THE Memory FAQ

I
think what he is saying is that it is better to have a 1:1 ratio (FSB 220/Memory 220) than it is to overclock beyond the memory’s ability (FSB 250/memory 200).
Click to expand...

I understand now. In this scenario one probably wouldn't get a significant oc since the ram dividers aren't brought down.

I guess if one bought ddr 500 and brought the divider to ddr 400 and then oc to (FSB)250 that would be ideal? :confused: Since FSB = 250 and the ram is running at 250.

Intersting thread nevertheless. Thanks Mcseiam :)
 
I have a question:

I have a P4 3.0 GHz 800 MHz 200 MHz FSB, running 2 sticks of 512 MB 2.5-something-something, all at stock speed.

I am planning to overclock and I have the opportunity to upgrade the memory, so:

Which is better?

2 x 512 MB PC3200 2-something-something or
2 x 512 MB PC4000 3-something-something?

I am planning to reach 250 FSB on water cooling.
 
Could anyone explain DDR2's actual clock speed/effective transfer rate and JEDEC rating system a little bit more to me?
 
MC FLMJIG said:
I'm so confused now. If Ocing, like the example above, has no benefit or doesn't help overall system performance on AMD's then why do it? I thought that OCing made your computer faster.

Before you mentioned that increasing the FSB 500 MHZ(250x2) and the ram being 400 MHZ would have no benefit but here you say OCing the computer would make it faster? I'm somewhat confused...
:confused:
Click to expand...

<Obi-Wan Kenobi voice>
Use the multiplier Luke
</OWKV>
 
I have asked various people and gotten various answers...

I am looking to build a new box with 2GB of RAM, which is better or faster, 2 1GB sticks, or 4 512 sticks?

Thanks in advance,

Bill
 
As far as I know latency is better with a smaller number of sticks, but if you have a dual channel system then I would go with 2 sticks so you get the added bandwidth of dual channel config without the extra lag of 4 sticks.

 
this sticky is in sore need up updating.. i shall refer you to the other one, where i have explicitly stated that 4 sticks is unwise for an a64 rig, but doesn't really matter for p4
 
(cf)Eclipse said:
this sticky is in sore need up updating.. i shall refer you to the other one, where i have explicitly stated that 4 sticks is unwise for an a64 rig, but doesn't really matter for p4
Click to expand...


yeah i learned my lesson
 
so instead of getting two more 512 sticks of platinum rev 2, i should sell mine, and get a 2 x 1gb set if i want to move to 2gb?
 
Schmiggy_JK23 said:
so instead of getting two more 512 sticks of platinum rev 2, i should sell mine, and get a 2 x 1gb set if i want to move to 2gb?
Click to expand...

yes

or get 4 sticks and intel cpu
 
I have an odd problem. On my A7N266-C mobo (Nf420 chipset I think) The manual specifies that in 2 of the slots, I may use Dual-bank DIMMs, but in the third, only single-bank. Well I finally decided to go past my 1gig and use that 3rd slot, either for a 1G or 512MB stick.
My problem is this: When I'm shopping for RAM online, how can I tell if it's Single or dual bank? I know you can easily tell when its in your hand by checking to see if there are chips on both sides, but I don't have that luxury when shopping online. Can you tell from the model number? Is there a reference of sorts I can refer to?
 
Guys how many errors are acceptable if any in memtest86?

A PC locks up from time to time just in windows and were thinking its the memory, it got 3 errors in memtest, is that bad ?
 
3 errors in how much time?

i find that even with 0 errors for an hour or so, my ram can still be a bit unstable. :(
 
(cf)Eclipse said:
3 errors in how much time?

i find that even with 0 errors for an hour or so, my ram can still be a bit unstable. :(
Click to expand...


3 errors in 24+ hours, the first error showed up like 3 hours into the test.
 
This is from a post in another thread I wrote. Thought I'd paste here for posterity. It's worded differently here that what's in the FAQ. Perhaps it'll enlighten some that wonder WHY CAS2 vs CAS3 doesn't matter much.

(from a programmer's point of view).



Regardless of what L33T G4m0rZ may tell you, CAS latency doesn't "really" matter. It's a gimmick, except if you want the mostest bestest hax0rz numbers in a synthetic benchmark. Here's why:

If memory (!) serves right, CAS latency refers to the number of cycles the memory needs to take before it completes a request from the memory controller.

For example, if the CPU needs to fetch some data somewhere in the memory, it first tries to see if it exists already in the cache. If it does not, it will instruct the memory controller to get it from the main memory (that's what's called a "cache miss", btw). The memory controller will instruct the RAM circuitry that it needs "X" amount of memory at spot "Y". RAM will wake up, wait "CAS" (2 or 3) cycles to set up things, and when it's ready it will send data back at whatever speed (mhz) your RAM is set to, towards the CPU.

