Memory Tweak shows scores that are not reflected on the efficiency of system memory

[sandro_c]

Hi people,

I checked with some evidence that the score that assigns MemTweak has some problems of reliability.
I was suspicious of the large difference in score between my memory, and those shown on the Global Leaderboard MemTweakIt where scores are shown 140000-120000 for members who use my own hardware and freq.di memclk. and my HW that instead receives a score of 42,000 to 45,000! I could understand a little different but more than twice as absurd.
I checked my settings (submitted by sandro on ID 458), and in fact I found a timing value on the secondary timings (-Dram refresc-interval equal to 7936) ...absurd value, instead one that has high scores is about 250. I applied this value of 250 to my MemTweak and in fact my score has risen to 129,781, in line with the highest values​​.
The problem is that by doing various tests the result was better when there was value in 7936: Intel.BT fell from 63 sec. step for 5 to 72 sec, and Sandra Mem.BW fell from 54 MB / sec to 21 MB / sec. : i must come back!
In addition, the item-Dram refresc-interval does not appear among the voices in my bios, so if I wanted to I could not make it stored fixed! Someone can explain these strange mysteries?
thanks
sandro c

-----------------------
MB: Rampage IV Extreme 2105 bios
CPU: i7 3930K @ 4500 mhz stepping C2-1320 v vcore
RAM: 16GB Corsair Dom.GT 2133 - CMT16GX3M4X2133C9 @ 2133 mhz
Cooler: Noctua NDH14-2011
PSU: Enermax 625 MODU82 + silver
GFX: HD-6970 2GB DDR5
Disks: 2xOCZ Vertex4 Raid0-2xOCZ Agility2 Raid0 1TB Caviar-1WD
Monitor: Samsung T240HD 1900x1200
Pioneer Blu-Ray Burner
Windows 7 Professional 64-bit

[Dartmaul]

It's all about DRAM Refresh interval. Bu default it has some huge value (7936 for me) and I've got 47319 score.
Guys who have 120 000 score have just reduced it to 250-300 and that's it.
Here's my screenshots. The only setting I've changed was DRAM Refresh interval.

I don't know what is it but with 250 interval real bandwidth drops dramatically.

[Nodens]

I'm fairly certain that this is a bug in Memtweaker. Performance-wise you want higher tREFI values. tREFI controls how often the DRAM cells are refreshed. The less often (higher value) they are refreshed, the better they perform BUT there's a limit to how high you can set that value because eventually the DRAM cells lose their charge (hence their contents) if they are not refreshed in time. The maximum value of tREFI depends on a great many deal of things, IC type, VDIMM, memory density etc. So don't go and tweak your tREFI to the roof hehe.

[sandro_c]

Nodens wrote:
I'm fairly certain that this is a bug in Memtweaker. Performance-wise you want higher tREFI values. tREFI controls how often the DRAM cells are refreshed. The less often (higher value) they are refreshed, the better they perform BUT there's a limit to how high you can set that value because eventually the DRAM cells lose their charge (hence their contents) if they are not refreshed in time. The maximum value of tREFI depends on a great many deal of things, IC type, VDIMM, memory density etc. So don't go and tweak your tREFI to the roof hehe.

Hi Nodens,

now that you 've explained what this means timing, I tried to change it without compromising too much the characteristics of BW of memory: I found that the value - 1800 - (... but I could also choose 1500 or 2500),
gave me a result of BW to 52MB/sec, 2 lower than my normal, a value of Intel BT similar to my normal .... and a score of MEM-TWEAK of 97,000. Of course I put my old initial value of 7936, partly because I do not know how to fix the Bios!
Hello

sandro c

-----------------------
MB: Rampage IV Extreme 2105 bios
CPU: i7 3930K @ 4500 mhz stepping C2-1320 v vcore
RAM: 16GB Corsair Dom.GT 2133 - CMT16GX3M4X2133C9 @ 2133 mhz
Cooler: Noctua NDH14-2011
PSU: Enermax 625 MODU82 + silver
GFX: HD-6970 2GB DDR5
Disks: 2xOCZ Vertex4 Raid0-2xOCZ Agility2 Raid0 1TB Caviar-1WD
Monitor: Samsung T240HD 1900x1200
Pioneer Blu-Ray Burner
Windows 7 Professional 64-bit

[sandro_c]

Hello Dartmaul,

is exactly what I said and how I see is what happens to you, but I'm sorry, what does that mean?
Someone who likes to show a high score that parameter changes on the timing for the sole purpose of having a good score and reduce the memory to work with BW halved ... then why he bought a platform X79 quad-channel?
I feel stupid!
Besides, I have not figured out how to store that value (which I hope someone will explain to us what it is) on the Bios because that item does not exist anywhere.
However, I prefer my miserable score of 45,550, but with a BW of 54.8 MB / sec
Anyway thanks for the answer.

