iXBT Labs - Computer Hardware in Detail

Platform

Video

Multimedia

Mobile

Other

Dual CPU VIA chipsets based motherboards

February 11, 2001



Wide spreading of multifunctional OS on the x86 platform gave a possibility to increase intensively performance of the computation systems of this architecture: with a simple increase of processors in quantity in the system. Intel, having developed and pantented the SMP realization on their own platform, started developing a new chipset. The first one was i430HX, also known as Triton II. From that time practically all manufacturers might produce inexpensive dual-processor computers. Moreover, such configuration is acceptable even in case of home assembling. But powerful systems with more than ten processors required (like in the case of the processors of the previous generation, e.g. i486) special architectural solutions; for the majority of tasks they are superfluous, unlike relatively cheap dual-CPU machines.

Many users of work stations have appreciated new possibilities, but they still unsatisfied with the performance gain. And it was due to low scalability of any CISC-architectures and, in particular x86. In order to get a significant increase in performance - apart from two processors and appropriate OS (Windows NT, OS/2 Warp Server, different UNIX-systems, Novell Netware) it was necessary to have an optimization of applications counting on multiprocessor feature. In the vast majority of cases a work station is not a multitask system (unlike servers). The number of processes is low and limited, even one processor never loads completely, that's why OS is not able to divide tasks among processors, and application optimization for SMP is a paramount problem.

Intel, when releasing their first multiprocessor chipset, faced the same competitors in the chipset production field as now: ALI, VIA and SIS. But to none of these companies the license for SMP was sold, that's why in this market sector Intel possessed exclusive rights. Though I should correct myself: some companies (such as Micron, IBM, DEC, AMI, ALR) received a license and had their own goodies in this field, but they had no right both to deliver the license to somebody, and to give out their chipsets to the third parties. AMD representatives, with the release of K6 processor, announced a potential possibility of its operating in multiprocessor configurations, but it got failed because of the license limitations.

The first chipset not from Intel which could be shipped to anybody and which supported SMP was Aladdin Pro II from ALI, however, like the i440LX/BX, it supported two CPUs only with an additional chip provided. Intel reacted adequately - shipping of these chips to motherboard manufacturers would mean a withdrawal of license from ALI for chipset production for Intel processor forever. Nevertheless, a practice of cross licensing played with Intel a spiteful joke - VIA concluded an agreement with the greatest semiconductor manufacturer in the USA - National Semiconductor (their logo can be seen on any chip from VIA). National Semiconductor was never interested in processor wars, that's why Intel didn't worry about the cross-license with that company, but that was wrong… VIA had received a license access to all processor developing works of Intel, and the latter didn't risk to start legal proceedings with National Semiconductor. so, Intel lost its last exclusive right - for the production of budget SMP-chipsets (Serverworks, which is easily selling their own chipsets, does not position them on the market of inexpensive high-efficient work stations). VIA gave out a response quickly - in April of the last year they announced a support for two CPUs by VIA Apollo Pro 133A chipset. So, the epoch of unlimited ruling of Intel chipsets, starting with a release of rather successful Triton, is coming to end.

Today, the huge majority of manufacturers offer dual-processor boards on a VIA chipset. Interestingly that Tyan - one of the leaders in shipping of the boards for servers and work stations to the retail market - became a pioneer.

Participants

Today we have 5 boards in question:

Abit VP6

A well-designed yellow-brown board with a legacy RAID-controller from HPT. Despite sound features support by 686B south bridge, the board lacks for AC'97-codec.

As for some downsides, I can mark only rather big dimensions and periphery position of the main IDE connectors, but only one dual-processor board on the Apollo133A can be called little - Q-Lity CPV4-TD, that's why Abit is not guilty.

Presence of additional USB-panel and UATA-cable is a pleasant feature of all modern boards from this company. And this is not surprising: Abit works only for the Retail-market, and thus, only for an end-user.

