DFI X48 LT T2R: Floats like a Butterfly…
by Rajinder Gill on April 28, 2008 4:00 AM EST- Posted in
- Motherboards
Still on our feet (BIOS)
Using AUTO for all sub-timings is generally acceptable until we start pushing for every last MHz of bandwidth and latency. The BIOS defaults are generally tight enough for reasonable performance on 24/7 PCs. We used Memset while running Windows XP to check where the AUTO DRAM timing defaults set the memory parameters. The timings shown by Memset were noted by us and entered manually into the relevant BIOS functions. We then began to tighten these values one at a time, noting the effects of any changes on bandwidth, latency, and stability. For benchmarking, we tend to use Memset once the XP operating system has loaded rather than entering overly tight settings directly in the BIOS. We can then begin to tighten the sub-timings far beyond 24/7 settings.
Additional Setting Ranges:
Ref to Act Delay (TRFC): The range of use for 1GB modules in 2GB setups is anywhere from 20-40 depending on overall memory speed. Lower values are more aggressive and give boosts in Read Bandwidth. For double density (2GB) modules, a higher tRFC range is required. For 4GB of memory running up to 1100 MHz, a value of 42 and above is usually required (54 is more compatible). For 8GB of memory a minimum value of 54 should be used.
MCH ODT Latency: ODT (On Die Termination) is used to match the MCH output impedance to the termination resistance of the RAM. Use of 1-2 is fine in most cases. Even for 4GB overclocking we did not find that more than a setting of 1 was required.
Write to PRE Delay (tWR): Range is from 10-13. For high RAM speed, use 12 and above for stability; for benchmarking 10-12 will be faster. Lower is faster.
Rank Write to Read (tWTR): Range is 9-11. Use 11 for stability and 10 or 9 for benchmarking. Lower is faster.
ACT to ACT Delay (tRRD): Range is 2-5. 3 is a good setting that allows high enough FSBs for most users. Higher numbers result in less memory read performance.
Read to Write Delay (tRDWR): Range is 8-9. 8 is ideal for most users; higher values are slower.
Ranks Write to Write (tWRWR): AUTO
Ranks Read to Read (tRDRD): AUTO
Ranks Write to Read (tWRRD): AUTO
The ranges for all these settings are quite well manipulated by SPD and BIOS. For those wishing to experiment, 4 is the lowest and most aggressive setting. 5-4-5 may work for SuperPi type benching. There is a slight write/copy speed increase with tighter settings, but this is only useful for extracting the last ounce of performance. For stability these are best left on AUTO
Read CAS# Precharge (tRTP): 2-3 works well on this board for most users. 2 is used for benchmarking while 3 and upwards are best for stability.
ALL PRE to Refresh: Range is 4-5. A setting of 4 is fine for most benchmarking, unless shooting high over 500FSB.
Facebook
Twitter
RSS
24 Comments
View All Comments
orangeblue - Friday, June 13, 2008 - link
hi,can anybody tell me if the ocz flex 9200 4gb-kit(2x2048) workz too instead the four 1024 plz.
Zoomer - Saturday, May 3, 2008 - link
I miss the analysis of other features than just plain memory latency.Max fsb overclock, max mem overclock, etc?
Running the board at 400fsb seems boring - lots of P965 boards can best that. Tons of people don't buy these multiplier unlocked chips.
Rajinder Gill - Sunday, May 4, 2008 - link
Hi,Dual core 45nm's top out at around 510-515FSB using air cooling. Most of the quad CPU info is in the review.
We are looking at perhaps adding some more content tomorrow. Mainly 2x2GB and 4X1GB maxmimum stable overclocks and best operating points - using air cooling.
When Using 45nm dual core CPU's X48/X38 chipsets are really only 'good' to around 500FSB tops for 24/7, after this VMCH requirements and total stability with low tRD becomes futile rather quickly. SOme of the DDR3 boards are a little better in this regard, though this overhaed always comes at a high voltage price when used with a suitably high CPU multiplier.
65nm CPU's in dual core form may cruise upto 550fsb with some work, perhaps a 67% GTL table will help for that. Many of the benches in the review were run at 8x500FSB just to reflect the maximum 24/7 somebody is likely to be able to achieve and/or use consistently. FSB values over this are prone to failure in applications like PC Mark Vantage or 3D in some instance (even with lots of GTL work). Our E8500 processor managed 510FSB with 4X1GB maximum stable on this board, while using 2x1GB memory allowed me to boot at a around 520FSB - though this was far from being called stable.
regards
Raja
Zoomer - Wednesday, May 7, 2008 - link
Thanks for taking the time to respond.Unfortunately, I have not been keeping up with the latest tweaks on the best boards. Thanks for the detail, though. :)
amalinov - Friday, May 2, 2008 - link
Am I imaging things or something fishy is going on here?The board has 2 PCIe 2.0 x16 slots - from the X48 MCH. OK.
The board has the following PCIe devices connected to the ICH9R:
1. PCIe x1 slot
2. PCIe x1 JMB363 SATA/PATA controller
3. PCIe x1 Marvell 88E8052 Gigabit Ethernet controller
4. PCIe x1 Marvell 88E8053 Gigabit Ethernet controller
5. PCIe x4 slot (physical x16)
IMHO in this situation the x4 slot can be utilized as x4 only if at least 2 of the PCIe x1 devices are disabled? (if so, then wich exactly?) But this is not mentioned anywhere - neighter in the review, nor in the official DFI specifications or manual. Also, in the manual there is no BIOS settings description section - so I can't find what does the "PCIE Slot Config - 1X 1X" option stand for? (I assume for putting the x4/x16 slot in x1/x16 mode when the other devices are not disabled?)
n7 - Tuesday, April 29, 2008 - link
Great stuff.I have one concern though.
What are you using to test stability for 8 GB?
I have found that what people like to consider "stable" is only stable because of the use of applications like P95, Everest, OCCT, & Memtest86+, which don't tend to really stress 4+ GB RAM & the Northbridge to the same extent as multiple instances of HCI Memtest or as LinPack.
Just curious is all.
Rajinder Gill - Monday, May 5, 2008 - link
HCI memtest seems the way to go for me for memory NBGTL related stuff.For CPU core testing Prime/PC Mark Vantage (Blu-Ray test).regards
Raja
aldy402 - Tuesday, April 29, 2008 - link
great review RajinderIt was very in-depth and well written. A big chunk of the memory portion seemed to focus on micron D9 chips. I have the 4gb Gskill(powerchip ICs) and would really like to see a write up on this board involving "other ICs" and maby a guide with 65nm quads.
also have you tested the performance differential between the intel and jmicron sata ports?
well done
-Aldy
Rajinder Gill - Tuesday, April 29, 2008 - link
HI Andy,Although the guide says Micron I did add a comment on TRFC and double density modules. tRAS adn tRFC are the 2 tiimings that need to be changed for the most part. The boards will hold upto 450FSB and a dual core 45nm 'stable' with most of the chipset registers set to fast/more aggressive. Comments are already in the guide for 65nm CPU's pertaining to leaving the GTL controls at defaults - nothing more is really needed tbh..
regards
Raja
coolbluewater - Tuesday, April 29, 2008 - link
Replacing two of the six USB ports ith two eSata ports would seem to make sense. Not sure why they didn't do this on a board in this price range.