Meanwhile, the CPU is sleeping because it's waiting for the data to arrive (or it may play Solitaire or Strip Poker with the RealTime Clock, and usually wins). Eventually, the data arrives, and the CPU wakes up (or lays down its cards) and resumes processing.

(Before I go on, I'll just open a short sidebar and say that if a processor had no cache, having CAS2 memory instead of CAS3 would have a HUGE impact on performance, because the CPU would always be asking the main memory to send data (and it would be sleeping most of the time). If memory transfers happened instantly, CAS2 vs CAS3 would yield a 30% overall improvement. But of course, CPUs have a large cache and limited bandwidth, so the situation is different. End sidebar)

So, modern processors have a lot of cache (1-2 megs), and as I explained when a "cache miss" occurs, the processor requests the data from the memory controller. However, more than just the "requested data" comes back to the CPU. What comes back is enough data to fill a small portion of the cache memory (called a cache line) so that the next time the CPU asks for memory, it it will probably be in close proximity to previous memory request. If so, the memory will be readily available in the cache and it won't need to go out and fetch it (smart huh). So the bigger the cache memory, the less often the CPU will need to bother to go out and ask the memory controller for some data, wait for the CAS overhead to happen, and for the time it takes for the data to arrive.

So now that we know that data tends to be processed sequentially (big chungs at a time) or in close proximity (from one memory request to another) cache misses occur relatively infrequently, so there's no CAS overhead and no memory transfer from main RAM to the cache. In "Real World applications" the only time where CAS latency will matter is if the program is requesting small chunks of memory from ALL OVER THE PLACE in physical RAM. When this happens, the memory controller loads too much memory, sends it to the CPU, and the next memory request isn't where the CPU predicted it would be, so it flushes the cache, asks the memory for data elsewhere, uses it, the next memory request is again far, cache gets flushed, (etc etc) so CAS overhead plays a little role there.

In a case like this yeah CAS overhead could be a small issue, but less than you'd think. What is more important than CAS is the speed of your RAM and the time it takes for the data to travel to the CPU. To help understand, you'll agree that it's better to "load a train" (CAS overhead) in 15 minutes and send it at 300 mph (memory mhz) towards it's destination (CPU) than to be able to load the train it in 10 minutes and send it at 50 mph towards it's destination. All in all, memory bandwidth (speed) matters more.


Okay, fine, so CAS2 memory will give you better results in SiSoft Sandra for synthetic benchmarks, whoop-didoo. And you may get that extra 2 FPS more in Quake3 that already runs at 450 fps. Who cares. It's certainly not worth the premium you'd pay for that CAS2 timing. In fact, if you _notice_ better performance from CAS2 vs CAS3 memory, I'll cut off my penis. I'm not talking here that a 3DS Max render finishes in 33 mins 22 seconds instead of 33 mins 28 seconds. I'm talking here stuff that matters.

Oh and BTW 2G ram is enough, you won't be able to use 4G in XP-32 bit anyway (you'll waste a gig).

I hope I've explained things better.
 
I agree with sticking to CAS 3 memory if you're not overclocking. Which isolates a majority of [H] members. I see this is overlooked in discussions refering to this subject lately.

I applaud your explanation of CAS latency but advising avoiding CAS 2 RAM I disagree with. Buy low latency memory, loosen the timings and use it to help attain your max cpu overclock. That's if you're overclocking. If not, get yourself some Value RAM and use the extra dough to take your old lady out.
 
great FAQ, little daited tho...

"todays newest comps use Nforce 2 platforms..." not with 590 out.

anywho. Tier Command rates. does anyone have any idea what they are? no one has ever fully explained this phonomia to me. on Nforce 4 platforms its generally accepted that 4X512s are bad because they force a teir 2 (T2) opperation, where as only one or two sticks can give a teir 1 (T1) opperation. in most cases 2X 512mb sticks outperformed 4X 512mb sticks, according to Xbitlabs, this was cuz of Teir command rates. they didnt go into further detail then that.

anyone? do the newer chipsets allow for better performance at Teir2 command rates?

thanks
 
MrWizard6600 said:
great FAQ, little daited tho...

"todays newest comps use Nforce 2 platforms..." not with 590 out.

thanks
Click to expand...

Worked out well for me, because I'm still stuck on an nForce2 and I won't be able to upgrade any time soon :D

But, yeah, cool guide. Very helpful - that is, if any of it even applies anymore.
 
I'm seeing dicussions of D9 memory being the stuff to have for the new dual core and core 2 duo CPUs. What exactly is D9 memory? How would I know it when I see it? When do I really need it?.
 