Elitegroup D6VAA

This board just slightly differs from PC Chips M790MR in layout: Elitegroup is already one and half a year a department of the greatest motherboard manufacturer HsingTech, "famed" with its trade mark PC Chips (and some others). Nevertheless, there are some differences: M790MR can be produced also in MicroATX format and has no an optional possibility of installation of an external ATA100-controller. In fact, the sample of Elitegroup which I had, also lacked for ATA100, but it had a place for the HPT370 chip and two additional IDE connectors.

Like all other boards, this one supports CPU VIA Cyrix III, but interestingly that its installation is done without any change in position of jumpers - in a uniprocessor configuration the first socket is used for Intel, and the second - for VIA.

Downsides are the same as in the Abit board. On the other hand, I think it makes no sense to use cheap small cases for dual-processor boards.

EPoX EP-D3VA

A young (for Taiwane) manufacturer of EPoX motherboards is well-known for high quality design and perfect bonding. For example, only one board series on the VIA Apollo MVP3 can work correctly with TV-tuners on the BrookTree/Conexant chips, and these boards are produced exactly by EPoX.

Peripheral position of IDE-connectors of the main controller doesn't prevent installation of the board due to small dimensions. However, a bit too close location of the first connector of the main disc controller to the fourth DIMM slot makes no good, but it's unlikely that you would have to install memory in the last slot.

Usage of the outdated 596B south bridge, which lacks for a controllers of the peripheral ports, keyboard and thermo control, is compensated by the corresponding external chips. As for control of temperature, voltages and fan rotation this solution is positive - standard possibilities of the VIA south bridges are limited (ASUS, for example, uses additional chips even on the boards with 686A, where it is necessary).

Iwill DVD266-R

The layout of the board is not that good, on the both available revisions (0.9 and 1.0). The both variants are engineering samples (0.9 whose photo is above, and 1.0 which is traditionally green).

Power connector located between a processor socket and AGP slot surprised me a lot. It's still good that I used ADDA fans of standard dimensions, and not the ORBs which could cause some troubles. Huge dimensions together with peripheral position of the main IDE-connectors and a single capacitor is a weight on my mind. Location of the floppy-connector on the edge of the board behind the ACR-slot is just awful.

The KK266-R board lying in front of me has, however, a good layout. But I think that every product of a respected company that sells their products at rather expensive price must be above all!

The only thing that can reassure you is that all these are pre-release variants.

The VIA VT8233 south bridge lacks for controllers of peripheral ports, that's why there is used an external chips from Winbond.

The Iwill board is the only that is equipped with a full-value sound controller from Cmedia - a department of the powerful Hsing Tech, what arises some thoughts (relations between Elitegroup and PCChips started from a decent installation of Cmedia 8330 sound chips on the ECS boards).

On the other hand, though, these chips are also installed by Asus and Chaintech on their boards.

The chipset is realized according to the single scheme (without deletion of a codec and the chip), but it is too cheap and allows to equip it with full-value sound without lifting the prime cost.

MicroStar 694D Pro

MicroStar, which is one of the three leaders of Taiwanese motherboard manufacturers, seems to be aimed at confusing users:

  • There are at least three (and maybe more) modifications of dual-processor boards on the VIA Apollo 133A with 6321 marking
    • VIA686A+RAID
    • VIA686B+RAID
    • VIA686B+RAID+IEEE1394
    • It's possible that there are some variants without RAID
  • There are at least two variants with 6362 marking
    • VIA686B
    • VIA686B+SCSI
    • The both PCBs have a place for IEEE1394-controller

But in price-lists all are named "MSI 694D Pro". Besides, any of the boards instead of the CNR-connector can have ISA slot.

I had an outdated 6321 sample with 686A south bridge (today 686B is installed) and RAID-controller from Promise. The layout is rather good: the main IDE-connectors are a bit far from the edge but are too close to the fourth DIMM slot, a power connector is on the edge. In my opinion, though, in all boards with AMR/CNR/ACR-connector the latter should be put between AGP and the first PCI slot. It's connected with the fact that some AGP-adapters have huge heatsinks with fans, what in case of too short distance doesn't let us to install any board in the first PCI. A communication slot between AGP and PCI usually solves this problem.

