Video and Memory Settings
Our ASUS 8800 GTS 640MB graphics card has always shown excellent potential for overclocking when using sub-zero cooling. It was a 'no brainer' to put the combination of the QX9650 and the 8800 GTS together, in order to see what could be achieved. Our initial goal was to aim for around 450FSB while keeping memory speeds around DDR3-1800MHZ using CAS 7 timings. We tried in vain to get the board to clock our OCZ modules at CAS 7 or 6 when running the CPU over 4.5GHz. Eventually we decided to leave the primary memory timings loose at CAS 8-8-8-20 using the 1:2 memory divider and the 1333 Northbridge strap. It seemed that the board prefers to overclock at CAS 8 at this point in development. We eventually settled for 8-7-7-20 timings using a command rate of 1N and the 11X multiplier. This allowed the memory to clock above DDR1900 easily, and the board held speeds up to 485FSB stable for benchmarking. We fully expect to hit 500FSB with this particular processor once Gigabyte has tuned the BIOS.
Our other option was to use the 12x multiplier and a lower FSB to raise CPU speeds over 5GHz average, again, this did not work out well for us - the board either would not boot or would crash as we raised FSB rates within XP or the BIOS. Admittedly, we ran out of time trying to work out what combination of chipset/memory parameters are required to run the higher CPU multipliers. In general, current 12MB cache processors seem to have a lower FSB limit than previous generation 65nm 8MB cache CPU's. Gigabyte has done a remarkable job in maintaining FSB scalability on the larger cache processors even when benchmarking at high CPU speeds. Although the 640MB 8800 GTS is an aging graphics card, it is still capable of some very respectable benchmark scores. As we are relatively new to extreme cooling, we would have been happy on the mediocre side of the fence, but armed with this board we managed a clean sweep of current 3D benchmark records with a single 8800 GTS 640MB. We do not expect our position to last long; there are a number of benchmarkers breathing down our necks with LN2 cooling. Here are the results we managed with the GA-X48T-DQ6...
Super PI 1M
3DMark 2001SE
3DMark 2003
3DMark 2005
3DMark 2006
Aquamark 3
Please note Aquamark 3 seems to exhibit a 'bug' when FSB changes are made using OS level overclocking software. Hence the CPU and GPU scores are listed as 'N/A'. While these synthetic benchmark scores do not represent results from actual applications, they are a very good indicator of a board's true performance capability. As we will find out in our follow up article, Gigabyte is bringing one very fast board to market in the next 60 days. The only question left is if it is faster than the new ASUS X48 boards, we will answer that in the very near future.
22 Comments
View All Comments
Bjenkin - Monday, July 21, 2008 - link
I bought this mother, did a bios upgrade to ver. F5 and all went ok... but after a reboot entered a power cycling boot.Im still looking for a solution
k3NzO - Friday, January 11, 2008 - link
support gForce graphic or only ATI?ahron - Saturday, December 29, 2007 - link
I thought that those screen shots were also of poor quality, they should have cropped them.JKflipflop98 - Saturday, December 29, 2007 - link
Geez man, people will bitch about anything. I for one thought this lil quick and dirty article was a neat change of pace. It's not everyday anandtech rigs up a cascade cooling solution and breaks every world record out there.But. . . OMFG! I had to click an extra time to see the pictures! Whaaaaaa. Motherboards that aren't even relevant to this article don't work on 45nm because my friend says so. Whaaaaa. Douchebags.
4Linux - Wednesday, December 26, 2007 - link
The bigger story is why doesn't 31/33 chipset boards work with 45nM CPU's. It doesn't pass the smell test.Didn't anyone notice? Care?
These boards were made for 45nM CPU's and they don't work. Isn't that a "story"?
SRoode - Friday, December 28, 2007 - link
The 3DMark 2006 screen shows two different scores. 15628 and 16128...How did this happen?
wingless - Sunday, December 30, 2007 - link
6106 + 5881 + 8265 = 20252?!??!Those scores are wrong to begin with. WTF?
Rajinder Gill - Tuesday, January 1, 2008 - link
http://service.futuremark.com/compare?3dm06=434375...">http://service.futuremark.com/compare?3dm06=434375...That is the ORB link proving the score to be genuine.
Secondly, 3D MArk 06 uses an algorithm to work oput the final system score, it is not a direct addition of the 3 scores as one would
assume.
Regards
Raja
Rajinder Gill - Friday, December 28, 2007 - link
The score in the £d mark original is form the previous bench. That score will not update until the current score window is closed.regards
Raja
OccamsAftershave - Tuesday, December 25, 2007 - link
This stateoftheArt board has a floppy connector? Instead of, say, a second IDE?