Yes, our skin is pasty white, and sunlight hurts our eyes.
But you get to benefit! We’ll show you just how this new chipset contender, which packs a wallop with AGP 8X and unofficial DDR400 support, stacks up.
Introducing the 648 chipset
The 648 chipset has nearly every one of the latest features you might want in a new chipset. The 648 north bridge chip has a revamped AGP interface with AGP 8X support andsurprise!twice the bandwidth of AGP 4X solutions. The front-side bus supports the latest Pentium 4 chips with 533MHz bus speeds. And the reworked memory controller is faster than in the 645/645DX chipsets, with the ability to host three DIMMs of DDR266 memory or two DIMMs of DDR333. Unofficially, the 648 will support DDR400, as wellthe necessary bus-to-memory clock ratio is there, and it worked flawlessly in our tests with a single stick of Corsair DDR400 memory. (When the time comes, I expect SiS to release a “648DX” chip that’s unchanged in silicon but has official support for DDR400.)
In between the 648 north bridge and the new 963 south bridge chip is SiS’s proprietary MuTIOL chipset interconnect. This 16-bit interconnect runs at an effective rate of 533MHz, delivering a total of 1GB/s of bandwidth. That’s twice the speed of VIA’s “Enhanced V-Link” and four times as fast as Intel’s Accelerated Hub interconnect.
The extra bandwidth will be needed, because SiS’s new 963 south bridge chip can push a whole lotta bits at once. The chip supports up to six USB 2.0 ports at 480Mbps a pop, plus three IEEE 1394a (also known as Firewire) ports at 400Mbps each. The 963’s updated disk controller can support dual ATA/133 channels, as well. Like its predecessors, the 963 also supports the full range of south bridge three-letter acronyms, including PCI, LPC, and ACR. The PCI controller can sustain six PCI master devices at once.
The AC’97 audio built into SiS’s 735/745 chips for the Athlon sounded much cleaner to my ear than, say, VIA’s south bridge audio. This time out, SiS has added more AC’97 channels, so the 963 can support six channels of audio plus a V.90 modem. (Yep, AC’97 is also used in chipsets for modem support.) As a result, the 963 can drive 5.1-channel surround audio speaker systems, if mobo makers choose to implement all six channels of SiS AC’97 sound.
All told, these feature additions and updates give the 648 chipset nearly every cutting-edge feature one could imagine, with a few possible exceptions like Serial ATA. In short, the 648 is loaded.
Since we’ve included results for the VIA P4X333 chipset below but not the new P4X400, I should explain a few things. First, the P4X333 is one of the oddest (and least vaporous) cases of vaporware I’ve seen. VIA sent us a P4X333 reference board for testing with a PR rep attached, and we proceeded to review the chipset in anticipation of P4X333-based products hitting the market soon. Yes, there’s an ongoing legal fight between VIA and Intel, but VIA itself and key partners like Shuttle were selling VIA-based Pentium 4 boards regardless. The chipset performed well, and our review was fairly positive as a result.
Weeks passed. Readers wrote us asking where they could buy P4X333 boards, and no one had an answer. The P4X333 was AWOL.
Now, the P4X400 has apparently launchedthis time for real. Shuttle and Soltek boards are rumored to be hitting the channel, and some publications received VIA-branded P4X400 boards (not reference boards, but full-blown products) to review. Unfortunately, we weren’t one of those publications, so we aren’t able to include the P4X400 here today. P4X333 test results appear below as an interesting little curiosity. We should have a P4X400 board soon, and we’ll review it then.
The method to our madness
We tested with a lot of different configurations for this review, so hold on. You’re about to see a lot of info here, and the table below will be important to understanding the results. For instance, we tested the Intel 845G chipset in three configurations: with DDR266 memory, with DDR266 memory using the 845G’s built-in graphics, and with DDR333 memory. (Although the 845G doesn’t officially support DDR333, the chipset is entirely capable of running the memory at DDR333 speeds, unlike the 845E.)
That said, I wish we’d had time to include a few other configurations. For instance, take the Intel 845E. Simple time constraints prevented us from testing with the 845E and including results. For what it’s worth, the 845G results should be very, very similar to what you’d get out of an 845E.
Also, please keep in mind that chipsets, when they’re put together well, shouldn’t affect overall system performance too drastically, all other things being equal. The key to better performance in current chipsets is usually the memory controller; chipset makers tend to optimize their memory controllers over time to squeeze out more performance. Support for faster memory types like DDR400 and PC1066 RDRAM will really make a difference, too. The performance results you’re about to see may not show the kind of eye-opening mega-differences you’d expect out of a new CPU or graphics chip. That’s expected with chipsets.
