Oh my AT8-32X

csimon

I just wanted to start recording some stability tests so that I'd know where I left off. I'm running out of tablets to jot notes on.

So far I've done a 10 hour successful overnight run of large fft's (mem) stress at the following settings:
ExtClk: 313MHz
PCIEClk: 100
VCORE: 1.425
NB 1.8V: 1.90
NB 1.2V: 1.30
DDRV: 2.65
HTV: 1.25

Now I'm at this running 10 hours successful so far:
ExtClk: 315MHz
PCIEClk: 100
VCORE: 1.425
NB 1.8V: 1.90
NB 1.2V: 1.30
DDRV: 2.65
HTV: 1.25

Anyway specs are in my sig and here are some detailed A64Info settings that I will begin updating as I go. I will get my bios details later when/if I am forced to reboot.

Does anyone know why the "settings" tab option will only let me set my vcore to 1.40v max? I understand that this is max but why can't I raise it beyond if my bios allows it?

Also, I've gotten some help from gdogg over at mushkin about settings for dram and so far it's making a nice difference. I'll know just how much difference it makes as I approach 260MHz & over.

Thanks

csimon

Nice and stable @2833/11=257.5!
ExtClk: 315MHz
PCIEClk: 100
VCORE: 1.425

Now to bump it up a notch and stress it ...10 hours!
2844/11=258.5
ExtClk: 316MHz
VCORE: 1.425

csimon

Nice and stable @2844/11=258.5!
ExtClk: 316MHz
PCIEClk: 100
VCORE: 1.425

Now to bump it up a notch and stress it ...2 hours!
2856/11=259.6
ExtClk: 317MHz
VCORE: 1.425

Edit:\\ 317 failed after 2 hours.

csimon

Now to tune down HT & voltage.
2808/11=255.3
ExtClk: 312MHz
VCORE: 1.4000

nonstop301

You're looking good at over 3,1 GHz then

Nice job

I'm nice and steady at 2,0 GHz thanks to your good help and there's always that reserve to 2.4 GHz if I need some more juice

csimon

nonstop301 wrote:

You're looking good at over 3,1 GHz then

Nice job

I'm nice and steady at 2,0 GHz thanks to your good help and there's always that reserve to 2.4 GHz if I need some more juice

Sorry for the confusion. That's 2.8ghz ...the 312 is HTT. 312x9=2808.
The HTT is reported this way in the uGuru software.

Glad you're doing well with your overclock nonstop!

csimon

Up the HTT a notch.
2820/11=256.4
ExtClk: 313MHz
VCORE: 1.4000

nonstop301

My bad there csimon I misread the value you recorded and it is indeed the HT in MHz

mirage

csimon wrote:

Nice and stable @2844/11=258.5!
ExtClk: 316MHz
PCIEClk: 100
VCORE: 1.425

Now to bump it up a notch and stress it ...2 hours!
2856/11=259.6
ExtClk: 317MHz
VCORE: 1.425

Edit:\\ 317 failed after 2 hours.

Can you tell what component is failiing, RAM or CPU? Maybe you can try HTT=317MHz again with memory divider at 14.

csimon

mirage wrote:

Can you tell what component is failiing, RAM or CPU? Maybe you can try HTT=317MHz again with memory divider at 14.

I'm not sure but I think it's ram failing at 317. I'm not sure if the 133 divider is equal to 14 or not but that's my only divider option next. For some reason they left off the 150 divider on this board. They only just added it to the Manta bios a few weeks ago and it was present on my A8R-MVP. So I'll just use a lower CAS latency to go along with that to make up the difference.

So far large fft prime stable at 9 hours.
2820/11=256.4
ExtClk: 313MHz
VCORE: 1.4000

After this run I'll try 315@1.4v and if it fails I'll bump up to 317@1.45v and throw down the 133 mem divisor.

edcentric

So, how does it fold at this speed?
Looks like a killer.

csimon

edcentric wrote:

So, how does it fold at this speed?
Looks like a killer.

Folds great especially with those 116x and other fast high pointers. It's been rock stable until I reach around the 319/320 mark so I'm working my way up slowly findng all the kinks and so far so good. Lowering the vcore isn't doing much in the way of temps cause the voltage flexes and droops pretty far but mostly flexes on the vcore. All of the other voltages droop mostly so I've had to bump them all as slight as possible. This in turn really heats up the PWM so I have a fan pointed right onto it and it's ok as long as I keep it around 320ish and just use f@h for stress.

csimon

Now going for 315@1.4v large fft prime stable.
2832/11=257.5
ExtClk: 315MHz
VCORE: 1.400

I suspect that 315 will fail at 1.4v but I already know that it is stable at 1.425v.
So if this is the case then:
312-313 = 1.400v
315-316 = 1.425v
and possibly
317-319 = 1.45v
and so on possibly ...we'll see.
314 & 318 are skipped over because of the "rounding up" of the decimals.

