You are not registered yet. Please click here to register!
DFi UT x48-t3rs, X3360 - Page 20 - i4memory.com - different look at memory
Tags Register Blogs FAQ Members List Calendar Mark Forums Read

Notices

Reply
 
Thread Tools
Old 16-03-2012, 08:01 PM   #153
Offler
Senior Member
Offler's PC Specs
 
Join Date: Nov 18 2009
Location: Bratislava
Posts: 278
Code:
 
CPU Clock Ratio: 8.5x
CPU N/2 Ratio: Enabled
CPU Clock: 400
DRAM Speed: 400/1600
- Target DRAM Speed: DDR3-1600
 
 
CPU VID Control: 1,27500
COY VID Special Add Limit: Enabled
CPU VID Special Add: 100,23%
DRAM Voltage Control: 1,712v
SB Core/CPU PLL Voltage: 1.51
NB Core Voltage: 1,492v
CPU VTT Voltage: 1.270v
Vcore Droop Control: Disabled  (underclocking with multiplier is disabled, so the CPU always run on 3400Mhz - thus this feature is not needed)
Clockgen Voltage Control: 3.45v
GTL+ Buffers Strength: Weak
Host Slew Rate: Weak
 
GTL REF Voltage Control: Disable
x CPU GTL 1/2 REF Volt: 113
x CPU GTL 0/3 REF Volt: 100
x North Bridge GTL REF Volt: 100
 
DRAM Timing
- DRAM CLK Driving Strength: Level 3
- DRAM DATA Driving Strength: Level 8
- Ch1 DLL Default Skew Model: Model 5
- Ch2 DLL Default Skew Model: Model 5
(other fine tune settins are set to "more aggresive")
 
- Enhance Data transmitting: FAST
- Enhance Addressing: Fast
- T2 Dispatch: Enabled
 
Common CMD to CS Timing: 1N
 
CAS Latency Time (tCL): 8 (xmp)
RAS# to CAS# Delay (tRCD): 8 (xmp)
RAS# Precharge (tRP): 8 (xmp)
Precharge Delay (tRAS): 24 (xmp)
All Precharge to Act: AUTO
REF to ACT Delay (tRFC): AUTO (XMP) - but currently testing value 74
Performance LVL (Read Delay) (tRD): 8
 
XMP Support: Profile 1
Just to note, i have 8giga ram, 4x 2gb dimms, Adata XPG 2.0.

4gb ram:
1600Mhz, 7-7-7-21, 1,712v and 1,265v on northbrige

8gb ram
1600Mhz, 8-8-8-24, 1,712v and 1.492v on northbridge

My goals during all this testing were:

400Mhz FSB, RAM 1600Mhz while:
NB voltage on default
Mem voltage on default

Tests with 4GB of memory (2 memory modules):
NB voltage on default 1,265 was stable at 400Mhz.
Mem voltage was determined 1,712v as most universal. Chipset expects 1,9v at 1600mhz, memory usually expect 1,5v. That value is in the middle and usually is set from XMP profile.
Overall success at any point.

Tests 8GB of memory (4 memory modules)
At 400Mhz FSB the NB voltage has to be higher. 1,33v for Windows to start, 1,41v for windows to run, 1,49v for stability under load. Gaining stability with so many memory sticks is very hard. Almost nothing except default XMP setting is not working. Even standard SPD settings are not working. Big surprise was fact that command rate remained at 1. At 333Mhz FSB default NB voltage worked fine.
Overall sucess, but only because there exist stable solution. I wanted better performance.


Currently I have 8 gb of ram. Not as fast as I want and I wanted to have less voltage on NB, but it seems to be not possible. If you have only 4 gb of ram the system runs much smoother and overclocking potential seems much better.

On the other hand if you manage latencies like 8-8-8-24, with 4 memory sticks, more memory chips with better Command Rate 1 can generate similar performance as 7-7-7-21 with 2 memory sticks and Command Rate 2 at cost of memory read (9500mb/s with CL7, and 9200mb with CL8).

Current cooling on Northbridge is Thermalright HR-05 with 8cm Noctua fan.

By the way I am suprised that even somebody reads this long and boring topic


Edit:
I removed the NB fan again, and placed 12cm on CPU cooler instead. Something changed because the system is now working fine even with lower latencies... But i have no idea what else has changed...

