Why tighten memory timings?

Tried my first ever overclock yesterday (P5K-E WiFi, PC2-6400, E6750, Artic Cooling Freezer 7 Pro) and went from stock 2.66GHz to 3.2GHz by doing nothing apart from following first part of clunk's guides. I set memory timings to what CPU-Z shows in SPD (5-5-5-18) and then bumped up CPU frequency little by little to 400. Everything else, including DRAM frequency and voltage, was left on Auto.

I get the impression from the guide that I should look at tightening my memory timings. But it doesn't really explain why. How much performance is it going to give me? Or is it to aid further overclocking? Please keep replies at the 'noob' level!

What, if anything, can I do next to get more speed before I start OC'ing my GPU? My 3DMark06 was 9266, now it is 9690 (stock BFG 8800GTS OC 320M at stock). I'm willing to change my memory sticks and maybe my CPUcooling (it is fine so far) but not looking at changing anything else.

Cheers,
Dave

to be honest i have wondered the same thing. i have the same cpu and its brilliant, mines at 3.25GHz on stock cooling and its at 51C at load, so i could go higher but am happy with it. Have you OCd your RAM yet?? that will always speed things up. i also have an 8800GTS 320 and have found using rivatuner very easy to overclock it.

fresh

keep a 1:1 ratio with ram and fsb. Tightening memory timings will i believe,decrease data load time because its faster xD, same idea as say uping the mhz of it.

I have an 8800GTS from bfg (g92/512mb) i got it to 760/1070, couldnt get the shaders up as i couldnt find them xD, fresh any idea how you unlock it in rivatuner?

You probably won`t see a huge difference by tightening the timings, it`s more a case of extracting every little bit of performance that is possible.

You may be able to clock your memory higher than 1:1, technically you should get more performance.

Looser timings aid overclocking as tighter timings may cause the overclock to fail, get to the overclock that you want then try to reduce them and see if you think it helps.

My understanding is that tightening timings will give you marginal gains in performance but also lead to more system instabilty.

Thanks guys. Sounds like I've not really got much more to gain unless I want to become a super-tweaker, but haven't really got the time I'm afraid.

Quote:

---

Originally Posted by fresh

i also have an 8800GTS 320 and have found using rivatuner very easy to overclock it.
fresh

---

Fresh, what have you managed to get up to with your 8800GTS 320? According to RT monitoring, I'm at 576/900 which actually differs from the 580/908 I set in RT. Any value in adjusting shader clock seperately?

Cheers,
Dave

tighter timings will improve performance, but it's so small that it's not worth losing any sleep over. You'll probably also have to increase the voltage to make it stable, but that's only if you push things reasonably far.

But it's free, so you may as well try to get better timings and test for stability. If you do decide to increase the voltage to get better timings (along with higher frequency) direct some air flow on the sticks. Most people will put a fan blowing over them just to make sure they don't overheat.

Regarding the Rivatuner issue. Have you got the latest version and the latest drivers? I can see the shader clock (although I have left on default, which keeps it in sync with the core). I found this site useful: RivaTuner 2.0 Guide at .:: Derek’s Blog

Quote:

---

Originally Posted by Thorsson

Regarding the Rivatuner issue. Have you got the latest version and the latest drivers? I can see the shader clock (although I have left on default, which keeps it in sync with the core). I found this site useful: RivaTuner 2.0 Guide at .:: Derek’s Blog

---

Yes and yes. I can see the shader clock, just wondering how much is to be gained by adjusting it seperately. I used that guide to get me started with RT - it was very useful.

Cheers,
Dave

if you lower the timings, it'll take say 33ms to write the data and be ready for the next data cycle at 5 - 5 - 5 - 18, but at 4 - 4 - 4 - 12 it'll take 24ms which is quite an improvement, but it's usually easier to lower the MHz of your RAM and decrease the timings, 'cause the higher the MHz the harder it is to tighten the timings, and also.. as some people have said, keep your RAM and FSB at 1:1 with each other, so a 266MHz FSB you'd want 533MHz RAM.

Quote:

---

Originally Posted by demonicd

Yes and yes. I can see the shader clock, just wondering how much is to be gained by adjusting it seperately. I used that guide to get me started with RT - it was very useful.

Cheers,
Dave

---

I didn't see much difference, but then I always stay a way short of the maximum OC, which I'm fairly convinced shortens the life of the card.

Tightening timings doesn't make a huge difference for core 2 based systems as they employ a whole host of tricks to get around the fact they have to use a northbridge memory controller (caching, pre-emptive reading etc.)

On AMD64 based systems tighter memory timings have a more significant effect on performance.

Quote:

---

Originally Posted by demonicd

Thanks guys. Sounds like I've not really got much more to gain unless I want to become a super-tweaker, but haven't really got the time I'm afraid.


Fresh, what have you managed to get up to with your 8800GTS 320? According to RT monitoring, I'm at 576/900 which actually differs from the 580/908 I set in RT. Any value in adjusting shader clock seperately?

Cheers,
Dave

---

i have core at 685, shaders at 1500 and mem at 999. when i isntalled rivatuner i just opened it and there everything was, in terms of overclocking the shaders serperately; i found that it did give a performance increase when checking my 3dmark06 score. I did have the core at 700 but it wasnt stable and temps werent safe. I am only new to this and basically followed clunks overclocking guide and various other things, but this website Overclock.net - Overclocking.net will help you with everything. its american based so forgive them, but really useful.

fresh

Quote:

---

Originally Posted by kalniel

Tightening timings doesn't make a huge difference for core 2 based systems as they employ a whole host of tricks to get around the fact they have to use a northbridge memory controller (caching, pre-emptive reading etc.)

