• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

Memory Timings in an AMD system

F

fryea

n00b
Joined
Jan 29, 2003
Messages
46
How important are timings in relation to an AMD system? Will i see a huge difference between 2-4-4-6 and 3-7-7-8 (or something to that extent) And what about the difference between the 1t timings? In an athlon system is it worth it to spend the extra money on it for the performance difference? Im more interested in high FSB's, but is it worth sacrificing the timings? someone enlighten me
 
Originally posted by fryea
How important are timings in relation to an AMD system? Will i see a huge difference between 2-4-4-6 and 3-7-7-8 (or something to that extent) And what about the difference between the 1t timings? In an athlon system is it worth it to spend the extra money on it for the performance difference? Im more interested in high FSB's, but is it worth sacrificing the timings? someone enlighten me
Click to expand...

Well first off, good luck trying to (although I'm not sure why you would want to) get any ram to run as slack as 3-7-7-8 but YES, you will see differences going from 2-4-4-6 (or 2-2-2-5 for that matter, if your dimms can manage it) to 3-4-4-8.....the slackest timings you'll be able to set it to.

Timings are far more important to AMD setups then they are to myself as an Intel guy but they play a big part in your performance irregardless.

A general rule of thumb for a nice AMD setup is to never, ever settle for anything slacker then 2.5-3-3-6.....looser then that and it just get's stupid in my eyes.

You can kinda tell how all the high speed ram out there (PC4000 and up) is being made solely with P4 setups in mind.....seeing as no hardware AMD enthusiast in their right mind would run any ram at 3-4-4-8 if they had the choice.

Intel Pentium 4 rigs benefit far more from the raw bandwidth and speed of the modules whereas with AMD timings are more important because you don't have the luxery of running the ram at 500MHz DDR to make up for CAS3. ;)

Hope I could help. :cool:
 
With AMD you want BOTH timings and high FSB. You don't need to scarifice timings for High FSB. Get your hands on some BH-5 PC3200 or PC3500 and you'll be able to get BOTH high speeds and tight timings with extra voltage like i did with my Level II

l2_250.jpg


If you don't have a mobo that supplies enough voltage, then MOD it or get a motherboard that does.
 
You do make a valid point Drisler but you have to realise that not everyone has the time/resources to find/pick up some nice BH5 PC3200 and so they are left looking at the vast majority of consumer memory that isn't necessarily capable of 250FSB @ 2-2-2-11.

If someone doesn't have an arm and a leg to spend on their ram it's good to shoot for something in the viscinity of 2.5-3-3-6 or 2-3-3-6 if you can get some slightly nicer stuff. There are a lot of dimms out there that are going to be rated at some pretty nice timings that aren't gonna break the bank made by Kingston (value), OCZ, and others.....you've just gotta shop around and find the best price/performance for YOU. If YOU think it's worth the extra $50 (or whatever) to get some 2-3-3-6 over some other stuff that's 2.5-3-3-6 then by all means, go for it.

The reality is that the average joe isn't going to be able to boot up his computer and tell right away if it's CAS 2 or 2.5 just from sitting in windows.

If you're the kind of person that wants every last little bit of performance out of your comp then yes, you probly will spring for the SLIGHTLY "faster" ram. However, a lot of people are a whole lot more practicle then that and will buy things only if they're going to be able to TELL that it's better when gaming or doing whatever.

A really good example of the first type of person is myself. When I'm benchmarking, i want every last little number to be as high/tight as possible. if that means doing extra testing to move from 2-3-3-6 to 2-2-3-6 then so be it. However, when I'm gaming.....I just don't care. :)

Figure out which kind of person you are and you'll know if it actually makes sense to go buy some "faster" PC2700 or something that's better on paper.
 
Originally posted by cornelious0_0
Well first off, good luck trying to (although I'm not sure why you would want to) get any ram to run as slack as 3-7-7-8 but YES, you will see differences going from 2-4-4-6 (or 2-2-2-5 for that matter, if your dimms can manage it) to 3-4-4-8.....the slackest timings you'll be able to set it to.

