New WR MAXXMEM with CORSAIR GTX2

[hiwa]

that's nice guys, if i am not wrong u can bench cl6 @ same freq under cold?

[_mat_]

that's nice guys, if i am not wrong u can bench cl6 @ same freq under cold?

We hope so.

But CL6 is also very demanding for the cpu, so it will depend very much on how good our sandy handles the memory.

[miahallen]

We hope so.

But CL6 is also very demanding for the cpu, so it will depend very much on how good our sandy handles the memory.

Very true....good luck guys

[hiwa]

We hope so.

But CL6 is also very demanding for the cpu, so it will depend very much on how good our sandy handles the memory.

for sure u are bclk limited but Cl6 with tight timing giving you a good boot too

[M.Beier]

1140mhz 6-6-5 might be too hard, I keep fingers crossed for you guys! I fear it might start performing less effecient due level of errors around ~1130

[_mat_]

1140mhz 6-6-5 might be too hard, I keep fingers crossed for you guys! I fear it might start performing less effecient due level of errors around ~1130

I never got past 2340 MHz CL6 with my GTX2 modules. Not on Nehalem, Lynnfield or Sandy Bridge ... not even with -75 °C and more than 2 volts.

I did not get any performance scaling problems yet ... only when a module decides to shutdown, because of voltages/cold/too tight timings. But that only happened on Nehalem. The performance goes back to dual channel in that case.

[Hondacity]

is the red part of the ram pot ...copper?

[rewarder]

is the red part of the ram pot ...copper?

The base plate is made of copper, the "pot" itself, if you want to say so is made of aluminium. If you like ... there is a small review of it:
http://www.ocaholic.ch/xoops/html/mo...l_lang=english

[splmann]

Is this a joke?

__________________

Download Software

What is your Problem ?

[aslanwas]

is this a joke ?

very very congrattsz on tht nice score, what Ram freq are you running, i can maximize the screenshot

[rewarder]

So Roger ... tell us ... what have you done yesterday?

[_mat_]

Is this a joke?

__________________

Download Software

Thanks for the amazing software!

So Roger ... tell us ... what have you done yesterday?

Did he go for cl6?

[rewarder]

Thanks for the amazing software!

Did he go for cl6?

I don't know ... I tried to call him today but he didn't respond till now

[SF3D]

More I look at your score, more convinced I am, that your 32,5ns latency test is bugged. I can produce similar / better scores.
If you bench same settings like 10x one or two of those will give you bugged results, which look almost normal, but are not real.

Looka at this new screen, where I tweaked subtimings a lot and got nice direct scaling from my previous score. I can reproduce this as many times I want.
Also aida is scaling directly with the memory speed and I have not seen any bugs in it even it is runned multiple times in a row?



So, I think that is the highest real score so far.

[chew*]

I'm not saying your run was bugged but i do have to agree that maxmem can definitely just slightly bug latency, I had to do my runs a few times and on several boots before i was content that mine were not bugged as I did get some barely bugged latency runs myself.

[SF3D]

I'm not saying your run was bugged but i do have to agree that maxmem can definitely just slightly bug latency, I had to do my runs a few times and on several boots before i was content that mine were not bugged as I did get some barely bugged latency runs myself.

Was the start of your reply ment to go for splmann or me?

My run is not bugged, like I tried to explain and you explained too.
The difference might be small, but scaling must happen in all test, not just latency.

That is why I would like to see splmann to fix his score himself

[chew*]

Was the start of your reply ment to go for splman or me?

My run is not bugged, like I tried to explain and you explained too.
The difference might be small, but scaling must happen in all test, not just latency.

That is why I would like to see splman to fix his score himself

My reply was to splman, and agreeing with you that it can bug slightly.

[Leeghoofd]

Now we need Massman to give points for Maxxmem scores... good job all, drooling stuff ( out of my voltage league :p )

[_mat_]

More I look at your score, more convinced I am, that your 32,5ns latency test is bugged. I can produce similar / better scores.
If you bench same settings like 10x one or two of those will give you bugged results, which look almost normal, but are not real.

Looka at this new screen, where I tweaked subtimings a lot and got nice direct scaling from my previous score. I can reproduce this as many times I want.
Also aida is scaling directly with the memory speed and I have not seen any bugs in it even it is runned multiple times in a row?

So, I think that is the highest real score so far.

We also got (much) higher scores, that we ignored because of too low latency. I don't know if this run was reproducible, but I am sure splmann can answer this. But now that we have your values to compare with, I guess you're right ... the run could be bugged after all.

Any way, congrats SF3D for your achievements. 107 bclock with that cpu frequency is just

Maybe we should find a solution to make our scores compareable again. I don't now ... like 3 screenshots of maxxmem, that have to be done at different times (time is display in the maxxmem window, so that would be no problem). Because when it bugs, the score is very unsteady and not reproducible. Or we ignore Maxxmem and use Aida 64 ... to make it directly compareable, we have to calculate some score out of it. Maybe ask the developers to work on that.