Now you can look at their characteristics in a single table:

Abit Elitegroup EPoX Iwill Microstar
Links
Model
VP6
D6VAA
D3VA
DVD266-R
694D Pro
Chipset
694X/ 686B
694X/ 686B
694XDP/ 596B
8633/ 8233
694X/ 686A
AGP 4x (Pro)
Yes (No)
Yes (No)
Yes (No)
Yes (No)
Yes (No)
Number of memory slots
4
3
4
4
4
PC133/VCM/DDR support
Yes/Yes/No
Yes/Yes/No
Yes/Yes/No
No/No/Yes
Yes/Yes/No
PCI/ISA/AMR slots
6 / 0 / 0
5 / 0 / 1
5 / 1 / 0
5 / 0 / 1 (ACR)
5 / 0 / 1
LPT / COM connectors
1 / 2
1 / 2
1 / 2
1 / 2
1 / 2
USB external / internal
2 /1
2 /1
2 /0
2 /2
2 /1
UltraATA connectors
4
2
4
4
4
External ATA100 RAID
Highpoint HPT370
-
Highpoint HPT370
AMI MG80649
Promise PDC20265
Sound subsystem
-
AC'97 (ALC100P)
-
AC'97 (W83971D)
Temperature / fan rotation control
Yes / Yes
Yes / Yes
Yes / Yes
Yes / Yes
Yes / Yes
Fan-connectors
3
4
3
4
3
Soft Menu
Yes
Yes
Yes
Yes
Yes
FSB frequency range (in 1 MHz steps)
66-178 (66-178)
66-160
66-180 (66-180)
66-200 (66-200)
66-150
Clock-generator
Phaselink PLL202-11
ICS 9250BF-18
IC Works W204H
ICS CA02B-027
IC Works W225H
VCore voltage adjustment
Yes
Yes
Yes
Yes
Yes
BIOS
AWARD 6.0
AWARD 6.0
AWARD 6.0
AWARD 6.0
AWARD 6.0
Standard complete set (description, FDD- and UltraATA-cables , CD)
Yes
Yes
Yes
Yes
Yes
Additional features
USB-panel, Diskette for HPT, UATA-cable
-
Diskette for HPT, UATA-cable
Diskette for AMI, UATA-cable
Diskette for Promise
Dimensions, mm
245*305
244*305
235*305
260*304
252*305

Comments:

  1. Voltage adjustment is not available in all the boards with usage of two processors.
  2. Four memory slots can work simultaneously only with usage of PC100.

Iwill DVD266-R board or
Some words on DDR-memory, dual-processor solutions and GTL+

Less than a month ago it was announced that the latest DDR chipset from VIA (Apollo Pro 266) supports dual-processor configurations and now we have received confirmation of this fact in the form of a mainboard from Iwill.

Today Iwill ships to the market rather exotic products. First they introduced a pair of the motherboards on the Ali Magik 1 for Athlon CPUs with 266 MHz bus support and DDR-memory - KA266/KA266-R with RAID. Now the DVD266-R motherboard is also important creation. It is based on the abovementioned Apollo Pro 266, and supports two Pentium III processors.

Now arises a question why other manufacturers are only starting a production of uniprocessor boards on the Apollo Pro 266, and Iwill managed to create a dual-processor sample. It turned out that Iwill was not even going to produce a uniprocessor mainboard with DDR-memory support for CPU Pentium III, which have an outdated 133 MHz FSB (it only brings to nought all advantages of a possibility to transfer twice more data on the part "chipset-memory". Instead, all efforts of Iwill were concentrated on development of a dual-processor monster, which realizes all possibilities of the progressive chipset, and therefore suits a role of a platform for a high-efficient server or work station.

The dual-processor board used completely the same (and at the same price) Apollo Pro 266 chipset as uniprocessor models of other manufacturers. And the difference in the price is defined by trifles such as the second Socket-370 and additional layout. For dual-processor systems on the base of the predecessor, Apollo Pro133A, the difference makes from $20 to $50 depending on a manufacturer, what is not too much for an actual possibility to increase efficiency by 100% depending on a definite task. In this case the situation is similar - the Iwill DVD266-R board, which has RAID controller at that, is hardly to come out of the price frames set by its uniprocessor brothers. We have to learn out whether only this feature was a reason for creation of dual-processor product. A reserve Socket-370 is, though, is a telling argument in favor of the DVD266-R.