Last edited by Offler; 19-03-2012 at 06:42 PM.
Offler is offline   Reply With Quote
Old 19-03-2012, 03:16 PM   #154
fluXX
Junior Member
fluXX's PC Specs
 
Join Date: Mar 16 2012
Posts: 3
At first sry that I haven't replied earlier - I have reinstalled windows etc. over the last days.
Now I must say it seems you have a quite nice board compared to mine - or it is cause of my RAM.
Have tested the settings you're using for 4GB RAM (thx for posting) but can't get the board run stable at default NB voltage.

But by reading through your settings I decided to squeeze every little thing out of my system.

So I'm finally using the following settings:

4gb ram:
1600Mhz, 7-8-7-24, 1,772v (and trd 7) and 1,555v on northbrige

Hoping my hardware (especially my RAM Moduls) won't get damaged

Code:
CPU Clock Ratio: 8.5x 
CPU N/2 Ratio: Enabled 
CPU Clock: 400 
DRAM Speed: 400/1600
- Target DRAM Speed: DDR3-1600     

CPU VID Control: AUTO 1,2950 
COY VID Special Add Limit: Enabled 
CPU VID Special Add: AUTO 
DRAM Voltage Control: 1,772v (quite high for the G.SKill ECO) 
SB Core/CPU PLL Voltage: 1.51 
NB Core Voltage: 1,555v (have to set so high) 
CPU VTT Voltage: 1.100v (default for Q9550 E0) 
Vcore Droop Control: Enabled   
Clockgen Voltage Control: 3.45v 
GTL+ Buffers Strength: Weak 
Host Slew Rate: Weak - set like you did 

GTL REF Voltage Control: Disable 
x CPU GTL 1/2 REF Volt: 113 
x CPU GTL 0/3 REF Volt: 100 
x North Bridge GTL REF Volt: 100 
  
DRAM Timing 
- DRAM CLK Driving Strength: Level 3 
- DRAM DATA Driving Strength: Level 5 
- Ch1 DLL Default Skew Model: Model 6 
- Ch2 DLL Default Skew Model: Model 6 
(other fine tune settins are set to "more aggresive") - set like you did 

- Enhance Data transmitting: Fast 
- Enhance Addressing: Fast - set like you did 
- T2 Dispatch: Disabled   

Common CMD to CS Timing: 2N   

CAS Latency Time (tCL): 7 (xmp) 
RAS# to CAS# Delay (tRCD): 8 (xmp) 
RAS# Precharge (tRP): 7 (xmp) 
Precharge Delay (tRAS): 24 (xmp) 
All Precharge to Act: AUTO 
REF to ACT Delay (tRFC): 88 (XMP) 
Performance LVL (Read Delay) (tRD): 7   

XMP Support: Profile 1
again thank you for doing this long test run and posting every detail ...
and no it wasn't boring - it was quite interesting for me (good old x48 user)

greatz

Last edited by fluXX; 19-03-2012 at 03:27 PM. Reason: reformating text etc.
fluXX is offline   Reply With Quote
Old 19-03-2012, 06:50 PM   #155
Offler
Senior Member
Offler's PC Specs
 
Join Date: Nov 18 2009
Location: Bratislava
Posts: 278
All you need is to find correct Dram Skew mode for your memory and CPU VTT voltage.

Also I would reccomend to set CPU VTT to value 0,05v lower than is CPU voltage. For this I set CPU voltage manually to 1,275v, set CPU to run at 3400Mhz all the time and CPU VTT was set to 1,27v.

This will move some of the load from the memory and northbridge to the CPU.

Also I have something to cooling...

8cm Noctua Fan was previously mounted on my HR-05 - together cooling the northbridge. Its temperature was 51. I removed that fan, and added other 12cm to the CPU so the NB cooler is directly in front of it.

NB temperature got much higher - 59-62 degrees, but overall stability of the system greatly improved - even with tighter latencies. Maybe airflow is better afterall, or some counterpart of the board is now in the airflow so its not getting so hot...

Edit: Something other for reference
http://www.xtremesystems.org/forums/...he-lowest-tRD&

and since DFI's product page for this board is also gone another page with some description:
http://www.overclock3d.net/reviews/c..._motherboard/1

I was searching for information how many layers the board has. 6 is quite good. Anyways my Adata memory dimms has 8 layers

Yesterday's heat testing:
On CAS 8-7-7-23 the system was failing at NB temperature around 60 degrees after some hours. Later I did some tests. Once system failed immediately during boot after being powered off for some minutes immediately after torture test (when system was hot but not cooled by fans). this shows that problem is related to accumulation of heat somewhere... 8-8-8-24 seems stable anyways. Just to note - the memory has XMP timing 1866 8-8-8-25 and this has been always reached under standard settings.

