Speed and timings

I am semi-familiar with his work. He's a sort of sub-Hegalianist who focuses mostly on the anthropological side of things and believes in genetic altruism. I am rather fond of that general area of philosophy and I used to argue genetic altruism a lot as an undergrad. It has its weaknesses though, specifically in that it fails to address why we sometimes act in ways other than how we should be genetically programmed. My gut tells me that a combination of this naturalist account of morality (which is virtually Hobbesian) and a more transcendentalist appreciation of human reflection will yield better explanations about what constitutes our moral make-up.

Out of interest, how did you manage such a good overclock? What are you using for cooling? And how did you get your X850XT to score 8,000 in 3dmark05?
 
Well, having waited too long for my finances to get in shape it looks like I'll now be unable to get hold of the Opty 146, leaving me with the option of a 144 or a 148. When I've figured out what's the best option between those I'll probably pick up the RAM man_from_uncle mentioned in this thread.

Unless I shun gaming and go dual-core.
 
Well the ram i recomend, the Corsiar XMS3500LL is probably, in my opinion, the best 2 gig set you can currently get, but considering you can get some PC3200 G.Skill that is not a lot worse for nearly £90 cheaper, i'd only go for the Corsair if you've got money to burn.

As far as processors go i've loved my opty 144, but i've not got the same massive overclocks that some other people have got from their opty 146s. I have had it up over 2.8ghz, but not that stabley and not cool. I think it is the HTT limit on the board rather than the chip though that is crapping out, so if you want to have a shot at 3ghz you really need a chip with a x10 multi. If you are happy at FX-55 speeds (maybe up to 2.7) though then it'll do you nicely.
 
man_from_uncle said:
Also, in very basic terms, 'dividers' are the ratio that you run your ram at compared to your FSB (or HTT in AMD64 terms). It is not essential that the calculation for determining CPU speed be 'RAM speed' X Multiplier. You can run your ram out of time with your FSB at certain ratios. Your ram might run at 200, but if you are running a divider of 5:4 then your calculation will be 240 x multiplier. Your ram will be running slower, which will affect your processing speed in real terms, but the inhanced CPU speed will more than make up for it.

Now, on these modern boards dividers are often not represented as simple ratios like 5:4 or 6:5. Instead they are represented as as 200, 166, 133 etc. I really don't know why this approach has been taken but the numbers are themselves a reflection of the comparitive speed that the ram will run at. If you pick 166 as your divider (it's often called something like memory index in the bios) your ram will not actually run that fast, but it will run at a ratio of 166:200 to the FSB. So if you push you ram back up to 200 it will push the FSB to 240. This use of dividers allows you to reach the maximum speed of your processor without the need to run your ram at ultra high speeds.
Click to expand...
Sorry to dredge this up from the depths of page 4, but I wanted to ask a few more questions as I've been doing some reading up and wanted to see how if I've begun to get my head round this yet.

First off - dividers.

Now, mfu mentions above that dividers are often represented as an X:200 ratio. And by mooching around the web I've managed to find a screenshot of the relevant options in my motherboad's BIOS:



So that looks like I've got 2:1, 3:2, 6:5, 1:1, etc etc. I am right in saying that the values in that screen relate to the ratio rather than the actual memory speed itself? Because I can't see an option to actually set the speed of the RAM, so I'm assuming that the speed is set when you change the HTT frequency and use one of the dividers. Is that right?

I think we've sort of established that dividers don't affect performance as much as people might think, but I'd imagine that they do come into play when you want to get the best out of your memory - you want to run a divider that will allow you to get the best from both your processor and your memory, but not necessarily by running them at 1:1?

Second part - timings.

I'm beginning to understand what the four numbers relate to, but what I don't understand is what makes a good combination and how the memory speed comes into it. If you had memory running at 2-2-2-2 (theoretically speaking, obviously) would memory doing those timings at 200Mhz be worse than memory running at 250Mhz? Assuming for the sake of this example that some would!
 
Right....see what i can do.

1) Dividers. Yes they relate to the ratio rather than the actual speed. You want to run a divider that will both get you to your maximum overclock and give you the best performance that the ram can offer. However, sometimes even if you can make do without a divider you might not be running your ram as well as it can be run. For example if your max overclock was 2.5ghz with a processor with a multiplier of 10, and your ran could run at 250 with 3-4-4-8 timings you would still find that you would be better off running a 5:4 divider and have your ram at 200 with 2-3-2-5 timings. Also you are correct that the ram speed is altered when you change the HTT (in proportion to the divider).

2) Timings: Yes, faster frequency still makes quite a difference. 250 is quicker than 200 at the same timings. When people say that timings are more important than frequency, they really mean that timings are more important compared to frequency than they used to be. Previously, before AMD64s, the frequency was a lot more important than the timings and the aim was always to get the frequency as high as possible; timings were secondary. Now the best thing is to have a good combination.
 
Thanks for popping in, mfu.

OK, so with memory you need to find a set that will run with tight timings and a high frequency to get the best out of it. However, I notice from my travels that with most RAM the faster the speed gets, the slacker the timings get.

If you're buying with the intention to overclock, are you buying blind or should you already know (where possible) what your CPU is likely to overclock to and how, and then work out what RAM would partner up the best? I'm sort of getting my head around the overclocking and timing issue, but I can't figure out what people do when buying their kit. Go for the best they can afford and hope it pans out?

