Geekbench 4 single and Geekbench 3 Multicore

[Splave]

Can we add a rule that you have to have at minimum 2gb's of memory available in the bench window?

There is a reason to run low memory available, its to trick the bench. At this point maybe its irrelevant and time to use newer version of the bench. 

 

[SparkysAdventure]

less than 2gb maxmem isn't relevant for gb3 multicore overall scores

it tanks integer and float scores, memory score is not enough to catch up

it doesn't make "lucky runs" more likely either, runs where memory score is randomly 13k when it's supposed to be 10k is just windows having a fit, had that happen on daily settings and no maxmem just as often if not more than bench

Edited June 17, 2021 by Sparky's__Adventure

[Rauf]

Can't we just call it a tweak and keep it for now. Can switch to newer versions later if necessary. It doesn't help Intel, I had same score with 2,3g compared to 1g. On amd maybe it helps more, but it doesn't bug the benchmark as far as I know, it still runs as it was designed to... But in a different way that wasn't intended for that type of CPU.

[Splave]

check 7740/7700 and 7350k scores all top scores are sub 1.5gb 

 

also not talking just memory score high which is easy to see, its more so when some CPU test subs are high.

 

keep it more succinct between the two geek benches that get global was more of the point, not that one is worse than the other. 

 

[Guest endpunkt]

1 hour ago, Rauf said:

Can't we just call it a tweak and keep it for now. Can switch to newer versions later if necessary. It doesn't help Intel, I had same score with 2,3g compared to 1g. On amd maybe it helps more, but it doesn't bug the benchmark as far as I know, it still runs as it was designed to... But in a different way that wasn't intended for that type of CPU.

https://hwbot.org/submission/4715079_rauf_geekbench4___single_core_core_i9_11900kf_11122_points

That is leading score.

Later.

You and 3 others went onto bug it higher than 11122. 25% of top scores are done this way above 10,000

So ask the question then. Why would moderator let you keep geek points for this?

In general. Larger amd processor bugs it the most, boost on scores.

Geek 3 + 4 + 5 all look affected.

So it is overclocker who is deliberate to run low max mem + freak benefit of large amd cpu's that destroys your global points earned.

 

[yosarianilives]

3 hours ago, Rauf said:

Can't we just call it a tweak and keep it for now. Can switch to newer versions later if necessary. It doesn't help Intel, I had same score with 2,3g compared to 1g. On amd maybe it helps more, but it doesn't bug the benchmark as far as I know, it still runs as it was designed to... But in a different way that wasn't intended for that type of CPU.

The workload is sized depending on memory size, under 1gb the workload is 50mb and mostly fits into cache which is why it is so much faster. So now the memory latency test doesn't hit memory, it's as much a tweak as running ram cache for a storage bench thus it must be banned. 1-2gb gives workload 100mb, 2+ gives 200mb workload which will hopefully be at least a year or so before that fits in cache. We need to remove all scores with less than 2gb end of story. 

[yosarianilives]

2 hours ago, endpunkt said:

https://hwbot.org/submission/4715079_rauf_geekbench4___single_core_core_i9_11900kf_11122_points

That is leading score.

Later.

You and 3 others went onto bug it higher than 11122. 25% of top scores are done this way above 10,000

So ask the question then. Why would moderator let you keep geek points for this?

In general. Larger amd processor bugs it the most, boost on scores.

Geek 3 + 4 + 5 all look affected.

So it is overclocker who is deliberate to run low max mem + freak benefit of large amd cpu's that destroys your global points earned.

 

Problem cannot be truly fixed until geek6 because the changing workload size was not designed with the understanding that cpu cache would get so big. Anyways needs to be 2gb minimum end of story. 

[Guest endpunkt]

3 minutes ago, yosarianilives said:

The workload is sized depending on memory size, under 1gb the workload is 50mb and mostly fits into cache which is why it is so much faster. So now the memory latency test doesn't hit memory, it's as much a tweak as running ram cache for a storage bench thus it must be banned. 1-2gb gives workload 100mb, 2+ gives 200mb workload which will hopefully be at least a year or so before that fits in cache. We need to remove all scores with less than 2gb end of story. 

Who will volunteer their time looking at all the scores worth points anyway

Have looked at geek4 single. So could knock out 20 obvious ones there. Then look at geek5 single. Even that is a push of time wasting evenings

You need the 4 others looked at out of 6 benches.

Could be 90% effective, you won't find them all of course.

Then up to moderate if he is going to approve all the score checking and follow on with reinstating points.

1.5GB maxmem as it is now, then a follow on check of 2GB later 

 

[yosarianilives]

25 minutes ago, endpunkt said:

Who will volunteer their time looking at all the scores worth points anyway

Have looked at geek4 single. So could knock out 20 obvious ones there. Then look at geek5 single. Even that is a push of time wasting evenings

You need the 4 others looked at out of 6 benches.

Could be 90% effective, you won't find them all of course.

Then up to moderate if he is going to approve all the score checking and follow on with reinstating points.

1.5GB maxmem as it is now, then a follow on check of 2GB later 

 

Yeah definitely have to have users help report them