Run Memory at CAS 2 @183 or CAS 2.5 @220?

[marks70]

I have my X2 3800+ running at 2.2Ghz with my OCZ Platinum Edition PC-3200 2x1GB running on the 333 divider and at 2-3-2-5 1T. My memory is thus running at 183Mhz at CAS 2. Would it be better to change the memory to run 1:1 at 220Mhz but then have to change the CAS to 2.5?

Thanks for your input!

 

[Budman]

the high the raw speed the better, you might even be able to run 250mhz if you losen to 3-4-4-8 and add a bit more voltage .

 

[Bobthelost]

That's a rather small OC for the CPU there...

 

[marks70]

Originally posted by: Bobthelost
That's a rather small OC for the CPU there...

I prefer leaving voltages at stock and minimizing heat than getting a couple more FPS in gaming.

 

[Hacp]

can your memory run at 2-3-3-7 at 209-210?

 

[imported_Seer]

Why don't you just bench your cpu at the various settings? If you want the speed for games, then run a few gaming benchmarks. IF you're doing encoding, time so encodes.

 

[Bobthelost]

Originally posted by: marks70

Originally posted by: Bobthelost
That's a rather small OC for the CPU there...

I prefer leaving voltages at stock and minimizing heat than getting a couple more FPS in gaming.

Wierdo

If you really want to know which is better then benchmarking software might help, bench @ the 1:1 and then @ CAS 2.

I'd go with 1:1 myself though.

 

[marks70]

Originally posted by: Hacp
can your memory run at 2-3-3-7 at 209-210?

Are you saying to try running it like at 250FSB on a divider?

 

[Megatomic]

Originally posted by: marks70

Originally posted by: Bobthelost
That's a rather small OC for the CPU there...

I prefer leaving voltages at stock and minimizing heat than getting a couple more FPS in gaming.

Same here. I'm a stock Vcore kind of overclocker.

 

[Hacp]

Originally posted by: marks70

Originally posted by: Hacp
can your memory run at 2-3-3-7 at 209-210?

Are you saying to try running it like at 250FSB on a divider?

Try the 2:3 divider. 266FSB.

Reduce multiplier to 7

 

[marks70]

Originally posted by: Hacp

Originally posted by: marks70

Originally posted by: Hacp
can your memory run at 2-3-3-7 at 209-210?

Are you saying to try running it like at 250FSB on a divider?

Try the 2:3 divider. 266FSB.

Reduce multiplier to 7

Sorry, I'm a little confused by your recommendation...I'm sure I'm just reading it wrong. You want me to try the 166 (or 333) divider, run the FSB at 266, and change the multiplier to 7? That means my CPU would only be running at 1.86Ghz and my RAM would be running at 178Mhz.

 

[TrevorRC]

220, no doubt. Though that CPU SHOULD be able to get higher on stock votlages.

 

[5t3v0]

Only way to find out is to invest some time to benchmark. Here's some comparisons I did with 3DMark05 on my previous 1Gb kit. For game performance, it doesnt matter a toss. For CPU performance, higher frequency with slower timings out performs lower frequency with faster timings.

MHz...Timings...CPC...3DMarks....CPUmarks
180....2-3-3-6.....1T.......6453.........4869
229....3-4-4-8.....1T.......6453.........4975

 

[Hacp]

Originally posted by: marks70

Originally posted by: Hacp

Originally posted by: marks70

Originally posted by: Hacp
can your memory run at 2-3-3-7 at 209-210?

Are you saying to try running it like at 250FSB on a divider?

Try the 2:3 divider. 266FSB.

Reduce multiplier to 7

Sorry, I'm a little confused by your recommendation...I'm sure I'm just reading it wrong. You want me to try the 166 (or 333) divider, run the FSB at 266, and change the multiplier to 7? That means my CPU would only be running at 1.86Ghz and my RAM would be running at 178Mhz.

Run the 2:3 divider. It should also be called 266 IIRC.

Then, set the ram to run at 210 2-3-3-7 and lower your multiplier to 7. That should give you around 2190.

 

[5t3v0]

You're not the only one confused, Marks70. Hacp, how on earth do you get 2190 from a x7 multiplier? And more to the point, how do you get RAM at 210MHz on a 2:3 divider with that cpu frequency? The sums just dont add up. 313x7 = 2191 but with a 2:3 divider, RAM frequency = 199. To get RAM at 210MHz on a 2:3 divider with a x7 cpu multiplier, the HTT has to be 330MHz, but 330x7 = 2310, not 2190. I'm afraid you're talking pish there big man (said in Glaswegian accent).