Mcseiam said:
Buying Memory


Memory Buying

Why are you recommending PC3500, when my motherboard only supports PC3200?

Memory modules really have no fixed speed. Like the tire to a car, there is a "rating" on it. When a tire is rated to be 150mph, it means it can run as fast as 150mph maximum. It also means that it can run at any speed lower than that. It is also quite safe to say that the tire should also withstand at 160 mph, just not as "safe" according to the Government's test environment.

Memory is very much similar in this way. Many people ask if a PC3500 or PC3700 module would run/blow up/be compatible in a motherboard originally designed to use PC3200 or PC2700. The answer is, hell ya! JEDEC (the “government”) has only approved PC3200. This reason, coupled with the fact that no processor needs memory rated higher than PC3200, are causes for motherboard manufacturers not stating support for newer, faster modules. But higher rated speeds of DDR are always ‘backward compatible’ so to speak, or capable of running at lower speeds. Older systems stand to gain from newer and faster modules. Even if they can't run the module at its top supported frequency, you can still tweak the timing parameters to maximize performance at lower clock speeds, that otherwise would not be possible with lower-rated modules.

.
Click to expand...

So does this mean that I can buy this board which only supports DDR2 533 and use DDR2 667 or DDR2 800 instead? If so, is there any benefit to doing that?
 
maybe someone here can help me, if not ill creat a new post, id really like this resolved.

EVGA 680i, loven every minute of it. just about to install a thermalright Ultra 120 w/ antec fan on my E6300. I managed to get it to boot stable @ 3.2Ghz. Im running this mobo in unlinked mode, setting it to its stock 666.6MHz (yeah this comp is a bet to prove I can outperform an alienware comp w/ half the money). I wanna maintain 1T because of some of the things I've heard about 2T operations and how poorly they perform on this board, according to sandra anyways. The ram is PQI Turbo DDR2 667, 4-4-4-12, stock on this stuff is 1.9V. I've got two 1024MB sticks of it.
Ok so now that i've ripped through the basics...

The problem:
I cant overvolt.
I am one of those people who is overclocking their ram for the sake of overclocking their ram. I was hoping to squeze DDR 800 out of this ram. I took a chance, which I dont usually do (I'm a wuss) on this ram, but only because i was able to get it off e-bay for a great price.
The first thing I tried to do was up it to a respectable 2.1V. System didnt post, infact it let out one of those scary elongated beeps that can scare the shit out of you if your not ready for it.
"C1" my mobo says. I look it up: "Failed to pass Memory Test". greeaaaaaaaaaaattttttt.
Cold Reboot, still no post. Clear CMOS time. That works, I boot to windows successfully at stock settings. Doesnt appear as though I've damaged anything (how could I). So I go for a more subtle Overvolt. 2.0V. Wont do it. Will not, for the life of it do it. Same thing. Long Beep, "C1" etc etc.
So i try for 1.95V, which it does. 1.975 it does, but it takes a longer time when on the "C1" indicator on my board.

So I'm wondering, WTF? this is not my first time overvolting ram, and I've never seen stuff this annoying. Did i just do it to fast? does it need to be weasled into 2.0 and later 2.1V?

any suggestions mucho appriciated.
 
Aratech said:
So does this mean that I can buy this board which only supports DDR2 533 and use DDR2 667 or DDR2 800 instead? If so, is there any benefit to doing that?
Click to expand...

yeah you can, theoretically you would see a little more heat on your northbridge then its ready for (and I mean like, 1C, 2C more). You'll be fine.

A good example of this is Intels 975X, which officially only supports DDR2 667. but EVERYONE is putting DDR2 800 on it. Some have put some Dominator DDR2 1066 or OCZ Flex DDR2 1133 on it. Stabillity might be a bit of an issue with the Dominator and Flex ram, but people buying that are enthusiasts, and theres always a work around.

The benifit would be better ram, lol. It would give you that little extra performance that extra memory bandwidth can do. 1-5%, be it in FPS form, loading times, or w/e, thats about what going from DDR2 533 to DDR2 800 would do. This is asuming timings are 5-5-5-12 on both sticks.
 
Short question on ram for C2Ds

im up to speed on the sticky, and ive been AMD for a while now, but im contemplating on switching to intel

now i know if i just buy DDR2-800 ill be golden for OCing, but what i want to know is how sensitive C2Ds are to timings, IE, is it worth it to pay more to get 4-4-4-12 instead of 5-5-5-15 ram?

and one more question, how are intel chips at handling 4 dual sided dimms? my current amd 939 chip has to drop back to 2t to run my 4x512 sticks at ddr400. Now i plan on getting 2x1gb initially, but im pretty sure i will be dropping in another 2gb down the lane

thanks for any advice
 
yeah it would be nice to see this updated for DDR2 knowledge. Like how important tRAS is and what ideal calculation is for it on DDR2 (Mushkin said CAS + tRCD +2 = tRAS for DDR). What's the formula for DDR2? How about any anomaly for NForce6 series?
etc.
 
