TR reviews the Via KT133A chipset

Well, it’s been a long time coming, but it’s finally here. Our review of Via’s KT133A chipset is later to market than a DDR motherboard.
Bah-dum-bum.

Seriously, folks, the KT133A has been around for a while, and so has the skeleton of this review, on which I’ve finally found time to put some flesh. Heck, my own computer has been running on a KT133A motherboard for a while now, come to think of it. And we just published a review of a KT133A mobo, the Abit KT7A-RAID. So I darn well ought to have published this review already, instead of running around in DDR land.

Thing is, even though DDR chipsets and motherboards are finally starting to hit the market in earnest, the KT133A is increasingly looking like it’s going to be around for a while. Read on to find out why.

Before we go much further, though, I should stop and point out that the Asus AV7133 motherboard we used for this review was provided by TR’s excellent hardware sponsors, Dr. John and the gang at KickAss Gear. If you happen to decide you want to buy a KT133A based mobo, you couldn’t do much better than to pick up an A7V133 from the folks at KickAss Gear. As KT133A boards go, it’s the best.

What you’re getting
The KT133A chipset is essentially the Via KT133 chipset, only tweaked to support a 266MHz front-side bus (FSB) clock frequency. That is, a 133MHz FSB that sends data twice per clock, actually.

We like to mix it up to keep a little confusion going.

But there is a little more to the KT133A than that. For one thing, the VT8363A north bridge chip—the one modified to support the faster bus, and the real heart of the chipset—has been widely reported to handle bus speeds quite a bit higher than the official 133/266MHz spec. KT133A north bridge chips manufactured very late last year or in 2001 have hit speeds of 150MHz and above. That’s a big deal, because previous Athlon chipsets just weren’t friendly to bus overclocking. Since KT133A motherboards are, most often, slightly revised versions of KT133 boards, they’re mature products, and many offer robust, menu-driven overclocking options. Taken together, these things could add up to one seriously fast Athlon platform for the overclockers among us.

Way back when, most manufacturers chose to pair the KT133 north bridge chip with Via’s VT82C686A south bridge, which was the official KT133 south bridge chip. (Via offers several south bridge chips that communicate with the north bridge using the PCI bus, so they’re interchangeable.) Via has since revised the 686 south bridge chip to support newer, ATA-100 storage devices. Most KT133A boards now ship with the new 686B revision of this chip.


A typical PC system configuration using the 686B south bridge

Beyond that, the KT133A still includes all the features that have made the KT133 the most popular Athlon chipset around.

But there are a few things the KT133A can’t do, and we’d best recount those. For one, the KT133 chipset doesn’t support DDR SDRAM. (DDR memory support is coming from Via in the form of the KT266 chipset, which we’ll be reviewing soon. The KT266 is much like its Pentium III-oriented sibling, the Pro 266.) DDR memory promises double the bandwidth of conventional SDRAM at a small price premium, but it’s been much slower to market than anticipated. Also, DDR DIMM prices have only recently begun to drop to reasonable levels, while PC133 SDRAM is exquisitely cheap. We’ll be testing the KT133A against AMD’s 760 chipset, which support DDR SDRAM, to see whether the price premium is worth it.

Also, the KT133A still links the north and south bridge chips via the PCI bus. All of Intel’s 800-series chipsets and Via’s newer chipsets, the Pro 266 and KT266, use a decicated, low-latency interconnect with 266MB/second of bandwidth—twice that of the PCI bus, and the PCI bus isn’t a dedicated link. In theory, the KT133A will be at a comparative disadvantage in certain scenarios where the system has to move large amounts of data from several sources and destinations at the same time. However, the KT133A’s main competitor at present is AMD’s 760 chipset, which also uses the PCI bus to link the north and south bridges.

