Gigabyte P45 UD3 review

[K404]

Gigabyte P45 UD3 overclocking + benching review.


Intro.

Since the introduction of Intels Core2Duo in Summer 2006, Gigabyte have put a lot of work into their motherboard design and BIOS coding in order to be seen as a serious purchase for high-end overclocking. From the initial C2D boards based on P965, through P35, X38, X48 and now to P45, Gigabyte have made leaps and bounds which have forced many users to completely re-think their opinion on the companys offerings. P45 represents Intels final mainstream 775 chipset before Nehalem and its associated family members move performance and design in a different direction. With this finality in mind, reviewing a P45 chipset should be an excellent way of seeing what can be done with a matured and fully understood architecture, as well as seeing, (hopefully,) the fruits of Gigabytes 775 work.

P45 is a revision of the earlier and very successful P35 chipset which was used in such boards as Gigabytes P35 DQ6, as well as popular boards from other companies- for example, the Asus P5K series, which stole most of the limelight. The P45 series has introduced a long list of extra BIOS options available for tweaking- this will be brought up again later on.

The subject of this review is the Gigabyte P45 UD3. UD3 stands for Ultra Durable 3. It’s a design development which increases the copper content of the PCB layers to reduce signal noise primarily, but will also have an effect on thermal stability, given that copper is a very good heat conductor. More board details will be given further on in the review.



Box and Contents:






The box is predominantly white with some green fade. Component and speed support logos are present as well as the technologies and design features present on the board. The back of the box goes into detail about the capabilities of the board and the design features. Looks are always subjective but I like it. The info is useful and the box looks clean and fresh. Gigabytes marketing have paid attention to the effect of colour choice!







Opening up the box, we find: IDE and floppy cables, S-ATA cables, the I/O plate, a driver CD, the mobo manual and a multi-lingual guide to the basics of building a computer, with an addendum. Basic, but it’ll do.



The Board.






As previously mentioned, the board is part of Gigabytes UD3 series. This point aside, the UD3s design is fairly pedestrian.







There’s a fairly well specced 6-phase PWM design for the CPU, along with a 2-phase VDIMM control circuit and 2-phase for the P45 Northbridge as well.

The 8-pin CPU power plug has plenty of space around it and the clip faces outwards, so unplugging it shouldn’t be a challenge for the majority of users. This can be seen above.


Because this is the entry-level UD3 board, there is only a single PCI-E x16 slot and the FETs for the CPU PWM are not heatsinked. (again, can be seen above.) On the plus side of cost-cutting, there are no heatpipes on this board (Hooray!!,) so changing the Northbridge cooling will be very easy. The cooling I would rate as a plus point (in theory)- given the lack of heatpipes, its easier and cheaper to add to a cooling solution than have to change everything in order to change one.





The I/O panel offers 8 USB ports, a LAN port, two Firewire, 6.1 sound with optical + co-ax S/PDIF out and the standard legacy keyboard and mouse ports.

Four fan headers are onboard, which is a slight disappointment, although fan controllers aren’t expensive if more are needed.





The RAM slots are JUST far enough away from the PCI-E slot to allow RAM removal without interference from the GPU. Better than a large percentage of boards available so credit where it's due.



6 S-ATA ports are included for the ICH10 southbridge along with an extra 2 ports that run from a dedicated chip. The ports are placed in a way that even dual-slot GPUs will not cover them, but they run across the board instead of the ideal placement of being angled along the edge.





The CMOS battery is directly under the PCI-E slot, so even single-slot GPUs will partially obscure it. The CMOS clear pins are open and need shorted by a convenient metal object if necessary. No jumper is present, although to be honest, fiddling inside a case to move a tiny jumper is never convenient. A screwdriver shaft should be easier to use if the user has steady hands.





The NB heatsink is designed in a way that isn't great for airflow. Theres a plate over the top of it that covers top-down airflow and there are two "spines" that block airflow coming in from the side- the way most air would move in a computer case ie- front to back.








There are phase activity LEDs in the top corner of the board. The software-based power saving features require EIST and C1E enabled in BIOS, so I didnt persue it!

[K404]

BIOS


All kinds of clock skews and additional RAM delay timings to squeeze every possible MHz out of the components. Theres towards 70 timings in Gigabytes BIOS and adjusting one will have an effect on others. Getting the best of these will require several long sessions in BIOS for testing.

The BIOS is not all that great at recovering from being pushed too far. The board has a habit of starting up for a seconds, then powering down, continuously until power is removed. A full system flush with a 5-10 minute wait normally did the job, but an occasional CMOS reset was needed.








The clock skews can be adjusted in steps of 50ps, the two drive voltages in steps of 100mV









I feel absolutely no shame in saying...I have very little idea what all those channel timings do!




Max voltages. Interestingly, only the VDIMM options flash red. VDIMM starts flashing from 2.32V.




Minimum voltages. Some real scope for extreme undervolting/low-power computing there!

