Athalon 64, X2 or not

Larry:

Looser memory timings are going to permit you to push it further but you also have to find the ratio between memory speed and CPU speed that will keep both stable. Don't get too worried about the bigger numbers here. I doubt that you actually have 2-2-2-5 XMS because its pretty special and still pretty pricey. Thats not a big deal at this point. Again, I would advise against running anything except stock core voltage during the initial stages of finding the "right mix ". Your boy who was bragging about the 12.5 x 208 is blowing smoke or has a modded CPU because the multiplier is indeed locked at the upper limit. You have to do your work with that frequency (lets call it FSB even though its a misnomer) and balence it out with a memory ratio that will give you stability. Once you find the max FSB for your processor at stock voltage and the right memory timings at something no more than 210-215 for starters, then we can start playing a little and pushing that puppy. I'm sending you a link for reading up a bit. it should take some of the mud out of this.

 

Hi, I am going to ask for specifics as having dual-core is messing with the numbers (200 vs 400MHz) and BIOS terms/settings are all different. Please answer, confirm or correct:

1. I would be a lot more comfortable if I had an answer for my question regarding the HT Freq multiplier (1-5x, Auto). What do I do with it, just leave it at Auto?

2. My XMS is rated at 2.5-3-3-6 for AMD. I tried 2-2-2-5 and 2.5-2-2-6. Both checksumed out. So, my timings will have to go up instead of down... say 2.5-4-4-7 or 3.0-4-4-8? If I loosen the timings to, say 3-3-3-7, will I see any kind of a change in CPU-Z? or am I just creating the potential for raising the FSB? Lowering the Memlock Freq will do the same thing? Same with taking the memory from 1T to 2T?

3. I understood your 5:4 goal in theory but my BIOS is labeled weirdly. My CPU is running at 240MHz while my DRAM Config/Memlock is set at 400MHz (is that really 200MHz and is 240:200,6:5)?). So, you are saying that I should lower the Memlock freq and try to boost the CPU and see what happens? Also, the timings in #2, too.

4. I am now at 240MHz FSB at stock vCore and stock mem timings/voltage. I assume that 400MHz is stock, too (see #3).

I am tired of clearing that BIOS. Will I wear it out?? I just had to reflash it as the ASUS CD flashed it backwards from 1016 to 1014.

Honestly, the issue is that all of these web pages have different nomenclature, different BIOS screenshots with different settings. Things I have never heard of and I don't know if they are the same as mine, just named differently (like FSB vs HTT vs LDT). Even the website you linked to is like that so what they recommend or explain serves to confuse, and enlighten. Many of my concerns above relate to those issues. I am unstoppable when I know which is the gas, the brake, the clutch, the brights, the wipers and the signals. Until then, I just have a lot of questions. So...

If I post screenshots of MY BIOS, will that help our cause?? And if so, how do I do that in DOS? (more questions, sorry.)

Lar

 

I might be able to help you with that one

From Windows, do a "Restart" and go into the BIOS. When you are at the screen you want, press the Printscreen key. After you Exit the BIOS and get to Windows, open an imaging program and "paste" it onto a new document. You will probably need to upload the image to a hosting website.

-----------------------------------------------------------
Comment:
I know you want a HOT system, but environmental factors may not let you get to the "pinnacle" (example, if you want to push the RAM really hard, you would probably need a higher "model "). AFAIK, if you push the frequency up, you need to back off on the timings.

You need to find the "sweet-spot" where your system will run at maximum and be stable at the same time. The system is a balance of CPU, RAM and the capabilities of the motherboard. You will need to probe to find that sweet-spot.

I would guess that the computer is running very fast now. Do your research. See how "sweet" you can get that "spot"

Matt

 

Lar:

Am pushing off in a few minutes and will be travelling for the next week but will try to get to each of your concerns in some detail within the next day or so. (Actually now have my laptop setup in my car with complete WLAN internet, email, Bluetooth GPS and Bluetooth printing functions available at all times - hehehe).

I will say you are closer than you think and yes, you might want to put a tailed jumper on those CMOS pins so its easier to get to - sorry, but that goes with the territory and can be a real PIA.

