How memory timings on A64's affect performance - The truth.

NoGodForMe said:
Great thread, finally, the truth has come out.

I bought fancy Corsair PC3200XLPro (2225) for my FX53 because of the hype. Guess what, I can't run at those timings. I have to run my memory at 2336, and I've tried two sets. I've seen many posts of people who must run at 2.5,3,3,5 and so on.
Click to expand...

Really? What mobo is he using? I must be lucky cause I can run 2.2.2.5-1T up to 215 HTT. I finally got to 250 the other day. Too bad I can't cold boot off of it though.
 
Auger said:
If you were asking me, yes...I read that X-X-X-10 was better on the Nforce 3 boards, so I tried that and got over 7K memory bandwidth (both readings) on SiSoft Sandra and am currently at 240X11 @ 1:1, I ran 245X11 for a while, but it wasn't super stable, 240X11 is.
Click to expand...


I know this sounds dumb, but how do I specify the memory timings???? There is way too many options in the BIOS for the memory settings.

Which ones do I have to change after I specify manual settings???

I have a LAN Party UT 250 NF3. I know I have shitty memory, so I would like to set the speeds manually.
 
Toasty said:
I know this sounds dumb, but how do I specify the memory timings???? There is way too many options in the BIOS for the memory settings.

Which ones do I have to change after I specify manual settings???

I have a LAN Party UT 250 NF3. I know I have shitty memory, so I would like to set the speeds manually.
Click to expand...

In your BIOS memory settings you'll find CAS, RAS to CAS (tRCD) and RAS.

CAS = 2, 2.5, 3
RAS to CAS = 2, 3, 4
RAS = 2, 3, 4

If you have more settings, just ignore them as they're for hardcore tweaking. You can also download A64 Tweaker (by CodeRed) and change the settings while in Windows and set it to boot to those settings. All timings are usually set in the BIOS. If you have shitty memory, it will probably be operating at it's rated and default timings, and it will probably crap out if you try to tighten it.
 
What a strange experimnt. What are you even trying to prove here? That timings don't affect latency as much as clock? That latency is CPU-clock dependant? Does this even need to be stated? It all seems pretty obvious. By the way - there are some pretty big gains to be had running 1-0-0-0 as opposed to 2-2-2-10 :p But that's a whole different ballpark - here we're turning dynamic memory into what in essece is static memory :).
 
Neat stuff. It would be good to see a comparison between 2-2-2-5 and 2-2-2-10, however. Your ASSUMPTION that going from 5 to 10 makes no difference is plausible, but unproven here. ;)

Interesting how RAM timings are very clearly reflected in an Aquamark 3 score. Another reason to like that benchmark as a tool for testing your PC.
 
NoGodForMe said:
The problem I have with this, is if you put Corsair PC3200XLPro in an Asus A8V, the MB will try to run at 2225, and you'll crash when loading windows (happened to me). When you go on the CorsairMicro forums, they'll tell you to raise the voltage and get the newest bios. Only after you try and fail, will they tell you to adjust your timings. They'll never admit their memory can't run at those 2225 timings.
Click to expand...

I know when I talked to the Ramguy he made it clear hat for my pc-3200 that the cas was to be set at 2.5 for AMD systems and 2 for Intel. Not sure why but that's what he suggested.
 
iddqd said:
What a strange experimnt. What are you even trying to prove here? That timings don't affect latency as much as clock? That latency is CPU-clock dependant? Does this even need to be stated? It all seems pretty obvious. By the way - there are some pretty big gains to be had running 1-0-0-0 as opposed to 2-2-2-10 :p But that's a whole different ballpark - here we're turning dynamic memory into what in essece is static memory :).
Click to expand...