Let's see if my understanding of the concept is correct or complete rubbish!

Let's say I overclock a Opteron 165 (9x multi) to 275 HTT. That would work out at 2475Mhz. Let's also assume that I have a set of your Corsair XMS3500LL sticks, which run at 2-3-2-8 at 433DDR (217Mhz HTT?).

I'm guessing here that I'd either need to run a divider that allowed this (can't see one to fit exactly though) or either run a divider that gets me as close to the 433DDR as possible, or that will get the speed of the memory up a little and hope that the timings don't suffer as a result?
 
glitch said:
I'm guessing here that I'd either need to run a divider that allowed this (can't see one to fit exactly though) or either run a divider that gets me as close to the 433DDR as possible, or that will get the speed of the memory up a little and hope that the timings don't suffer as a result?
Click to expand...

Spot on, if its a choice of slightly higher freq or slightly lower latencies do some benchies as what works for other people may not be the bst for your individual system.

You can always bump up the voltage to get the settings you want, but dont go over 2.9 (assuming its tccd based ram)

Remember always run 1t, this is the most important factor with memory on the 939 chipset
 
man_from_uncle said:
I am semi-familiar with his work. He's a sort of sub-Hegalianist who focuses mostly on the anthropological side of things and believes in genetic altruism. I am rather fond of that general area of philosophy and I used to argue genetic altruism a lot as an undergrad. It has its weaknesses though, specifically in that it fails to address why we sometimes act in ways other than how we should be genetically programmed. My gut tells me that a combination of this naturalist account of morality (which is virtually Hobbesian) and a more transcendentalist appreciation of human reflection will yield better explanations about what constitutes our moral make-up.

Out of interest, how did you manage such a good overclock? What are you using for cooling? And how did you get your X850XT to score 8,000 in 3dmark05?
Click to expand...

Re: Failing to address acting in other ways - How much biology have you done? I'm also a Biochemist and many non-scientists see genetics as an unbreakable link between DNA and behaviour. I.e. 'I have a gene for selfishness, so I cannot act in any other way'. That's simply not true as we have evolved to such a level where, in many ways, our brains, memes (you've probably come across this term) or social effects overrule the gene for that behaviour.

Re: overclock - I'm fully watercooled and have volt-modded my trusty X850 :D
 
glitch said:
I think we've sort of established that dividers don't affect performance as much as people might think, but I'd imagine that they do come into play when you want to get the best out of your memory - you want to run a divider that will allow you to get the best from both your processor and your memory, but not necessarily by running them at 1:1?

Second part - timings.

I'm beginning to understand what the four numbers relate to, but what I don't understand is what makes a good combination and how the memory speed comes into it. If you had memory running at 2-2-2-2 (theoretically speaking, obviously) would memory doing those timings at 200Mhz be worse than memory running at 250Mhz? Assuming for the sake of this example that some would!
Click to expand...

Have a read of this and decide for yourself how much impact memory overclocking has on the AMD64 platform:

http://www.neoseeker.com/Articles/Hardware/Guides/athlon64oc/

Re: Timings.

Each number refers to a clock cycle. The higher the frequency of the memory (200Mhz, 250 Mhz etc) the shorter the clock cycle.

e.g.

2 clock cycles @ 200 Mhz = 2 * (1000/200) = 10ns

2 clock cycles @ 250 Mhz = 2 * (1000/250) = 8ns

Another thing to remember is that lower numbers don't mean better performance for all the timings. Some are more stability settings and there will be an optimal number:

tCL (more commonly called CAS) = 1.5 (rare), 2, 2.5 or 3 (slow). Recommend 2 if possible, 2.5 if that's not possible.

tRCD (or RAS to CAS delay) = 2, 3 or 4. Recommend 2 or 3. 4 is pretty slow.

tRP (Row Precharge) = 2, 3 or 4. Recommend 2 or 3. 4 is pretty slow.

tRAS (Active to Precharge Delay) = large variety of numbers. Highly recommend using the following formula:

tRAS = tCL + tRCD + 2

or

tRAS = tCL + tRCD + tRP


In an ideal world, you'd be running the memory @ 300 Mhz or so @ 2-2-2-6 timings :D :D :D

However, often, running the memory at a high frequency can hold back your maximum stable overclock on the CPU (on-die memory controller). On the AMD64 platform, you should never sacrifice CPU clockspeed for memory clockspeed.

e.g. in my sig, I have quite a high CPU clockspeed (not as high as I'd like ;)) but a relatively low memory clockspeed with tight timings.

Hope some of this helps.
 
Thanks, Defcon; that confirms what I've been reading up on these past few weeks.

Jimbo; that's exactly the sort of information I've been looking for. Obviously not been looking hard enough, eh?

Off to read up some more now, cheers gents.
 
glitch said:
Jimbo; that's exactly the sort of information I've been looking for. Obviously not been looking hard enough, eh?
Click to expand...

Nah, there's just too much bad information out there with people just blindly assuming lower numbers = better without bothering to research or decide for themselves.

I guess it's up to them, but I hate doing things without understanding WHY I'm doing them.

:D

GL and HF as they say...
 
You must log in or register to reply here.
Back
Top Bottom