Our testing methods
Our benchmarking suite is getting a little long in the tooth, since newer versions of a whole lot of these tests and applications are available now. However, we’ll run through ’em all one more time before putting the ol’ test suite out to pasture. Heck, in many cases, the older benchmarks will arguably do a better job of reflecting current—rather than future—application performance. We chose to test in Windows 2000 rather than in Win9x/ME for a simple reason: Win2K is much, much better than Win9x/ME. Once the next rev of Win2K, named Windows XP, makes it out the door, Win9x/ME will finally be put out to pasture. Yes, even for gamers. Win2K is making big strides in this area, and we expect Windows XP to dominate the desktop market in six months to a year (unless Linux finally makes an earnest run at the desktop). No enthusiast buying a new system today ought to use it for any length of time with WinME or the like.

As ever, we did our best to deliver clean benchmark numbers. All tests were run at least twice, and the results were averaged.

The KT133A test system was built with the following parts:

Processor: AMD Athlon 1.2GHz CPU on a 266MHz (DDR) busMotherboard: Asus A7V133 – Via Apollo KT133A chipset – VT8363A North Bridge, VT82C686B South Bridge, plus integrated Promise PCI-ATA-100 I/O controller – Courtesy of KickAss Gear

Memory: 256MB PC133 SDRAM memory in one 256MB DIMM

Video: NVIDIA GeForce 2 Ultra 64MB (Detonator 3 version 6.31 drivers)

Audio: Creative SoundBlaster Live!

Storage: IBM 75GXP 30.5GB 7200RPM ATA/100 hard drive

Because of some problems with the ATA-100 implementation on the Via south bridge chip in Win2K, we decided to test with the hard drive connected to the Promise controller. Microsoft is promising a fix in Win2K Service Pack 2, which is coming Real Soon Now, and has been for months. Via has their own ATA-100 driver for Win2K, but it was very immature at the time of the test. No, it’s probably not entirely fair to use a different hard drive controller in a chipset review, but then most decent KT133A boards now ship with extra ATA-100 controller chips on board, while most (all?) AMD 760 boards do not. And, in single-drive configuarations, we’re confident the performance difference is minimal. So there.

Our comparison systems varied only with respect to the motherboard, memory, and CPU. The Athlon DDR box looked like this:

Processor: AMD Athlon 1.2GHz CPU on a 266MHz (DDR) busMotherboard: Gigabyte GA7-DX motherboard – AMD 761 North Bridge, Via VT82C686B South Bridge

Memory: 256MB PC2100 DDR SDRAM in two 128MB DIMMs

The Pentium 4 system was built using:

Processor: Intel Pentium 4 processor at 1.4 and 1.5GHzMotherboard: Intel D850GB – Intel 850 chipset – 82850 memory controller hub (MCH), 82801BA I/O controller hub (ICH2)

Memory: 256MB PC800 DRDRAM memory in two 128MB RIMMs

For the Athlon/KT133 system, we used:

Processor: AMD Athlon 1.1GHz CPU on a 200MHz (DDR) bus – Courtesy of KickAss GearMotherboard: Abit KT7-RAID motherboard – Via Apollo KT133 chipset – VT8363 North Bridge, VT82C686A South Bridge – Courtesy of KickAss Gear

Memory: 256MB PC133 SDRAM at 133MHz

Finally, we included a Pentium III test system—though only at 800MHz, we thought it would be a useful reference point—using these components:

Processor: Intel Pentium III 800EB (Coppermine) CPU at 800MHz on a 133MHz busMotherboard: Asus P3V4X motherboard – Via Apollo Pro 133 chipset – VT82C694X North Bridge, VT82C596B South Bridge

Memory: 256MB PC133 SDRAM at 133MHz

We used the following versions of our test applications:

  • SiSoft Sandra Standard 2000.3.6.4
  • Compiled binary of C Linpack port from Ace’s Hardware
  • ZD Content Creation Winstone 2000
  • LAME 3.70
  • SPECviewperf 6.1.2
  • ps5bench 1.1 Intermediate
  • Adobe Photoshop 5.5
  • POV-Ray for Windows version 3.1g
  • 3DMark 2000 Pro build 335
  • Quake III Arena 1.17
  • MDK2 Internet demo
  • Expendable Internet demo

In the Quake III Arena timedemo tests, we used the game defaults for “Normal” and “High Quality” rendering, with a few exceptions. For the “High Quality” tests, texture detail was set to maximum and the “high” geometry settings were enabled, as well.