I don't find it strange. He is simply illustrating what effect CAS has on real world results at stock clocks. For the most part, it has proven to be insignificant. If, however, you bring clock rates versus latency timings into the picture, then the low latency versions of the RAM are significantly better. This review is a simple 2D look at RAM timings. It would be interesting to see benchmark results based on RAM timings at various FSB settings. For instance, does 250 FSB at 2-3-3-6 bench better than 270 at 2.5-3-3-8? That would be the thrid variable not covered by this review.
 
If you can increase your clock, your actual latency will drop, even if you have to increase delay measured in CPU cycles. If you can scale the CPU to make the delays even smaller, more power to you. You could try bringing the timings down to 0, to kill off a lot of delay, then you would see some very substantial performance gains. There are so many factors affecting memory pereformance. It's not possible to arrive at any conclusion conducting these not two-, but one-dimensional tests. And he doesn't even measure his latency. He also doesn't run several tests every time and average the results (as sometimese, these things fall well under the margin of error).
 
iddqd said:
If you can increase your clock, your actual latency will drop, even if you have to increase delay measured in CPU cycles. If you can scale the CPU to make the delays even smaller, more power to you. You could try bringing the timings down to 0, to kill off a lot of delay, then you would see some very substantial performance gains. There are so many factors affecting memory pereformance. It's not possible to arrive at any conclusion conducting these not two-, but one-dimensional tests. And he doesn't even measure his latency. He also doesn't run several tests every time and average the results (as sometimese, these things fall well under the margin of error).
Click to expand...

well, if you can explain how to set cas lower than 2, ras to case lower than 2, ras lower than 2, and so forth, i'm sure he'd be glad to test like that. however i have never in my life heard of a bios capable of setting those timings. and that would probably also need some major voltage running through it.
 
binkgle said:
well, if you can explain how to set cas lower than 2, ras to case lower than 2, ras lower than 2, and so forth, i'm sure he'd be glad to test like that. however i have never in my life heard of a bios capable of setting those timings. and that would probably also need some major voltage running through it.
Click to expand...

A64 Tweaker needs to be used for those settings. I've seen high FSB with tight timings like 1-1-1-10 before. From the people I spoke to who have reached those timings, it apparently doesn't affect performance all that much. A few of the top benchers have ram capable of 1 and lower timings, but just never use those settings, as it doesn't provide a large enough increase to warrant it.
 
Another problem is that today's DDR memory simply isn't made to handle or run at timings of 1-1-1.....and it can actually hinder efficiency and speed of the ram at times.
 
Rob94hawk said:
Really? What mobo is he using? I must be lucky cause I can run 2.2.2.5-1T up to 215 HTT. I finally got to 250 the other day. Too bad I can't cold boot off of it though.
Click to expand...



I run at 2-2-2-5 with my 3200+ 939 @ 2.2ghz

I have passed ALL test. I ran everest with it and got 6200read and 2100write with very low latency. Also passed the 3dmark benches. Now I wish I could figure out how to pass the 2.3 barrier with this ram, with slight looser timings. Say 2.5 3 3 6(8) or (10) but I cant seem to pass any of the test. Crashes in windows or 3dmark when I try. Is there anyone out there with my specs.

A8V Rev2
3200+ 939
1GB Dual Channel LowLat XMS Corsair (rated 2-2-2-5)

If you have passed 2.3 or 2.4 please post your bios specs so that I may run benches and stability test with your settings. Please include the memory timings and also what the multiplier is etc................... Any help would be great. Thanks in advance.
:)
 
I was just gonna say maybe it's the AGP/PCI speeds....but you've got a rev2 board......double checked that the lock is actually applied properly right? You could download Clockgen and check it at 1fsb intervals from 215 and up with clockgen cus it states the AGP/PCI clocks......just to double check. ;)
 
cornelious0_0 said:
I was just gonna say maybe it's the AGP/PCI speeds....but you've got a rev2 board......double checked that the lock is actually applied properly right? You could download Clockgen and check it at 1fsb intervals from 215 and up with clockgen cus it states the AGP/PCI clocks......just to double check. ;)
Click to expand...


