-2/2/2 is low latency.
-Memory's SPD is JEDEC-compliant.
-High FSB on P4s is favored to low latencies.
-Get dual channel.
-PC4400 is sweet.
-CMX is Corsair's value line. (<b>Ed Note:</b> It sucks.)
-Modules that didn't pass XMS speed-binning might get bumped to lesser sticks.
-Those sticks might be capable of far more than what they're rated for.
-Any manufacturer is capable of making good, bad, and ugly modules.
12 x 200mhz = 2.4ghz. A 2.6 CPU has a multiplier of 13, 2.8 has 14 and so on.
0
Straight_ManGeeky, in my own wayNaples, FLIcrontian
edited December 2003
Even more simple Thrax, the ugly are the ones that did not pass even the bad test, most often. The good are all good modules usually, not perfectly always. You pay for QC(Quality Control Testing), more QC more payee. All the modules can come from same silicon wafer build.
Not only is FSB emphasized, but Latency is tuned for auto sync. Adjust only FSB for RAM and CPU, LET rest be auto, set by SPD, and you can go async more as far as RAM bus speed (RAM subbus portion of bus structure) on a P4 board with a good chipset than you can on most AMD boards.
Some Intel based OC hints: Easisest start point is to find a stable FSB OC, then tweak RAM bus speed and set BIOS for spread spectrum as much as will be stable. THEN, see if can sync RAM base speed more with same FSB, work for a stable running box. THEN write down settings and experiment, so you can reset things after wiping CMOS or in the case of Intel baords pulling jumper off and running it once into setup in RECOVERY mode is needed and configure mode does not work. Intel uses a tri-condition-state option, 3-PIN CMOS jumper, typically pin one to two jumpered is run\normal, pin 2-3 jumpered is Configure, and no jumper causes a complete CMOS table reload to defaults on poweron. Do not bother trying to configure after a RECOVERY in same session, simply shut down box completely, put jumper in CONFIG state, and start up and configure, when sure you have it right then shut down completely, and jumper to normal. FIRST thing I do is shut off the dang Intel splash screen, it blocks you seeing 80-95% of the POST messages.
John.
John-- who got a GOOD CMX Corsair stick in the Barton buy.
And to end my day, while taking off the heatspreaders I must have scratched the connector part of the ram and now my 256mb of pc3700 is dead and since warranty is voided if you remove the heatspreader I can't RMA it. I can just hear the Kingston reps saying "That'll teach him". Oh well, thank god it wasn't anything good... and my p4 system has been used very lightly lately. Only F@H no one has touched it or run any apps on it so really it shouldn't affect me. Just gives me all the more reason to get new memory for my P4 2.6C.
Time to make up an outline for my English final tommorow.
-2/2/2 is low latency.
-Memory's SPD is JEDEC-compliant.
-High FSB on P4s is favored to low latencies.
-Get dual channel.
-PC4400 is sweet.
-CMX is Corsair's value line. (<b>Ed Note:</b> It sucks.)
-Modules that didn't pass XMS speed-binning might get bumped to lesser sticks.
-Those sticks might be capable of far more than what they're rated for.
-Any manufacturer is capable of making good, bad, and ugly modules.
Unfortunately, nobody has RAM that will run CAS2.0 1:1 at high FSB 230-235+ without voltage mods to motherboards. All the stuff that will run 250+ (which a 2.4 + 2.6 will/should easily be capable of) are CAS 2.5 or most often 3.0.
So the solution is generally to run 5:4 ratio so that 250 fsb = 200 memory speed, and so on. The benefit to this is that THEN you can run 2-2-2-5 (as I do) and absolutely yes, the P4 loves low latency memory timings. Plus, you can buy a little cheaper memory, PC3500 is lots and PC3200 usually sufficient if it will overclock *a little*. Sure, PC4400 is sweet if you got the bux but it's gonna be CAS3 at those speeds, so the actual gain is relatively small.
As for NON-XMS sticks being capable of more, I wouldn't count on it. CMX is (as I understand it) entirely different memory modules - it's cheaper for a reason.
