-2700 6-10-6-27 was not a problem, but not stable enough to pass 32M. Hope I will test this memory kit with Haswell soon, to see if it goes for higher clocks. Some frosty pics: Edited May 7, 2013 by S_A_V 1 Quote
Christian Ney Posted May 2, 2013 Posted May 2, 2013 That's some serious insulation work Nice scores too Quote
TaPaKaH Posted May 2, 2013 Posted May 2, 2013 BD must be very light on stability (due to low bandwidth, I guess) if you managed to screen 1500+ tRCD 10. Even with a good kit (1200+ tRCD 10 on air) that would require a fair bit of luck. Quote
l0ud_sil3nc3 Posted May 2, 2013 Posted May 2, 2013 This is what I like to see, excellent results SAV. Cas 5 is bugged unless you can get tWCL 5 to run, but the 2500 6-9-6 time is very nice. How are you getting the dimms that cold or is that pot temp you are measuring? Quote
S_A_V Posted May 2, 2013 Author Posted May 2, 2013 Cas 5 is bugged unless you can get tWCL 5 to run, but the 2500 6-9-6 time is very nice. Tried to set tWCL 5 in BIOS and got error code 55. Also noticed that 6-9-6 tWCL 6 was faster than 5-9-6 tWCL 6 (on first loops, both runs with waza). How are you getting the dimms that cold or is that pot temp you are measuring? Yes, it's pot temp. Next time will try to install second probe directly to memory module. 1 Quote
Splave Posted May 3, 2013 Posted May 3, 2013 intresting way of using the ney pro, how do you manage a tight fit on the sticks? I always feel as hard as I try they are still fairly loose surrounding my memory heat spreaders. nice results Quote
Hyperhorn Posted May 3, 2013 Posted May 3, 2013 Very clean insulation and interesting results! How much Vdimm for the 5-9-5-20 @ 32M try? Quote
S_A_V Posted May 3, 2013 Author Posted May 3, 2013 intresting way of using the ney pro, how do you manage a tight fit on the sticks? I always feel as hard as I try they are still fairly loose surrounding my memory heat spreaders. Not sure what you mean, Allen. May be this scheme and couple of pics will explain it better: Very clean insulation and interesting results! How much Vdimm for the 5-9-5-20 @ 32M try? 2.15V was for both 32M runs Quote
Splave Posted May 6, 2013 Posted May 6, 2013 Not sure what you mean, Allen. May be this scheme and couple of pics will explain it better: 2.15V was for both 32M runs ahh I think I see now, what I mean is with the ney pro I find it hard to tighten the plates around the memory. perhaps the eraser is helping you hold it to the sticks while you tighten Quote
chispy Posted May 7, 2013 Posted May 7, 2013 Excellent work S A V , insulation is top notch , awesome results and frosty pictures ,like ! Quote
S_A_V Posted June 2, 2013 Author Posted June 2, 2013 Tried G.Skill 2000C6 on Haswell with 1L of LN2 and half hour of time. 2400MHz limited by Intel board, subs not optimized, memory voltage is 2.00V. Looking for better board... Super Pi 32M - 6.00.062 Quote
Recommended Posts
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.