Jump to content
HWBOT Community Forums

UCBench11: Tweaks and Luck

GENiEBEN

By GENiEBEN
in Benchmark software

 Share

Recommended Posts

  • Replies 207
  • Created
  • Last Reply

Top Posters In This Topic

Popular Days

Top Posters In This Topic

Popular Days

ObscureParadox Collaborator

ObscureParadox

Posted
Posted
It wasn't personal. It was a gentle nudge to get some help to resolve this bench. I hope you understand. :)

 

Oh I know, like I said, glad someone found it and helped get rid of it because I'm not always 100% sure what is bugged and what isn't.

Bogd4n Newbie

Bogd4n

Posted
Posted (edited)

I don't know how to recognize bugged runs, so please explain to me and I will check/delete mines. Or feel free to report bugged ones. Thanks.

 

Edit: I have read the first post, I think I got it...

Edited by Bogd4n
xsuperbgx Newbie

xsuperbgx

Posted
Posted

I will work on our submissions if I was sure about which ones should be removed. Look and see if passwords checked is is in ascending order for the given instruction set? For example:

 

http://hwbot.org/submission/2347268_edward2_ucbench_2011_4x_opteron_6180_se_5842_mpt_score

 

This one shows the second to last "passwords checked" is not in order and therefor this sub should be deleted?

or do I have to look at thread counts and only erase if #thread in result.txt are not divisible by actual thread count of cpu?

knopflerbruce Apprentice

knopflerbruce

Posted
Posted

Two and a half years ago I said that this was not normal (# of threads selected >> # of threads seen by windows => huge improvement). So, in the end I was right - but it doesn't help me now, does it... I think it's nuts to remove scores using a tweak that was legit for more than two years, it would've been better to just leave things as they were (given the time it's not a bug that destroys the competition as long as everyone knows about it) - like we've been doing with LOD for ages (technically it's not legit). Anyway, I'm not touching my own subs. This should've been taken care of about 900 days ago before people started posting bugged runs for hundreds of different CPUs and thinking legit tweraks were actually legit. Sorry if this causes extra work for good people, but there's no way I'm wasting my spare time deleting my own scores because other people screwed up. Period. I wasted enough time running the benchmark all for nothing :)

Mr.Scott Proficient

Mr.Scott

Posted
Posted (edited)
Two and a half years ago I said that this was not normal (# of threads selected >> # of threads seen by windows => huge improvement). So, in the end I was right - but it doesn't help me now, does it... I think it's nuts to remove scores using a tweak that was legit for more than two years, it would've been better to just leave things as they were (given the time it's not a bug that destroys the competition as long as everyone knows about it) - like we've been doing with LOD for ages (technically it's not legit). Anyway, I'm not touching my own subs. This should've been taken care of about 900 days ago before people started posting bugged runs for hundreds of different CPUs and thinking legit tweraks were actually legit. Sorry if this causes extra work for good people, but there's no way I'm wasting my spare time deleting my own scores because other people screwed up. Period. I wasted enough time running the benchmark all for nothing :)

 

Everybody's in the same boat bro.

Model attitude for a moderator here BTW. :rolleyes:

When a result moderator refuses to moderate even their own bugged subs, maybe it's time to let that ship sail.

Edited by Mr.Scott
CL3P20 Enthusiast

CL3P20

Posted
Posted (edited)

*the number of threads can be greater than the actual # of cores\threads CPU supports... that has never been an issue for this bench

*the number of selected threads needs to be evenly divisible by the # of CPU cores though... otherwise nearly 99% time produces a bugged result

 

If your benching 4c/8t, your likely going to end up with best results using thread sets like 32,48,64 ... always run the longer thread set last though.

 

 

---

-I understand this is quite possibly the only bench on the bot where you can select default settings and get a bugged result (OC aside)

-I understand that most people do not take the time to familiarize themselves with the rules associated with submitting a bench result

 

What I dont understand - what all the fuss is about, concerning this bench. If people would put time in on the bench to learn it better (isnt that what we do as benchers anyhow... ) they would come to learn what produces bugged results, as well as valid ones; quite easily too! Obviously this was not done thoroughly enough previously, which is why there is so much confusion regarding valid results. If anyone were to follow any of the info I have posted previously they would be hard-pressed to make a bug result !

 

**the only changes the owners of the software need to make; the bench needs to detect how many cores CPU has and disable 'oddly divisible thread sets' from selection.