Hmmmm interesting you brought that up. I set the value of the lock in the bios to .......
66/33 value but in the clockgen it shows that its at 76/36 AGP/PCI. If the lock works than it should be reporting a higher value right?
 
CounTDookU23 said:
Hmmmm interesting you brought that up. I set the value of the lock in the bios to .......
66/33 value but in the clockgen it shows that its at 76/36 AGP/PCI. If the lock works than it should be reporting a higher value right?
Click to expand...

If the lock works it should be reporting 66/33 no matter what HTT you have set.....

Is there an option in the BIOS that is simply called "locked" or something? If not, I'm afraid that your AGP/PCI lock might not be working....and once you hit a certain frequency it makes perfect sense that you'd even have issues getting into windows.
 
cornelious0_0 said:
If the lock works it should be reporting 66/33 no matter what HTT you have set.....

Is there an option in the BIOS that is simply called "locked" or something? If not, I'm afraid that your AGP/PCI lock might not be working....and once you hit a certain frequency it makes perfect sense that you'd even have issues getting into windows.
Click to expand...



I figured out what was wrong. I set the OC to auto a few days ago when i gave up messing witht the memory timings. The Lock isnt turned on on Auto Overclock. I have tried to OC with the AGP/PCI lock on though. I tried 2.5 - 3 - 3 -6/7/8/9/10 without any stability even at 2.3. What gives. I have the memory running at 2:1 and T1. :confused:


*Update* No wonder my games were crashing with the OC set on auto. That higher AGP/PCI bus must cause crashes easily.
 
CounTDookU23 said:
I figured out what was wrong. I set the OC to auto a few days ago when i gave up messing witht the memory timings. The Lock isnt turned on on Auto Overclock. I have tried to OC with the AGP/PCI lock on though. I tried 2.5 - 3 - 3 -6/7/8/9/10 without any stability even at 2.3. What gives. I have the memory running at 2:1 and T1. :confused:


*Update* No wonder my games were crashing with the OC set on auto. That higher AGP/PCI bus must cause crashes easily.
Click to expand...

Interesting..........
 
This also brings me back to the question. Which is better? 2.2ghz at 2-2-2-5 with obviously higher benches(memory benches anyways)


or 2.4 at 2.5 - 3 - 3 - 8 (2.5 @ 2.5 -4 -4 8) with less bandwidth depending on how loose the timings are and frequency.


What gains are there for pushing the hertz on these cpus. I know that on the prescott 2.8 running at 3.7 it boosted my game performance. I did not notice any boost in execution of programs or say opening a winrar file. With this athlon64 i did notice a huge increase from 2ghz to 2.2 (it had more zip to it) with the same 2-2-2-5 timings. Before I kill myself trying to OC this bish to 3.6 ............ is it worth it.
 
You'll never ever get that to 3.6 Ghz.

But to answer your question, it's something the further investigations will look into. Everyone asks these questions, I just want to answer them. And I will with future updates.
 
joecuddles said:
You'll never ever get that to 3.6 Ghz.

But to answer your question, it's something the further investigations will look into. Everyone asks these questions, I just want to answer them. And I will with future updates.
Click to expand...



Ooooooooooops. I meant 2.6.................... :eek:
 
joecuddles said:
You'll never ever get that to 3.6 Ghz.

But to answer your question, it's something the further investigations will look into. Everyone asks these questions, I just want to answer them. And I will with future updates.
Click to expand...


Joe what do you advice i do to hit 2.3 or 2.4 stable. What am I doing wrong? Should I use an MSI board?
 
cornelious0_0 said:
Another problem is that today's DDR memory simply isn't made to handle or run at timings of 1-1-1.....and it can actually hinder efficiency and speed of the ram at times.
Click to expand...
Why not? SDRAM could do it just fine. Yes, you don't see huge performance gains when going from 2-2-2 to 1-1-1, but by comparison, you would see a HUGE performance gain just by setting tras to 0. If you can kill your read delay, and the write delay, that translates into huge gains for your latency, and A64s like latency :D. Some guy on xtremesystems tested 1-0-0-0 vs 2-2-2-10 and gained about 25% in memory performance alone (that was at 100MHz), and about 3-12% in benchmark/game performance.
 
