The Official Geekbench3 Memory Challenge thread

[Rauf]

Hi

 

I think multi core memory score would be better. Singe core scales strangely and favours 4dimm T-topology boards too much.

Edited November 28, 2016 by Rauf

[Christian Ney]

I chose the single core score to not favor multi-core processors

[scannick]

Interesting challenge, I never liked this bench, sure this time I can learn how to do it better

Lokt at this my nice score 9314 @45/45 and ddr4@4000cl12:

 

just a little bit bugged

 

on the left rerun with the same ram setting and cpu @47/47

Edited November 29, 2016 by scannick

[websmile]

For all contestants, please submit a complete screenshot complying with geekbench rules. This also makes life easier for staff when we check results

[Rauf]

Interesting challenge, I never liked this bench, sure this time I can learn how to do it better

Lokt at this my nice score 9314 @45/45 and ddr4@4000cl12:

 

just a little bit bugged

 

on the left rerun with the same ram setting and cpu @47/47

Interesting, I have never seen bugged geekbench scores before.

[websmile]

Easy to see when you watch the multi-thread result at 4 cores of 9k

[scannick]

Easy to see when you watch the multi-thread result at 4 cores of 9k

 

9314 is the Memory Performance in Single core

[websmile]

9314 is the Memory Performance in Single core

 

9436 is your general score (Multi-core Score) at 4,5ghz with 4 cores - normally you beat this with an i3 at same frequency^^

[scannick]

9436 is your general score at 4,5ghz with 4 cores - normally you beat this with an i3 at same frequency^^

 

Ok, but the memory score is bugged or can I sub it in the challenge ?

[websmile]

Bugged runs are not OK for submission, if general score is bugged we have to assume subtests are bugged as well

[scannick]

Bugged runs are not OK for submission, if general score is bugged we have to assume subtests are bugged as well

Of course

In your first post I got you said that I read the wrong score, but you mean that the general score was bugged too so is easy to check out the bugged run

Now is all clear:)

Edited November 29, 2016 by scannick

[GtiJason]

Memory challenges are a great idea, but I can beat a 5g 4133c12-11-11 tight sub with a 5.1g 4000c12-12-12. Kind of defeats the point here if fastest cpu wins. If you want to keep it fair, cap cpu freq to 6.0-6.2g Max to shift the focus onto mems. Not a lot of guys have high binned Haswell cpu (ddr3 challenge) and since I do would it be fair if I placed Top 3 becuse my 4770k at 6.45g with mems on XMP ?

 

EDIT: Please disregard, after reading the rules I realized I'm an idiot. That is all

Edited December 8, 2016 by GtiJason

[speed.fastest]

Joined on DDR4, sadly i dont had any DDR3 platform this time...

[mllrkllr88]

Why are points disabled for this competition? I still want to compete no matter if we have the possibility of points or not, just curious.

[rtsurfer]

Besides from the question above, which would be nice to have an answer for. Is Kaby/Z270 legal for this Comp? I know typical rule for Comp is no new hardware released after the Comp start time is allowed. But how new is Kaby/Z270 really? Its mainly a CPU clockspeed bump, which if you look at the DDR4 rankings, means very little.

[websmile]

This is no listed benchmark, because of this it can get no points, I thought this was already answered. If the tests might work like the competition now, we consider listing it and maybe later it will be awarded points, but this is now no topic for this test.

Let´s keep Kaby and z270 out of this, I am simply not sure everyone is able to get it on all places in the world, we saw problems on z170 with new BIOSes and the new chipset Z270 might not be fully available everywhere

[rtsurfer]

Okay. Thank you for the answers.

 

Sent from my Nexus 5X using Tapatalk

[funsoul]

No competition points for an ocsports competition? As someone who still hasn't gotten 10, 'tis a bit of a bummer. Ah well. It was still fun to learn a bit more about this bench but not worth too much more testing/tweaking time.