Edited by CL3P20
Christian Ney Newbie

Christian Ney

Posted
Posted (edited)

Bugged results can happen with any benchmatk.

Thing is we discovered that when using higher thread count than the cpu total, result ends up "bugged".

 

"In essence, the problem is that the M (64) threads are forced to run on N (4) actual cores, which the OS disperses into an incoherent mess, that can't be timed exactly. Basically, M > N is a recipe for inaccurate value in the Time field, which I guess happens as soon as M > N; all results with this number of threads are compromised.

 

One thing to note is that this sort of issue can never happen if M ≤ N."

 

In short: Every single result with thread higher than cpu total is "bugged" (understand not correct/accurate).

 

There will be a v2 of the bench (there are several ways to fix this).

Now the idea was to delete the old ranking and only allow the v2 of the soft. But the author hasn't made the v2 yet so what we can do in the meantime is freeze and clean the current ranking, allowing only subs with default thread settings in the future the time v2 is out. When v2 is out since results will not be affected (score with v1 8t = score with v2 8t = score with v2 64t), they can be in the same ranking. V2 will be mandatory.

Edited by Christian Ney
I.nfraR.ed Collaborator

I.nfraR.ed

Posted
Posted

In fact every score (if the actual thread count is selected together with other "random" threads) contains the valid score too.

For example, this score: http://hwbot.org/submission/2292481_i.nfrar.ed_ucbench_2011_athlon_64_4600_x2_windsor_186.9_mpt_score

It's bugged, but #1 and #2 are ok and #2 is the valid score.

 

What about this one: http://hwbot.org/submission/2260455_i.nfrar.ed_ucbench_2011_athlon_64_4200_x2_manchester_152.4_mpt_score

Screenshot and images are lost, but the link shows that I've only used sse2 switch. Is it valid?

It's not my fault the images are gone.

Christian Ney Newbie

Christian Ney

Posted
Posted
In fact every score (if the actual thread count is selected together with other "random" threads) contains the valid score too.

For example, this score: http://hwbot.org/submission/2292481_i.nfrar.ed_ucbench_2011_athlon_64_4600_x2_windsor_186.9_mpt_score

It's bugged, but #1 and #2 are ok and #2 is the valid score.

 

That was discussed internally. When we see something similar, we will edit to the correct score.

 

What about this one: http://hwbot.org/submission/2260455_i.nfrar.ed_ucbench_2011_athlon_64_4200_x2_manchester_152.4_mpt_score

Screenshot and images are lost, but the link shows that I've only used sse2 switch. Is it valid?

It's not my fault the images are gone.

 

Screenshot is our fault yes. But you got the valid link so that's cool :)

I.nfraR.ed Collaborator

I.nfraR.ed

Posted
Posted (edited)

Ok, thanks. So I have 1-2 scores to remove then. Too bad we can't edit the scores ourselves.

Oh, another question. What about scores linked to competitions?

 

For example this one: http://hwbot.org/submission/2296871_i.nfrar.ed_ucbench_2011_core_2_e7500_%282.93ghz%29_310.4_mpt_score

 

I don't care about it, but think it's a valid point in case some submissions have only bugged thread counts.

Deleted my FX-9370 submission, #2 global 8-cores, but it had to go...

Edited by I.nfraR.ed
ObscureParadox Collaborator

ObscureParadox

Posted
Posted (edited)

I think I'm getting a firmer grasp of what a bugged result is and isn't now so I'll go through all my results now and delete those I believe are bugged.

 

I won't be bothering to bench the benchmark anymore though until we get a definite answer and what will be happening to the benchmark.

 

Edit, these are the ones I've deleted so far, the rest look ok to me :

 

http://hwbot.org/submission/2433312_obscureparadox_ucbench_2011_a10_4600m_331.4_mpt_score

 

http://hwbot.org/submission/2499465_obscureparadox_ucbench_2011_pentium_4_2.8_ghz_b_northw._(133_fsb)_87.7_mpt_score

Edited by ObscureParadox
Bobnova Newbie

Bobnova

Posted
Posted

Lost a E450 sub to this new development this morning. Oh well.

Noticed last night when it was reported that the only person in front of me was Geniben, also using threads tweak.

Noticed this morning that his score is gone too.

 

I appreciate the equality, HWBot mods. Thanks!

 

I also appreciate that the ruling has become much simpler and easier to look at.

 

 

 

For clarity and to make sure this post survives translation, there is no sarcasm contained in this post. I really do appreciate it.

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

×
 Share
Go to topic listing
×
×
  • Create New...