For those of you who were waiting  for a motherboard from Abit that is based on the Apollo Pro 133A chipset from VIA, your wait is over. Abit's new VT6X4 is like the VA6, but with AGP 4x support.  While the VIA 133A chipset is not perfect, it took VIA to give us the features that Intel has denied us. In fact, you can count on Intel continuing to make funky chipsets with Rambus support, and no special overclocking features, while VIA is hell bent on pleasing it's loyal customers with all the goodies you could want.

  The VT6X4 comes with Softmenu II, which is very nice, but Softmenu III would have been greatly appreciated.  It is unclear if Abit was unable to, or was talked out of including Softmenu III on it's line of VIA Apollo Pro motherboards. Intel clearly does not like motherboard makers producing VIA chipset-based motherboards for their processors that have lots of overclocking features.

  The VT6X4 has 5 PCI slots, 1 AGP slot, 2 ISA slots, and on-board Yamaha audio.  Fortunately, no space was wasted on an audio-modem riser slot.  The on-board audio can be quickly disabled in the BIOS, under the 'advanced chipset features' menu. The manual that comes with the VT6X4 is quite comprehensive. Abit has included an interesting disclaimer in the motherboard specifications section of the manual.  It states: "The 66MHz/100MHz/133MHz bus speeds are supported, but not guaranteed due to the PCI and chipset specifications".  I wonder what Abit knows about the VIA chipset that we don't?

  I wanted to see how well the Abit VT6X4 would overclock the Pentium III 550E in Slot-1 format (550E S-1), as well as the 550E and 650E Flip-chips.  I will discuss the Flip-chips and slot-1 adapters I tried below. The VT6X4 natively accepts Coppermine Pentium IIIs, so the Slot-1 CPU plugged into the socket and booted normally. 

The Setup: The 550E slot-1 version was from the newer B0 stepping, as indicated by the updated part number of (550/256/100/1.65v S1) with the product code ending in SL3V5 (packed 02/14/00). I was using Siemens PC-133 SDRAM in the test system, with an Elsa SDRAM GeForce 256 video card set at default core and memory speeds. The test system was running Windows 98SE. I was using the 5.13 reference drivers from NVidia, because the newest 5.16 drivers can disable 4x AGP mode if a VIA chipset is detected. When testing motherboards for overclocking capabilities I am always more interested in stability than final benchmarks.

  Overclocking The first speed to try is the safe one: 133MHz, with the memory set to the host clock (133MHz), and the core voltage set to 1.75volts (the default is 1.65v). The setting of 133MHz is considered "safe", because that's where the 1/2 AGP and 1/4 PCI clock dividers automatically kick in.  This brings the AGP and PCI slots back to their normal speeds of 66MHz and 33MHz respectively. The system speed rating at 5.5 x 133MHz comes to 733MHz. The test system booted right into Windows 98SE (with Direct X 7.0a installed), and was rock-solid in applications like 3D Mark 2000. 

  The next available speed in Softmenu II is 140MHz, so a quick trip to Softmenu II and a reboot got the system running at 770MHz. I left the memory clock at the Host Clock setting.  The system booted perfectly, and went through the benchmarks without any problems.

  Finally, it was time for the big 150MHz.  Leaving the core voltage at 1.75v, and the memory set to Host Clock, I upped the front side bus to 150MHz.  The system booted just fine, and ran the benchmarks without a glitch. 

Some Quick Benchmarks: The benchmarking was done with the Elsa GeForce card at it's default clock settings. Performance could have been boosted by another 5 to 10% by overclocking the GeForce card. I was using the NVidia 5.13 reference drivers.

  First off, the CPU speed numbers below are from Norton Utilities 2000 processor rating utility, and the values are an average of 3 runs each.

Speed:        100MHz    133MHz     140MHz      150MHz

Norton          284         378.8         387.2        414.9
Rating         (base)     (+33%)      (+36%)     (+46%)

  The benchmark increase is almost linear with the bus speed increase.

   Benchmarking under Direct 3D was done with 3D Mark 2000 at a resolution of 1024 X 768 in 16 bit color.  The scores shown below are an average of 2 test runs each.

Speed:        100MHz     133MHz     140MHz      150MHz

3D Marks       3832.5      4341         4405         4496.5
                    (base)    (+13%)     (+15%)      (+17%)

  The smaller increases above 133MHz on the front side bus are due to the GeForce card becoming the limiting factor when the system speed rises above 700MHz.  At the 133MHz setting, the system had an overall rating of 733MHz, which is where the benchmarks begin to level off with GeForce cards.  The only way to boost speed significantly further would be to overclock the GeForce card itself.