[SF3D]

Yes, you can not "tweak" timings to get only gain in latency test. Also the CPU speed was very low for the latency, cause there is correlation between those two.

So Matthias, if you also think that current higest score is bugged, please ask splmann to remove it? I am sure he can run it again with ln2 on memory etc.
We all should post just the correct scores from now on. We can compare the data from these screenshots we have here and in HWbot.org. Scaling is qute direct and good, so we easily know if some score is out of the line.
My next step will be ln2 as well, cause I will target 6-6-6 settings.

Keep up the good work guys and no hard feelings hopefully.

[youngpro]

i think its fair to say maxxmem is a fairly buggy bench and it is not always as clean cut as this is bugged or not,

if we want to accept a bench like this into our community and use it in rankings then i think we need to accept he results that are bugged within the realms of actual capacity of the platform... this result is a plausable result, sure it is slightly out of line but it is something that is possible..

this is just a by product of what we are doing, pushing benchmarks and hardware to places they were not designed to go, we will always find some results like this

[SF3D]

i think its fair to say maxxmem is a fairly buggy bench and it is not always as clean cut as this is bugged or not,

if we want to accept a bench like this into our community and use it in rankings then i think we need to accept he results that are bugged within the realms of actual capacity of the platform... this result is a plausable result, sure it is slightly out of line but it is something that is possible..

this is just a by product of what we are doing, pushing benchmarks and hardware to places they were not designed to go, we will always find some results like this

Well, we can not go in to that. We have to see how some benchmark behave and learn how it scale with timings and clocks. When we have enough data to evaluate, it is easy to see the odd results.

Personally I think that the score should be removed from Hwbot and all the newsites around the internet could write their news right I was amazed to see how much talk there was from this score, which is bugged after all.

But, if all will agree, that we can post scores which are slighty off from scaling, I will put my 31.9ns score in to bot

P.S I wont do that for real

[Astennu]

Compared to some powerchips on 6-9-6 or something the values don't seem to be lots faster. But the latency sure is and that is most important for MaxxMem to get higher scores. For some reference you can compare dinos 2220 Mhz CL7-8-7 with Hypers and Pt1t's 2200+ MHz CL6 with PSC.

splmann, it was a pleasure to "remote bench" with you. Here are a few screens from my perspective

Well. The second one also has a higher CPU speed (300 MHz higher). Higher CPU speed on SB lowers latency quite a bit.
But the results are not stable. When i do multiple runs sometimes one is about 1,5ns slower. And yesterday i got 1,2 ns lower by tweaking the sub timings. These settings down show the whole picture. You would need to see a shot of all the dram timing settings and a more reliable test (lets say 5 runs and take the average of those 5) before you can say that 6-9-6 is not much faster.

[Monstru]

i think its fair to say maxxmem is a fairly buggy bench and it is not always as clean cut as this is bugged or not,

MaxxMem still needs some work, if you ask me. We cannot judge anyone for pushing hard and trying to get the best score, sometimes the program itself is too wacky. There is slight bug, where things can be argued, and way off bugged, where it is very clear there is a bug. Problem is, I don't think it is fair to the overclocker's work to try and spot the "slight bugged" ones, I think the real solution is to work some more on MaxxMem to be sure this kind of things don't happen anymore. And if we talk about repeatable results, here's a bunch of results from me...Not very healthy if you ask me

[miahallen]

Any serious OCer wouldn't make a big deal over a CPU-Z bugged run, and wouldn't post it at the bot either. Even though CPU-Z is usually accurate, there are situations where Bloomfield bugged to the moon.

The situation here is more difficult, and if bugging is occuring it's less noticable, but a slight bug is still a bug, and once confirmed...I think a self respecting OCer wouldn't/shouldn't post it any more than a bugged Bloomfield CPU-Z score. If the score was posted and the OCer was not sure, once identified they should then remove the score.

But, if all will agree, that we can post scores which are slighty off from scaling, I will put my 31.9ns score in to bot

P.S I wont do that for real

@ splmann, keep pushin mate, you're obviously doing very well ....but if it's bugged, you should try again for a legit run

MaxxMem still needs some work, if you ask me. We cannot judge anyone for pushing hard and trying to get the best score, sometimes the program itself is too wacky. There is slight bug, where things can be argued, and way off bugged, where it is very clear there is a bug. Problem is, I don't think it is fair to the overclocker's work to try and spot the "slight bugged" ones, I think the real solution is to work some more on MaxxMem to be sure this kind of things don't happen anymore. And if we talk about repeatable results, here's a bunch of results from me...Not very healthy if you ask me

^^^ this bench is obviously not ready for prime-time (HWBot points)