The test systems’ Windows desktop was set at 1024×768 in 32-bit color at a 75Hz screen refresh rate. Vertical refresh sync (vsync) was disabled for all tests.

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.

Memory performance
Since the major performance and feature differences between today’s PC chipsets are often about memory, we’ll kick it off with a couple memory tests. Behold the purty Linpack graph:

To better understand what we’re getting here, let’s concentrate on the red and purple lines, representing the KT133A and AMD 760 DDR Athlon rigs, respectively. These two overlap almost exactly in the left half of the graph, where Linpack is processing data matrices small enough to fit into the Athlon’s L1 and L2 caches. The Athlon 1.2GHz whups all comers when Linpack data fits into its caches. At 320K, Linpack exhausts the L1 and L2 caches, and main memory comes into play. At that point, the lines begin to diverge, with the DDR system taking a slight lead over the KT133A. The difference between the AMD 760 board with DDR SDRAM and the KT133A board with PC133 SDRAM is pronounced, but not terribly dramatic. Despite DDR SDRAM’s theoretical ability to deliver twice the bandwith of conventional SDRAM, even the memory-oriented Linpack test can’t tease much extra performance out of DDR SDRAM.

Sandra’s Stream benchmark measures memory bandwidth a little bit differently than Linpack, but the results are actually rather similar:

The relative performance of the different configurations here looks like a cross-section of the Linpack graph results at, say, the 1660KB matrix size. Once again, the DDR system comes out ahead, but not radically so. Let’s talk for a moment about how the KT133A stacks up against the rest of the field—and why. Compared to the 1.1GHz Athlon running on the KT133 chipset, the KT133A delivers a slight performance increase. This is almost entirely due to the bus speed difference (100/200MHz DDR vs. 133/266MHz DDR), not the 100MHz processor clock speed disparity. Although both chipsets can run PC133 memory at 133MHz, the KT133 has to resort to a bit of a trick, running the memory and front-side bus at different speeds. This trick can yield a bit of extra performance—a KT133 with a 100MHz bus and 133MHz memory is a little faster than a KT133 with a 100MHz bus and 100MHz memory. But running these clocks at different speeds adds latency and constricts bandwidth. Have a look at this excellent explanation of the problem at Ace’s Hardware, if you’re curious about why that is.

The KT133A, on the other hand, can run the front-side bus and memory synchronously, so it’s faster all around than its predecessor. The front-side bus, CPU-to-memory latency, and memory bandwidth all benefit from this more optimal arrangment.

Speaking of more optimal arrangements, it’s hard not to notice how alarmingly fast the Pentium 4-based system rips through these tests. Both the P4 processor and its memory subsystem are designed for excellent memory access performance, and they deliver in spades. Memory bandwidth isn’t everything, however, as we’ll find out in our real-world tests below.

POV-Ray 3D rendering
POV-Ray stresses x86 floating point performance more than anything else. Will the memory bandwidth benchmark results matter at all here?

Nope. The KT133A is in a virtual tie with the AMD 760 box for the lead. Moving on… LAME MP3 encoding
LAME is an open-source MP3 audio encoder, and we used it to encode a 64MB WAV file ripped from an audio CD. Compressed to fit into an 128Kbps stream, the resulting MP3 file was about 5.8MB, or just under one-tenth the size. You’ve gotta love MP3s.

LAME, another FPU-intensive program, likes the KT133A just as well as the AMD 760. These two programs demonstrate how, in certain situations, DDR’s extra memory bandwidth just isn’t useful. But then these are kind of “worst case” scenarios; not all real-world apps are so utterly CPU dependent.

Photoshop image processing
Photoshop does bang on the FPU, but it also uses integer, MMX, and SSE instructions, where it can, so it’s fairly well-rounded test. The ps5bench intermediate suite we used handles large enough images to tax system memory considerably, as well.

Here the KT133A falls a little behind the 760, but not by much. Let’s bust it out into individual tests…

As you can see, no single Photoshop filter made all of the difference. Instead, the KT133A lags behind the DDR board by just a teensy bit in nearly every test. Comparatively, neither chipset has a single great weakness—or strength.