The pricey ram does hold it's value. I actually made money on the high end PC3200 ram that I bought. You can usually get back what you paid when you sell it.
 
yacoub35 said:
yeah it would be nice to see this updated for DDR2 knowledge. Like how important tRAS is and what ideal calculation is for it on DDR2 (Mushkin said CAS + tRCD +2 = tRAS for DDR). What's the formula for DDR2? How about any anomaly for NForce6 series?
etc.
Click to expand...


  • CAS (Column Address Strobe or Column Address Select) - is the number of clock cycles (or Ticks, denoted with T) between the issuance of the READ command and when the data arrives at the data bus. Memory can be visualized as a table of cell locations and the CAS delay is invoked every time the column changes, which is more often than row changing.
  • tRP (RAS Precharge Delay) - is the speed or length of time that it takes DRAM to terminate one row access and start another. In simpler terms, it means switching memory banks.
  • tRCD (RAS (Row Access Strobe) to CAS delay) - As it says it's the time between RAS and CAS access, ie. the delay between when a memory bank is activated to when a read/write command is sent to that bank. Picture an Excel spreadsheet with a number across the top and along the left side. They numbers down the left side represent the Rows and the numbers across the top represent the Columns. The time it would take you, for example, to move down to Row 20 and across to Column 20 is RAS to CAS.
This is from a noob's prospective NOT TO BE TRUSTED . tRCD is, I think, tRP + CAS. so add both the timings + any timing you want. Just be careful not to lower it too low especially for DDR3 or high latency RAMs. Hope that helps...
 
Is there a thread or a guide somewhere on the world-wide-web with a list of the fastest DDR2 memory kits?

I'd like to know the top 3 fastest DDR2 memory available (2GB and 4GB kits) within a price range of $50-100 and $100-$200 respectively. If there's not a list, is there somebody "[H]ard" enough to provide a quick list here (LOL.. :p)??
 
Bo_Fox said:
Is there a thread or a guide somewhere on the world-wide-web with a list of the fastest DDR2 memory kits?

I'd like to know the top 3 fastest DDR2 memory available (2GB and 4GB kits) within a price range of $50-100 and $100-$200 respectively. If there's not a list, is there somebody "[H]ard" enough to provide a quick list here (LOL.. :p)??
Click to expand...

I'd very much like to see a list similar to this, but exclude price categories... (let buyers decide for themselves).
 
I am running Vista Business 64 bit edition with a Phenom 9500 processor, 2 radeon HD3870's and 4 gigs of OCZ reaper but for some reason vista says that I only have 2 gigs of reaper. Does anyone know how to get Vista to recognize 4 gigs?
 
Manic_Midwestern said:
I am running Vista Business 64 bit edition with a Phenom 9500 processor, 2 radeon HD3870's and 4 gigs of OCZ reaper but for some reason vista says that I only have 2 gigs of reaper. Does anyone know how to get Vista to recognize 4 gigs?
Click to expand...

I am running 64bit ultimate and it recognized 8gb out of the box.
 
Yesterday I added a 1gb module to my motherboard. I already had 2 512 modules. The computer shop guy suggested i buy just the one module rather than two. I had intended to replace the 512 modules with two 1gb modules. I put the new 1gb module in the empty middle slot and left the two 512's in the outside slots. The computer see's the added module.
Now reading the FAQ here I am beginning to think I should pick up another 1gb module and take out the 512 modules. Whats the opinion here?
I have 3 slots, if i pick up another 1gb module, does it matter which slots I use. Could I leave one of the 512's in? The mob will handle 3gb ram.
I know this sounds like a silly thing, much more interesting stuff going on around this forum, but I'm getting this machine ready to pass on to my son after I build a new desktop. I'll post the new build info before I start. I have lot's more research to do first.

Thanks
Ed
 
What motherboard? Assuming that 2 of the slots are on 1 channel and the other slot is channel 2, then ideallu you'd want a 1 gig stick on both channels and the odd 512 in the extra slot.
 
Hi HemiRick,
The motherboard is a Chaintech, about 2yrs old. 7n something.
Thanks for the tip, The two 512's are in the outer slots where the guy who added them put them when the computer was upgraded.
If I get another 1gb module you suggest I put both 1gb's in the outer slots and the other 512 in the center? Two slots are close and there is a gap between them and the 3rd slot.
Maybe I should get two more 1gb modules as my son plays games a lot.

Ed
 
You must log in or register to reply here.
Back
Top