In order to get an answer, let's turn to AMD Athlon processors and chipsets for them. Their initial difference from Intel competitors is in a more perspective DDR EV-6 bus "processor-chipset" which recently reached base 133 MHz and effective 266 MHz against "hard" 133 MHz FSB of Pentium 3 CPUs, which remain 133 MHz ones both for data and for all other things. Intel delivers a response blow - the latest CPU Pentium 4 and i850 chipset transfer data via 400 MHz Quad-Pumped Bus with a possibility of transmission not only 4 data blocks, but also two addresses per clock. It didn't help to Pentium III, though.

It turned out that for SMP systems the EV-6 bus is preferable - its organization is implemented so that processors can use a bus and apply to a chipset at the same time, unlike GTL+, where processors use a bus one after the other. That's why we can't get approximately equal transfer "needs" of two processors and DDR memory possibilities. You can find a consolation only in the fact that realization of dual-processor configuration on the GTL+ base is rather simple and doesn't require huge investments, unlike the same variant on the EV-6 base. Absence of chipsets with SMP support for Athlon and delays in release of the 760MP chipset prove it. That's why Iwill was guided mainly by a low cost of dual-processor solution on the Apollo Pro 266, than by a higher performance increase from usage of DDR exactly in SMP systems relatively to uniprocessor solutions - on the base of Pentium III.

Test platform. Stability factors

The test system is the following:

  • Processors:
    2 Intel Pentium III 800EB, SL4 MBytes series (CC0 stepping, 1.7V rated core voltage) and SL464 series (cB0 stepping, 1.65V rated core voltage)
  • Memory:
    Kingston 512 MBytes SDRAM PC133 ECC or Samsung 2*256 MBytes DDR SDRAM PC2100
  • Video adapter:
    ATI Radeon 32 MBytes DDR
  • HDD:
    2 Western Digital WD200BB, 20 GBytes each, ATA100
  • Modem:
    Lucent WinModem
  • Sound card:
    Yamaha YMF744
  • Netcard:
    Digital/Intel 21143 and CNet Pro120C
  • SCSI-controller:
    Tekram DC390U2W
  • DVD-drive:
    Pioneer U03S (SCSI)
  • Mouse:
    Logitech M-UN58A (USB, HID-compatible)
  • Standard floppy-drive and keyboard
  • External ATA100, COM and LPT ports are off

OS Windows 2000 Server Eng with installed Service Pack1 was used.

All tests were carried out at 1024*768, Hicolor, 85 Hz. AGP4X was off on all boards, memory settings - "by SPD". HDDs were on one cable, the first one contained the operating system, the second one was not formatted.

In general, all the boards ran flawlessly. PCI-sharing worked ideally. While simultaneous wondering on the Net with the modem, playing MP3-files on another machine (connected with the test ones via the netcard) and watching DVD the system remained stable. The problems were concluded in the fact that two CPUs were with different stepping (the main thing is that the rated core voltage differs).

Besides, the integrated IDE-controllers played up from time to time. In Windows 2000 there is a problem of change of a south bridge - in the most cases you have to reinstall this OS, when replacing a motherboard by a new one, with another south bridge. In particular, this problem is connected with a change of a disc controller, and with an inability to recognize a new controller by the OS during booting. Among all IDE-controllers that I had (686A, 596B, 686B, VT8233, HPT, Promise, AMI) the OS could correctly cope only with replacement of 686A by 686B and vice versa and VT8233 by 686B, (i.e. VT8233-686B-686A-686B-VT8233 refused working only at the last stage).

As I already mentioned, there were processors with different stepping and core voltage. Strictly speaking, such system shouldn't work, but this configuration didn't worry me on all boards but one. All boards except Elitegroup required an installation of a CPU with CC0 stepping into the first socket. Below you can see a table of Vcore automatically set by the boards for a dual-processor configuration.