As ever, we did our best to deliver clean benchmark numbers. Tests were run at least twice, and the results were averaged.
Our test systems were configured like so:
|Athlon XP||Intel 845||Intel 845G||Intel 850||Intel 850E||SiS 648||VIA P4X333|
|Processor||AMD Athlon XP 2200+ 1.73GHz||Intel Pentium 4 2.4GHz||Intel Pentium 4 2.4GHz||Intel Pentium 4 2.4GHz||Intel Pentium 4 2.4GHz||Intel Pentium 4 2.4GHz||Intel Pentium 4 2.4GHz|
|Front-side bus||266MHz (133MHz double-pumped)||400MHz (100MHz quad-pumped)||533MHz (133MHz quad-pumped)||400MHz (100MHz quad-pumped)||533MHz (133MHz quad-pumped)||533MHz (133MHz quad-pumped)||533MHz (133MHz quad-pumped)|
|Motherboard||Shuttle AK35GT2/R||Abit BD7-RAID||Abit BG7||Intel D850MD||Asus P4T533-C||SiS 648 reference||P4X333 reference|
|Chipset||VIA KT333||Intel 845||Intel 845G||Intel 850||Intel 850E||SiS 648||VIA P4X333|
|North bridge||VT8367||82845 MCH||82845G MCH||82850 MCH||82850E MCH||SiS 648||VT8754|
|South bridge||VT8233A||82801BA ICH2||82801DB ICH4||82801BA ICH2||82801BA ICH2||SiS 963||VT8235|
|Chipset drivers||VIA 4-in-1
|Intel Application Accelerator 6.22||Intel Application Accelerator 6.22||Intel Application Accelerator 6.22||Intel Application Accelerator 6.22||AGP 1.10.03
|INF update 1.50 beta
AGP 4.11 beta
IDE filter driver 1.20a
|Memory size||512MB (2 DIMMs)||512MB (2 DIMMs)||512MB (2 DIMMs)||512MB (4 RIMMs)||512MB (4 RIMMs)||512MB (1 DIMM)||512MB (2 DIMMs)|
|Memory type||Corsair XMS3000 PC2700 DDR SDRAM||Corsair XMS2400 PC2100 DDR SDRAM||Corsair XMS2400 PC2100 DDR SDRAM
Corsair XMS3000 PC2700 DDR SDRAM
|Samsung PC800 RDRAM||Samsung PC800 RDRAM
Samsung PC800 RDRAM at 1066MHz
|Corsair XMS3000 PC2700 DDR SDRAM
Corsair XMS3200 DDR400 SDRAM
|Corsair XMS3000 PC2700 DDR SDRAM|
|Graphics||NVIDIA GeForce4 Ti 4600 128MB (Detonator XP 28.32 video drivers)|
|Sound||Creative SoundBlaster Live!|
|Storage||Maxtor DiamondMax Plus D740X 7200RPM ATA/100 hard drive|
|OS||Microsoft Windows XP Professional|
Thanks to Corsair for providing us with memory for our testing. This memory is undoubtedly some of the best stuff on the market. Their XMS3000 DIMMs were actually able to overclock and run stable at DDR400 with conservative memory timings, and the new XMS3200 memory was able to run at 400MHz using moderate timings (though still at CAS latency 2.5, of course).
I should note that we’re using the Intel Application Accelerator drivers instead of the older Ultra ATA drivers. We elected to go this route because Intel is replacing its Ultra ATA drivers with IAA. In addition to providing support for Ultra ATA modes, the Application Accelerator does some prefetching to improve I/O throughput, so products based on Intel chipsets may have a slight advantage as a result. But then, that’s the point. We’re hopeful other chipset manufacturers will incorporate similar performance-boosting measures in their drivers, as wellif they haven’t already.
The test systems’ Windows desktops were set at 1024×768 in 32-bit color at an 85Hz screen refresh rate. Vertical refresh sync (vsync) was disabled for all tests.
We used the following versions of our test applications:
- SiSoft Sandra Standard 2002
- ZD Media Business Winstone 2001 1.0.2
- ZD Media Content Creation Winstone 2002 1.0
- Sphinx 3.3
- ScienceMark 1.0
- LAME 3.91
- MadOnion 3DMark 2001 SE
- Codecreatures Benchmark Pro
- Comanche 4 demo benchmark
- Serious Sam SE v1.05
All the tests and methods we employed are publicly available and reproducible. If you have questions about our methods, hit our forums to talk with us about them.
One of the key determinants of chipset performance is the memory controller in the north bridge chip. Sandra’s bandwidth tests measure one of the key elements for memory performance.