Timings are far more important to AMD setups then they are to myself as an Intel guy but they play a big part in your performance irregardless.

A general rule of thumb for a nice AMD setup is to never, ever settle for anything slacker then 2.5-3-3-6.....looser then that and it just get's stupid in my eyes.

You can kinda tell how all the high speed ram out there (PC4000 and up) is being made solely with P4 setups in mind.....seeing as no hardware AMD enthusiast in their right mind would run any ram at 3-4-4-8 if they had the choice.

Intel Pentium 4 rigs benefit far more from the raw bandwidth and speed of the modules whereas with AMD timings are more important because you don't have the luxery of running the ram at 500MHz DDR to make up for CAS3. ;)

Hope I could help. :cool:
Click to expand...

Are you on crack, dude?

http://www6.tomshardware.com/motherboard/20040119/index.html

Don't waste your money-- you will barely (if at all) notice the difference. According to the above review, even actual benchmarks noticed only minimal differences between CAS3 and CAS2 RAM.

While getting better RAM isn't really snake oil, as it DOES improve performance (by however little), you just have to ask yourself if that little bit of performance is worth it to you. Now by no means am I advocating buying shitty no-name RAM, but these days CAS latencies really don't mean much. Just read the article-- it's spelled out pretty clearly there.

As for high-performance RAM, such as anything higher than DDR400, that may well be a different story. It is important to run your RAM in synch with your FSB, especially on AMD systems, as not doing so translates into a loss of performance. If you're running your FSB above spec, you need to have RAM to match, for example DDR433 RAM will run at 216.5 MHz, which will be fine if that is what your FSB is. And these days, with motherboards like the DFI Infinity, seeing FSBs like 220 is becoming pretty common.
 
why on earth are you reading tomshardware? :eek:

and there is a very noticeable difference between cas2 and 3...
 
Originally posted by Jason711
why on earth are you reading tomshardware? :eek:

and there is a very noticeable difference between cas2 and 3...
Click to expand...


I have not noticed a difference at all.
 
thank you thank you thank you

that article helped alot in my decision :)

I think my next ram purchase for newers systems will focus on higher speeds rather than latency settings, because latency can be a selling point to alot of people, but it just wont be me next time. And yes, synthetic numbers are fun to tote, but in real world applications, it really doesnt makes THAT big of a difference. Now if i was getting like 10 more FPS in halo (or some other slow arse FPS) then yes, lower latency would be for me, but i see the larger gains in the fsb/mem speed area.

thank you again all of you for the info/discussion
 
Originally posted by Visable-assassin
I have not noticed a difference at all.
Click to expand...

run things at cas 3 for a while then switch to cas 2.. i always see a difference.
 
Originally posted by fryea
thank you thank you thank you

that article helped alot in my decision :)

I think my next ram purchase for newers systems will focus on higher speeds rather than latency settings, because latency can be a selling point to alot of people, but it just wont be me next time. And yes, synthetic numbers are fun to tote, but in real world applications, it really doesnt makes THAT big of a difference. Now if i was getting like 10 more FPS in halo (or some other slow arse FPS) then yes, lower latency would be for me, but i see the larger gains in the fsb/mem speed area.

thank you again all of you for the info/discussion
Click to expand...

Oh boi..you'll regret that:(
 
If people want to be ignorant and ignore the cold hard facts I can't do anything about that except pray that one day they'll realise the truth and change their ways.

Don't waste your money-- you will barely (if at all) notice the difference. According to the above review, even actual benchmarks noticed only minimal differences between CAS3 and CAS2 RAM.
Click to expand...

First off, I can't believe you're giving any credibility to THG.....'cus THAT'S the exact kind of bull that comes out of their site. I have never heard anyone try to make the point that CAS rating don't matter but I'm not surprised that it came from Toms when it did. I wouldn't take that article with a grain of salt.....it's complete nonsense.

why on earth are you reading tomshardware?
Click to expand...

Exactly what I'm saying, I'm glad to see that SOMEONE knows what they're talking about. :rolleyes:

I have not noticed a difference at all.
Click to expand...