[K404]

Voltage readpoints:











I appreciate that the most accurate reading points would be from the ouput of the relevant chokes, but given that all of those would need wires soldered from the back of the board, I went for points that could be measured from the top. The voltages were all consistent for me during testing.

[K404]

Test setup:

Gigabyte P45 UD3, BIOS F4
Intel C2D E8600 E0
9800GX2

2x1GB Crucial Ballistix PC6400. 64MB ICs (twin-sided…the good stuff!)

Simple watercooling for CPU




Test methodology, drawbacks, assumptions, notes.

The stock NB cooler was used. It was not removed to change the thermal paste or to modify for extra contact pressure.

All voltages were measured with a multimeter. The I/O cases were used as ground.

Twin low-speed 80mm fans were positioned over the RAM at all times.

The NB did receive some airflow from a low-speed 80mm fan positioned over it and the back of the GPU.

This review is primarily about the overclocking ability and performance efficiency of the board. If you want FPS from the latest games, file transfer rates or CODEC abilities….sorry, you are in the wrong place. Most similar boards will perform very similarly to one another and I don’t have a hoard of available boards to use for comparison.

As previously mentioned, there’s a LOT of BIOS options. I chose to avoid most of these, and left as many as possible set to auto. Specific values changed will be noted where applicable. While the BIOS is being discussed here… two major flaws jump out. The voltages can be keyed in using the keyboard BUT…. If a value is keyed in that doesn’t match a BIOS option, the voltage moves to the maximum available. Do not key in values if you are not completely paying attention!! The other flaw is that, if leaving most settings to Auto, some will default to un-necessarily high values. Of course for RAM latencies this only means a loss in performance. For voltages values, this can hurt the lifespan of components


Using Memset in Windows, changing Performance Level is a hit n miss affair. It appears to work in jumps on this board. Setting it from 13 to e.g. 11 has no effect on stability but has no boost to performance. Setting it from 11 to 10 results in insta-freeze. (from ~590FSB.)

The BIOS has normal, turbo and Extreme options for the RAM. Changes this setting has an effect on tRFC, tRAS and PL. Setting it to Turbo is fine for values over 600FSB, I suspect Extreme will be best around stock FSB.


FSB.

The important one! FSB can be summed up very easily: vTerm (some companies call this vTT). This one setting decides how far the FSB goes and an overclocking session will be determined by how far a user is willing to push this. Gigabyte (and Asus) have recently published new charts of what voltages are considered safe, but interestingly, they are significantly different to the accepted safe levels for P35. Several P35 users found that vTT values over 1.35V would quickly degrade or kill their C0 45nm chips. The new safe value is over 1.60V. Seeing as this is a termination voltage primarily for the CPU, I’m not sure how the chipset revision can have such a difference to the safe values. I appreciate a lot of users will need time to come around to this new approach, if they come around to it at all. Intel have not released any updated information in a Whitepaper that I have seen, so user beware!

As an example of vTT and its effects: setting it to 1.60V got SuperPi 1M stability to 615FSB. 1.64V got that up to 635FSB. No slouch at all! SuperPi32M stability appeared to be around 580FSB at 1.64vTT. (PL11 for all)

FSB Benching:

With 1.64vTerm, 1.61vPLL, 1.50vNB, 1:1 @ 5-5-5 RAM timings I managed:

590FSB 3DM05
610FSB AM3
635FSB SPi 1M
580FSB SPi32M
610FSB 3DM01
580FSB for XP Pro installation








RAM

Gigabyte have made a big deal out of the RAM capabilities of these UD3 boards- they are even running a competition for the highest DDR2 values obtained using boards from the UD3 series. This competition, despite having two MONTHS(!) Left to run, has already seen several new DDR2 World Records set.

In the same way that vTT is squarely in charge of raw FSB, vNB and vDIMM are completely in control of the RAM. Realistically, for high FSB values, the RAM will run 1:1 so some of this impressive ceiling will go to waste.

RAM ratios:

200 strap: 3:4, 3:5, 1:2
266 strap: 4:5, 2:3, 1:2
333 strap: 1:1, 5:6, 5:8, 1:2
400 strap: 1:1, 3:4, 2:3
Reports persist that the 3:5 ratio is fundamentally bugged, so kudos to Gigabyte for limiting its presence.


Quad core settings depend on the FSB limit. For a limit up to ~480FSB, it will be possible to use the 200 strap with the 3:4 ratio. For 65nm quads with FSB limits over 500, the best mix of strap and ratio will probably be 266/ 4:5. For 500FSB, that will give 625MHz on the RAM, which should be fairly easy with good sticks.

For duallies with FSB ceilings towards 600, the 333 strap will be a necessity due to the RAM limit imposed by a 266 strap with a 4:5 ratio. 560FSB would give 700MHz on the RAM- ok for SuperPi 1M and maybe AM3 but not much else. I tried ~560FSB on 266/ 4:5 and the board wouldnt POST. Maybe better NB cooling and more volts would help here, as the load on the NB at those settings would be pretty heavy.