 

[Hacp]

313 Times 2= 626
626 divided by 3= 208.6666666666

There you have it.

If you want me to explain it even more,

2 divided by 3 is .66666666666 THIS IS THE RATIO THAT YOU WANT TO ACHIEVE.

210 divided by 330= .636363636363 THIS IS YOUR RATIO

209/210 divided by 313= .667/.671 THIS IS THE RATIO THAT I SUGGESTED

Which one of the numbers is closer to .666666? Yours or mine?

 

[5t3v0]

Hcap, it is not a simple as that. The divider changes slightly as you change cpu multipier. The memory speed is calculated in this case by cpu speed/11 not HTT x 2/3, hence 2191 / 11 = 199. There are tables available on the web demonstrating this if you dont believe me or the simplest way to prove this is to download A64MemFreq1.1.

Here's a table on i4memory. CPU frequency assumes default 200MHz. Look at the RAM divider for cpu multiplier 7 and DRAM freq 133. It is 11.

If its any consolation, my BIOS reports the RAM speed incorrectly at 233 (280x5/6), cpu-z reports it correctly at 229 (2520/11).

 

[lopri]

I would first go with the 20MHz rule for CAS and TRCD. If the max OC is higher than 20MHz using one step looser timing, I'd go with the clock. But if the max OC is not even 20MHz when loosened the timings, it's not really worth it. Of course one should bench himself/herself with his own setup with main apps, but it's been generally the case for me.

 

[Tweakin]

Originally posted by: marks70
I have my X2 3800+ running at 2.2Ghz with my OCZ Platinum Edition PC-3200 2x1GB running on the 333 divider and at 2-3-2-5 1T. My memory is thus running at 183Mhz at CAS 2. Would it be better to change the memory to run 1:1 at 220Mhz but then have to change the CAS to 2.5?

Thanks for your input!

To answer your question is a simple fashion, I would suggest that you go back to a 1:1 setting meaning that you run your memory at the full speed of the bus, set the multi to 10, fsb to 220 and memory timings to 3-3-3-8 1T with 2.7vdimm and see how that goes. That's what I would do just for the fact the AMD's memory controllers like higher lat's with dual 1GB sticks.

 

[edmundoab]

overall bandwidth rocks, cas latency is minor

 

[5t3v0]

Not wanting to be out done by patronising and blatantly incorrect formulae, here's how you really calculate memory frequency:

Memory Frequency = (HTT x CPU mutilplier) / CEIL(CPU multiplier x (HTT : DRAM ratio))

= (313 x 7) / CEIL(7 X (3/2)

= 2191 / CEIL(10.5)

= 2191 / 11

= 199.1818

For more info, refer to this great guide on DFI Street

Edit: BTW, CEIL = ceiling, which means round up to the nearest whole number

 

[Tweakin]

Originally posted by: 5t3v0
Not wanting to be out done by patronising and blatantly incorrect formulae, here's how you really calculate memory frequency:

Memory Frequency = (HTT x CPU mutilplier) / CEIL(CPU multiplier x (HTT : DRAM ratio))

= (313 x 7) / CEIL(7 X (3/2)

= 2191 / CEIL(10.5)

= 2191 / 11

= 199.1818

For more info, refer to this great guide on DFI Street

Edit: BTW, CEIL = ceiling, which means round up to the nearest whole number

That would require thinking

 

[robertk2012]

Memory example.

This is why i dont recommend high priced memory. Look how small the gains are. The last example is an increased clock speed. That is why it appears to be so good. you have to watch for that in reviews.

 

[marks70]

Thanks 5t3v0 and everyone else for their explanations. Just out of curiosity, some of you are recommending and are actually running your memory at CAS 3 with timings such as 3-3-3-8. If my memory will run 1:1 at 220Mhz with CAS 2.5, is there any reason why I wouldn't want to run it at something like 2.5-3-2-5?

 

[5t3v0]

You need to change 1 setting at a time and test it with memtest86 & Prime95. Its a lengthy process mind. My ram is rated at 3-4-4-8 at 250MHz, but I have it at 3-3-3-6 at 229MHz. There's a whole bunch more settings than those 4 that I tightened up as well. If you're looking for a good site explaining all the timings & recommended settings, there's this one on DFI Street which I used. I'm a bit paranoid messing around with memory timings in case I corrupt my data so each time I make a change, I boot into memtest from a floppy (or cdrom) & run a complete pass. If there's no errors, I boot windows and run an overnight test of Prime95 blend test. I found that Prime95 was much less forgiving and would find a failure where memtest wouldn't if run for the same period of time.