Ram speed higher , bench results lower ?

McstylisT · 4 May 2014 at 10:19

Hi all

Just picking some brains and asking for some view and insight really.

I had some 8gb GeIl leggera 1866mhz ddr3. I over clocked this most of the time to 2133mhz and run @ 10-11-10-36 1T @ 1.68v perfectly.
I used Sandra lite most of the time and memory bandwidth scored well. Showing 24.8gbs for standard 1866mhz - and a fine 28.3gbs when running at 2133mhz.

Now loving components and buying as we do , saw the Kingston beast 2400mhz c11 kit on offer so gobbled that up for a laugh. Reviews show it is serious ram, and powers through most things while O'Cing as we'll.
So I sold on the GeIl ram and installed the Kingston. All fine straight in XMP 2400mhz 11-13-13-30 . So benched these at rated speed. I got an uninpressove 26.4gbs bandwidth.

from my experience increasing the speed always improves bandwidth, but seemingly not in this case. So I continued to play and tighten timings. They got a bit better 10-11-11-30 1T and managed to go upto 27.5gbs , so a bit better. However I tried and tried and could not improve the timings at all so could not achieve anywhere near the 28+gbs bandwidth the Geil memory achieved at 2133.?
So went onto Oc the memory. I managed 2600mhz with 1.76v , but timings loosened slightly to 11-12-12-30 2T. Otherwise not stable or no boot. Bench ? A disappointing 26.8gbs bandwidth ? :mad:

Ok so maybe I'm missing something here. As tbh using the beast around windows seems rapid and snappier. Things scale very well and instantly especially with faster speed. Though the benches show lower bandwidth. ?

So first question, does overall speed make up for less bandwidth and looser timings ?

Second question - since installing the Kingston memory my system boots ever so slowly. Before in 7 seconds ,, now sometimes 1minute. Until windows it's fine.

So feel free to give me your opinions and any relevant comparing Sandra lite bandwidth tests to show me what your getting at same speed.

Thanks all.

wazza300 · 4 May 2014 at 12:15

only the latency improves from my experience when you push past 2133mhz

the read/write/copy figures tend to wane

pgi947 · 4 May 2014 at 12:17

Leave the beast at 2400 or drop to 2133 and tune. I've had my kit at 2400 9-11-11-28-1t & 2133 8-10-10-24-1t

Silent_Scone · 4 May 2014 at 13:09

Tuned 2133 is the sweet spot on a 24/7. I've seen some users on LN2 even in fact, running 2700c11, my 2133c8 obtains me higher physics scores in Futuremark and majority of benches. Unless you're going very tight past 2400 it's absolutely pointless.

McstylisT · 4 May 2014 at 20:03

Thanks guys , really appreciate the input.

I will just do my best to reduce them toward what you gus mention. Can I ask what voltage those timings were ?

Cheers chaps.

McstylisT · 4 May 2014 at 22:14

Ok peeps , been playing with these again but cant seems to get them c9

best ive managed is 10-11-10-28 1T which looks and seems good. But benches of bandwidth score only 27.46gbs using sandra lite.

Now whats odd to me is that the Geil was running slower 10-11-10-36 1T @ 2133mhz but scored 28.3gbs

With very similar timings now and more mhz speed, how is that possible ?

any pointers welcome.

Silent_Scone · 5 May 2014 at 08:49

The other kit probably had tighter subs.

McstylisT · 5 May 2014 at 09:09

MjFrosty said:
The other kit probably had tighter subs.

Yeah think your exactly right here frosty.

Some of the higher clocked sub timings were like double that of the Geil, and don't like to be brought down much.

I am now looking to go down to 2133mhz and see what I can squeeze out if this for testing.

Silent_Scone · 5 May 2014 at 10:35

Tbh if you're that worried about bandwidth that's where quad channel comes in

I get 65Gb/s out of my 2133cl8

McstylisT · 5 May 2014 at 19:04

Thats amazing bandwidth. But pretty expected from quad channel.

Im just looking to optimize my dual channel setup that was all. My conception is that improving speed , or tightening the timings would results in a more efficient ram, giving greater bandwidth results. Always liked GeiL memory, should have been happy with my Leggera TBH , though the beast looks the part with the black pcb.

I tried lower clocking of 2133mhz. 1.65v seemed enough and got tight 9-10-9-27 1T from them. All seemed well but again using benchmarks a dropped 24.8gbs bandwidth result.

I must say im not really impressed with the kingston beast, they dont seem to live up to their name well.

Anyone feel like uploading a MaxMem2 or sandra lite bench so i can see what your ranking generally ? 1866 - 2133 - 2400 and maybe anyone rocking 2666.

Especially those with team group setups

Ill add mine to show. Thanks

setter · 5 May 2014 at 19:21

8gb Team group vulcan red, 2400mhz, 10-11-11-31. 2T.

pgi947 · 5 May 2014 at 19:30

I'll put up some 2666, the results will be poor though as its a cheap single sided kit I picked up on the mm to do some high MHz runs. Just clawing them up to 2933 with great difficulty.

McstylisT · 5 May 2014 at 20:11

setter said:
8gb Team group vulcan red, 2400mhz, 10-11-11-31. 2T.

You the man setter

thank you.

Your clearly scoring higher than me, but your 4770K might be the factor here. At least it shows my rams performing close to what it should be. Thanks mate

pgi947 said:
I'll put up some 2666, the results will be poor though as its a cheap single sided kit I picked up on the mm to do some high MHz runs. Just clawing them up to 2933 with great difficulty.

Be good to see the spectrum mate, good or bad. At least I and anyone else will know what's performing and what's not.

wazza300 · 5 May 2014 at 20:14

anyone have aida64?

McstylisT · 5 May 2014 at 20:22

Nice one wazza

Some very decent scores.

I dont have AIDA as id happily do a test. Paid for program right ?

Heres my sandra lite test guys. Though im begging to see that the beast does perform. I mean its doing ok by the scores i think.

pgi947 · 5 May 2014 at 20:27

I derp'd hard, had my timings still at 12-13-13 from running at 2800/2933, will set them back to XMP and run it.

For now here are some of my Aida runs on those 2400 hyperX beast kits.

McstylisT · 5 May 2014 at 20:37

Nice one PGI

Jut wish i had AIDA to compare better , especially with same ram.

Could you do a maxmem2 possibly ?

Rroff · 5 May 2014 at 20:45

The sub timings do seem a bit lax on the 2400 beast - tRFC especially seems to be twice or there abouts what the RAM can handle.

McstylisT · 5 May 2014 at 20:48

Rroff said:
The sub timings do seem a bit lax on the 2400 beast - tRFC especially seems to be twice or there abouts what the RAM can handle.

yeah i agree Rroff , seen this myself. I think its something insane like 312 clocks on tRFC using XMP on these. I have mine down at 238 clocks but as your aware many kits do sub 150 clocks here.

pgi947 · 5 May 2014 at 20:49

McstylisT said:
Nice one PGI Jut wish i had AIDA to compare better , especially with same ram.

Could you do a maxmem2 possibly ?

I'll be throwing the ram back in later in the week to test a newer bios so I'll run it then

Rroff said:
The sub timings do seem a bit lax on the 2400 beast - tRFC especially seems to be twice or there abouts what the RAM can handle.

Yep, iirc at stock they were something like 217, went down to 96 with very little fuss.