As for timings Vs MHz, the standard DDR2 limitations still apply. Towards 600MHz with 4-4-4 timings will be fine if the RAM can handle it. Over 600MHz on the RAM will need 5-5-4 or 5-5-5 timings unless VDIMM is high.


RAM benching:

With vNB at 1.44V and VDIMM at 2.44V, 700MHz booted fine at 5-5-5 timings on the 333 strap. SuperPi was fine up to 720MHz. Spi32M stability was far lower- around 655MHz.The box advertises 1366+ speeds (683MHz,) but I believe this will be strongly tied to the RAM sticks used.







CPU-based benches.

SuperPi was average. At 4GHz, the best 1M time I could pull was 11.594 seconds (572x7, 333 strap, 1:1 ratio, MAXMem) and the best 32M was 11M 29.890. (471x8.5, 200 strap, 3:4 ratio, copy-waza, MAXMem.) Using a competitive 4Ghz leaderboard on OCX for reference (http://www.ocxtreme.org/forumenus/showthread.php?t=961), it is clear that better times can be had with DDR2 based setups, although most need ~600MHz on the RAM @ 4-4-4 timings. Motherboard BIOS cannot be very well tuned for all benches, it looks like Gigabyte have went for improved performance in "bigger" benches. Again, credit where its due. Each DFI mobo has at least one BIOS available that is brilliant for SPi1M at the performance cost of all other benches.







GPU performance.

Technically, a mixed bag. Testing with an nVidia 7-series showed FPS and performance to be bang on but some results with a 9800GX2 were a little lower than expected. The numbers for comparison came from an nVidia 790I reference board, so it is entirely possible that an nVidia mobo + GPU combo unlocks some driver+ interface magic and gives better scores. The 750I and 780I probably offer the same boost if the FPS are THAT important, but their problems with FSB and heat are well documented.

A check of GPU-Z showed that the GFLOPs for the cards were spot on, so PCI-E bandwidth is fine. The BIOS has a "Robust Graphics Booster," which has Auto, Fast and Turbo settings. Changing this setting made no apparent difference. The P45 chipset appears to have limits on PCI-E frequency, especially with an overclocked FSB. Values of 105MHz and higher had a negative effect on the boards ability to POST.


Sub-zero prep:

The PCI-E slot is very close to the CMOS battery, so insulating around that is a bit of a challenge. There are also a 2 caps right above the slot. The Southbridge is nearby as well.




CPU: Enough space between the CPU socket and the capacitors. Insulation was a breeze.



Sub-zero benching.

This is a let-down. The VCore accuracy gets progressively worse as the values are raised in BIOS. 1.75V in BIOS was giving me at least 1.85V real. 1.65V BIOS gave me at least 1.70V real. I did not take a full range of numbers or test under these conditions, I moved back to BIOS and lowered volts.


The CPU temp was unstable and higher than expected with these high VCore values. That is a logical conclusion, but the temperature rise was much more dramatic than it should have been and worse than I have seen on other boards at similar loads.

The maximum benchable MHz for the board were disappointingly low. I did not even see 5.6GHz for SuperPi1M under dry ice. The CPU has been tested on a P5K-E under dry ice and only stopped at the boards known FSB limit of 575-580, giving me 5.75-5.8GHz.

These facts combined lead me to believe the PWM is not up to the task of high MHz and high voltages at all.

At lower volts (to at least 1.40V BIOS) there is slight overvolt at the FETs- around 0.025V. The PLL feedback loop goes the other way and suggests an undervolt of around 0.025V. Nothing all that serious either way at lower voltages. These were measured at low MHz however and should be considered separately to the high voltage behaviour.




Conclusion.

The raw FSB and RAM numbers are nice, but for extreme benching they are wasted if the board cannot do the business when cranked.

The extra copper in the signal and ground layers doesnt appear to have any obvious effect on bench stability, unless somehow its contributing to the boards poor MHz scaling, though I dont think it is.

Good or bad under extreme cooling, that represents a very small percentage of the market. Users with watercooling should have a great time with this board. Maybe my review board was slightly dodgy. Maybe I screwed up somewhere else, but on an entry-level board, Granite-hard PWM is probably not to be expected. Component makers are realising that overclocking is an ever-growing sport and will want to give us incentive to buy better quality parts. That might not be the case here of course.

The lack of a second PCI-E slot might limit its appeal slightly, but for those who are completely happy to stick with a single GPU, the board should do the job very well.

My thanks to Gigabyte UK for the board sample, and also to Andrew Borkowski for his patience and assistance.

[p0inT]

Good review, its nice to see reviews and overclocking abilities of the other Gigabyte UD3s borads then just the UD3P as you see all the time.

[kemo]

great review you got there m8 , it shows how reviews should have been made , i only wish if you would include Quad core FSB overclocking if you have a chip lying around

[K404]

Yea, I appreciate some quad results would have been good to include, but I dont have a quad any more.

[OC4/3]

Great result there,really enjoy the review.
How i wish there are more people who write review of this type and nature.

[Oliver]

you did a good job Kenny

keep pushing it