And some other info:
http://www.tomshardware.com/reviews/...ling,2342.html

Overall most applications in the test used only CPU for acessing memory. Therefore most applications are showing no difference. Adobe Premiere and selected games seems to utilize the memory for the concurrent CPU and GPU operations. There is visible the difference. However its only at rate of 10-15 percent.

This improvement is done mainly by write speed of CPU. This value usually revolves around 8500mb/s no matter of memory frequency, however single CPU can send data at speed at 5400mb/s (at FSB400). If only one channel is used, or memory frequency is 800mhz this value drops to 5000mb/s.

However most such tests cannot show big performance drop which might occur during heavy memory/pcie operations when used highspeed SSD if 800mhz or single channel 1600mhz is used.

At all those 15 percent are added to CPU, GPU and SSD performance if the memory is well configured.

Last edited by Offler; 20-03-2012 at 09:26 PM.
Offler is offline   Reply With Quote
Old 29-03-2012, 07:15 PM   #156
Offler
Senior Member
Offler's PC Specs
 
Join Date: Nov 18 2009
Location: Bratislava
Posts: 278
Ok some results from X48 vs Elpida hyper testing which are related only to the board.

settings:
CPU - 1,275v
NB 1,41v
Mem 1,712v
CPU VTT - 1,245v

This setting was stable when computer was stored in cool environment. When the ambient temperatures went high the system crashed. I was searching for the exact point where is the problem.

Near PSU power connector on the board is quite big aluminium cooler. Its on one chip which seems to provide power regulation for the 3.3v and 5.0v rails. It means memory, PCI-E...

I placed fan near this area. After torture tests the system remained stable. I removed the fan and system crashed almost immediately. Later I also covered memory with paper to be sure that its not memory what is getting trouble due overheating.

Generally the problem is located somewhere in this area. Adding more heatsinks did not help me to identify the exact part (at least not yet). I suspect that some chokes are having trouble with temperature over 50 and input voltages.

There is also other way how to overcome this instability problem - Raising vNB to 1,49v. This also can explain why this board hates quad cores or more memory...
Offler is offline   Reply With Quote
Old 04-04-2012, 01:49 PM   #157
fluXX
Junior Member
fluXX's PC Specs
 
Join Date: Mar 16 2012
Posts: 3
Well sadly I have to say that I'm not able to maintain stability on 400/1600 or
333/1333 (DRAM 1601) Strap at a moderate voltage. Have tried all possible skew mode combinations
and different VTT voltages. It's deffinetly not possible to get the board stable unless by pushing a lot
volt to the RAM and the NB.
(Maybe this is due the fact that my ECO DIMMS are for extreme low voltage)

So I decided to roll back - and give a f*** to DDR1600 ...
(maybe I will buy some other modules for testing puposes in near future)
but for now I go with DDR1333 CL6 CR1 and low voltages .... cause the summer comes to the city

Another strange thing is that I have to tweak my GTL's manually as soon as I use
a VTT other than default.

For now I use:
VTT: 1,270
GTL: 76/71/69 (61% / 61% / 63%)

- post some screens to show mem-bandwidth which of course is much lower than DDR1600
Attached Images
File Type: jpg 1.jpg (142.7 KB, 9 views)
fluXX is offline   Reply With Quote
Old 05-04-2012, 02:22 AM   #158
Offler
Senior Member
Offler's PC Specs
 
Join Date: Nov 18 2009
Location: Bratislava
Posts: 278
Yep. I would try Dimms which are binned as 1.65v or 1,9v.

Try different memory voltages. Everything between 1,35-1,5v is undervolting according to JEDEC.

According to Intel 1,9v and 1600Mhz is stock, while northbridge is on stock voltage. But for me works 1,712v because my memory chips are manufactured to 1,5v.

this is true If you find the correct Dram Driving strenghts.

Edit: Try performance level 7 and 6. it should work on that CAS.