Paired with DDR400, the SiS 648 screams ahead of other DDR solutions and even outruns the 850E chipset with PC800 RDRAM. Only PC1066 RDRAM is faster. Of course, we’re only measuring one aspect of memory performance here. (I’m currently pondering a new memory latency test.) Our real-world tests will stress the chipsets’ memory controllers in different ways.
The SiS 648 shows up very strong in Business Winstone, bested only by the PC1066 RDRAM solution and Intel’s 845G chipset, which has always done especially well in this test. Interestingly enough, the gap between the top and bottom spots here is over 10 points. Content Creation Winstone
Yet again, the 648 is near the top, both with and without DDR400 memory. LAME MP3 encoding
Obviously, chipsets don’t make a great deal of difference during MP3 encoding.
With the super-intensive Codecreatures benchmark, we measure differences in tenths of a frame per second. The SiS 648 with DDR400 matches the 850E/PC1066 RDRAM combination stride for stride. 3DMark 2001 SE
We see a little more separation between the contenders in 3DMark. The 648 with DDR400 comes in just behind the 850E/PC1066 combo yet again. With DDR333, though, the 648 falls behind both the 845G and the P4X333. Serious Sam SE
The 648 puts in another strong showing in Serious Sam SE. Comanche 4
The SiS 648 takes spots two and three here, just behind the perennial leader. Obviously, the 648 has what it takes to handle 3D gaming and graphics well. We’ve seen reservations expressed over SiS AGP implementations in the past, but this latest one seems to work very well.
Here’s a great chipset test: Sphinx speech recognition. Sphinx’s high-quality speech recognition algorithm is performance limited by memory bandwidth, but recent Pentium 4 chipsets have been managing to run Sphinx at speeds faster than real time. The true goal here is to process speech at about 0.8 times real time, so there’s sufficient CPU time available for the other overhead associated with speech-enabled applications. So far, none of our test systems has ever reached the 0.8 goal.
Our PC1066 test rig hits the 0.8 mark, and the 648 with DR400 is right behind it! Impressive stuff. The 648 isn’t quite as quick with DDR333 memory, performing just a tick slower than an Intel 845G chipset with DDR333. ScienceMark
Oddly, the 648 actually scores lower with DDR400 here instead of DDR333. Of course, we’re only talking about a hundredth of a point, so we can probably chalk it up to the test’s margin of error. Overall, the 648 is the strongest of the DDR-based Pentium 4 chipsets. Only the 850E with RDRAM is faster, and even then, not by much.
The Primordia test, which is just one component of the overall ScienceMark score, stresses memory pretty intensively, so we’re looking at its results separately. As you can see, the 648 is near the top of the pack yet again.
In the 648 chipset, SiS has delivered one of the best Pentium 4 chipsets available. Paired up with DDR400 memory, the 648 is faster than anything we’ve tested except the 850E chipset with PC1066 RDRAM. With DDR333 memory, the 648 isn’t always fastest, but it’s often close. Not only that, but the 648 showed no significant weaknesses in any of our tests, which is an especially important consideration in third-party Pentium 4 chipsets. (The stillborn VIA P4X333, for what it’s worth, was inexplicably slow in both Winstone tests.)
Beyond raw performance, the 648 comes with more of the latest features than any of its competitors, including AGP 8X, USB 2.0, Firewire, six-channel audio, a 1GB/s chipset interconnect, and ATA/133. Competing solutions from Intel offer much less: no official support for DDR memory over 266MHz, no ATA/133 support, AGP 4X, and a 266MB/s chipset interconnect. SiS 648 boards ought to be cheaper, faster, and offer more features than Intel DDR motherboards.
And unlike VIA’s P4X333/P4X400, the 648 should be widely available soon from top-tier manufacturers, with motherboard partners like Asus and Abit offering boards. In fact, the last SiS board I used extensively was Abit’s 645-based SD7-533, and it was nothing short of excellent in terms of stability, performance, and compatibility. Abit has already announced its successor, the SR7-8X.
I should note, also, that the 648’s performance with DDR400 memory is the first proof we’ve seen of DDR400’s viability. Naysayers have speculated that DDR400 wouldn’t offer much of a performance increase over DDR333 because of the latency penalties exacted by the slower memory timings required to make 400MHz DDR stable. Not so. In bandwidth-sensitive apps, DDR400 is quite a bit faster than DDR333. And it’s as fast as or faster than DDR333 even in what we’ve traditionally considered latency-sensitive apps, like Business Winstone. Our PC1066 RDRAM rig did put on quite a show today, but it had the ring of a swan song. DDR400 can take us to the next level of performance, and that’s probably good enough.
Chipsets like the SiS 648 will help. The 648 is excellent. I wouldn’t hesitate to put one of these into my own system, given SiS’s solid track record with the 640 series of P4 chipsets. And having said that, there’s really nothing more to say.