Hmmm, after reading this I might not spend an extra $60 to get 2-3-2-6 timing ram and just go for 2.5-3-3-6 and use the money I save to get a better video card.
 
iddqd said:
Why not? SDRAM could do it just fine. Yes, you don't see huge performance gains when going from 2-2-2 to 1-1-1, but by comparison, you would see a HUGE performance gain just by setting tras to 0. If you can kill your read delay, and the write delay, that translates into huge gains for your latency, and A64s like latency :D. Some guy on xtremesystems tested 1-0-0-0 vs 2-2-2-10 and gained about 25% in memory performance alone (that was at 100MHz), and about 3-12% in benchmark/game performance.
Click to expand...

Except going any lower then 5 on TRAS has literally or next to no effect on performance using apps like A64 Tweaker and others.....which is the only way to do it.
 
CounTDookU23 said:
This also brings me back to the question. Which is better? 2.2ghz at 2-2-2-5 with obviously higher benches(memory benches anyways)


or 2.4 at 2.5 - 3 - 3 - 8 (2.5 @ 2.5 -4 -4 8) with less bandwidth depending on how loose the timings are and frequency.


What gains are there for pushing the hertz on these cpus. I know that on the prescott 2.8 running at 3.7 it boosted my game performance. I did not notice any boost in execution of programs or say opening a winrar file. With this athlon64 i did notice a huge increase from 2ghz to 2.2 (it had more zip to it) with the same 2-2-2-5 timings. Before I kill myself trying to OC this bish to 3.6 ............ is it worth it.
Click to expand...
Well, at 285fsb on my ram (2.5-3-3-10 1T) and 2565MHz with my 3000+ I was getting 8033mb/s read, 3147mb/s write and 35.8ns latency in everest so MHz can make up for slightly looser timings.
 
cornelious0_0 said:
Except going any lower then 5 on TRAS has literally or next to no effect on performance using apps like A64 Tweaker and others.....which is the only way to do it.
Click to expand...
The "magic" tRas value is cas-dependant. For tCl=2, the optimal tRas=10 on nForce 3 boards; for tCl=1 the optimal tRas=0 :). It doesn't do much; but you can notice it in latency/bandwidth benchmarks.
 
iddqd said:
It doesn't do much; but you can notice it in latency/bandwidth benchmarks.
Click to expand...

Which is how a lot of this ends up turning out......I dunno, but I could have sworn I saw my AM3 score drop a bit when I went from a TRAS of 7 down to 5 with my PC4000EL at CAS2.5.....but it was probly just due to the general margin of error in the test so who knows. :p
 
also i have done some tests with how 1t and 2t effect game performance, namely in hl2. though it makes huge differences in andra, etc., using 2t onyl droped my numbers by an average of 2 fps (all tests were run three times and their results were averaged, then the number of fps dropped on average were averaged out). that means that it has no real game perf difference but greatly boosts ocability and stability. i know i'll never use 1t again unless i see a true use for it.

because i have ram only one step up from value, and it's only rated at 2-3-3 @ 200 mhz, i've set mine to 2.5-3-3-10-2T and htt is raised from 200 to 225. i see a perf boost in this :)

i'll try for more speed another time.
 
binkgle, even though I refuse to "cave in" and buy an A64, I can still see where you're coming from.....although I can't say I would stand alongside you on this 100%. If someone isn't obsessing over every little number then sure, use CAS2.5 or 2t for your ram and boost your overall overclock that much more.....but there are still those of us who WOULD strive to attain those same overclocks WITH 1t.....even if it meant getting different ram to do the job.