FSB vs HTT vs LDT - consider all of them different nomenclature for the same thing.

#2,#3 & #4 in your post - yes to all across the board - you know more than you are giving yourself credit for. And yes, 200=400 because its 2x200 on AMD's firing on the upstroke and the downstroke so to speak.

The HT frequency multiplier is going to come into play a bit later as we change that to stay as close to original rated speed (terminology only). As an example 200x5 =1000 which is your rating. 250x4 =1000 which is also your rating. I've been trying to take you through this in stages and will continue to do so but I also understand that you are no dummy and need to understand the theory so you can differentiate between the brake, clutch and gas peddles - more on this later.

Regards,

 

Hi Matt, I tried the Printscreen/Cntl-V/Paste method but the computer always reboots after accessing the BIOS, therefore, clearing the clipboard. I must be doing something wrong. Please enlighten!!

Also, I actually am not looking for a hot system. I am just looking to make the most of what I have (without reducing its life expectancy).

Rock, sounds like a heck of a mobile setup. Disco ball, too?? Look forward to hearing from you on the road. Thanks for keeping me in mind.

Lar

 

Here are some random comments from my A64 overclocking experience...

You want to keep your HTT at around 1000. Examples of HTT multipliers and HTT bus speeds that achieve that are 5 x 200 (default) and 4 x 250. Boosting your HTT beyond 1000 yields no performance benefit and can make your system unstable.

Being able to change those two values (HTT multiplier and "bus speed ") allows you to increase your HTT bus speed and team this up with your cpu multiplier for a larger overclock. Thus, with a certain cpu multiplier, a higher HTT bus speed will yield a higher overall cpu overclock.

I've heard that the goal with A64s is to get your cpu speed as high as possible and that memory speed contributes the least to overall performance in an Athlon 64 system. So to put it simplistically, the higher your HTT bus speed, the faster your cpu will run but you don't want to stray too far from that 1000 HTT figure.

When I was experimenting, I relaxed my RAM timings very much and made sure that the memory divider I used didn't stress the RAM too much. I did this so I wasn't "stressing" to many variables at the same time. If I did, I couldn't be sure which value was causing the instability.

After I sort of figured out what cpu and HTT values my rig could handle, I then moved on to memory. What was difficult was that I couldn't just ignore my RAM while working on my cpu/HTT. Finalizing all these settings is sort of like a "dance" because of all the interdependencies. Here's how I ended up overclocking my 2.2Ghz cpu to 2.7Ghz:

cpu multiplier: 10
HTT "bus ": 270
HTT multiplier: 4
Memory Divider: something that resulted in my PC3200 ram running in the 210-220Mhz range I think.
Vcore: 1.45v (max available on mainboard)
Vdimm: 2.7v (max available on mainboard)

This resulted in a 2.7Ghz cpu speed and an HTT of 1080 - not very far out of spec.

BTW, I use Prime95 and SuperPI to stability test my overclocked system. One of them uses a "maximum heat torture test" and the other does zillions of calculations and will report anything other than perfect accuracy. And yeah, CPU-Z is awesome for seeing the resulting values.

And BTW, even though I can run my cpu at 2.7Ghz, I currently run it only at 2.5Ghz. Most of the games I play are at least 3 years old - they just don't make 'em like they used to!

 

Chiles, that was very helpful. Thanks.

With HTT Multiplier at 4x (CPU multiplier is maxed at 9), my CPU is now 270MHz at a core speed of 2428, I dropped my mem divider (Memclock) to 266 from stock 400 (2x133) and loosened the timings from 2.5-3-3-6 to 3-4-4-7 at 1T. I am in the process of hiking the HTT 5MHz at a time. What I have found is that if I do it this way instead of 10MHz jumps, and hit a wall, the BIOS will still load but Windows will falter. Much easier recovery that way. As an initial indicator of stability, I run JV 5.0 for the 5 minute CPU heat up period. If it survives the Windows and 5 Min JV test, I go for another 5Mhz. When windows falters, I back off and run JV all night.

Still at all stock voltages.