VCore1, V VCore2, V
Elitegroup D6VAA
1.69
1.63
Abit VP6
1.68
1.37!!!
EPoX EP-D3VA
1.71
1.66
MSI 694D Pro
1.69
1.69
Iwill DVD266-R
1.68
1.52!!!

Because of impossibility to control VCore in a dual-processor configuration, testing of the Abit and Iwill boards (on the whole, the latter worked good, but it refused to pass some tests from synthetical packets; in case of installation of equal processors the problems disappeared) were done with two CPUs from one series (SL464 with the rated 1.65 VCore).

Besides, I tested the EPoX board with processors both with different stepping and with the same one - we didn't notice any difference in performance.

Testing

Applications used for tests:

  • WCPUID, by H-ODA v.3.0 beta1 (a utility showing a real CPU frequency)
  • Testmem (memory read/write speed testing program)
  • Bapco Sysmark2000 (a synthetical test based on real applications)
  • Kinetix 3D Studio MAX r3.1 (3D-modelling packet which is the best optimization for SMP)
  • WinZip 8.0 (a popular archiver)
  • GO-GO v2.39 (MP3-coder optimized for SMP)
  • Madonion Video2000 (video applications' performance test packet, in particular for MPEG2 coding)
  • VGA Speed Benchmark v1.80 from CyberLink PowerDVD 3.0 (a program for testing DVD decoding speed in overlays)
  • ACDSee v2.41 (program for viewing graphics files, convenient in case of usage of slideshow mode with zero delays in time, as a test of successive read/write of HDD)
Real CPU frequency
Elitegroup D6VAA Iwill DVD266-R Abit VP6 EPoX EP-D3VA MSI 694D Pro
Real FSB Clock, MHz
803
797
798
802
798

Some lag of the Iwill board in the "record to memory" mode is mainly connected with the fact that in testing was used not a serial sample, but an engineering one.

Bapco Sysmark 2000

This test was carried out on each board 5-7 times, after that we calculated the geometrical mean. The relative error of the test is 1.5% (for the Content Creation Winstone 2000 the error was more than 2.5%, what brings to nought the sense of testing of the boards with similar performance). The Iwill board doesn't show a considerable gain (at least, it doesn't correspond to the difference in memory price). Note that the Abit board in a uniprocessor configuration lags behind the competitors by a great margin, and in dual-processor it is practically equal. A noticeable performance increase was achieved only in coding of audio-file (see the diagram), in other applications the increase from SMP is either minimal or absent at all.

Testing in 3DS packet came down to rendering of "an anisotropic wheel", a scene included in the standard complete set - the most popular scene in performance testing.

A perfect gain in case of SMP usage proves a developed optimization for multiprocessor configuration. As for performance increase of the Iwill board, it is again minimal.

With the WinZip 8.0 we archived a .MOV file, 580 MBytes in size in the "best compression" mode. There is some increase. Traditionally, archiving requires a lot of processor time, here we can see exactly the OS dividing tasks among processors, on an optimized but a "hard" application. And usage of a board with DDR yielded little.

MP3-coder Go-Go has an optimization for 3D Now!, SSE, SSE2 and SMP. We implemented packing of a .WAV-file (44.1 KHz, 16 bit) 31 minutes long (350 MB) into 320 Kbps MP3-file.

No comments.

Results of the Iwill are absent because of incorrect work of the program with this board (the results are 8-10 times worse than of the others).

MPEG2 coding results were taken from Video2000 packet.

20% performance, or 1/5 computer time… Isn't it too little?..

DVD decoding is one of those tasks that induce users to lift performance of home systems. That's why, apart from packing of different files, I decided to show you these results as well. (VGA Speed Benchmark shipped together with DVD-player from CyberLink gives such opportunity). Normal YUY2 mode was used.

Well, high increase in case of SMP usage, and the highest with DDR SDRAM... I was surprised at such efficiency gain of the Iwill board.

Quake III, or why there are no its results

You might know that SMP didn't give a high performance gain in Quake III. Each of you can calculate how much it will be beneficial to purchase a powerful computer, a video adapter, additional memory etc. instead of a dual-processor board and the second CPU.

