Dual-channel SiS and VIA Chipsets for Pentium 4: SiS655TX and PT880
|
VIA had the honour to announce the first DDR chipset for Pentium
4 with its semi-legal P4X266. SiS had a lot of innovations too, like
the FireWire MAC controller integrated in the south bridge. And of
course, Intel announced a lot of novelties as well. In the spring
of 2003, i875P and then i865PE
chipsets appeared and remained unachievable for all their rivals in
terms of performance. As for the functionality, it wasn't until last
autumn that SiS and VIA caught up with them. We are not taking ALi
into account as ALi
1683 has a very poor functionality and a very low performance.
ATI's first more or less successful desktop chipset (Radeon
9100 IGP) is a long way from i865PE in what concerns performance
of discrete work (which can't be said about its graphic core performance
compared to i865G). Besides, ATI south bridges look really deplorable.
And now, a year later, the market sees a pile of boards based on modern
dual-channel Taiwanese chipsets. Will they surmount the 865/875 height?
SiS655TX
SiS655TX crowns the current line of SiS chipsets for Pentium 4, as it has the richest functionality (though it doesn't differ from SiS655FX in this aspect) and [presumably] the highest performance. Let us look at its functional scheme:
All the things we said about SiS655FX in our previous
article hold good here, so we'll only enumerate its capabilities
briefly:
- support of 400/533/800MHz CPU bus;
- dual-channel DDR266/333/400 memory controller (up to 4 modules/4 GB);
- AGP 4x/8x bus for external videoaccelerators;
- MuTIOL 1G bus (~1 GB/sec bandwidth) down to the south bridge;
- SiS964 south bridge;
- 2 IDE channels for 4 UATA66/100/133 devices;
- 2 SATA ports for 2 devices with RAID (0, 1, JBOD) function;
- 8 USB 2.0 ports;
- up to 6 PCI Bus Master devices;
- AC'97 interface for 6-channel audio codecs;
- Fast Ethernet MAC controller (up to 100 Mbit/sec);
- support of low-speed peripherals.
Both bridges have almost standard functionalities today and feature no "luxuries", such as integrated FireWire or Gigabit Ethernet controller. The company is certainly planning to introduce the support of PCI Express and DDR2, but there's no information as to when SiS656-based products will appear on the market. In fact, SiS655TX+SiS964 is all but a spitting image of i865PE/i875P+ICH5R, the only difference being that the former appeared a year later.
SiS655TX has a feature that fits very well its low price: the chipset allows to apply the dual-channel mode not only to two/four identical DIMMs (as is the case with other dual-channel solutions including those by Intel and VIA), but also to unsymmetrical pairs, and even to three modules. Thus, SiS655TX-based boards keep their advantage over single-channel chipsets in terms of performance, but provide users with more flexibility and save them the trouble of choosing an operable pair of DIMMs for the dual-channel mode.
The promised difference from SiS655FX consists in a revised memory controller.
The new technology is called Advanced HyperStreaming. But if we visit
the company's site, we can read that the Advanced
HyperStreaming Technology in SiS655TX "enhances the transfer
rate on the path from host block to memory controller block".
Does that ring any bell to you? It definitely reminds us of last-year
press releases made by Intel (about its i875P) and mainboard manufacturers
(concerning their own acceleration technologies in i865PE-based boards).
But instead of going on repeating
about this "coincidence", we'll try to estimate once
again how much SiS655TX is better than its predecessor and if the
couple can compete with i865PE/i875P.
VIA PT880
PT880 is VIA's first ever dual-channel chipset (KT880 has been announced
but obviously, we'll have to wait for boards based on it to appear).
PT800, its predecessor,
though having set no performance records, showed a surprisingly good
functionality that could compete (and how) with top rival products.
Nothing except the memory controller has been revised in this VIA
chipset, so the functional scheme of PT880 must seem familiar to you:
These are PT880 capabilities, in brief:
- support of 400/533/800MHz CPU bus;
- dual-channel DDR266/333/400 memory controller (up to 4 modules/8 GB with ECC supported);
- AGP 4x/8x bus for external videoaccelerators;
- Ultra V-Link bus (~1 GB/sec bandwidth) down to the south bridge;
- VIA VT8237 south bridge;
- 2 IDE channels for 4 UATA66/100/133 devices;
- 2 SATA ports for 2 devices with RAID (0, 1, JBOD) function;
- 8 USB 2.0 ports;
- up to 6 PCI Bus Master devices;
- AC'97 interface for 6-channel audio codecs;
- Fast Ethernet MAC controller (up to 100 Mbit/sec);
- support of low-speed peripherals.