If I understand what John is saying about sync/async correctly, I disagree. My P4 tweaking guide/P4C800-E Deluxe review is in MediaMan's hands and should be published shortly, all is explained therein.
Yes. I know what timings can/cannot be used at what bus speeds.
Generally 2/2/2 tops out at 220MHz on even the nicest of DDR modules in production. Most all of the 250+ is 3/4/4 in timings, which for a Pentium 4 is not that bad.. If it were an Athlon, you'd be murdering system peformance.
Pentium4s also don't mind asynchronous busses, as it's quad-pumped architecture provides better support for it than the Athlon's crappy-narrow bus.
PC3200 (Good Corsair) will do around 225MHz at 2/2/2, and PC3500 will do around 230/235 at 2.5/3/3.
The actual gain for the Pentium4, even at 3/4/4, is significant at 250 FSB.
I was just summarizing John's post, I wasn't pointing out if it was right or wrong; points 7 and 8 are.. VERY wrong.
0
Straight_ManGeeky, in my own wayNaples, FLIcrontian
edited December 2003
Yes, this is for others too. Point was, not all "bad" labelled series need be bad, there are such things as overproduction of modules and sometimes you do get a better module set than you expect. But with cheaper it is the luck of the draw or pruchase, and rarer that it happens.
So, was stressing likelyhood, not theroy certainty, and you got the summary perfect, Thrax.
I was expanding to show scope and why and some how details for folks less um, suddenly grasping than you of concept sets, not lecturing, get it??? AND showing a pattern that is not based on theory, but based on market reality that has held for decades. Besides, outlines are nice and breif, but mask the why and how of underlying casues for patterns, which 90% of time are NOT pure standards based, in fact violation of standards is more common, in some ways, than perfect adherence. Cheaper is more likely to be more majorly violating of IEEE standards than more exspensive, because of the costs of intense QC. That was the underlying theme, the facts were exemplary (as in being illustative examples) and on-topic, but deeply knowing(groking for shorthand) the market involves knowing who has good QC as standard in production also.
---> THRAX: EXCUSE ME, but to my mind discussion should be beneficial to all thread readers and some enthusiasts do not know this stuff deep enough that checking it before buying is second nature, and it should be if you each and all want to OC and still do what YOU each want most to do productively. So if I seem to say things you know, confirm it, but touchy is out as this was not an attack it was a fillin of logic behind and reasons why. That is true for most of my posts.
EDIT: Reasons why include market reality, and market reality has held as a very real dragging force for the history of IT.
For Thrax in particular but not solely, you might look at the fairly new IEEE Computer Society, it is a relatively inexpensive way to get a foot in the door there. They have online classes on basic and more advanced topics, the advanced ones are basicly would-be hardware engineer content level, some few up to master's level content as they want college grads and knowledgeable (VERY) advanced high school age folks for future members and judge invites in part by judging contributions(intellectual) to the Computer Society.
Comments
-2/2/2 is low latency.
-Memory's SPD is JEDEC-compliant.
-High FSB on P4s is favored to low latencies.
-Get dual channel.
-PC4400 is sweet.
-CMX is Corsair's value line. (<b>Ed Note:</b> It sucks.)
-Modules that didn't pass XMS speed-binning might get bumped to lesser sticks.
-Those sticks might be capable of far more than what they're rated for.
-Any manufacturer is capable of making good, bad, and ugly modules.
12 x 200mhz = 2.4ghz. A 2.6 CPU has a multiplier of 13, 2.8 has 14 and so on.
Not only is FSB emphasized, but Latency is tuned for auto sync. Adjust only FSB for RAM and CPU, LET rest be auto, set by SPD, and you can go async more as far as RAM bus speed (RAM subbus portion of bus structure) on a P4 board with a good chipset than you can on most AMD boards.