I have a few tricks to try as I reach the upper limits of the HT. If 9ns max async doesn't work I'll kick down the mem divider to 133 and up the cas latency ...I suspect cas 2. A few other variables can be lowered as well when running at this lower frequency.

csimon

These are the DRAM settings that I've been using all along:

CL: 3.0
tRCD = 3
tRP = 2
tRAS = 8
CMD = 1T
Interface = 128bit
tRC = 7
tRFC = 15
tRRD = 2
tRWT = 3
tWR = 2
tWTR = 1
tWCL = 1
tREF = 4708
DQS Skew Control = off
DRAM Skew Value = 0
Digital Locked Loop = Enabled
Max Async Latency = 8ns
DRAM Drive Strength = Weak
Read Preamble = 5.0ns
DRAM Data Drive Strength = L3 (Reduce 15%)
Idle Cycle Limit = 16 cycles
32 bit Granularity = Disabled (8 bursts)
Dynamic Counter = Disabled
R/W Queue Bypass = 16x
Bypass Max = 07x

Also ...vdimm is always set at 2.65v

This will need to be tweaked again when I kick down the frequency.

csimon

csimon wrote:

I suspect that 315 will fail at 1.4v but I already know that it is stable at 1.425v.
So if this is the case then:
312-313 = 1.400v
315-316 = 1.425v
and possibly
317-319 = 1.45v
and so on possibly ...we'll see.
314 & 318 are skipped over because of the "rounding up" of the decimals.

HTT 315@1.4v did in fact fail.

Now going for 317@1.45v large fft prime stable.
2832/11=259.6
ExtClk: 317MHz
VCORE: 1.450

mirage

csimon wrote:

These are the DRAM settings that I've been using all along:

CL: 3.0
tRCD = 3
tRP = 2
tRAS = 8
CMD = 1T
Interface = 128bit
tRC = 7
tRFC = 15
tRRD = 2
tRWT = 3
tWR = 2
tWTR = 1
tWCL = 1
tREF = 4708
DQS Skew Control = off
DRAM Skew Value = 0
Digital Locked Loop = Enabled
Max Async Latency = 8ns
DRAM Drive Strength = Weak
Read Preamble = 5.0ns
DRAM Data Drive Strength = L3 (Reduce 15%)
Idle Cycle Limit = 16 cycles
32 bit Granularity = Disabled (8 bursts)
Dynamic Counter = Disabled
R/W Queue Bypass = 16x
Bypass Max = 07x

Also ...vdimm is always set at 2.65v

This will need to be tweaked again when I kick down the frequency.

You can try using systool to enable additional memory dividers on the fly. Our systems are almost identical, I also have 2x1GB DDR500 but timings are 3-3-3-8-2cmd and the brand is Ballistix. I was not able to use 1cmd but I admit I did not try aggressively. Do you want to benchmark against each other ?

csimon

mirage wrote:

Do you want to benchmark against each other ?

Sure throw some benchies down and when I get done testing I'll do the same. I've got about 6-8 hours left.

I like to bench with 3dMark06 and PCWizard2006 and sometimes Aquamark. PCWizard2006 has a good mem bench that I gauge my speed and bandwidth. Also the cpu bench is pretty good too.

csimon

Test failed after 2.5 hours. I'm going to retest @ 9 max async latency. So while I'm down I ran a few benchmarks. I can do some cpu-z shots if you want. This was at 317HTT@1.45v and 260MHz mem.
Not too bad but the video scores on 3dMark06 are really sucking with the newest drivers. I should be in the mid 1800's on those scores. I'll look for another driver soon as I'm dusting out the system.

mirage

csimon wrote:

Test failed after 2.5 hours. I'm going to retest @ 9 max async latency. So while I'm down I ran a few benchmarks. I can do some cpu-z shots if you want. This was at 317HTT@1.45v and 260MHz mem.
Not too bad but the video scores on 3dMark06 are really sucking with the newest drivers. I should be in the mid 1800's on those scores. I'll look for another driver soon as I'm dusting out the system.

I will post my benchies tonight when I can access the console. But 3DMark06 will suck with X800XL. Is your opty165 stable at 3 GHz when you lower the RAM frequency?

csimon

mirage wrote:

Is your opty165 stable at 3 GHz when you lower the RAM frequency?

Idunno I tried it once but it wasn't orthos stable so I haven't tried it again. This time I plan on going up as high as I can til orthos gives out or temps get too high. Then I'll probably leave it there and fold my ass off!
Just before I cut the mem frequency I ran pcwizard2006 one more time. The results were a little better despite the fact that I lowered the max async latency to 9ns.

mirage

csimon wrote:

Idunno I tried it once but it wasn't orthos stable so I haven't tried it again. This time I plan on going up as high as I can til orthos gives out or temps get too high. Then I'll probably leave it there and fold my ass off!
Just before I cut the mem frequency I ran pcwizard2006 one more time. The results were a little better despite the fact that I lowered the max async latency to 9ns.

Here are my benchies, I am about 2-3% behind you. But considering I have HTT=311, I think the results are consistent.

csimon

mirage wrote:

Here are my benchies, I am about 2-3% behind you. But considering I have HTT=311, I think the results are consistent.