  To check out how much of a boost we can expect by overclocking the GeForce card, as well as the Pentium III, I ran 3D Mark 2000 at two CPU speeds; 733MHz and 825MHz (bus speeds of 133MHz and 150MHz respectively). I boosted the GeForce core from 120MHz to 130MHz, and the memory clock from 166MHz to 180MHz.  To do this with the reference drivers, you need to enable "Coolbits" by editing the Registry. Make this key value with 'regedit'

HKEY_LOCAL_MACHINE\Software\NVIDIA Corporation\Global\NVTweak

  Now make a DWORD value in the NVTweak key called 'Coolbits' (no quote marks) and set the value to 3.  When you reboot, a new hardware options tab will appear in the GeForce advanced menu (see AGP 4x section below for more detailed instructions on editing the Registry).  Don't play with the Registry if you are not comfortable with 'Regedit'.

Direct 3D performance with the GeForce overclocked:
System Speed:               133MHz                     150MHz

3D Marks                      4606.5                       4761
                                   (+6)                         (+6%)

  In both cases, overclocking the GeForce card yielded another 6% increase over the value obtained with the GeForce card at default speed settings.

  When you do a comparison between the 3D Mark 2000 scores with the system at 550MHz and the GeForce at it's default settings vs. the system at 825MHz with the GeForce overclocked to 130/180MHz, you get a pretty significant difference:

3832.5 vs. 4761 3D Marks = 24% increase

  With 3D Marks approaching the 5000 Mark on a standard SDRAM GeForce card, I'm not complaining. 

Flip-Chips: The first Flip-Chip and adapter combo I tried was the 550E (product code ending in SL3R3, packed 03/13/00), and the Abit SlotKET!!! version 1.0.  The test system failed to boot.  Checking all connections and retrying got the system to boot to Windows, but it quickly crashed to DOS with the error message "Windows protection error, restart system".  This happened several more times.  I decided that the slot adapter might be a problem, so I switched to the Gigabyte GA-6R7+ Flip-Chip adapter.  Sure enough, the Gigabyte adapter worked fine, and the system booted right into Win98SE. After a little checking, I found that the Abit ver. 1.0 adapter does not work properly with Coppermine chips.  We tried 3 different adapters, and all either failed to boot, or gave Windows protection errors.  Abit was kind enough to send us a version 1.1 SlotKET!!!, and it worked perfectly.  In the two pictures below, you can see that Abit made numerous changes to the circuit elements, including the addition of an integrated circuit chip to ver 1.1 of the SlotKET.  So if you are getting a slot adapter from Abit, make sure it's the 1.1 version.

Abit SlotKET ver. 1.0 (note missing integrated circuit)

Abit SlotKET ver. 1.1

Overclocking the 550E Flip-Chip: Using the Gigabyte adapter, the test system booted up without incident. Raising the core voltage to 1.75 and boosting the bus frequency to 133MHz worked perfectly, yielding a system rating of 733MHz.  Setting Softmenu II next to 140MHz, and then to 150MHz showed that they worked as well.  The system was stable at 150MHz with the DRAM clock set to the host clock. Stability in 3D mark 2000 was very good.  So if you can't get a 550E SECC2 CPU, you can get the same results on the VT6X4 with a good FC-PGA slot adapter and a Flip-Chip CPU. If you get an Abit SoltKET, make sure it's version 1.1 (with the extra chip).

Overclocking the 650E Flip-Chip: I also wanted to try the Intel 650E Flip-Chip Pentium III to see how fast it would overclock. We had one 650MHz Flip-Chip to play with, which had a packing date of 03/18/00 and a product code ending in SL3NM.  I put it in a Gigabyte GA-6R7+ FC-PGA adapter and installed it on the same VT6X4 motherboard that I tested the other two CPUs with. 

  In the past, Intel's slower chips have overclocked relatively better than their higher speed counterparts, which were, in essence, already partially overclocked at the factory by Intel.  So the question was, would we be able to get the 650E Flip-Chip up to 150MHz on the front side bus, giving us a speed rating of 975MHz? If the Coppermine chips act like all previous Intel chips, we should not expect that the 650 or 700MHz models will run at 150MHz on the front side bus, just because the 550 model does. So how did the 650E Flip-chip do?

Bus freq.   Results                        Speed rating

112MHz:   Booted, stable               728MHz
120MHz:   Booted, stable               780MHz
124MHz:   Booted, stable               806MHz
133MHz:   Booted, not fully stable   866MHz
140MHz:   Did not boot                   910MHz
150MHz:   Did not boot                   975MHz

  Therefore, using the same motherboard and GeForce card that had previously run at 150MHz on the front side bus with the 550E processor, we could only manage 120MHz on the front side bus if we wanted complete stability using the 650E chip. The 3D Mark scores with the 550E chip running at 825 were better than anything I could get with the 650E chip.  So keep this in mind when shopping for a new Coppermine Flip-Chip, not all of them will be stable at the higher bus speeds.

AGP 4X mode?: The whole point of getting the Apollo Pro 133A chipset is to get 4X mode AGP support.  Not that it has turned out to be much of a speed boost when implemented, but I still wanted to find out what would happen to benchmarks with AGP 4x mode enabled and disabled.  The GeForce card is AGP 4X-capable, so in theory, it should work on the VT6X4.