SPECviewperf workstation graphics
The viewperf suite is a bit of a departure, because it stresses a chipset’s AGP implementation, among other things. Here’s how they stack up.

Notice the two Via-based systems (KT133 and KT133A) posting scores very close together on almost every test. Awadvs, in particular, is a big win for the Via systems.

However, the KT133A and 760 are neck-and-neck otherwise, with the 760 holding the slightest of leads in each of the other tests.

Quake III Arena OpenGL gaming

Q3 is known for being bandwidth intensive, and it runs like gangbusters on a Pentium 4. As you can see, the DDR system is appreciably faster than the KT133A, and the P4/RDRAM system is faster yet again. The pack tightens up as the screen resolution increases and the system starts to bottleneck on the video card. Frickin’ GeForce2 Ultra. 🙂

MDK2 OpenGL gaming

MDK2 is the first place we see the KT133A at a pronounced disadvantage in a real-world scenario. Here the DDR rig beats out the Pentium 4, but the KT133A falls behind. Expendable Direct3D gaming

Athlons own this test, but even so the DDR box leads the KT133A by a fair margin.

3DMark 2000

Once more, DDR SDRAM makes the difference for the Athlon between beating and losing to the Pentium 4. Games and graphics applications are obviously more memory bandwidth intensive, as these last few tests show. 3DMark also generates a CPUMark score by running some of its tests in very low resolutions. Here’s how that one came out:

The KT133A trails the DDR system by 15% here. Oddly, the Pentium 4 rises to the top, but it’s slower in overall 3DMarks.

Each of the indvidual game tests tracks pretty closely with the overall 3DMark scores. All in all, it’s clear why AMD pushed (some would say rushed) the first DDR systems out the door before the Pentium 4 launch. The KT133A is very fast, but for games and graphics, memory bandwidth matters.

Overclocking
We’ve seen where the KT133A matches its DDR-based competition, and we’ve seen where it falls a little behind. For many of us, this chipset’s ace in the hole may be something else entirely: the quality of the enthusiast-oriented motherboards that incorporate the KT133A. Few DDR boards offer the kind of robust multiplier and bus speed overclocking options available with the KT133A.

Just to illustrate this point, I decided to play with overclocking my own PC—where the Asus A7V133 mobo used in the tests above now lives—and record a few scores. Keep in mind that these scores are not from our test system. My system differs in several ways. Notably, it has 512MB of RAM, not 256MB. It was two 75GXP drives in a RAID 0 array. And it only has a 32MB GeForce2 GTS, not a 64MB Ultra.

I’ve also dropped in a 1.33GHz “AXIA” Athlon, which runs happily at nearly 1500MHz. Here’s what it can do at various bus/processor speeds:

I had hoped to include some really exotic bus speeds, but either my “cheap ‘n’ plentiful” RAM didn’t want to cooperate or this KT133A board isn’t good much past a 136MHz front-side bus. Plus, I didn’t care to corrupt my ATA-100 RAID array. The thought made me a bit touchy. RAID 0 is fast, but dual points of failure are a daunting prospect at times, especially when it comes to overclocking the FSB. Still, these are some decent performance gains from overclocking, even with my (relatively) slower video card. Keep that in mind when you’re weighing a KT133A system against a DDR-based box. The fact of the matter is, you may not be able to overclock your processor much at all with many first-generation DDR motherboards. The extra headroom you get out of a well-made KT133A motherboard might just tip the performance scales away from DDR.

What’s the frequency, Kenneth?
One more note before we wrap this all up: both the Asus A7V133 we used here and the Abit KT7A-RAID that Andy tested recently reported front-side bus frequencies slightly higher than 266MHz, even when the bus was set to its default frequency. Have a look:


The Gigabyte AMD 760 board is accurate with the clock frequency


KT133A boards seem to run a few ticks faster than we ask

At the 266MHz FSB setting, both the Asus and Abit KT133A boards seem to run the FSB at 268MHz. We’ve seen this kind of behavior before, but it’s beginning to become a problem. As you might surmise, this problem is exacerbated by higher CPU multiplier values. At what should be 1466MHz, this KT133A system reports an actual bus speed of 268.61MHz and a CPU clock of 1477.34MHz.

