UCBench11: Tweaks and Luck

[havli]

I just noticed this thread... and I am not sure I understand the issue completely.

 

1.

Pwds. Checked are in ascending order

So the "Pwds. Checked" column must contain all the numbers strictly in ascending order - i.e. the lowest number is first in the list and the highest one is last. No high numbers "in the middle".

 

 

2.

Make sure that thread selected divided to total available threads (cpu threads) is not an odd number.

The thread count could be like this?

dual core CPU -> 2, 4, 8, 56, 60, 64

quad core CPU -> 4, 8, 56, 64

 

and how about single core CPUs? Does this bug apply for them as well?

For example this score is done on single core http://hwbot.org/submission/2367371_havli_ucbench_2011_duron_1.0ghz_24.3_mpt_score Is it considered as bugged because of the Pwds. Checked number?

[Splave]

so what if you only run 1 setting? automatically accepted?

[Mr.Scott]

Except it never is bugged on default settings, I know what you meant.

Easy to control then if that's true. Bench must be run at default settings only. What's the issue? Default or get rid of it.

[ObscureParadox]

Easy to control then if that's true. Bench must be run at default settings only. What's the issue? Default or get rid of it.

 

 

I agree, not everyone is going to fully understand the whole bugged or not bugged situation and the only way to guarantee good results every time is to force everyone to run stock.

[Moose83]

Yes only Windows tweaking

 

And what about non retail OS like Neptune

[Strunkenbold]

Easy to control then if that's true. Bench must be run at default settings only. What's the issue? Default or get rid of it.

 

+1

 

But also keep in mind that, when we only accept default thread count, probably 40-50% of results have to get blocked. And than we have to think about if it makes sense to keep this bench.

[Mr.Scott]

+1

 

But also keep in mind that, when we only accept default thread count, probably 40-50% of results have to get blocked. And than we have to think about if it makes sense to keep this bench.

Totally agree. :celebration:

[havli]

Well, if Ucbench will be removed from HWbot, we should get some other benchmark (with HW points) as a replacement. It shoud be capable of running on Pentium class CPU (without MMX / SSE).

[Mr.Scott]

Well, if Ucbench will be removed from HWbot, we should get some other benchmark (with HW points) as a replacement. It shoud be capable of running on Pentium class CPU (without MMX / SSE).

Again, I agree.

[ObscureParadox]

Again, I agree.

 

 

Not sure if I do, we should be looking more towards benchmarks which really do test the likes of the 4960X fully Multi-threaded to give points to. Old benchmarks like wPrime 32M really aren't good enough any more as the CPUs are just too good for the benchmark. If we concentrate on getting benchmarks to run on older platforms then we run the risk of not furtureproofing for the latest chips.

 

Imho we need a new benchmark which was as hard as SuperPi 32M for the old P4 CPUs but is fully multi-threaded.

[Calathea]

Well, if Ucbench will be removed from HWbot, we should get some other benchmark (with HW points) as a replacement. It shoud be capable of running on Pentium class CPU (without MMX / SSE).

 

I really liked UCbench, opened up for more activity with interesting choices, clarkdale, wolfdale, even mobile cpus. A new benchmark which works 100% and is similiar to UCBench would be the best IMO.

Edited May 8, 2014 by Calathea

[Moose83]

Ucbench works good in default settings and is an nice benchmark too

[havli]

I like UCbench very much - it has wide support of CPU instructions and thus scales well with many CPU generations. There are not much (multithread) benchmarks working on stuff like dual socket 5 Pentium and newest Core i7 at the same time

 

If default settings is 100% bug-free, it would be best to keep it. I personally have no problem rebenching all my stuff to to meet the new rules.

Edited May 8, 2014 by havli

[Bobnova]

Sounds like it needs a wrapper, and then to have the database purged of all pre-wrapper results.

 

I don't think we can really ask the mods to check every single result, because:

1) That's a crapload of results and therefor work.

2) The mods won't do it anyway.

[GENiEBEN]

^ Started working on it, just on hold until I can deal with real life stuff. Yes we (I) would, it's just the crapload of PM's you have to answer after you delete the scores that's going to take time :\

[Mr.Scott]

I recant my position slightly. Default is NOT the only way to run this bench and fulfill El Genieben's requirements. It just takes a little more work. UC Bench should stay.

[[GF]Duke]

So, is this a bugged score? Will global points be restored when the wrapper is finished?

[Mr.Scott]

Duke;309620']So' date=' is this a bugged score? Will global points be restored when the wrapper is finished?[/quote']

 

Bugged. Odd number used in every second thread.

[I.nfraR.ed]

I'd say run at default settings (as we did in our local forum in the past) and only allow manual instruction set selection.

[[GF]Duke]

What happens when you run a single core cpu?

[ObscureParadox]

Duke;309636']What happens when you run a single core cpu?

 

If my testing is anything to go by it's extremely bugged with odd numbers amongst other things.

 

Single core brought more consistent bugged runs for me than with a dual for example.

[Christian Ney]

[Christian Ney]

Wasn't expecting it to work...

[ObscureParadox]

[[GF]Duke]

rofl