As far as the HTT Multiplier goes, 1000 div by 4 = 250, 1000 div by 3 = 333. What do I do in between where it is mathmatically over 1000 (as high as 332 x 4 = 1,328? Do I split the difference and drop it to 3x at (333+250 div by 2 =) 292?? Hope that was logical. I don't want to add this factor to any perceived errors. I am already at 1083.

Questions:

1. As far as what to do next..... I don't think that I want the memory div to be any lower, do you???? (2x133MHz) Should I actually plan to raise it to 300 or 333 or 336?

2. Once I peak at this config, what do I do next, play with the timings a little more or will that make much more difference? I am not looking for every little bit of MHz. Or do I go into the voltages at that time?

Thanks, Lar

 

I think we might be out of luck. I read that several years ago and there were some provisos on how it would work. I did a few experiments on mine and could not get it to work either.
If you still want to post a screenprint, check the m/b manual, if there is a screenprint there, you could get it from the manual's download at Asus.
Some of the items in the BIOS may be "new" or an Asus "thing" for any of us to know about and the Rojackpot site Rockster referenced only has generalized BIOS details. The m/b manual will have a reasonable explanation of the various settings and the Asus Knowledgebase is about the best. See if you can find any references at their website. If you can post the explanation of the setting back here, someone may be able to decipher it. There's always trial and error, which you seem to be familiar with already

I think we have a different makeup. I was wondering why you seemed so impatient to get it set up so quickly. I would be pushing it along an inch at a time. You have picked up so much about all this I find it amazing. You are making me jealous! I would like to find the resouces to get an Opteron myself (I just have to sell three computers first ).

Stretch out the experience if you can, me, I would do a little at a time as the free time became available and just focus on those areas that needed attention in the sequence.

If you jump to the last page you miss out on all the good stuff in between

Matt

 

Hi Matt, I do love to tweak but I don't want to break it and it just takes a lot of time. This new puter has chewed up a lot of time over the last few months and part of me loves it, part of me wants to watch "House ".

Also, my head is in cyber-mode right now and if I wait, I will have another learning curve. I just got the gum chewing down pat, too!!! It's midnight and I am motivated!! The price of the AMD just now seem to be dropping. We expected something in late July but it never came. Some prices seem to be down 15-20%.

So, I finally hit the next wall at HTT 280Mhz. I will do the JV5 all night but I am sure that it will be fine. I did drop the HTT multiplier down to 3x just because it was getting way high over 1000. Now its 840, better than 1120?? I tried going back to 4x just to see if it was just a thrill-seeker of a mobo. Alas, it made no diff. the 280MHz wall remained unchanged.

Temps are peaking at 38*C mobo, 53*C CPU, during the JV Whetstone testing. Core Speed is 2526.

NOW-
1. Should I mess with the timing more or at this point will it really make a difference?

2. Is there any advantage/disadvantage to its current 2x133MHz mem divider status (stock is 2x200)?

3. Adj timing or go on to something else??

4. With the faster core speed vs slower memory, how would video rendering be affected?

Thanks, Lar

 

Lar:

Have you checked the mail lately? Rumor has it that Cal Poly, Rensselaer and MIT have been panning this board and recruitment letters have already been launched in your direction. There are no advanced curriculum scheduling conflicts with "House" at any of the three.

Glad to see Gary and Matt offering some great guidance too. I think you are very very close with what you have already done and the direction you have taken. Unfortunately, I don't have the opportunity to look at my own BIOS right now but for potential next steps, you might want to try bringing that memory back up closer to its intended speed in MHz. As Gary referenced, something along the lines of 210-215 should be very acheivable. You can try dropping your timings to 4-4-4-8 and get away from that 2x133 experiment. I also think with a current core speed of 2526, you can now try bumping your core voltage but a word of caution here - go very slow and use minimal incremental increases. The objective here is to use just enough to achieve stability. Same concept applies to memory but check max voltage on your specific memory modules at the Corsair site and don't try to push beyond that.

I really think Gary summed it all up quite accurately when he stated, "Finalizing all these settings is sort of like a "dance" because of all the interdependencies. "

Lar - your thirst for knowledge and the amount of research you have put into this are truly amazing. You've certainly earned my respect.

Regards,


note: I'll be taking a correspondence course learning to spell pedal.

