Worth taking my Q6600 higher?

[McCartney]

Hi guys,

Currently I've pushed my q6600 to 2.7ghz at A 1200mhz fsb. I could push it further but I'm having my doubts... Why?
well I have a 2x2gb pc6400 kit running 800 Mhz at 4-4-4-12 and 1T. That's right, I'm running 1T Stable on DDR2!

Now is it worth taking my CPU to 3.4 and going to 2T? I have been prime stable at that setting but I really like the whole 1T getup.. and I'm not sure if it's worth clocking my ram down for faster CPU speeds..

Also I 3dmark06'd my PC (2 8800 GTXs with forceware 169.28, and nForce 680i with latest drivers, vista x64 and 4gb ram) and I only got 12.8k 3dmarks! What gives?

Thx

 

[fluffy01016]

Seems strange the 3dmark06 score you got. I'm running a Q6600 oc'd at 2.7ghz and a 8800GT oc by around 10%. I also have 3gb ddr2 667 in vista 32bit and I have acheived a score of 12,100.

 

[Idontcare]

Originally posted by: McCartney
Now is it worth taking my CPU to 3.4 and going to 2T? I have been prime stable at that setting but I really like the whole 1T getup.. and I'm not sure if it's worth clocking my ram down for faster CPU speeds..

Higher CPU frequency will always trump increased ram latencies.

Don't forget that while you will be relaxing the timings on your RAM, you are doing so in order to enable a higher FSB which in itself will in turn decrease your ram latencies.

So is a 1T setup at 300MHz FSB (or DDR2-600 speeds) all that superior to a 2T setup at 400MHz FSB (DDR2-800 speeds)?

I expect you will find the ram latencies are very much second-order trivialities compared to the overall system speedup you will experience in going from a 2.6Ghz CPU to a 3.6GHz CPU in this example. (your actual max overclock will vary)

Truthfully you should clock it and determine for yourself, either via synthetic benchmarking or by using your actual applications of interest.

 

[MetaDFF]

From my understand of how RAM works, since you are running on a 300 MHz FSB (1200 MHz quad pumped) with 1T timings, that means the CAS latency that your RAM can handle is 1 clock cycle at 300 MHz or 3.33 ns. That means if you want to maintain the same effective latency with a 2T timing, you need your FSB at 600 MHz (i.e. 2 x 1.67 ns = 3.33 ns). Of course, a Q6600 (or even any motherboard for that matter) will not overclock with/to a 600 MHz FSB.

If you loosen your ram timings to 2T at 400 MHz FSB (1600 MHz quad pumped) then you give your RAM 5 ns delay (when technically only 3.33 ns is needed for the RAM to fetch the data requested and place it at the outputs). This would represent a 50% increase in RAM latency. Of course, this is all just calculations.

But just as Idontcare said, you will definitely see a greater benefit from having a higher CPU clock speed than having insanely tight RAM latencies.

 

[Idontcare]

Use Everest to measure your ram latency insofar as applications see it.

You will find it varies from as low as ~45ns for the best evar to 60ns typical on tight timings to upwards of 80ns for very loose timings.

1T != 1 clock cycle to complete a transaction.

2T is 1 clock cycle adder to an already lengthy series of steps, each with their own clock adders.

So to complete a transaction, at 1T or 2T, you are talking about whether you want to have the transaction take 14 memory cycles (1T) or 15 memory cycles (2T).

All my numbers here are made up for example purposes to communicate the point. Specific numbers will vary by chipset and RAM.

 

[MetaDFF]

Originally posted by: Idontcare
Use Everest to measure your ram latency insofar as applications see it.

You will find it varies from as low as ~45ns for the best evar to 60ns typical on tight timings to upwards of 80ns for very loose timings.

1T != 1 clock cycle to complete a transaction.

2T is 1 clock cycle adder to an already lengthy series of steps, each with their own clock adders.

So to complete a transaction, at 1T or 2T, you are talking about whether you want to have the transaction take 14 memory cycles (1T) or 15 memory cycles (2T).

All my numbers here are made up for example purposes to communicate the point. Specific numbers will vary by chipset and RAM.

Sorry, my bad! You learn something new everyday
I guess I should have said, "That would represent a 50 % increase in CAS latency."

 

[McCartney]

So in brief I don't care, if my ram is running 800 mhz at 1T 4-4-4-12 and my cpu 300 mhz quad pumped, what's my best solution?
I mean yeah I agree CPU speeds trump latency, but I thought that 1T/2T made somewhat of a difference (dependong on the app) and increased ram performance significantly (Atleast it did in teh s939 days ).

So I should clock it back down to 2T at 4-4-4-12 and crank it up? I can hit 1333 fsb np at 2T with the same timings.

Thanks

P.S. Why such a low 3dmark?

 

[Thor86]

Originally posted by: McCartney
P.S. Why such a low 3dmark?

3Dmock is cpu based scoring. So higher cpu = higher scores.

DDR2-1000 at 5-5-5-15 2T would still be faster, as always, Intel processors are bandwidth starved, and tighter latencies are only good at lower speeds.

 

[Tempered81]

1t command rate vs. an extra 1000mhz should be a no-brainer