The AMD Phenom II OC'ers Club

[mastrdrver]

I was referencing to the MSI 890GXM-G65 I've got. Were you talking about that or the G70 board?

 

[eidairaman1]

The GD70 is getting some praise, also it seems to be a decent overclocker, the other point being- the Biostar TA890FXE w/e it is maybe the highest overclocking board out there, but biostar has had a rough past with bios updates apparently, along with ECS, then again when a brandnew product is launched it is expected for it not to be perfect. That happened to me back in 2002 with an Asus P4S8X board and a video card.

 

[wojo]

I just got the Gigabyte 890GPA-UD3H I still need a few more parts ( memory-PSU )before I put it back together and when I do I will let you know how it go's if no one else post's anything before I get everything I need, I should get the parts for my B Day this July 24th.
Thanks Bob

 

[mastrdrver]

This is ridicules. I can run 4x2gb with 4x ram multi and cas 7 but 3.33 multi and cas 6 doesn't boot. I'm done. Finishing up moving folders/files around and going over to my Gigabyte 790x. I'd rather deal with it than this thing.

I've also found some weird quirks with my X3 435 on memory benches. Is it normal for an Athlon II to be faster in memory benches with cas 7 1666 compared to cas 6 1333 given no other changes except ram speeds and their respective timings so as to be comparable to each other given the clock difference?

 

[crazyeyesreaper]

well i ordered all the water cooling parts i needed
barbs
8 feet of tubing 2 enermax magma turbine fans 1 danger den fillport extra fittings barbs etc

once it arrives ill start fiddling lets just hope everything works as planned im gonna be seriously bummed if this water cooling goes horribly wrong after dropping $110

 

[eidairaman1]

This is ridicules. I can run 4x2gb with 4x ram multi and cas 7 but 3.33 multi and cas 6 doesn't boot. I'm done. Finishing up moving folders/files around and going over to my Gigabyte 790x. I'd rather deal with it than this thing.

I've also found some weird quirks with my X3 435 on memory benches. Is it normal for an Athlon II to be faster in memory benches with cas 7 1666 compared to cas 6 1333 given no other changes except ram speeds and their respective timings so as to be comparable to each other given the clock difference?

Your talkin about memory bandwidth with timings.

 

[mastrdrver]

833 cas 7

667 cas 6

Once I enable ACC for hybrid to unlock it to a fully working Phenom II X4, everything drops slightly but latencies increase big time. That 40.5ns at 833 becomes 53.5ns and the 43ns becomes 55.9ns. Ran the program multiple times especially when unlocked because I couldn't believe it. Everest memory bench shows similar but with only a 3ns increase. Though, with that one you can see that the L3 and memory have almost exactly the same copy and read bandwidth speeds.

Worse still is I'm pretty sure my 955 couldn't match the bandwidth or latencies given same clocks and timings.

Running with just the default X3 435 clocks and cores, the system feels like a rocket sled with the overclock and higher memory speed. When unlocked, it and even my 955 on the MSI with 8GB of memory feel sluggish. That's with the memory, cpu-nb, and cores overclocked higher on the 955. The 435 is even using an EIDE drive as its main where as I have a 1tb black on the MSI.

I need to do more testing. Of course any program always performs better with the fully working Phenom.

I've never seen anyone mention it (so maybe I'm just catching up with everyone else ) but maybe the reason the Phenoms are sensitive to memory timings is because of the sluggish L3 cache and no so much the memory controller. Like I said though, I need to do more testing and most importantly get my 955 on that board again.

 

[AphexDreamer]

I'm so confused as to whats better for RAM low timing or higher Clock? Or is it just best to do whatever yields the best performance/Benchmark?

 

[Mussels]

I'm so confused as to whats better for RAM low timing or higher Clock? Or is it just best to do whatever yields the best performance/Benchmark?

both. say, CL6 1333Mhz would out perform CL9 1600Mhz - so get the best you can in both categories.

Higher MHz makes FSB OCing easier, so usually its best to reccomend that first.

 

[Velvet Wafer]

a few question aside, mussels :

did you ever saw how fast something under 1333 with cl5 is for phenoms, and if its more beneficial than higher bandwith with looser timings?
Do Phenoms really scale to the bottom of their memory controller?
And how does the NB relate to that?
Does NB speed really has to be x4 memory speed, when timings influence that fixed number,from what i know? And when they do, how big is the effect of the timings on the needed NB speed? Are there any charts existant, that test from cl5 to cl11 and ddr3-800 to ddr3-1600?
I looked for those infos quite a few times, but was unable to find something out!

I would count these questions to the search for the holy grail of phenom OCing knowledge. I
f i got answers to that, my Puzzle would be completed by about an additional 25%!

 

[Mussels]

a few question aside, mussels :

did you ever saw how fast something under 1333 with cl5 is for phenoms, and if its more beneficial than higher bandwith with looser timings?
Do Phenoms really scale to the bottom of their memory controller?
And how does the NB relate to that?
Does NB speed really has to be x4 memory speed, when timings influence that fixed number,from what i know? And when they do, how big is the effect of the timings on the needed NB speed? Are there any charts existant, that test from cl5 to cl11 and ddr3-800 to ddr3-1600?
I looked for those infos quite a few times, but was unable to find something out!