On AMD64 based systems tighter memory timings have a more significant effect on performance.

---

Yup, if you can tighten the timings, do so, as it does increase performance. But the gains in going from 5-5-5-15 to 4-4-4-12 at the same frequency won't be that great. I'm tempted to benchmark it later, so may get back to you and see if it is worth it. I run at 4-4-4-12 because I can and there is no reason not to...

There are far greater gains to be had from clocking your cpu/gpu, I've never found that tweaking RAM makes a great deal of difference. Reviews I've read in the past comparing exotic/expensive RAM to normal RAM show extremely marginal differences.....

this is all true, BUT tighter timings make faster maths :)

When a CPU chucks out a request for something to be done, the BANDWIDTH is vital (thats the bus speed) and so is the clock speed (as to how fast the cpu makes the requests)

BUT,......imagine ram timings as the speed it takes to confirm the job is done and send back a confirmation to the CPU. The quicker it repsonds the faster the next one gets done. Not important? Maybe not, if it was once per second, and the difference was a millionth of a second at either end...but you've got MILLIONS and MILLIONS per second.....get those timings wound down :)

I believe in tighter ram timings. If it'll do it, use it.

Quote:

---

Originally Posted by fresh

i have core at 685, shaders at 1500 and mem at 999. when i isntalled rivatuner i just opened it and there everything was, in terms of overclocking the shaders serperately; i found that it did give a performance increase when checking my 3dmark06 score. I did have the core at 700 but it wasnt stable and temps werent safe. I am only new to this and basically followed clunks overclocking guide and various other things, but this website Overclock.net - Overclocking.net will help you with everything. its american based so forgive them, but really useful.

fresh

---

I got RT to show me the shader clock, but have left it linked to core for now. Those are very high clocks compared to mine, especially the core. How are you cooling the VGA? What do you consider to be safe/dangerous temps and what drives up those temps most? Core or memory?

Quote:

---

Originally Posted by demonicd

I got RT to show me the shader clock, but have left it linked to core for now. Those are very high clocks compared to mine, especially the core. How are you cooling the VGA? What do you consider to be safe/dangerous temps and what drives up those temps most? Core or memory?

---

just the normal cooler and i have it set to 85%, my case is the antec 900 which has 3 80mm and 1 200mm tri cool fans. these i have set to mediuma dn the air flow is brilliant. at its hottest ive seen the gpu get up to 65C but my tower is under my desk and the heating was on. i have not checked which raises the temps more, i just kept increasing the speed until i got some glitches or whatever you call them in 3dmark then i backed it down.

fresh

When you raise the RAM freqs the timings in ms will lower.

Timings of 5 at say 400 MHz will be lower than timings of 4 at 200 MHz.

Aren't the timings (5-5-5-12 etc) measured in cycles? So increasing your bus speed and keeping ram 1:1 will mean tighter timings anyway. I've read an article somewhere that showed in benchmarks/games/general usage, tighter timings make very little difference.

*bump*

didn't want to start a new thread for something trivial (i know... it's not like me at all :laugh:)

but would i benefit from dropping the freq and tightening timings as ATM i'm on 5 4 4 12 @ 400mhz won't boot if i drop down further on CL @ 400mhz

but i'm wondering whether to drop to 200mhz and go for 3 3 3 9 or tighter. what do you reckon guys??

This thread is pure internet...

Fanastically detailed knowledgeable quality information but with no definate answer

In an Intel based system, timings make very little difference. In my experience it has always been better to go for raw CPU and memory bandwidth speeds over an exchange with tighter timings.

AMD based systems with the memory controller on the chip will be hurt a bit more by lax timings. As for performance? That depends on what you do with your system.
Also its dependent on how much you have to lower the timings by. If you do +1 across the main ranges but gain an extra 300mhz core speed from it, its almost certain that its a better solution every time. The very nature of overclocking means that you will need to find that threshold between lowering them and CPU core/bus gain and benchmark the two in the programs you use though.

Thats as close as your going to get to "definite" because by and large, it depends on the system and its use in question.

Everyone is always saying keepiong timings to 1:1 on the CPU RAM timings but once the CPU is overclocked is it fine to negate 1:1 to get a higher RAM speed to make full use of what you've paid for and possibly get more or will it make things worse, somehow :P??????

I personally have never found the 1:1 ratio to be crucial, it is a widely disseminated view, but as long as you are stable, then run some benchmarking software at the different ratios your RAM can handle and look at the numbers. :)

Quote:

---

Originally Posted by ikix

Everyone is always saying keepiong timings to 1:1 on the CPU RAM timings but once the CPU is overclocked is it fine to negate 1:1 to get a higher RAM speed to make full use of what you've paid for and possibly get more or will it make things worse, somehow :P??????

---

I'd like to know this too, for beginner OC'ing do you just OC at 1:1 to find the CPU limit and then return the RAM to stock (assuming that the OC requires below stock ram speed)?


Also if you have say a CPU with a 10x multiplier and wanted to get to 3.2ghz, using a board with a 1600fsb and 800mhz PC6400, would it be theoretically better to lower the multiplier and do a 400mhz x 8 clock on the CPU so that the FSB & RAM are at stock, or is it no different to just do a 320mhz x 10 (assuming the components allow for either of these options)

I know it all comes down to real world in the end but I'd like to know in general terms whether one should aim to just get the fastest overall CPU speed (irrespective of mhz & multiplier used) or trying to balance CPU, FSB & RAM frequencies.