You have to take a really powerful magnifier to see the differences from rivals' solutions. The PT800/880 memory controller supports twice as much memory and modules with ECC. The VIA chipset has no integrated FireWire or Gigabit Ethernet controllers, and it won't be before the company's next product (PT890) appears that PCI Express and DDR2 will be supported. The only functional difference between PT880 and PT800 except the number of memory controller channels is that the newer one uses a twice faster Ultra V-Link bus to the south bridge. But why should one increase the bus's bandwidth up to ~1 GB/sec is anyone's guess.
Performance
Testbed:
- CPU: Intel Pentium 4 3.2 GHz (16x200 MHz), Socket 478
- Mainboards:
- ASUS P4R800-V Deluxe (version BIOS 1006) on ATI Radeon 9100 IGP
- ASUS P4P800 SE (version BIOS 1002) on Intel 865PE
- ASUS P4S800D
Deluxe (version BIOS 1002) on SiS655FX
- ASUS P4S800D-E Deluxe (version BIOS 1003 beta 001) on SiS655TX
- Soltek PT880Pro-FGR (version BIOS M1.1) on VIA PT880
- RAM: 2x256 MB PC3200(DDR400) DDR SDRAM DIMM TwinMOS, CL 2
- Videocard: Manli ATI Radeon 9800 Pro 256 MMB
- HDD: Western Digital WD360 (SATA), 10000 RPM
Software:
OSs and drivers:
- Windows XP Professional SP1
- DirectX 9.0b
- ATI SMBus Driver 5.10.1000.2b
- ATI GART Driver 1007b
- Intel Chipset Software Installation Utility 5.1.1.1002
- SiS AGP Driver 1.17
- SiS IDE Driver 2.04a
- SiS SATA Driver 1.05
- VIA Hyperion 4.51
- VIA SATA Driver 2.10a
- ATI Catalyst 4.2
Test applications:
- CacheBurst32 0.91.07
- 7-Zip 3.13
- WinRAR 3.30
- Canopus ProCoder (Demo v1.25)
- Adobe Photoshop 7.0
- Gray Matter Studios & Nerve Software Return to Castle Wolfenstein v1.1
- Croteam/GodGames Serious Sam: The Second Encounter v1.07
- Digital Extremes/Epic Games/Atari Unreal Tournament 2003 v2225
Board |
ASUS P4R800-V Deluxe |
ASUS P4P800 SE |
ASUS P4S800D Deluxe |
ASUS P4S800D-E Deluxe |
Soltek PT880Pro-FGR |
Links |
ASUS P4R800-V Deluxe
|
ASUS P4P800 SE
|
ASUS P4S800D Deluxe
|
ASUS P4S800D-E Deluxe
|
Soltek PT880Pro-FGR
|
Chipset |
ATI Radeon 9100 IGP (9100IGP + IXP150)
|
Intel 865PE (RG82865PE + FW82801ER)
|
SiS655FX (SiS655FX + SiS964)
|
SiS655TX (SiS655TX + SiS964)
|
VIA PT880 (PT880 + VT8237)
|
CPU support |
Socket 478, Intel Pentium 4, Intel Celeron
|
Memory slots |
4 DDR
|
4 DDR
|
4 DDR
|
4 DDR
|
4 DDR
|
Extension slots |
AGP/ 5 PCI/ ASUS Wi-Fi
|
AGP/ 5 PCI/ ASUS Wi-Fi
|
AGP/ 5 PCI/ ASUS Wi-Fi
|
AGP/ 5 PCI/ ASUS Wi-Fi
|
AGP/ 5 PCI
|
I/O ports |
1 FDD, 1 COM on bracket + 1 connector, 1 LPT, 2 PS/2, 1 VGA,
1 S-Video, 1 Composite video
|
1 FDD, 2 COM, 1 LPT, 2 PS/2
|
1 FDD, 2 COM, 1 LPT, 2 PS/2
|
1 FDD, 2 COM, 1 LPT, 2 PS/2
|
1 FDD, 2 COM, 1 LPT, 2 PS/2
|
USB |
4 USB 2.0 + 1 connector for 2 USB 2.0
|
4 USB 2.0 + 2 connectors for 2 USB 2.0 each
|
4 USB 2.0 + 2 connectors for 2 USB 2.0 each
|
4 USB 2.0 + 2 connectors for 2 USB 2.0 each
|
4 USB 2.0 + 2 connectors for 2 USB 2.0 each
|
FireWire |
1 port + 1 connector for 1 port (VIA VT6307)
|
|
1 port + 1 connector for 1 port (VIA VT6307)
|
1 port + 1 connector for 1 port (VIA VT6307)
|
1 port + 1 connector for 1 port (VIA VT6307)
|
Integrated ATA controller |
ATA100
|
ATA100 + SATA RAID
|
ATA133 + SATA RAID
|
ATA133 + SATA RAID
|
ATA133 + SATA RAID
|