  The Elsa GeForce card worked well in Windows with AGP 4x enabled, but in 3D Mark 2000, the system became much less stable.  The symptoms included hanging in 3D Mark 2000, texture flashing and texture smearing.  The texture flashing problem is a well known occurrence when combining GeForce cards with some VIA and AMD chipset motherboards. One method of eliminating the problem is to force 2x, or in some cases, 1x AGP mode.  

  I decided to try a different setup that would let me run the video card in 4x mode without glitches. So I put the 650MHz Flip-Chip on another VT6X4 motherboard, also with a Gigabyte slot adapter, and paired these with a Guillemot 3D Prophet DDR GeForce card.  Fortunately, this setup worked well in all 3 AGP modes.  I could get to the 2x and 4x modes by using the BIOS setup menu, but to force 1x mode AGP, you need to do a quick Registry edit.  You can do this by using Regedit, which is built-into Windows.  Start it by typing regedit in the run command box, and then navigate to this key:

HKEY_LOCAL_MACHINE\Software\NVIDIA Corporation\Global\System

1) If the System key does not exist under the Global key, just make one by right-clicking on the Global key, and selecting 'New' from the menu, and then selecting 'Key'.  Now type in 'System', and hit enter (no quote marks). 

2) Now right-click on the System key, and select 'New' again.  Select 'DWORD Value' this time, and type in:

'ReqAGPRate', and hit enter (that's a Q, in Req, not a G).

3) Now right-click on the new ReqAGPRate key, and select 'Modify'.  In the value data box, enter the number 1, and hit the OK button.  

  You can quit Regedit now.  When you reboot, your GeForce card will be in 1x AGP mode.

AGP testing: The test system consisted of an Abit VT6X4 motherboard, Gigabyte adapter, 650E Flip-Chip, and 3D Prophet DDR GeForce card. The hard drive had Win98SE, Direct X 7.0a and the NVidia 5.13 reference drivers installed.  The system was running at 120MHz X 6.5 = 780MHz.  Settings included a 32MB AGP aperture, single AGP read and write wait states and default core and memory settings on the 3D Prophet DDR card. The 3D Mark results are shown below (average of 2 runs each).

AGP Setting:        1X              2X              4X   

3D Marks           4747.5        4843.5        4837.5
                                         (+2%)    (no increase)

  As you can see, enabling 2x did not offer much performance increase, and with this test system, no further increase was observed with 4x AGP enabled.  So even though 4X AGP support sounds good on paper, it has not proven to be very useful in practice, at least so far. But, based on the benchmarks numbers above, it appears that 4X mode did not even engage for some reason, even with the BIOS set to 4X AGP enabled.  Even using the Registry edit below, I still saw no indication that AGP 4X was enabled:

HKEY_LOCAL_MACHINE\Software\NVIDIA Corporation\Global\System

"EnableVia4x"=dword:00000001

Final Comments:   

  Intel has been saying for months that they are converting their entire processor line over to the Flip-Chip format as soon as possible, but there are very few motherboards that will take this new chip format without an adapter. Considering the Flip-Chips have been semi-available for 7 months, we are starting to wonder if something strange is going on here. Intel wants you to buy their new i810 and i820-based motherboards, but power-users don't want them for numerous reasons.  Most of the non-Intel motherboards that do have the so-called FC-PGA socket, do not have variable core voltage settings for the CPU. I hope Intel hasn't been twisting mobo company's arms about not making native Flip-Chip motherboards with variable core voltage settings!!!  grrrrrrrr

Summary:

  The VT6X4 has many features that are very desirable in a new "Coppermine-capable" motherboard, including AGP 4x support.  Unfortunately, the 4X mode does not appear to work with GeForce 1 cards, as indicated by the benchmarking.  This may or may not be an issue that can be fixed with drivers, or BIOS updates. 

  The VT6X4 is a stable board for overclocking 'Coppermine' Pentium III chips, and Softmenu II makes overclocking simple.  I would have liked to see a few more options in Softmenu II, including separate menu choices for the AGP and PCI clock dividers at each bus speed, and the ability to set the AGP mode to 1x, as well as 2x and 4x.  

  It is clear from these results that Coppermine-based Pentium IIIs have the same limitations on overclocking as previous Intel chips.  The lower speed versions overclock relatively more than the higher-speed versions.  It is not worth paying more for a 650MHz chips, than for a 550MHz chip, since both can be overclocked to near the same level.  In fact, in our tests, the 550E chip overclocked higher than the 650 chip. Keep in mind that overclocking results vary depending not only on each individual CPU tested, but also on the components in your system.  Some video cards and PCI cards overclock better than others, and at the 150MHz setting, both types of slot are overclocked.

  The VT6X4 is a great low-cost, Coppermine-capable motherboard with some nice features.  We give it a KickAss Gear thumbs up, Abit did another excellent job.

Copyright 2000, KickAss Gear