Riiiiiiiight. :confused:

that article helped alot in my decision
Click to expand...

Plz don't say that, and don't follow it. I'm not sure what kind of crack the guys at Toms are smokin' but CAS latencies always make a difference.....especially in AMD rigs, just because they don't have the massive raw bandwidth to make up for slack timings.

I can't remember what site did it (I'm pretty sure it was Anandtech) but they took an engineering sample P4 to test this out. They ran the P4 at 200FSB with some nice LL Corsair at stock speeds, and they ran it at 250FSB with some PC4000 at 3-4-4-8 while maintaining the CPU clock of 2.4GHz. What happened is that while there is still more raw bandwidth at 250FSB, games and 3dmark benefited much more from the tight timings and the fps's and scores were for the most part higher on the 200FSB system.....cus of the CAS 2 ram.

PLZ for the love of god do NOT ignore cas latencies, get some decently tight stuff.....or you WILL regret it. Don't take advice from people or websites that dont know what they're talking about or you'll just get burned later on. Seeing as Toms is the FIRST and ONLY site that I've ever seen make that statement in the last 2 years what do YOU think??? ;)

Do the right thing.
 
Originally posted by stub
Are you on crack, dude?

http://www6.tomshardware.com/motherboard/20040119/index.html

Don't waste your money-- you will barely (if at all) notice the difference. According to the above review, even actual benchmarks noticed only minimal differences between CAS3 and CAS2 RAM.

While getting better RAM isn't really snake oil, as it DOES improve performance (by however little), you just have to ask yourself if that little bit of performance is worth it to you. Now by no means am I advocating buying shitty no-name RAM, but these days CAS latencies really don't mean much. Just read the article-- it's spelled out pretty clearly there.

As for high-performance RAM, such as anything higher than DDR400, that may well be a different story. It is important to run your RAM in synch with your FSB, especially on AMD systems, as not doing so translates into a loss of performance. If you're running your FSB above spec, you need to have RAM to match, for example DDR433 RAM will run at 216.5 MHz, which will be fine if that is what your FSB is. And these days, with motherboards like the DFI Infinity, seeing FSBs like 220 is becoming pretty common.
Click to expand...

Don't even read TH.

Most biased site on the net.

Generally speaking, low latency memory timings are critical for AMD rigs prior to A64/FX51.

Intels run much more stable with looser memory latency.

Look for BH-5 cas 2 rated memory for Athlon XP's, currently Mushkin Level II pc-3500 or Hyper X pc-3200.

Hyper X pc-3500 is now ch-5, the bh-5 was used up awhile back.
 
Originally posted by ZigZagZeppelin
Don't even read TH.

Most biased site on the net.

Generally speaking, low latency memory timings are critical for AMD rigs prior to A64/FX51.

Intels run much more stable with looser memory latency.

Look for BH-5 cas 2 rated memory for Athlon XP's, currently Mushkin Level II pc-3500 or Hyper X pc-3200.

Hyper X pc-3500 is now ch-5, the bh-5 was used up awhile back.
Click to expand...

Yay, someone else with some sense to them to back me up!!! :D
 
Originally posted by cornelious0_0
Yay, someone else with some sense to them to back me up!!! :D
Click to expand...

i tried.. but nobody ever listens to me..

so it seems anyway.. :rolleyes:
 
Originally posted by Jason711
i tried.. but nobody ever listens to me..

so it seems anyway.. :rolleyes:
Click to expand...

Heh, don't take it personally buddy. Think of it this way.....you're a part of the winning team. :D

I know that sounds sad but whatever, I'm bored. :eek:
 
Suggesting that timings are of little value to AMD chipsets is bull. And for not only AXPs, A64s as well --> Mushkin's A64 report specifically says it's best to buy RAM with a tRCD of 2 for Athlon64s because of thier low latency memory controller. Only Memories that carry a tRCD of 2 (or are able to do a tRCD of 2) are those with BH-5 chips. CAS is not the most critical of the timings..tRCD is just as important.

