Advice re. memory speed and timings

[lahiruwan]

I have a C2D 2.2 GHz running at stock speed (FSB 200MHz) on an ASUS P5GC-MX (945GC) board with 2 X 1 GB DDR2 667MHz Kingston ValueRAM.

CPU-Z shows that the memory runs at 333MH FSBRAM to be 3:5 and the timings to be CL - 5.0, tRCD -5, tRP - 5, tRAS - 15, tRC - 21. The SPD shows that it can run at 266MHz and 200MHz too. I gues if I set the memory to run at 200MHz the FSBRAM will be 1:1.

Which will get me better performace? Is it better to leave the memory running at 333MHz or to set it to run at 200MHz so that the FSBRAM runs at 1:1?

If so should I manually set the timings to those indicated in the SPD. The timings for 200MHz are 3.0,3,3,9,12. Or should I let the system decide the timings?


Thanks

 

[nzxtlexa]

Hi lahiruwan,

If I was you I would leave the RAM at its standard 333mhz (DDR2 667mhz). Otherwise technically you would be underclocking you RAM from 333mhz down to 200mhz, and I personally would rather have faster mhz RAM than lower timings in this situation.

Hope this helps

 

[david__t]

Ahhhh the age old RAM ratio question. Personally I think this needs a decent article and some serious testing from THG but here's my 2 pence worth...

Previously when we were at the original DDR levels like PC2700 & PC3200, a 1:1 ratio was always better because if you changed to 4:5 for example, then the speed differential and resultant bottlenecks cancelled out any gains - especially when we were at CAS 2.0 (with decent memory). Also for a while, the RAM available matched the bus speed as far as technology was concerned, so you couldn't really get RAM that was much faster than the rated bus speed. Plus as we all know, raising the clock speed usually meant relaxed timings which was a trade off.

Now we have entered a new era where you can buy memory that runs at double the bus speed. Obviously when it gets this fast, the gains far outweigh the loss of tight timings and bottlenecks. This is why the new 45nm chips at 1600FSB are so important because you get the faster bus overall and the 1:1 ratio - ie the best of both worlds. These chips will be running 800MHz memory in sync and with a 1600MHz bus, we might return closer to the previous era where the 4:5 ratios are slower than 1:1.

Personally I configure all of my customer's systems to run at 1:1 - and who can argue when you can get 2GB of Corsair XMS memory for £37?! (before I get flamed, I understand that overclocking runs by a different set of rules)

 

[lahiruwan]

Ahhhh the age old RAM ratio question. Personally I think this needs a decent article and some serious testing from THG but here's my 2 pence worth...

Previously when we were at the original DDR levels like PC2700 & PC3200, a 1:1 ratio was always better because if you changed to 4:5 for example, then the speed differential and resultant bottlenecks cancelled out any gains - especially when we were at CAS 2.0 (with decent memory). Also for a while, the RAM available matched the bus speed as far as technology was concerned, so you couldn't really get RAM that was much faster than the rated bus speed. Plus as we all know, raising the clock speed usually meant relaxed timings which was a trade off.

Now we have entered a new era where you can buy memory that runs at double the bus speed. Obviously when it gets this fast, the gains far outweigh the loss of tight timings and bottlenecks. This is why the new 45nm chips at 1600FSB are so important because you get the faster bus overall and the 1:1 ratio - ie the best of both worlds. These chips will be running 800MHz memory in sync and with a 1600MHz bus, we might return closer to the previous era where the 4:5 ratios are slower than 1:1.

Personally I configure all of my customer's systems to run at 1:1 - and who can argue when you can get 2GB of Corsair XMS memory for £37?! (before I get flamed, I understand that overclocking runs by a different set of rules)

Thank you for the reply. I'm afraid I didn't get an answer to the question. I want to know if DOWNGRADING the speed of the memory on the bios (setting the DDRII-667 to run at DDRII-400) so the FSB matches and the ration becomes 1:1 will INCREASE the performance. What would you personally do?

Also, I see the words relaxed and tight often with regards to memory timings? I'm a bit confused! of the settings 5,5,5,15,21 and 3,3,3,9,12, what is the more RELAXED setting? What does 'relaxed' transalate to in terms of performance?

P.S: Thanks nz nzxtlexa!

 

[seabreeze]

Will reducing the ratio down to 1:1 (while not changing the FSB) will increase in performance?

Perhaps you'll gain 0.4% or loose 1.3% (random numbers), only some memory tests will confirm, but I don't think it will be significant. As long as your CPU-Multiplier-FSB specs stay the same, changing the ratios on DDRII-667 so it runs at 533 or 400 won't do anything you'll notice, either way.

The greater the number, the more relaxed the timings are, but only when comparing to the same speed. You can't compare timings alone of two different speeds and assume the smaller ones are better.

The 5,5,5,15,21 and 3,3,3,9,12 timings are memory clock multiples. You have the 667 bus speed operating at 1.67x more clocks over any given time period than the 400, however, the timings increase of 667 compared to 400 are (oddly enough) longer by 1.67x (5/3), 1.67x(5/3), 1.67x(5/3), 1.67x(15/9), 1.75x(21/15).

So there's really no difference there, assuming all other system specs stay the same, just the capacity of a RAM modules to be used at different bus speeds. Unless you manually tighten up the 667 timings one at a time to find something stable, but quicker, run either SPDs. Leave it on 1:1 or auto. For your system, unless you test both to find out which is better, it's not worth loosing sleep over.

 

[lahiruwan]

Thanks. Just the answer I was hoping for. Now I can forget about this and get on with it......