The cpu bench is really impressive. I think the mem bench could be improved with the 1T command rate.
This is just a preseption that I have ...but I think there may be a tradeoff between ram bandwidth/speed and cpu speed. I've noticed it on my previous mobo but it's not as prelevant on this one. It's just a hunch.

csimon

I've gone down to the 133 divider since 317 HTT failed 260MHz@2.65v. I really think I could have gotten it to work with stock timings but I'll deal with that later. Here are the latest settings ...and I'll move timings closer to stock after this run.

CL: 2.0
tRCD = 3
tRP = 2
tRAS = 7
CMD = 1T
Interface = 128bit
tRC = 7
tRFC = 14
tRRD = 2
tRWT = 2
tWR = 2
tWTR = 2
tWCL = 1
tREF = 1032
DQS Skew Control = off
DRAM Skew Value = 0
Digital Locked Loop = Enabled
Max Async Latency = 8ns
DRAM Drive Strength = Weak
Read Preamble = 7.0ns
DRAM Data Drive Strength = L3 (Reduce 15%)
Idle Cycle Limit = 16 cycles
32 bit Granularity = Disabled (8 bursts)
Dynamic Counter = Disabled
R/W Queue Bypass = 16x
Bypass Max = 07x

2/2 - 7 hours orthos large fft stable.
HTT = 317
VCORE = 1.450

That'll do I'm anxious to get moving up.

csimon

tRAS = 8
HTT = 319
VCORE = 1.450
2868 / 14 = 205

Done!

Now for this!

HTT = 320
VCORE = 1.450
2880 / 14 = 206

csimon

I hope the temps can hold out just a little bit more for testing.

HTT = 320
VCORE = 1.450
2880 / 14 = 206

csimon

HTT = 320
VCORE = 1.475
2880 / 14 = 206

This is too strange. I up the voltage and the temps drop. I guess it was straining something to make it hot idunno.

Trying to relax some timings just to take it out of the equation ...this is what I've come up with so far. I must admit I really don't know a whole lot about relaxing the ram so if you have any suggestions they would be appreicated.

CL: 2.5
tRCD = 3
tRP = 2
tRAS = 8
CMD = 1T
Interface = 128bit
tRC = 8
tRFC = 12
tRRD = 2
tRWT = 4
tWR = 2
tWTR = 2
tWCL = 1
tREF = 1032
DQS Skew Control = off
DRAM Skew Value = 0
Digital Locked Loop = Enabled
Max Async Latency = 8ns
DRAM Drive Strength = Weak
Read Preamble = 7.0ns
DRAM Data Drive Strength = L3 (Reduce 15%)
Idle Cycle Limit = 16 cycles
32 bit Granularity = Disabled (8 bursts)
Dynamic Counter = Disabled
R/W Queue Bypass = 16x
Bypass Max = 07x

csimon

See ...this makes no sense. I raise the voltage and the load temps drop. Look at the core temps a few posts above and compare. The above is 1.450v and this is at 1.475v.
This is what they mean by "Don't look a gift horse ....".

mirage

csimon wrote:

See ...this makes no sense. I raise the voltage and the load temps drop. Look at the core temps a few posts above and compare. The above is 1.450v and this is at 1.475v.
This is what they mean by "Don't look a gift horse ....".

Maybe the ambient temperature has dropped. If the room temperature changes even a couple of degrees, it directly affects the CPU temperature. At least, this is my experience.

mirage

csimon wrote:

Trying to relax some timings just to take it out of the equation ...this is what I've come up with so far. I must admit I really don't know a whole lot about relaxing the ram so if you have any suggestions they would be appreicated.

Leaving all the rest at default BIOS values the followings usually give me the most relaxed timings for stability. Hope this helps
CL= 3
tRCD = 3
tRP = 3
tRAS = 8
CMD = 2T

csimon

mirage wrote:

Leaving all the rest at default BIOS values the followings usually give me the most relaxed timings for stability. Hope this helps
CL= 3
tRCD = 3
tRP = 3
tRAS = 8
CMD = 2T

Right now I'm at this:
CL= 3
tRCD = 3
tRP = 2
tRAS = 8
CMD = 2T

If it fails I'll definitely try that though. Thanks!

csimon

This is a strange anomaly for sure.
I tried to replicate my old settings from all of these notes I've been keeping and absolutely nothing would match. I could'nt get anything to work then all of a sudden I remembered that I had changed something in A64Info. Under the "Settings" tab the "New" voltage defaults to 1.35 and I remembered changing this to 1.4 although I was already at 1.4 in bios.

BINGO! Now everything seems to be successful as I had kept in these notes. I do notice that when this is not set the voltages reported in uguru seem normal and stable but when I make the change in A64Info I get overvolting. So this must account for the being able to lower the voltages in uguru to get the same results. I'd say I get about a +.05 overvolt which makes sense going from 1.35 to 1.40. But it actually goes from 1.40 to 1.45.

I'll have to study and test these results a bit more just thought I'd mention it in case I need a reference.