Still amazes me how Toms can run their P4s at 2-2-2-5 but still give AMD 2-3-2-6. Aren't reviews supposed to be balanced? Check that 3400+ review ...you see them using some cheap 'never heard of before' memory brand with shitty timings ...and the P4 compare rig gets Corsair XMS.:rolleyes:
 
Originally posted by vegetash
is this good or should I change something?
url]
Click to expand...

Is that a palamino 1700+ you've got there? looks like it based on the memory clock but whatever. 2.5-3-3-6 is kinda the line in which you never wanna cross as an AMD junkie and make sure that you ALWAYS and I repeat ALWAYS have the ram running at 1:1.

I know you might not have a lot of room to work but I'm just lettin' you know for later that CAS2'd be even better. ;)

Still amazes me how Toms can run their P4s at 2-2-2-5 but still give AMD 2-3-2-6. Aren't reviews supposed to be balanced? Check that 3400+ review ...you see them using some cheap 'never heard of before' memory brand with shitty timings ...and the P4 compare rig gets Corsair XMS.
Click to expand...

If this isn't proof enough that they're a completely bull review site then I don't know what is. Down with THG, down with THG!!! :cool:
 
well, to all of you wondering, i have been running an XP 2100+ tbred b for about a year now at 400fsb in sync with the ram. The ram is the discussion, so thats why im bringing it up. Its samsung pc3200 that runs at 3-4-4-8. This has been for an entire year. I havnt noticed anything particularly "slow" about my system, but since i have never experienced any kind of ram besides it in my system, i wouldnt know the difference. The CPU is running at 11*200 (2.2ghz) in an abit nf7-s. What exactly makes cas 2 or lower tRCD better? I keep hearing that cas 3 "sucks" but what makes it so bad in regards to real world performance that makes it so bad?

btw, i picked that ram up last year at a small computer shop in seattle for like 50 bucks a stick (256mb sticks), and that was back when pc3200 was top notch, so i got the low end of the best spectrum (or however you want to put it)
 
Let me put it this way-- if you want that performance, and if it is worth $80 or more to you, then by all means go for it. Personally, however, I don't do much gaming, and for me that $80 would be better spent on a nicer monitor ($80 is the difference between a normal 19" CRT and the holy grail of CRTs, the P95f+), or a hard drive, or fuck it, a water cooling system.

You guys might have a point that LL RAM makes a difference, and I might decide to go for 2.5 instead of 3.0 RAM, since its not much of a price difference.

EDIT: I just remembered I had 200 bucks budgeted for a gig of RAM, and it seems like a pair of 512MB Mushkin Level One DIMMS cost that much, and are 2-3-2 (I think the last one is 6). This looks like a great deal, what do you guys think?
 
Always go with tight memory. 2-2-2-5 at like 190 will prolly be on par or faster compared to 3-4-4-8 at like 215-220.

ALWAYS TIGHT TIMINGS FOR THE LOVE OF BLACK BABY JESUS.
 
Originally posted by pduan87
Always go with tight memory. 2-2-2-5 at like 190 will prolly be on par or faster compared to 3-4-4-8 at like 215-220.

ALWAYS TIGHT TIMINGS FOR THE LOVE OF BLACK BABY JESUS.
Click to expand...

Yay!!! More smart people!!! :D

Definitely go for that Mushkin.....some of the nicest stuff you can buy, but you often end up paying for it. ;) Sure it's level one and not level 2 but whatever, 2-3-2-6 are great timings no matter what and are gonna help you out.....big time.

Go for it and let us know.
 
Well, since i hear THG is unreliable and biased, i took matters into my own hands, and ran a few tests w/ my new ram, giel ultra platinum pc3500, rated to run at 2.5-3-3-6 @ 433DDR.
For this test, i ran an athlon xp 2100+ tbred @ 11*200 in an nf7-s. 1.75 vcore, 2.9vdimm. Powercolor 9800se w/latest omega drivers and softmod. SLK-800 w/ smart fan 2. and our test subject, 2 sticks of giel ultra platinum pc3500 2.5-3-3-6. For these tests, i ran 3dmark 2001 (only that because im lazy, and doing this takes too long, i have a rise of nations campaign that needs tending).
Cas/Rpd/Rtc/Rad-----------------Score
3-4-4-7-----------------------------16552
2.5-4-4-7---------------------------16609
2-4-4-7-----------------------------16727
2-4-4-6-----------------------------16815
2-4-4-5-----------------------------16694 (?)
2-3-4-6-----------------------------crash
2-4-3-6-----------------------------wierd windows errors
2.5-3-3-6---------------------------16821