Some Intel based OC hints: Easisest start point is to find a stable FSB OC, then tweak RAM bus speed and set BIOS for spread spectrum as much as will be stable. THEN, see if can sync RAM base speed more with same FSB, work for a stable running box. THEN write down settings and experiment, so you can reset things after wiping CMOS or in the case of Intel baords pulling jumper off and running it once into setup in RECOVERY mode is needed and configure mode does not work. Intel uses a tri-condition-state option, 3-PIN CMOS jumper, typically pin one to two jumpered is run\normal, pin 2-3 jumpered is Configure, and no jumper causes a complete CMOS table reload to defaults on poweron. Do not bother trying to configure after a RECOVERY in same session, simply shut down box completely, put jumper in CONFIG state, and start up and configure, when sure you have it right then shut down completely, and jumper to normal. FIRST thing I do is shut off the dang Intel splash screen, it blocks you seeing 80-95% of the POST messages.
John.
John-- who got a GOOD CMX Corsair stick in the Barton buy.
Spread spectrum decreases system performance.
Time to make up an outline for my English final tommorow.
Annoyed,
Al
Unfortunately, nobody has RAM that will run CAS2.0 1:1 at high FSB 230-235+ without voltage mods to motherboards. All the stuff that will run 250+ (which a 2.4 + 2.6 will/should easily be capable of) are CAS 2.5 or most often 3.0.
So the solution is generally to run 5:4 ratio so that 250 fsb = 200 memory speed, and so on. The benefit to this is that THEN you can run 2-2-2-5 (as I do) and absolutely yes, the P4 loves low latency memory timings. Plus, you can buy a little cheaper memory, PC3500 is lots and PC3200 usually sufficient if it will overclock *a little*. Sure, PC4400 is sweet if you got the bux but it's gonna be CAS3 at those speeds, so the actual gain is relatively small.
As for NON-XMS sticks being capable of more, I wouldn't count on it. CMX is (as I understand it) entirely different memory modules - it's cheaper for a reason.
If I understand what John is saying about sync/async correctly, I disagree. My P4 tweaking guide/P4C800-E Deluxe review is in MediaMan's hands and should be published shortly, all is explained therein.
Generally 2/2/2 tops out at 220MHz on even the nicest of DDR modules in production. Most all of the 250+ is 3/4/4 in timings, which for a Pentium 4 is not that bad.. If it were an Athlon, you'd be murdering system peformance.
Pentium4s also don't mind asynchronous busses, as it's quad-pumped architecture provides better support for it than the Athlon's crappy-narrow bus.
PC3200 (Good Corsair) will do around 225MHz at 2/2/2, and PC3500 will do around 230/235 at 2.5/3/3.
The actual gain for the Pentium4, even at 3/4/4, is significant at 250 FSB.
I was just summarizing John's post, I wasn't pointing out if it was right or wrong; points 7 and 8 are.. VERY wrong.
So, was stressing likelyhood, not theroy certainty, and you got the summary perfect, Thrax.
I was expanding to show scope and why and some how details for folks less um, suddenly grasping than you of concept sets, not lecturing, get it??? AND showing a pattern that is not based on theory, but based on market reality that has held for decades. Besides, outlines are nice and breif, but mask the why and how of underlying casues for patterns, which 90% of time are NOT pure standards based, in fact violation of standards is more common, in some ways, than perfect adherence. Cheaper is more likely to be more majorly violating of IEEE standards than more exspensive, because of the costs of intense QC. That was the underlying theme, the facts were exemplary (as in being illustative examples) and on-topic, but deeply knowing(groking for shorthand) the market involves knowing who has good QC as standard in production also.
---> THRAX: EXCUSE ME, but to my mind discussion should be beneficial to all thread readers and some enthusiasts do not know this stuff deep enough that checking it before buying is second nature, and it should be if you each and all want to OC and still do what YOU each want most to do productively. So if I seem to say things you know, confirm it, but touchy is out as this was not an attack it was a fillin of logic behind and reasons why. That is true for most of my posts.
EDIT: Reasons why include market reality, and market reality has held as a very real dragging force for the history of IT.
For Thrax in particular but not solely, you might look at the fairly new IEEE Computer Society, it is a relatively inexpensive way to get a foot in the door there. They have online classes on basic and more advanced topics, the advanced ones are basicly would-be hardware engineer content level, some few up to master's level content as they want college grads and knowledgeable (VERY) advanced high school age folks for future members and judge invites in part by judging contributions(intellectual) to the Computer Society.
John.