External ATA controller |
SiS180 (SATA/ATA133 RAID)
|
|
SiS180 (SATA/ATA133 RAID)
|
SiS180 (SATA/ATA133 RAID)
|
Promise PDC20378 (SATA/ATA133 RAID)
|
Sound |
AC'97 codec Analog Devices AD1888, Coaxial and Toslink S/PDIF-Out
on bracket
|
AC'97 codec Analog Devices AD1985
|
AC'97 codec Analog Devices AD1980, Coaxial S/PDIF-Out
|
AC'97 codec Analog Devices AD1888, Coaxial S/PDIF-Out
|
AC'97 codec VIA VT1616
|
LAN controller |
Marvell 88E8001-LKJ (Gigabit Ethernet) on a special bus
|
Marvell 88E8001-LKJ (Gigabit Ethernet)
|
3Com Marvell 940-MV00 (Gigabit Ethernet)
|
Marvell 88E8001-LKJ (Gigabit Ethernet)
|
VIA VT6122 (Gigabit Ethernet)
|
I/O controller |
ITE IT8712F-A
|
Winbond W83627THF-A
|
Winbond W83627THF-A
|
Winbond W83627THF-A
|
ITE IT8705F
|
BIOS |
4 Mbit Phoenix AwardBIOS v6.00PG
|
4 Mbit AMI BIOS v2.51
|
4 Mbit AMI BIOS v2.51
|
4 Mbit AMI BIOS v2.51
|
4 Mbit Phoenix AwardBIOS v6.00PG
|
Form factor, dimensions |
ATX, 30.5x24.5 cm
|
ATX, 30.5x24.5 cm
|
ATX, 30.5x24.5 cm
|
ATX, 30.5x24.5 cm
|
ATX, 30.5x24.5 cm
|
Test results
It wasn't long ago that we examined SiS655FX and compared
all modern chipsets made for Pentium 4. Therefore, we decided this
time to confine ourselves to dual-channel solutions only. One of the
fastest boards on i865PE was the leader (as you remember,
the chipset enables to catch up with the fastest i857P-based boards
providing a certain optimisation on the BIOS level is done). And we
used the results of the same model with BIOS optimisation disabled
as a "typical" i865PE. The pair of dual-channel chipsets
is, in fact, represented by one and the same ASUS board, which is
supposed to exclude all possible manufacturer-related performance
variations. But as for the Soltek model, it may prove not the fastest
one on PT880. And we certainly shouldn't discount the integrated ATI
Radeon 9100 IGP. It evidently slower than its rivals, but is the difference
really so great?
Well, there's still no alternative to a "maximised" i865PE (named i865PE+PAT in the diagrams) in terms of memory read speed. Curiously, a standard i865PE is not so impressive at all: it is outdistanced by current SiS and VIA solutions and only leaves the ATI chipset behind. SiS655TX takes the second place here and looks better than its predecessor.
Both "newcomers" are winners in memory write speed. Intel ex-champion loses almost 10 percent to them, and even loses a little to itself with optimisations disabled — though it is due to setup peculiarities of the ASUS board. Radeon 9100 IGP takes the last place here. The test is also marked by the Advanced HyperStreaming Technology in the new SiS chipset outdoing the just HyperStreaming Technology in SiS655FX.
Moment of truth: memory latency test. It is obvious that an optimised i865PE is still the best, whereas a non-optimised one looks very poor here. SiS655TX and VIA PT880 are quite up to the mark while the ATI chipset loses even to most single-channel products. The new VIA chipset shows almost the same latency with a single-channel PT800 which is rather typical of such pairs (remember i875P/i865PE and i848P, the latter would quite take the second place in today's testing with its 221 clocks). As for SiS655TX, it wins over its predecessor once again.
