AMD64 Overclocking Guide (21/09/05) - discussion thread
This is a discussion on AMD64 Overclocking Guide (21/09/05) - discussion thread within the AMD motherboards / CPU forums, part of the Motherboards / CPU category; You'll find it very hard to screw anything up so long as you are careful of voltages. Backup your windows ...
| ||LinkBack||Thread Tools|
|15-08-2005, 05:43 PM||#82 (permalink)|
Join Date: Feb 25 2005
Hmm very nice rig you have there Dr.Clean, the only item I would not recommend is the Tt cooler though I am biased
It would be very hard to screw things up too badly if you have read and printed the relevent guides and with a specific guide for your mobo
To see how others are faring with the X2 I would recommend keeping up to date with the X2 thread over at Extreme Systems and please keep us all here updated with your progress
|16-08-2005, 12:32 AM||#83 (permalink)|
Join Date: Aug 15 2005
Well lets just say i was at uni doing Computing and i now hate building pc's =(
Im getting a local hardware guy to build pc for £35 and im trying to convince him to OC it too
I cant wait to play on this monster of a pc
|23-09-2005, 05:43 PM||#88 (permalink)|
Join Date: May 17 2005
alternative (more correct?) ram frequency chart to be made?
Hi eva2000 & others !
Here are some my opinions about all those charts presented with memory frequencies revealed with different memory dividers on "K8" platform /w_Cpu integrated memory controller/ & Cpu-Z & Bios showing them accordingly (& even about AMD datasheet stating similarly):
I believe that really nobody around the Net knows exactly, how "K8" internal Cpu PLL_circuitry logic for memory clock setting with different dividers (except when Fsb/ram=1:1 ; in_sync) works in reality & AMD keeps, I believe, that info reserved (by their own reasons) with just a datasheet too much simplified info. Yes, I was also reading stuff on the Net about guys who have done some Oscilloscope measurement & famous DFI MoBo designer released some memory frequency tables time ago, but I still have some serious doubts even if some "Net authorities" support those previous ones ... My attention was attracted to the problem after doing some benches with WinRAR first, which throw out results always higher than expected according to "known" mem.clock; later I switched to Sandra bench, but some other mainly mem.clock dependent bench could be also used instead ...
For other & older platforms was/is easy to determine the real memory clock (known logic inside clock generator chips on MoBos piloted by Bios), but it is not so with newest AMD platform & since the base for calculations was "simplified AMD info" & thus IMHO Cpu-Z & Bios-es [& those tables] are NOT showing/calculating properly true memory clock!
I could be wrong, but how the hell NOBODY TRIED (or got the idea) TO DO some comparative BENCHMARKING to verify those numbers ??? !!! Well, I did some of them & got also some partial results from "Net-friends", so I will try to explain how can anybody determine his memory clocks ratios for his own system & than calculate from results true memory clocks for own "k8" (A64 etc.) platform!
How to do such a comparative benchmarking & get some results:
- it is nice to do benchmarking in SAFE mode to avoid any excessive background programs/services impacting on results & to use a well known SiSoft Sandra Memory Bench, because its results (unbuffered!) in those upper first 2 bar graphs (RAM Bandwith Int_ALU & Float_FPU) are almost practically Cpu/Fsb independent & mainly proportional to only memory_Controller-Memory theoretical bandwith performance of it; the program is well known & free /Standard/ & almost every serious OC-er has it installed
- set in Bios CPU & Fsb frequency (200MHz) and memory latencies on defaults (also Htt & Cpu multi) & also be sure to set (if you do not have it already) Fsb/ram ratio in_sync 1:1 (200MHz); you need DDR400-Pc3200 sticks!. Do Benchmarking : sum together both results in MB/s & wrote down on paper your summed result you´ve got it for future DEFAULT REFERENCE [REF] on non-OC-ed machine!
- benchmark further with SAME! default CPU/FSB,Htt/mem.Latencies etc.(do NOT change! that) in Bios all other Fsb/MEM ratios & wrote down Sandra sum_results for each different setting/ram_Ratio/Ram_divider in Bios. (If you use 2ch mem.access also do not change that during all benches!)
- calculate from sum_results you´ve got according to 200MHz [REF] sum_result, memory ratios (in % or indexed) & revert/recalculate memory_ratios to show ram_clock instead in MHz. (ex.: if REF=10.000 & first lower mem.frequency result [9:10=180MHz /DFI_NF4 bios/ & for 2,2GHz (multi-11x) A64_3500+, Cpu-Z shows 170MHz -div.Cpu/13] is 9.400, means that real ram clock frequency is 9400/10000x200=188MHz since with equal ram latencies Sandra result is, as I believe, proportional to ram clock! So, Ram clocks revealed from other sources as stated are NOT correct, except when Ram runs 1:1 to Fsb (in_sync to it) & than are properly reported with different progs!
This way anybody can reveal himself mem.clock for own system IMHO better than blindly believe to other previously widely used numbers, formulas & discussions & this way could be explained also some OC-ing problems & problems with .5x Cpu multi´s etc. regarding calculating true Mem.clock.
Also if OC-ing CPU by Fsb raising it, naturally if not changing set ram_divider, Mem.clock also raises, but ratios stay the SAME, so mem.frequency is higher for the same amount (%) as Fsb raised! I also verified this my idea with real_life_mem.bench using built_in WinRAR bench with very similar ratio results for memory like with Sandra, but this last is for finding mem.clocks more suitable /measuring Ram bandwith/.
So, with some "common sense" & an afternoon benching & doing some math on single system (easy, but a little time consuming job - changing bios settings, booting to safe mode, rebooting to normal mode to see what Cpu-Z says etc.) & making some additional your_own table with calculated mem.clocks with different mem.dividers /& inserting also Bios ram clocks, Cpu-Z ones & from those mem.clock charts/, you can easily start developing certain doubts when comparing results, especially about Cpu-Z correctness of real Mem.clock revealing on newest AMD platform!
I could be wrong... but ... also previous widely used methods of knowing out true mem.clock could be wrong ... make your own opinion & before to much commenting my idea, do your own benching & a table of clocks & maybe post just final differences/results between different methods [like: Bios says 180, Cpu-Z says 170, chart says 183; I calculated-Sandra bench 188MHz] for discussion. So its your turn to do some "homework"; I did mine ... ... Summer holidays are over, engage your grey brain cells ...
PS.: sorry for too long post, but I was thinking to do it for 2 months ..
Last edited by Spajky; 23-09-2005 at 07:09 PM.
|amd64, discussion, guide, overclocking, thread|
|Currently Active Users Viewing This Thread: 1 (0 members and 1 guests)|
|Similar Threads for: AMD64 Overclocking Guide (21/09/05) - discussion thread|
|Thread||Thread Starter||Forum||Replies||Last Post|
|AMD64 Overclocking Guide (07/11/06)||eva2000||General AMD Motherboards / CPU||7||02-11-2006 05:42 PM|
|Opteron Socket939 Overclocking Discussion Thread||dinos22||DFI AMD motherboards / CPU||425||26-04-2006 07:19 AM|