Now, this could be nothing more than a reporting error instead of an actual clock frequency reading, but assuming the reported speed is correct, it’s a bit of an annoyance. That would mean all of the preceding tests at 1.2GHz actually spotted the KT133A an 8MHz advantage. It’s not the end of the world, and it’s not clear this is a chipset problem instead of a motherboard problem, but it is a little bothersome.

Summing it all up
At the end of the day, it’s clear DDR-based systems are, indeed, faster than the KT133A. In a number of our gaming and graphics tests, the 1.2GHz Athlon could only beat out the 1.5GHz Pentium 4 with help from DDR SDRAM. If it’s bragging rights you’re after, only a DDR-based system will do.

However, the KT133A’s relatively strong performance across the board solidifies a conclusion many of us have drawn over the past few months: the current Athlon and Pentium III processors can’t take great advantage of the extra bandwidth offered by exotic memory types like RDRAM and DDR SDRAM. The Pentium 4, obviously, is a whole other story, but we won’t go into that now.

What matters here is simple: if you’re looking to build a PC right now, there’s no shame in going with a KT133A motherboard and PC133 SDRAM instead of the “latest and greatest” stuff. PC133 memory is still about half the price of PC2100 DDR memory—about $30 vs. about $60, respectively, for a 128MB DIMM. In the real world, most of us would do better to grab a KT133A board and 512MB RAM than to buy a DDR board and 256MB of RAM. If you’re faced with that kind of choice, grab the KT133A.

That said, Via’s KT266 chipset is just now hitting the market, and DDR DIMM prices are still dropping. Once the dust settles, it’s very likely the KT266 will be the new king of Athlon chipsets. And it will sport some of those nifty, new features like V-Link. If you can’t stand the thought of your shiny, new system being ever-so-slightly behind the curve, you may want to pop for a DDR solution. Freak.

Comments closed
    • Anonymous
    • 19 years ago

    I agree with the conclusions reached. I actually bought an a7a266 system and then returned it because it wouldn’t run games properly and seemed to have immature drivers. I replaced it with an a7v133 with twice the memory and am very very happy with it.

    • Anonymous
    • 19 years ago

    Wonder these glowing comments still hold after the discovery of the Via southbridge bug . . .

    • TwoFer
    • 19 years ago

    I’m presently building a KT133A box for some friends; they needed a reliable, low-cost solution which would be upgradable for a few years, and this is honestly the only one I see.

    Sure, there’s better tech arriving at the table — and in another year we’ll know more about what it is. For hardware you can buy now and depend on, Damage is right: it’s KT133A.

    • Anonymous
    • 19 years ago

    #2. I agree. The memory benchmarks are good and higher than SDRAM but jump into Quake3 and the FPS are not much different than using SDRAM. The same goes for RAMBUS. But if I had the choice I’d rather have DDR.

    • Anonymous
    • 19 years ago

    *[

    • Anonymous
    • 19 years ago

    Yea, the CPU does have to be optimized to take advantage of the extra memory cycles.

    Remember Pentium III with Rambus? it sucked! No memory advantage. Pair Pentium 4 with Rambus, and the scores are off the charts.

    I doubt we’ll see the same performance boost with DDR and Palomino, but it looks like Pally is being built from the ground up with DDR in mind, and should take better advantage of DDR.

    • Anonymous
    • 19 years ago

    DDR is a bust. I know Crucial sells 256MB PC2100 CL2.5 DDR RAM for ~$100 and it’s about the same price as PC133 RAM, but I don’t see the results I’m looking for. The bandwidth is 2x that of PC133 and the synthetic benchmarks look impressive, but all the real world benchmarks show a ~6% improvement. Something is VERY VERY wrong. Maybe the CPU has to be optimized for the bandwidth???

    –k

    • Anonymous
    • 19 years ago

    *[

Pin It on Pinterest

Share This