 

Well, I think that I am about there.

Current status:
290MHz HTT
2626 Core Speed
HTT Multiplier 3x/ 875
MemClock (div) 2x166=333
Mem V 2.75 stock
Mem Timing 3-4-4-7
vCore 1.4 (stock is 1.36v)

1. HTT Multiplier- Going from 4x to 3x made no diff. Just less potential for trouble. Staning pat with 3x.

2. MemClock- Going from 266 to 333 made no stability diff at all. Going to 400 knocked me back to 240/2426. Standing pat.

3. Mem V- Stock is 2.75 and Corsair says nothing else onthe subject. Standing pat.

4. MEM Timing- Tried to go up to 3-4-5-8, no dif. The BIOS max for CAS is 3.0. Standing pat with 3-4-4-7

5. vCore- Best scenario with stock 1.36v vCore gets me to 280MHz/2526. Best scenario with 1.4v vCore gets me to 290MHz/2626.

Questions:

1. Temps still look OK (54/37*C) but I am now concerned about the vCore and at what point do the benefits outweigh possible reduction in the live expectancy of the CPU. Your thoughts on this?

2. Rock and Gary, re your comments on the MemClock (divider) rate and shooting for 210-215. Do you mean 2x210=420MHz?? Stock is 400MHz. Am I better off going back to 400MHz and sticking with HTT 240/2426, sticking with HTT 290/2626 at 333Mhz Mem or am I missing something??

3. Will the relaxed memory timings and the reduced divider have any downsides?

That is my tale. Thank a lot!

Lar

EDIT NOTE: While on hour 7 of JV5, my son was web surfing and the computer rebooted. Oh well. Back to 285 or 280.

 

Lar:
I'll fall back on Gary's Dance Commentary and drop it back in your lap. You can run Prime95 and SuperPI (I'm not familiar with this one) as he suggested for additional benchmarks to help you make a determination. I think you have such a good handle on this at this juncture that there is little if anything I can add. Personally, I wouldn't be worried about running at 1.4 vs 1.36 as I don't think a modest bump like that is going to have any kind of measureable effect on your CPU or its long term health. You can push this quite a bit higher and still be pretty comfortable but we need to keep in mind that the ultimate objective here isn't a speed record but is instead a nice safe overclock combining performance, stability and longevity. As to the memory, I think it can handle more but that is not going to be a make or break situation if you settle on 333mhz - again, benchmarking is going to give you a better answer on this than I ever could. I'll also be anxious to hear what Gary thinks in this regard.

One of the things I like to do is find out just how far I can push a system and then back off of those settings to reach a comfortable trade off. You can also head on over to some motherboard forums and OC forums to read what others have done with the same setup but don't believe everything you read - look for patterns of consistancy in the comments from several different users. And, as to your son's surfing at the same time you were running the whetstone & MMR tests, I don't run anything else when I'm running those tests so don't let the reboot deter you.

Now that you've hit the speeds you are getting and assuming that you are doing this with total stability - I'd say mission almost accomplished. Going way back to one of my first posts in this thread, I referenced 2.5 and 2.6 as possibilities - Again, you've certainly earned my respect and you personally deserve the credit for what you have accomplished. Nice Job Lar!

I'll defer to Gary for his thoughts and any additional tweaking that he might have up his sleeve. There may be others who would also like to comment and make suggestions.

Regards,

 

I'm not too thrilled with your temps. I think my cpu stays at around 43C even when greatly oc'ed. Mine's a Venice 3500+ and my vcore is at 1.45v. I'm guessing that you have the retail heatsink that came with the chip. I bought my cpu retail but I slapped this $12 heatsink on it with AS5. It's an AMD stock hsf but it's meant for faster dual-core chips and has heatpipes. I.E., I think you could do much better with an enthusiast's hsf. Undoubtedly, you could bump your vcore higher with that cpu but I wouldn't without a good hsf and good resulting temps.

Yes. I refer to RAM speed in simply Mhz, not the double-data-rate speed. Running your RAM at 333Mhz(166) seems a bit too slow for an A64 system. Are you sure you can't finagle all your settings so that your RAM runs close to spec speed but you still get a high cpu clock speed?

