Should I run Async?
Ok I have a KX7 which I here is good for 180+ FSB and a 1800+ JIUHB that can do over 2.2 GHz. Thing is I only have Kingston Value PC2100 to run with. Its up to 158 atm and I dont know how far it can go.
I was thinking that I could get closer to limits of the mobo by running Async CPU:RAM timings. Extra mhz. I know its usually not good to do this but RAM is really limiting my OC.
Should I go Async? Thoughts of wisdom please.
I was thinking that I could get closer to limits of the mobo by running Async CPU:RAM timings. Extra mhz. I know its usually not good to do this but RAM is really limiting my OC.
Should I go Async? Thoughts of wisdom please.
0
Comments
You can go faster RAM than FSB, but not the other way around.
you can look up the pins to mod it for higher than that, if it's unlocked...
Look @ the benchmarks where a DDR266 FSB was used and DDR333 RAM was used. It's often slower than DDR266 for both.
Look at the benchmarks for the current p4s that are seriously overclocked FSB-wise but not overclocked (or at least not as overclocked) RAM-wise.
The benchmarks seem to support that having the FSB faster than the RAM is more beneficial than having the RAM faster than the FSB.
Of course, synchronous is best, but if the RAM is the limiting factor, then he'll have to up the multiplier, or run async.
Geeky1: I don't know about the 1800+ but I am almost positive the 1700+ (TBred) is factory unlocked.
How do you think some of these p4 boards are getting close to (and sometimes reaching) 300MHz FSB? No DDR module that I've ever seen or heard of will touch 300.
AMD boards DO NOT RUN FSB faster than the RAM.
NF7-S v2.0
FSB set to 200
FSB/DRAM ratio set to 4/5 = DDR320 RAM
FSB/DRAM ratio set to 5/4 = DDR500 RAM
Seems like async works both ways.
But, whatever I'm through posting about async.
Edit: Multi now at 12. Now at 1.9 GHz. FSB isnt topped either. Think I can make it to 12.5?
The core reason not to is that if you set async and take your CPU more than 5% above balanced rates you chance having data interchange dropouts both ways. Try with RAM at exisitn rates, take CPU in smallest steps possible with 12-24 hour folding between steps to test data flow reliablity or run a continuous RAM aggressive multipattern tester-- Micro2000 used to be the name of a company who was so well known for good RAM testing many validated RAM versus CPU speed balance with it as well as machine validation tests but it loads continuous passes of variable data RAM load and verifies at bit level and does not exercise for balance as well as Folding does. This will take more patience but might get you 3% better overall system if PCI is not hardwired to RAM or CPU very tightly. Guess, low risk if proceed slow, low real return also.
The bigger the on-die L2 and L1 the better, if L3 also there go for it. CPU is more likely to have tis buffers full with RAM skewed slower, and a tied to RAM PCI will invalidate idea more than 1-2% gain.
You are on edge. Your choice. If you can async and fix PCI close to normal go for it-- 3-5% bus variance from normal on any one bus is within safety for all boards by standard minimum or they shuold be RMA'd immediately. More than 7% for any bus is likely to need that busses components component heatsinked and also will edge the cards if any or devices hooked in, if any, to allowances or over them.
10% bus over on any bus is going to need very aggressive cooling and an ariflow effective case design for what is in it. PCI is more sensitive than the RAM you have with Tt active cooling kits installed on at least every other dimm and passive Tt kits on alternate RAMS without fan. Fan is designed to cool two dimms heatsinked that way in alternate pairs, so if two DIMMs in then one active cooling kit and one heatsink for RAM only, and put overhung fan over heatsinked DIMM. You can stretch RAM that way 5%-15% extra depending on exactly what else is on board if PCI is not hooked and fixed directly to a RAM speed ratio.
If you want safety you can try pumping CPU 5% more async probably and see if RAM actually is overclockable well without burning RAM up at all in a good case. That is why OC is mostly art, you are trying for a balance above normal and very good airflow is needed plus probably south bridge and PCI card heatsinks and if IDE on seperate chip or SATA and\or NIC, then those get ehatsinked also if PCI is ratioed fixed to RAM-- in async this is common, sometimes video also is a PCI multiplier for clock tick .
If video is off north bridge you are more likely to have success with independent PCI from RAM with minimal heatsink extras. Typically South bridge provides PCI clock tick and uses multiples for faster busses and North bridge talks to CPU and frequently RAM as well as south bridge.
Forget the Async thing. It doesnt go in the direction I want to go. The settings up the RAM and not the FSB like I want it/able to use it.
And I was thinking about a FSB:RAM ratio of 5:4 or so I could run 166 FSB and 133 RAM and then OC a lil bit from there. But the Ratios are actually 4:5 and not 5:4.
It´s impossible to run the fsb faster than the ram on the KX7. Intel and Nforce2 boards has that possibility if you want.
Mmonnin, use multiplier instead, but use as high fsb as you can.