 

Going forward, if there are no boints for a challenge or comp, could that info be listed on the main ocsports page for that?

 

As an aside...as Rauf pointed out, 4 sticks does much better than 2 so, for example, z170 oc formula has a significant built-in advantage over the z170m ocf. Was going to try 8 sticks of psc in a 2011 board for comparison but, unfortunately, a couple are dead and I'm down to 7 Tried 4 sticks in 2011 (3930k/rive, 2400 7-11-7), ddr3 result wasn't even close to being competitive against z97 (g3258) at same cpu clock and ram sticks/speed/timings.

[Noxinite]

Yep, good old 4-dimm tweak works well here as it isn't as harsh on the sticks, so you can push much harder.

[jpmboy]

would be a better comp if the cpu core speed were limited in each DDR class.

[Noxinite]

would be a better comp if the cpu core speed were limited in each DDR class.

 

No full-out with no cooling restrictions is the only way to stop people from cheating. And here if your mem efficiency is very good then you can make up for a (very) large difference in core clock (as shown by John Doe's DDR4 score).

[jpmboy]

<blockquote><span style="font-weight:bold;">jpmboy said: </span>would be a better comp if the cpu core speed were limited in each DDR class.</blockquote><br/>

<br/>

No full-out with no cooling restrictions is the only way to stop people from cheating. And here if your mem efficiency is very good then you can make up for a (very) large difference in core clock (as shown by John Doe's DDR4 score).

 

JD's core clock is far from deficient. A memory benchmark, should be just that, not how high the cpu is clocked to. And how does one eliminate cheaters by allowing unrestricted cooling? You are conflating very different things. I never mentioned cheating, just looking for ways to deconvolute one's ability to tune for ram efficiency and separate out the benchers that rely upon just overclocking the snot out of a cpu in a memory benchmark.

[rtsurfer]

JD's core clock is far from deficient. A memory benchmark, should be just that, not how high the cpu is clocked to. And how does one eliminate cheaters by allowing unrestricted cooling? You are conflating very different things. I never mentioned cheating, just looking for ways to deconvolute one's ability to tune for ram efficiency and separate out the benchers that rely upon just overclocking the snot out of a cpu in a memory benchmark.

 

What Nox meant with the Cheating thing is that, it's a common theme on Hwbot these days where whenever there's a competition with a clock restriction, people cheat. Run at higher clock & then downclock for Screenshot. 32M LCC is ruined because of it.

 

Yes, a Mem contest should only focus on Mem, but Geekbench single core mem score scales really really poor Mem clocks. The benchmark by design itself seems to have very miniscule response to core clocks. This can be evidenced by the fact that as far I know, the top 2 scores in the comp right now, beat out all geekscores on Hwbot right now, save for Dancop's world record. When you can beat almost all of ln2 Mem subs with an AIO, I think u can safely say core clocks matter very little to none.

 

5.2 is not deficient, but 5G can be done with a good chip & an AIO. The bench's inherent clock scaling nature combined with the Cheating aspect with limited clock comps, makes me think that the current setup is not bad.

[jpmboy]

yeeh not bad. Seems tho, there has to be a way to cull out cheats in this benchmark since the hardware setting info is in the verification link. From what I've seen so far, single core memory performance (using the same cpu) scales quite well with core freq, cache freq and memory freq from the few (a dozen) runs I did here with a KBL. IDK, anyway it's pointless ;-)

[marco.is.not.80]

Well, this sucks BobaFet's rocket sausage. Just read in this thread that KabyLake isn't allowed. Wish that was in the rules or something so I didn't waste my time figuring out how I was going to get this 790i Ultra SLI board up with a QX9650 in it to contribute to the DDR3 stage... Went out a bought a PSU for it, etc.

 

God this overclocking shit keeps costing me but I'm like "just another 0.1 hw point."

 

Anyways, grats to whoever wins this thing. I gotta take my CrabbyLake and go sit in a corner.