Last edited by Offler; 05-04-2012 at 10:50 PM.
Offler is offline   Reply With Quote
Old 10-04-2012, 06:19 PM   #159
Offler
Senior Member
Offler's PC Specs
 
Join Date: Nov 18 2009
Location: Bratislava
Posts: 278
And now for some other boring calculations:

Memory read (real) 9200mb/s
Memory Write 8500mb/s

L2 Cache read 24800mb/s
L2 Cache write 18070mb/s

Theoretical memory bandwidth is 25600mb/s

So if the memory is writing to CPU cache at speed of 9200mb/s CPU can write to the cache at approximately same speed and in next step the data are exported to the memory at same speed.

PCI-E can send data to memory at speed of 6700mb/s, however my current harddrive can read data at speed of 750mb/s. Since the memory has such wide bandwidth CPU and PCI-E can write at same time with speed 8500+6700 mb/s

Standard system bandwidth is fluctuating from 750 thought 6700, 8500 and 9200mb/s, but the system still has potential of lower the bandwidth differences. Current average bandwidth is 6200mb/s. Much better harddrive, and bit PCI-E overclocking can grant average system bandwidth around 8100mb/s.

So... E7500 vs Q9550 will have exactly same Memory read in everest on FSB 400. Also the Memory read will remain same no matter 800Mhz memory or 1600Mhz. Whats the difference?

Memory bandwidth should be at CPU bandwidth + PCI-E bandwidth. At 400Mhz fsb its 12,8gb/s + 8gb/s. This configuration allows to write from CPU and PCI-E simultaneously. The lowest memory frequency at this FSB speed is 1280Mhz dual channel to comply, while 1600Mhz will improve performance only because of its synced frequency. Beyond 1280Mhz of memory speed is no measurable improvement.

Why not to set the memory frequency to 800mhz? This will be enough for CPU only. Concurrent actions from CPU and PCI-E at the same time can mean lowering of total bandwidth to half - 6400mb/s.Range of memory frequency between 800Mhz-1280Mhz at FSB400 will mean performance improvement in matter of lowering bottleneck at memory subsystem and thus lowering probability of memory overload.

Most tests with GPU-z and HD 5850 1gb also show that Video memory is not utilized much. For games like Deus EX its mostly 50%. Skyrim uses up to 90%, but video memory dedicated from system ram never crossed 200mb. Therefore even 3D engines are running fine. Constant data stream from DVB-T tuner to graphic card can therefore cause system to fail more likely than 3d engine.

Last edited by Offler; 10-04-2012 at 09:06 PM.
Offler is offline   Reply With Quote
Old 02-05-2012, 05:45 PM   #160
Offler
Senior Member
Offler's PC Specs
 
Join Date: Nov 18 2009
Location: Bratislava
Posts: 278
After few days of testing:


NB 1,492v
Mem 1,756v

Dram timing modes were set to 3-3/5-5.


Under stress test with Prime95 the memory voltage drops a lot to 1,66v. This stress is similar to any 3d engine especially when combined with video playback.


With those voltages system survived 7 hours in 35 °c ambient temperature, 60°c on NB, 70°c on CPU.

This chipset is beast. Therefore I decided to upgrade also my PSU. I removed the great Seasonic 500W and used Corsair 650watt PSU again. 5v and 12v rails stabilized a bit. 3.3v now shows 3.08v which is still not the best, but better than 3,02v with seasonic psu.
Offler is offline   Reply With Quote
Reply

Bookmarks

Tags
dfi, x3360, x48t3rs


Currently Active Users Viewing This Thread: 1 (0 members and 1 guests)
 
Thread Tools


Similar Threads for: DFi UT x48-t3rs, X3360
Thread Thread Starter Forum Replies Last Post
Initial notes, findings and tips DFI UT X58-T3EH8 eva2000 DFI Intel Motherboard / CPU 0 04-08-2009 03:17 PM
[Bjorn3d.com] DFI Lanparty LT X48 T3RS News Reviews & News Online 0 09-12-2008 07:41 AM
DFI UT X48 T2R - C1 Error sunjovasie DFI Intel Motherboard / CPU 3 19-09-2008 12:17 AM
[Anandtech.com] DFI X48 LT T2R: Floats like a Butterfly… eva2000 Reviews & News Online 0 11-05-2008 01:14 AM
DFI STREET IS NO MORE ++++ NEW DFI CLUB FORUM IS UP!!!!!!!! dinos22 General Hardware 52 26-01-2007 09:17 PM


All times are GMT +11. The time now is 03:18 PM.

no new posts