it looks like the stock timings are where its at, doesnt seem to like much else. Although from very loose (3-4-4-7) and somewhat tighter (2.5-3-3-6) its only a 269 point difference. Now i wonder if leaving it at looser timings and boosting my fsb/ram will yeild a performance gain that makes up for the looser timings. Stay tuned, and feel free to ask questions/nit pick me/tell me im a fucking nub.
 
Part II, an hour or so later. So in these tests, i focus on FSB and ram speeds, while keeping a sharp contrast between my lax timings (7-4-4-3) and my tighter timings (2.5-3-3-6).
I started off pretty much where i last left the tests. The only change is i bumped the vcore up to 1.8 (the highest supposedly thats "safe" for air cooled setups). Everything else is the same as it was (except for the experimental options listed below). and yes, ram is still in sync (6/6)

Mult / FSB / Timings / Result
11 / 200 / tight / 16821
11 / 205 / tight / 17160
11 / 210 / tight / blue screen startup
11 / 210 / loose / blue screen startup
10.5 / 210 / loose / 16741
10.5 / 210 / tight / 17066
10.5 / 215 / tight / 17325
10.5 / 220 / tight / blue screen startup
10.5 / 220 / loose / blue screen startup
10 / 220 / tight / 17121
10 / 225 / tight / dead startup
10 / 225 / loose / 17459
10 / 230 / loose / dead
10 / 230 / 3-5-5-8 / dead
10 / 230 / 3-6-6-9 / dead

so there you have it, my little experiment. Although this setup didnt have the most controlled environment possible, it did get the job done at what we wanted tested. The tight timings prevented it from starting up at 225 FSB, yet when we kicked the timings down a bit, we got a good 3dmark. I might have answered my own question for now, but im sure theres alot more for me to know on the topic of memory timings, but i feel ive got a better understanding of the situation.
 
You have done a fine job of bringing a sense of clarity to the issue and giving ppl a visual but remember that not everything is based on 3dmark and different apps/games will show different results. I'm not accusing you of anything, just stating the facts.
 
Originally posted by stub
Let me put it this way-- if you want that performance, and if it is worth $80 or more to you, then by all means go for it. Personally, however, I don't do much gaming, and for me that $80 would be better spent on a nicer monitor ($80 is the difference between a normal 19" CRT and the holy grail of CRTs, the P95f+), or a hard drive, or fuck it, a water cooling system.

You guys might have a point that LL RAM makes a difference, and I might decide to go for 2.5 instead of 3.0 RAM, since its not much of a price difference.

EDIT: I just remembered I had 200 bucks budgeted for a gig of RAM, and it seems like a pair of 512MB Mushkin Level One DIMMS cost that much, and are 2-3-2 (I think the last one is 6). This looks like a great deal, what do you guys think?
Click to expand...

that would be perfect...
 
We need to make a setup with the same CPU, Same MOBO, but different RAM sticks.

That RAM you tested had horrible timings, but could do reasonable FSB. You need to max out your FSB (with lowest possible timings at that FSB) then record your scores (which you have done).

Then you need to install ultra low latency RAM like Mushkin level 2, set your timings to 2-2-2-6(or 11) then raise FSB to as high as possible. Do tests then record the scores (which hasn't been done).

This is necessary because we want to test extremes - high FSB vs. low timings. You can't just do high FSB on one RAM and compare it with the so-so timings on that same RAM, because high speed RAM weren't made for ultra low timings.
 
Originally posted by Unit-0
We need to make a setup with the same CPU, Same MOBO, but different RAM sticks.

That RAM you tested had horrible timings, but could do reasonable FSB. You need to max out your FSB (with lowest possible timings at that FSB) then record your scores (which you have done).