Obviously these are two comPLETELY different viewpoints and are coming from opposite ends of the spectrum here, just so long as you know that there are still just as many people out there who would shoot you in your sleep for saying that you'll never use 1t again.....but it wouldn't be anything personal. ;)
 
P4s don't benefit from low memory latency as much, because they get "hidden" in their massive pipeline. Ask me, and if you have a P4, don't bother with the timings at all - just jack up the bandwidth as high as it will go :D
 
iddqd said:
P4s don't benefit from low memory latency as much, because they get "hidden" in their massive pipeline. Ask me, and if you have a P4, don't bother with the timings at all - just jack up the bandwidth as high as it will go :D
Click to expand...

This is where some people like you (no offense) need to actually look into things a little more, or really need to work on not sounding so definite. I'll agree that P4's don't benefit nearly as much from tight timings.....but they still do. I'll agree on the point that it makes more sense with a P4 to relax your timings to get my MHz out of your system then it does for an A64...but once you find your max fsb you can hit I'd still spend time figuring out your tightest timings at THAT speed.

The tight timings will still help....and if you're a benchmark whore like me then you'll know what I'm talking about. For instance....it might seem small....but going for 3-4-4-8 to 2.5-4-3-5 gains me about 200-300 points in Aquamark3, and while that only equates to .3fps overall....its still enough of a gain for me to justify spending the time looking for my tightest timings I can do.

Most people will never bother because they early on buy into the misconception that timings dont matter with Intel...which is too bad. While slack timings aren't nearly as detremental to a P4's performance as they are for an A64, it still doesnt hurt to do what you can with them. After all why not take all the free performance you can get?
 
300 poins in AM is well under the margin of error :). Just keep running it, untill you get the best score :D
 
iddqd said:
300 poins in AM is well under the margin of error :). Just keep running it, untill you get the best score :D
Click to expand...

I've honeslty never seen my score bounce around that much.

My margin of error with all my benchmarks is FAR FAR FAR lower then most other ppl's.....because I have a seperate stripped down partition for doing this with next to nothing installed, and NOTHING running in teh background....70MB commit charge on boot. It's a lot of extra things in a normal partition/setup of windows that can cause and amplify that margin of error in a lot of tests.....but my results are consistent enough in this partition I've found that if I DO gain 300 points from something in AM3 that its a genuine increase.
 
ok well from what i've seen in this thread I'm seeing lots of average fps, and scores and this that and the other thing, but what I'VE noticed on my intel rig is that with high FSB, 250 or so, and looser timings, I do get more MAX fps. But what you miss when you go grab a pop while the benchmark was running is that the FPS are more stable when you use tight timings, and PAT settings. I look at specific parts of aquamark, and watch in 3dmark, when i run 2-2-2-5 and fast PAT(GAT) settings the fps are much more smooth ,less drastic jumps in FPS, which in my eyes, is more important than you rmax.

Lets say with high bandwidth loose timing the average dip in fps overall is 30 or more(which it is, run aquamark, or 3dmark) and with tight timings its 20(which is about right, a low guess). at the end of aquamark it gives an average fps, thats not what you should look for if you want good gaming performance. You want as little dips in fps as possible. Say you're gettin 60 fps in half life 2, you drop 30 and thats half your fps gone, your eye will notice that. Drop 20 and ur down to 40, still above the 16-25fps the eye can see in action.

Lets say you make poor fps, 40, the dips are HiFSB-10, and Tighttimings-5(seems about right from what I've seen) you go from 40fps, bearly good fps, but playable, drop 10 and ur in 30, in lulls in action you'll notice its choppy, because your eyes can see more FPS. The less dip in the fps as a result of better timings, will keep you further from the gap.


overall I'm pretty sure you sacrafice a modest piece of overall fps and scores, for smooth, better looking benchmarks, and in reality better looking gameplay. Thats why I overclock and tune, to get the best possible gaming experience I can get. Maybe if you guys want I'll try to setup a test on it to prove it.
 