Are you using the Corsair XMS Cas2 Platinum stuff? With A64s, I've read that CPU speed is king so I wouldn't get too wrapped up in RAM timings until you've settled on a HTT bus speed, HTT multiplier and CPU multiplier. You don't want your memory "tail" wagging your cpu "dog ". If you have Corsair XMS, you should be able to go with aggressive timings but leave the RAM timings until all other factors have been settled upon. Introducing aggressive RAM timings before you've decided on the above variables would make me lose my marbles. 2.75v for Corsair XMS is fine. You could go higher to push extreme timings but remember that memory is a bit of a stepchild on A64 systems. I'd just be happy to find a memory divider than runs my RAM close to spec or a little above and then tweak the timings from there. Personally, I would not push the VDIMM higher than 2.75v for extremely tight memory timings simply because I don't think you'll see any benefit - at all.

I'm not even sure what benchmark to use to measure performance if your goal isn't max. gaming performance. SuperPI and Prime95 are stability testers only. I would guess that the only RAM setting that will significantly impact performance would be the Command Rate. I would accept any RAM timings no matter how relaxed as long as I could run a Command Rate of 1T with stability. Keep in mind that the Command Rate is known to greatly impact stability. You need to run Memtest86 for a good amount of time to see if 1T doesn't cause problems. As I mentioned somewhere, I cannot run 1T with two 1GB sticks. Pushing your VDIMM higher than 2.75v might let you run with 1T. With a 1T setting, you should start with CAS set to 3.0 and then lower the CAS if you can. Corsair XMS can probably run fine with 2.9v but pushing your memory that hard on an A64 just might not be worth it.

Include your command rate with your RAM timings when posting so we can get a better picture of it, e.g. 3-4-4-7-1T@2.75v.

That's if for now,

Gary

 

Gary:

Think Larry has a Zahlman Copper/Aluminum HSF and believe those temp spreads 54/37 were taken while running JV5 which isn't bad at all if thats the case - its designed to run CPU at 100% for several minutes to heat things up before running torture tests.

 

My bad - I thought those were idle temps.

 

We ALL Bad!!!....and that's good!!

Yes, my little Zalman is doing 2450rpm and keeping that CPU while stressed between 46 and 55*C, usually between 47 and 51.

I seem to have hit a wall. I have been stable at the following config-

280MHz/2525/1.3625v
3-4-4-7-1T@2.75v (stock is 2.5-3-3-6-1T@2.75v)
3x/875/166MHz (stock is AUTO/200MHz)

I have tried pushing the HTT to 285, 290, 295 and 300 at increasing vCore starting with 1.3625v stepped up to 1.425v. All of them increasing from slightly unstable to very unstable. I am not sure at this point what I could increase along with a high vCore that would be stable. It is not the HTT.

I cannot see bringing up the MemClock as at 200MHz I could not get the HTT past 240. It was only with the relaxed MemClock did I get even this far (280MHz).

My RAM is 2 sticks of the Corsair XMS TwinX c2 with a stock timing of 2-3-3-6 (seems to be some discourse on AMD being CAS2 or 2.5). I found it very unstable at CAS2.0. Got much more stable at 3.0. That is the max setting of my BIOS. I have seen a lot of 2-2-2-5 in groups on the net but that setting put me into Bios RAM Checksum Error... not fun. Aggressive timings just don't seem to be in the cards.

I am not sure if relaxing the timing even more would give any benefit. I beleive the XMS is supposed to be run at 1T and that is where I have left it. I have read the same about A64's not being very sensitive to RAM.

Thanks, Lar

 

Lar:

Maybe you've already been there but 3-4-4-8 may offer some relief.

There can often be some strange disparity even between the exact same memory. I use a lot of 2x256 2-3-2-5 OCZ Platinum in standard builds and have noticed considerable variance package to package. Also recall a particular DFI board a couple few years back where I tried XMS Pro 2-2-2-5 (the Black HS spreaders with leds) and it wouldn't run for squat. Same with XMS Platinum 2-3-2-5. The only thing that did the trick was XMS Platinum 2-2-2-5. This was an OC'ed 2500+ Barton - struggled with it for a few days.