This concerns practically all productivity results of multiprocessor systems based on the x86 and GTL+. Now we are waiting for a response from AMD 760MP.

Summary

First I want to consider overclocking - all boards feature rather huge possibilities in this field, but when building a work station, stability is a chief thing (then a couple of percents in performance over). If a rendering of some giant scene in 3DS can take several hours, with intensive loading both processors and memory, then I think that it's much better not to get a system failure a minute before the end than to gain ten minutes. That's why I didn't consider this aspect of boards' operation.

In general, boards showed that with intensive load and in a specially optimized applications you can get some (sometimes significant) performance gain, but today's situation with price for CPUs - when the price for each MHz slumps with increase in frequency, when there is much cheaper and not less efficient systems based on the AMD CPUs - doesn't leave a chance to multiprocessor systems to settle in the house of an end-user. The destiny of SMP systems now is servers and specialized work stations, preferably those which use optimized applications.

It's hard to mark out some card - their performance is very similar. Nevertheless, I'd like to stress the following:

Abit VP6

Fans of this manufacturer can sleep quietly - a high quality board with a good performance level (only uniprocessor mode made a mess). No unfavorable criticism, only a bit too price.

Elitegroup D6VAA

Everything in this board is good, especially absence of ATA100 RAID (I'm ready to overpay in order to get an external Promise, AMI or SCSI-controller). Nevertheless, even with it the board's price is minimal among the contestants. My choice.

EPoX EP-D3VA

Very similar to the Abit, but much faster, more compact and is pleasantly green :) And the main thing its cheaper price (the difference between these boards exceeds a decent 64 MBytes memory).

MSI 694D Pro

Relatively high price is compensated by presence of RAID-controller from Promise and the highest performance. In general, it's the best among the equal.

Iwill DVD266-R

This board deserves to pay attention to, but the purchase can have a sense only with considerable price decrease for DDR SDRAM.

Write a comment below. No registration needed!


Article navigation:



blog comments powered by Disqus

  Most Popular Reviews More    RSS  

AMD Phenom II X4 955, Phenom II X4 960T, Phenom II X6 1075T, and Intel Pentium G2120, Core i3-3220, Core i5-3330 Processors

Comparing old, cheap solutions from AMD with new, budget offerings from Intel.
February 1, 2013 · Processor Roundups

Inno3D GeForce GTX 670 iChill, Inno3D GeForce GTX 660 Ti Graphics Cards

A couple of mid-range adapters with original cooling systems.
January 30, 2013 · Video cards: NVIDIA GPUs

Creative Sound Blaster X-Fi Surround 5.1

An external X-Fi solution in tests.
September 9, 2008 · Sound Cards

AMD FX-8350 Processor

The first worthwhile Piledriver CPU.
September 11, 2012 · Processors: AMD

Consumed Power, Energy Consumption: Ivy Bridge vs. Sandy Bridge

Trying out the new method.
September 18, 2012 · Processors: Intel
  Latest Reviews More    RSS  

i3DSpeed, September 2013

Retested all graphics cards with the new drivers.
Oct 18, 2013 · 3Digests

i3DSpeed, August 2013

Added new benchmarks: BioShock Infinite and Metro: Last Light.
Sep 06, 2013 · 3Digests

i3DSpeed, July 2013

Added the test results of NVIDIA GeForce GTX 760 and AMD Radeon HD 7730.
Aug 05, 2013 · 3Digests

Gainward GeForce GTX 650 Ti BOOST 2GB Golden Sample Graphics Card

An excellent hybrid of GeForce GTX 650 Ti and GeForce GTX 660.
Jun 24, 2013 · Video cards: NVIDIA GPUs

i3DSpeed, May 2013

Added the test results of NVIDIA GeForce GTX 770/780.
Jun 03, 2013 · 3Digests
  Latest News More    RSS  

Platform  ·  Video  ·  Multimedia  ·  Mobile  ·  Other  ||  About us & Privacy policy  ·  Twitter  ·  Facebook


Copyright © Byrds Research & Publishing, Ltd., 1997–2011. All rights reserved.