XeroHouR said:
ok well from what i've seen in this thread I'm seeing lots of average fps, and scores and this that and the other thing, but what I'VE noticed on my intel rig...
Click to expand...

That's all very interesting. Unfortunately it doesn't apply to anything here. Since AMD64 and Intel are entirely different platforms, you'll get entirely different results, which shouldn't suprise anyone.

I haven't finished the testing recently because life does take precedence to overclocking and computers. Let's just say I haven't been home in 5 or 6 days, and won't be until maybe Monday night, so for now this must take a back seat.
 
CounTDookU23 said:
This also brings me back to the question. Which is better? 2.2ghz at 2-2-2-5 with obviously higher benches(memory benches anyways)


or 2.4 at 2.5 - 3 - 3 - 8 (2.5 @ 2.5 -4 -4 8) with less bandwidth depending on how loose the timings are and frequency.


What gains are there for pushing the hertz on these cpus. I know that on the prescott 2.8 running at 3.7 it boosted my game performance. I did not notice any boost in execution of programs or say opening a winrar file. With this athlon64 i did notice a huge increase from 2ghz to 2.2 (it had more zip to it) with the same 2-2-2-5 timings. Before I kill myself trying to OC this bish to 3.6 ............ is it worth it.
Click to expand...


More HZ on the CPU's is going to pretty much always be superior to better timings, unless it's a very small amount of HZ gain when going from CAS 2.5 to CAS 3. As noted in the original post the drop from 2 to 2.5 is practically non-existant.

Consider this.... going from 2 to 2.5 gives a 0.25% decrease in RAM performance and at most 1% decrease in gaming performance (usually less). Going from 2.2ghz to 2.4 is a 9% increase. Even if the actual gaming performance increase is half that (and it won't be unless your videocard is your bottleneck) then it's still a 4% increase minimum.... Much better than 1%... Go for the HZ!

Also most Motherboards allow you to clock down your RAM independently. For instance, I have an Abit KV8 pro with pretty mediocre RAM that allows me to run a 500mhz OC on my 3000+ (native 2.0ghz) because I can downclock the RAM to 333mhz instead of 400. With the 25% OC it is then raised back up to 416mhz.... easily within it's ability without any timing decreases. I also downclocked my HTT to 800, so it runs at it's native 1000 with the OC. So basically my CPU is OC'ed a ton while almost all the rest of my system isn't OC'ed at all (my PCI bus, AGP bus, and HTT aren't, and the RAM OC is only minimally OC'ed to keep me from having to relax any timings). The CPU changes jumped my 3dmark05 score up by about 15% I think.... Have to check when I get home.
 
iddqd said:
What a strange experimnt. What are you even trying to prove here? That timings don't affect latency as much as clock? That latency is CPU-clock dependant? Does this even need to be stated? It all seems pretty obvious. By the way - there are some pretty big gains to be had running 1-0-0-0 as opposed to 2-2-2-10 :p But that's a whole different ballpark - here we're turning dynamic memory into what in essece is static memory :).
Click to expand...
Curious question here: how are you able to validate the gains of SRAM at 1-0-0-0? If real-world validation of said timings with the injected correlated FSB variance (dependent upon chipset implementation, design efficiency, among others of course) could be proven, sustained on current hardware, and economically feasible, then I believe that is something the industry should be made aware of. If they are not already.
 
Whoa, long words there! Anyway, some boards have extra settings for the usual timings (along with some other timings). No, 1-0-0-0 is not possible with current specifications. For example, the 2-2-2-10 vs 1-0-0-0 test was done with BH-5 that ran at 100MHz at 3.2vdimm, IIRC. There was a substantial gain in gaming and benchmark performance, if not bandwidth :). This was done on an AMD system, by the way.
 
You must log in or register to reply here.
Back
Top