Suggestion - Maxxmem2 for Hwbot points

[Morphling]

Hi to everyone.

 

First of all, I must apologize for my bad English. I'm from Portugal, and I'm more or less new here.

 

I would like to suggest that Maxxmem2 should be quoted for HWBot points. This is an amazing 2D Benchmark, since it doesn't depends only on how much can you give with CPU and RAM, but also lots of RAM tweak and also OS tweaking.

 

One of the problems that anyone who knows how the benchmark works are the latency bugs. Of course there are lots of people who posts their results with bug, some thinking that it's normal, and others just for cheating.

 

The point is, if this is a quoted benchmark, then all the users who run this will always pay extreme attention on other users results, making the reports to staff when one of this situations happens.

 

I think that this is a benchmark that every 2D addicts just love. It's easy to run, and realy hard to make more and more points.

 

This topic is to have your view point, and also, of course, to try to go further with this

 

Come on, let us have one more 2D benchmark to fight. There are some great overclockers with realy good scores there, like Chew, Kintaro, Billy-the-Kid, MSIMAX, SF3D, slamms, and others that have their scores submited.

 

Please say what do you all think about this.

 

Regards to all

 

Morphling

Edited September 20, 2011 by Morphling

[Hyperhorn]

Hm, if the latency bug is the only reason this benchmark isn't rewarded with points, why not let make a hwbot version with latency measured shown but without affecting the final score, so all that counts would be the copy/read/write results?

[Morphling]

Hm, if the latency bug is the only reason this benchmark isn't rewarded with points, why not let make a hwbot version with latency measured shown but without affecting the final score, so all that counts would be the copy/read/write results?

 

Hi Hyperhorn, and thanks for participate on this thread.

 

Well, the point is that the latency is the "soul" of this benchmark. The "secrets" of this benchmark are on how you can make the latency as lower as possible, of course, without bugs.

 

Without the latency, this benchmark would be resumed to those who could get CPU and RAM as high as possible. That will give your maximum Read/Write/Copy, almost without touching on any other timings.

 

As I said before, if this would be a quoted benchmark, latency bugs are easily reported.

 

For example, and I'm sorry for this user, this is a typical latency bug:

 

http://hwbot.org/submission/2194818_diabolo_80_maxxmem_ddr3_sdram_maxxmem___read_mbytesec._30651

 

Because when compared to Kintaro's score:

 

http://hwbot.org/submission/2156559_kintaro_maxxmem_ddr3_sdram_2847.9_marks

 

We can easily see that Kintaro's settings are much better, and when comparing memory speed and timings, diabolo_80 shouldn't stand chance. He could actualy have at least the same copy/read/write, once he has almost plus 60mhz on the CPU, but never even similar latency. You can see that in my own score:

 

http://hwbot.org/submission/2208064_morphling_maxxmem_ddr3_sdram_2719.4_marks

 

I have more mhz on CPU and RAM, but still worse timings, so my latency is beated by Kintaro's settings. It's easy to find these bugs.

 

Come on, it would be great to have this benchmark to fight.

 

Regards to all

[knopflerbruce]

But... the latency of diabolo's score is .2s slower, which seems correct.

 

The bugs are just too hard to spot - we've had situations where we're talking about like 0.4ns - too little to be sure if you ask me. Removing the latency, and have like a combined score of the other 3 tests is an option, but a good solution? I dunno.

[Morphling]

But... the latency of diabolo's score is .2s slower, which seems correct.

 

The bugs are just too hard to spot - we've had situations where we're talking about like 0.4ns - too little to be sure if you ask me. Removing the latency, and have like a combined score of the other 3 tests is an option, but a good solution? I dunno.

 

Thank you for participating knopflerbruce

 

Ok, you are right For me, it would be reasonable to have two solutions:

 

- Either some latency bugs could be tolerated, when talking about 0.5ns~1ns;

 

or

 

- Remove the latency, and have like a combined score of the other 3 tests, as you have said.

 

But it must be quoted

 

Regards to all

[Sir_Mike]

I think that Maxxmem should count points, and I would leave the benchmark as it is now, with the latency included.

 

There are several benchmarks that are on HWbot and do also have lots of bugs too, like 2001, or aquamark.

 

My vote is positive, Maxxmem need to give points

 

Thank you for the iniciative Morphling.

 

Regards

[Billy-The-Kid]

Hello everybody!

This benchmarl program has already served wiht benchmark qualifier and evaluation in several contests and give recognition on the world of overcloking! But there exist the the bug of latency.....

 

In my opinion should only be assigned points based on the values ​​of reading, to avoid this bug! And I think all the overclockers on this community already ran this so that more than time to be recognized as such!

[knopflerbruce]

There are more bugs - but maybe the phenom NB bug has been fixed?

[Morphling]

Ok, so we are getting something here That's a start.

 

Well so I think that everyone is now looking more at the possibility of using Maxxmem2 without latency.

 

For me that's perfectly fine then. It's better than continuing to submit briliant scores without any beneficts.

 

So I ask, how can we make this go further and have Maxxmem2 giving points before September ends?

 

Come on let's do this another 2D benchmark to use, and a good one. There is not actualy any specific benchmark that is so directed to Memory, and HWBot realy needs one. The closest thing we got to use memory performance is SuperPi 32M, but no matter how good your RAM is, you are always overwhelmed from the stronger CPU's that can have more mhz. On Maxxmem things are not that way Of course CPU helps, but it's realy not the most important thing.

 

Please give me feedback

 

Regards to all and thanks for participating!

[pipes]

This program reads the bad memory bandwidth, I with a triple channel at 2100 mhz I only 20xxx mb/s?