I would count these questions to the search for the holy grail of phenom OCing knowledge. I
f i got answers to that, my Puzzle would be completed by about an additional 25%!

i dont know specifics, i was just giving an example.

I have no idea as to the answers to those questions. i've only just mastered getting my ram to run at 1600Mhz.

 

[Velvet Wafer]

i dont know specifics, i was just giving an example.

I have no idea as to the answers to those questions. i've only just mastered getting my ram to run at 1600Mhz.

ah, alright! then nevermind!
i just hoped that you got an insight on Phenoms, that was denied to me up until now

 

[Mussels]

ah, alright! then nevermind!
i just hoped that you got an insight on Phenoms, that was denied to me up until now

my phenom runs at 3.8GHz at 1.45v (working on 4GHz later, its 99% stable)

that sums up my knowledge. i'm new to phenoms.

 

[Velvet Wafer]

my phenom runs at 3.8GHz at 1.45v (working on 4GHz later, its 99% stable)

that sums up my knowledge. i'm new to phenoms.

Ah, alright, i never knew you havent dangled with them before! my bad!

that 99% stability sounds much like it errors after a few hours of linx, am i correct?

 

[Mussels]

Ah, alright, i never knew you havent dangled with them before! my bad!

that 99% stability sounds much like it errors after a few hours of linx, am i correct?

i had instability that turned out to be ram related, it was 2 hours linpack stable but never tested further due to ram issues. no time to finish til next week.

 

[Velvet Wafer]

i had instability that turned out to be ram related, it was 2 hours linpack stable but never tested further due to ram issues. no time to finish til next week.

my secret sauce for finding definite stability (at least crunching,gaming and everyday stable,with the occasional monthly BSOD) :

first: 2 hours linx, for CPU, NB and some memory stability

500-800% memtest stable (not memtest x86, the windows based memtest)
with 90+% mem on all cores/threads (just open 4 memtests for a normal quad,2 for a dual,8 for an i7 and so on, and split all your mem between them)
for definite memory stability.
believe me, if its not fully stable, it will give you at least one error in that time.
that method worked for me both on AMD and Intel

 

[Mussels]

my secret sauce for finding definite stability (at least crunching,gaming and everyday stable,with the occasional monthly BSOD) :

first: 2 hours linx, for CPU, NB and some memory stability

500-800% memtest stable (not memtest x86, the windows based memtest)
with 90+% mem on all cores/threads (just open 4 memtests for a normal quad,2 for a dual,8 for an i7 and so on, and split all your mem between them)
for definite memory stability.
believe me, if its not fully stable, it will give you at least one error in that time.
that method worked for me both on AMD and Intel

2 hour linpack, 30 minutes of memtest (bootable) on test 5, then gaming.

to be honest, i've NEVER had a system crash or error on me, even with that small amount of testing... the errors tend to show up fast, or not at all.

If i ever run into issues down the track, i may well perform longer tests - but they've never shown up instabilities that werent shown already.

 

[eidairaman1]

I have had errors show up several hours later after gaming and testing, Win would crash.

 

[AphexDreamer]

I just fold and play games to test my stability. For instance I'd leave my PC on for folding over night and bam crash in the morning. I raised the Volts and then next night my computer was running, played some games that same day still running. I'm not sure if its 100% stable but if I don't see it crashing in games folding and benching (and its been on 4 days straight before) then I consider it stable.

 

[mastrdrver]

When running memtest86+ it is best to loop test 4 and 3 for a few hours, reboot and repeat. Do this on and off and you might find errors that don't show up normally and will even pass LinX and Prime95.

A quick and dirty stability test though is 32M Hyper Pi. If you can pass that then your getting close.

 

[mastrdrver]

a few question aside, mussels :

did you ever saw how fast something under 1333 with cl5 is for phenoms, and if its more beneficial than higher bandwith with looser timings?
Do Phenoms really scale to the bottom of their memory controller?
And how does the NB relate to that?
Does NB speed really has to be x4 memory speed, when timings influence that fixed number,from what i know? And when they do, how big is the effect of the timings on the needed NB speed? Are there any charts existant, that test from cl5 to cl11 and ddr3-800 to ddr3-1600?
I looked for those infos quite a few times, but was unable to find something out!

I would count these questions to the search for the holy grail of phenom OCing knowledge. I
f i got answers to that, my Puzzle would be completed by about an additional 25%!

Let me sum up most of this with this: cas 5 at 1333 is faster than cas 7 at 1600, always on a Phenom. I believe it to be related to the "slow" L3 cache that is used on the chip and not the memory controller.

On how fast does your cpu-nb need to be, you need it fast enough so it doesn't bottleneck the ram. If L3 bandwidth is below ram, you're leaving performance on the table and you need to up the cpu-nb.

 

[erocker]

The NB should be about twice the speed of the RAM. Anything past that is not beneficial and the RAM will start to bottleneck the CPU's L3 cache.

 

[mastrdrver]

You're talking rated ram speed right E (i.e. 1333 not 667)?

 

[TheMailMan78]

The NB should be about twice the speed of the RAM. Anything past that is not beneficial and the RAM will start to bottleneck the CPU's L3 cache.

That means I have one BIG bottleneck.

 

[Velvet Wafer]

That means I have one BIG bottleneck.

i told you so once, when you first posted a cpu-z screen!
seems like you have overseen it,back then!