All of this may sound off topic but if you bought from NewEgg, you can probably still RMA that memory and get some 2-2-2-5 for a few extra bucks. NewEgg is very good about this kind of stuff. I've also played around with some Patriot memory lately and am quite impressed even though I had never heard of it 6 months ago.

The main reason I mention this stuff is because I have some suspicion that your bottle neck is related to your current memory settings and I'm not real sure how to advise you at this point.

One final somewhat unrelated thought, If you have been running constantly for the last several days, you may want to go through some extended shut down periods which permit your CPU to cool completely to room temp between running sessions. This permits Artic Silver to further embed in the HSF and CPU surfaces and will promote better cooling. This type of break-in will eventually occur regardless, but you can accelerate the process in this manner - thats one reason why things will run faster after extensive use.

 

I'm not sure where this is coming from. As long as your variables are not of of whack, I would have thought that the HTT would have little or no effect on the clock speed of your RAM aside from their obvious mathmetical relationship.

And are you saying that you cannot run your Corsair at stock timings at all with any HTT settings? That ram of yours should be kicking butt. IMHO, the only impact that your HTT has on your memory is the clock speed at which it will run - based on all that divider stuff (which I cannot remember the exact formulas ATM). Once that clock speed is determined you should be able to tweak the snot out of it.

My own setup has problems when I get much higher than 270 HTT but that's not limiting me. One thing that has me curious is if I had bought 500Mhz DDR memory so I could run the RAM at 250Mhz or even higher. With an Athlon 64 you have that flexibility. I know the latencies on the higher speed ram are larger but it would have been cool to see if RAM clocked at ~270Mhz would result in greater performance. What I read seems to say "no" but I'd love to test it.

Lar, can you tell me what you maximum multiplier on your cpu is and what multiplier you are using right now? As I said, I can't really go much higher than 270 on the HTT. Maybe you could go more conservative on the HTT "bus ", bump your cpu multiplier up one notch (or half notch) and then see if you can clock your ram speed where it should be. I still don't get why you can't use the stock memory timings. As long as the RAM is not overclocked, you should be able to run it at least stock and probably more since it's Corsair XMS.

Gary

 

Here's what I now know...(or is it know now?)

Back to stock settings.... HTT can be pushed to 240. Relaxed 1T/2.75v timings (3-4-4-7 vs 2.5-3-3-6) did not increase HTT, it only changed the crash point from freezing at the "IDE Channel 1" statement to a full auto-reboot 2 seconds later in to the boot up.

The 200 to 240 bump brought my core speed from 1800 to 2425. The bump to 280 (along with the mem div drop to 166, and changing the HTT Mltpr from 4x to 3x) bought me an increase in core speed from 2425 to 2525. If 290 could have stuck, it would have brought it to 2625. Apparently, the timing changes have had no benefit.

EDIT: I should add (I think I just did) that bumping the vCore has not brought added stability to any scenario to date.

1. Unless there is something that I am missing, other than dropping the memory divider from its 200MHz@2.75v spec, I am out of options, right?

2. Is the current drop from 200 to 166 worth the 100Mhz core speed increase?

3. I understand that other CPU's have CPU dividers that go up to 12x and above. Should the built-in CPU divider of 9x for the Opteron 165 be considered as a significant limitation?


Lar

 

Yes, absolutely because the effective memory speed is actually higher. I need to be in front of my machine right now to provide more guidance but unless you change the memory divider to compensate for the increase in HTT (lets still mislabel it FSB), you are running your memory faster than 166x2. I'd still like to see you try 3-4-4-8 at 166 and perhaps with a little bump to your memory voltage (hold off on that initially).

I'm not so sure about the "no benefit ". I think you are headed into an interesting area that warrants further pursuit. ..... Here's where I would first push core voltage beyond 1.4 and with 3-4-4-8 timings at 166, consider a bump in memory voltage. not trying to sound ridiculous but 300+ wouldn't surprise me under these conditions.

I don't want you to burn anything up and don't think you are even close to having to worry about that as we are still talking very safe settings.

No, just means you have to learn a few new dance steps - Waltzing with Matilda isn't going to cut it any more.