Anyway thanks for the answer.
sandro c

-----------------------
MB: Rampage IV Extreme 2105 bios
CPU: i7 3930K @ 4500 mhz stepping C2-1320 v vcore
RAM: 16GB Corsair Dom.GT 2133 - CMT16GX3M4X2133C9 @ 2133 mhz
Cooler: Noctua NDH14-2011
PSU: Enermax 625 MODU82 + silver
GFX: HD-6970 2GB DDR5
Disks: 2xOCZ Vertex4 Raid0-2xOCZ Agility2 Raid0 1TB Caviar-1WD
Monitor: Samsung T240HD 1900x1200
Pioneer Blu-Ray Burner
Windows 7 Professional 64-bit

[Dartmaul]

I'm sorry, didn't read first post well.
What does mean this interval? No idea actually.
Why does MemTweakIt show much score while actual performance drops?
This score means not actual performance, it's like max CPU frequency under LN2, you can reach increasable frequency but in fact it's so unstable so you cannot pass even a short benchmark.
In simple words this score is = (mem frequency / sum of mem timings)
The original idea is that if you drop some timings to low you'll BSOD, but this one is another story.
It should be some extra low score multiplier for this refresh interval to reduce it's impact on overall score but seems that there isn't one.

BTW X79 quad channel has less bandwidth then Z77 dual channel) (At least every test I saw is agree with that)

[Raja]

You can read Noden's post above for what tREFI does.

The software has a bug. When I grouped these settings for MemtweakIT, the algorithm I sent the R&D team was set to add a multiple of tREFI to the score, so something must have got lost in translation somewhere. I will check up with Shammy when he gets time to nudge them. May take a while...



The software is supposed to be for fun anyway. Shammy wanted me to score it without weighting it heavily on some of the primaries. I went for something more realistic first, but Shammy rightly wanted something everyone could play in without having to buy high performance memory kits. It's not supposed to be real-world reflective but is supposed to be fun. Treat it as such...

[sandro_c]

Raja@ASUS wrote:
You can read Noden's post above for what tREFI does.

The software has a bug. When I grouped these settings for MemtweakIT, the algorithm I sent the R&D team was set to add a multiple of tREFI to the score, so something must have got lost in translation somewhere. I will check up with Shammy when he gets time to nudge them. May take a while...



The software is supposed to be for fun anyway. Shammy wanted me to score it without weighting it heavily on some of the primaries. I went for something more realistic first, but Shammy rightly wanted something everyone could play in without having to buy high performance memory kits. It's not supposed to be real-world reflective but is supposed to be fun. Treat it as such...

Hi Raja,

As always, your answers sweep away all doubts: Nodens explained the function of the parameter DRAM refresh interval timing, and you have explained the purpose of the MemTweak which unfortunately also has this bug, but no matter, it's just a game and should be treated as such . If the bug is corrected the better otherwise it's okay!
However, what no one has yet explained is whether this timing -DRAM refresh interval- it is or not an item in the Bios or only one parameter to be set on the fly just for testing fast. I my Bios 2105 I did not find.
Thanks to all of you who have responded.
Ciao
sandro

-----------------------
MB: Rampage IV Extreme 2105 bios
CPU: i7 3930K @ 4500 mhz stepping C2-1320 v vcore
RAM: 16GB Corsair Dom.GT 2133 - CMT16GX3M4X2133C9 @ 2133 mhz
Cooler: Noctua NDH14-2011
PSU: Enermax 625 MODU82 + silver
GFX: HD-6970 2GB DDR5
Disks: 2xOCZ Vertex4 Raid0-2xOCZ Agility2 Raid0 1TB Caviar-1WD
Monitor: Samsung T240HD 1900x1200
Pioneer Blu-Ray Burner
Windows 7 Professional 64-bit

[sdmf74]

Raja@ASUS wrote:
You can read Noden's post above for what tREFI does.

The software has a bug. When I grouped these settings for MemtweakIT, the algorithm I sent the R&D team was set to add a multiple of tREFI to the score, so something must have got lost in translation somewhere. I will check up with Shammy when he gets time to nudge them. May take a while...



The software is supposed to be for fun anyway. Shammy wanted me to score it without weighting it heavily on some of the primaries. I went for something more realistic first, but Shammy rightly wanted something everyone could play in without having to buy high performance memory kits. It's not supposed to be real-world reflective but is supposed to be fun. Treat it as such...

Boy you wasnt kidding when you said it would be a while! Whats ol shammy up to these days? I wonder if he still has that task in the back of his mind "man I gotta tweak those memtweakit algoritms still" 😛

Overclocking is fun too but god forbid you use memtweakit as one of your tools to quickly check a quick oc. Unless of course you omit the trefi timing altogether!

[Raja]

tREFI is in the Z77 board BIOSes, probably not added X79 due to limited margin for adjustment in high density configurations and end-user memory module mix and match tomfoolery. The post training process performs over 1000 read and writes to check the DIMMs and de-skew any flight time variance between the signal lines to each bank. If this parameter is set wrongly, the board will not POST. You can actually make a more aggressive change in the OS, as it's not as stressful as the POST tests...lol

When we sit down and discuss what we need in each BIOS, we usually take out options that have little impact or little room for adjustment, as things are quite complex for most users as it is. Or we tie them in to settings that work with others for a combined performance boost. TBH the memory controllers in these newer CPUs are so well optimised for scheduling and opportunistic refreshes, these parameters have little impact for 24/7 setups unless you can make a very big change (which is difficult over 4 channels).