Summing up the low-level testing of chipset memory subsystems, we can be happy for the results of the "newcomers". Version TX exceeds SiS655FX in all the aspects and takes the second place, leaving behind even i875P/i865PE in write speed. VIA PT880 is not so eminent: it has the same unremarkable latency with its single-channel predecessor but makes up for it with a high read speed (typical of dual-channel solutions) and a record write speed. Thus, we think we can finally expect a more or less serious competition with top Intel products. ATI Radeon 9100 IGP confirms its rather low results and can only compete with single-channel products (and not even all of them) in speed.
To ilustrate what we have said above, we enlarge the archiving speed
test with the result of the best single-channel chipset, as this kind
of task is the most sensitive to memory latency. Here, i848P
(represented by an ASUS P4P800S-E Deluxe board) is just slightly inferior
to the second-best SiS655TX, though the SiS chipset is almost 1.5
times faster in memory read and 20 percent faster in memory write.
The latency VIA PT880 reaches leaves it on the same level with a seemingly
slower SiS655FX.
7-Zip archiving confirms the WinRAR results: i865PE exceeds the improved SiS chipset and a dual-channel VIA debutant by 6 and 10.5 percent, respectively. The breakaway distance is not so significant as Intel products had a year ago, but the difference is still visible.
We chose MPEG2 encoding with the help of Canopus ProCoder from all the tests of our video encoding methodology. The reason for that was rather banal: the test showed the widest gap between the contestants. Memory performance is much more important than CPU clock frequency for all the tests of the methodology, so don't be surprised that the maximal difference between the rivals is as small as 4 percent. You should also bear this in mind if video encoding is paramount for you. Speaking about specific results, the ATI chipset finally improved its position, although its third place was not quite fair. The thing is, the ASUS P4R800-V Deluxe board raises the FSB frequency a bit too high (the resulting CPU frequency was 3271 MHz in our case, which is exactly those 2 percent above the norm).
Concerning image processing in Adobe Photoshop, it is only i865PE and ATI Radeon 9100 IGP that make themselves conspicuous. The former exceeds the bulk of contestants by 7 percent, while the latter is 6 percent worse than the bulk.
The three games that we used show the same picture: i865PE is the leader, SiS655TX is second-best, VIA PT880 is third-best and equal to SiS655FX. However, the gap between the rivals is not so large: i865PE only exceeds SiS655TX and the top VIA chipset by about 4 and 6-9 percent, respectively. The integrated ATI chipset takes the habitual last place, losing 9-10 percent to the leader.
Here is a table that illustrates some of our today's results:
|
Test, supremacy in percentage |
Read, CB32 |
Write, CB32 |
Latency, CB32 |
7-Zip |
WinRAR |
MPEG2 |
Photoshop |
RtCW |
SS2 |
UT2003 |
SiS655TX vs. SiS655FX |
+9,0
|
+9,6
|
+13,0
|
|
+10,8
|
|
+1,5
|
+3,7
|
+2,8
|
+2,5
|
SiS655TX vs. VIA PT880 |
+9,6
|
+0,7
|
+14,3
|
+4,2
|
+9,1
|
+3,4
|
+0,5
|
+4,2
|
+3,8
|
+3,2
|
SiS655TX vs. i875P/i865PE |
-9,8
|
+9,0
|
-14,3
|
-5,8
|
-10,2
|
0,3
|
-5,7
|
-4,4
|
-4,3
|
-3,1
|
VIA PT880 vs. i875P/i865PE |
-17,7
|
+8,3
|
-25,0
|
-9,5
|
-17,7
|
-3,0
|
-6,2
|
-8,2
|
-7,8
|
-6,1
|
Conclusions
In brief, SiS655TX and VIA PT880 can be quite competitive with top Intel chipsets. The problem is, though, that SiS and VIA introduced their dual-channel solutions almost a year after i865PE/i875P appeared. Moreover, neither chipset offers anything really unique or useful for customers and both SiS655TX and VIA PT880 have worse speed results than Intel solutions do: 5 and 10 percent lower (respectively) in real applications. But of course, Taiwanese companies are likely to set lower prices for their products, as usual, which will help them to win back part of the market.
Speaking about the future, new Intel products codenamed Alderwood and Grantsdale are about to be released, while it is still unclear when SiS and VIA will be able to offer chipsets supporting DDR2 and PCIE. And although it is very doubtful whether end users will really need DDR2 (or, to be precise, current revisions of this memory type), there seems to be no stopping PCI Express now. Thus, we can face an already-seen situation where Intel sets the yardstick and others try to qualify for it. Well, we won't have to wait long to see what it all will be like in reality.
04.05.2004
Write a comment below. No registration needed!
|
|
|
|
|