Then you need to install ultra low latency RAM like Mushkin level 2, set your timings to 2-2-2-6(or 11) then raise FSB to as high as possible. Do tests then record the scores (which hasn't been done).

This is necessary because we want to test extremes - high FSB vs. low timings. You can't just do high FSB on one RAM and compare it with the so-so timings on that same RAM, because high speed RAM weren't made for ultra low timings.
Click to expand...

Agreed.....and a fair point.
 
if anyone wants to send me some bh-5, i would love to :) but until pigs fly and mcdonalds becomes healthy, i just cant help there. I know SOMONE out there wants to make a donation for the betterment of society.

my tests were done for 1.) see how so so timings and lax timings compare at different FSB's and 2.) max out this ram and see how much juice this 2100xp has left in it :)
 
Originally posted by fryea
if anyone wants to send me some bh-5, i would love to :) but until pigs fly and mcdonalds becomes healthy, i just cant help there. I know SOMONE out there wants to make a donation for the betterment of society.

my tests were done for 1.) see how so so timings and lax timings compare at different FSB's and 2.) max out this ram and see how much juice this 2100xp has left in it :)
Click to expand...

fine fine, point taken.

1 question though, why name yourself after someone in that game???
 
On Nforce2 systems there is a performance gain from running TRAS 11, plus it helps with stability some. That is a good pointer.
 
I was just messing around with my BH-5 and decided to do some testing. Keep in mind though that this is with a fresh install of xp, no service pack, not even directx 9b, using cat 4.1 and 3dmark2k1. This is with my 9700 Pro at stock for both.

Timings/3dmarks
8-4-4-2.5/13435
6-2-2-2/16146

8-4-4-2.5 was the lowest I could get the bh-5 to run. Don't take these as the final results though, I didn't think the gap was this wide, I will have to test again after I get everything installed. Oh and btw, don't listen to THG they are obviously paid by Intel and therefore their results are scewed in their reviews.
 
Originally posted by Drisler
With AMD you want BOTH timings and high FSB. You don't need to scarifice timings for High FSB. Get your hands on some BH-5 PC3200 or PC3500 and you'll be able to get BOTH high speeds and tight timings with extra voltage like i did with my Level II

l2_250.jpg


If you don't have a mobo that supplies enough voltage, then MOD it or get a motherboard that does.
Click to expand...

How did you get all those fields to show up, like Freq, FSB:DRAM Ratio and such, mine looks like this...

cpuz.jpg
 
Ok, I installed everything and re-ran 3dmark2k1 but I changed the slack timings.

Timings/3dmarks
7-3-3-2.5/15879
6-2-2-2/16219

Overall a difference of 340marks but obviously that can vary a bit.
 
Originally posted by RC7
Ok, I installed everything and re-ran 3dmark2k1 but I changed the slack timings.

Timings/3dmarks
7-3-3-2.5/15879
6-2-2-2/16219

Overall a difference of 340marks but obviously that can vary a bit.
Click to expand...

For sure it can vary, but 340 points is enough of a gap to make it more then just a fluke run that would be different next time.
 
Originally posted by KINgGh0sT
How did you get all those fields to show up, like Freq, FSB:DRAM Ratio and such, mine looks like this...

cpuz.jpg
Click to expand...

Get the new version of CPU-Z
 
Originally posted by RC7
I was just messing around with my BH-5 and decided to do some testing. Keep in mind though that this is with a fresh install of xp, no service pack, not even directx 9b, using cat 4.1 and 3dmark2k1. This is with my 9700 Pro at stock for both.

Timings/3dmarks
8-4-4-2.5/13435
6-2-2-2/16146

8-4-4-2.5 was the lowest I could get the bh-5 to run. Don't take these as the final results though, I didn't think the gap was this wide, I will have to test again after I get everything installed. Oh and btw, don't listen to THG they are obviously paid by Intel and therefore their results are scewed in their reviews.
Click to expand...
Huh? Do you mean the highest you could get it to run? Or did you mean to say 6-2-2-2?
 
You must log in or register to reply here.
Back
Top