Of course, all I could think was: comparison review.
What better way to test the merits of DDR SDRAM versus PC133 than to test a board that could run both types of memory? Now, many months have passed, and DDR products are finally becoming widely available. And we have here a review of Shuttle’s AV32, a hybrid DDR/PC133 board for the Pentium III and its ilk. As you might expect, we’ll be reviewing the AV32, but we’ll also let it serve as a kind of test mule for Via’s Apollo Pro 266 chipset. I think that’s appropriate, really. Keep reading, and I’ll show you why.
The AV32 is, in many ways, a “pure” implementation of Via’s Apollo Pro 266 chipset. That’s no bad thing, because that makes the AV32 one of the more advanced Pentium III motherboards on the market. It also makes the AV32 a study in the current state of the art for Via-based motherboards. Here’s a summary of the goods:
|CPU support||Socket 370-based CPUs, including Intel Pentium III and Celeron PPGA and FC-PGA processors|
|Chipset||Via Apollo Pro 266 (VT86333 North Bridge, VT8233 South Bridge)|
|AGP slots||1, 2X/4X AGP w/sidebanding and fast writes|
|Memory||2 168-pin DIMM sockets for PC100/133 SDRAM
2 184-pin DIMM sockets for PC1600/PC2100 DDR SDRAM
|Storage I/O||Floppy disk
2 channels ATA-100
|Ports||1 PS/2 keyboard, 1 PS/2 mouse,
2 serial, 1 parallel, 2 USB,
4 additional USB ports via 2 expansion headers,
1 line out, 1 line in, 1 mic in for AC’97 audio,
1 game port
|Monitoring||Voltage, fan status, and temperature monitoring|
The feature list includes several notables, not least of which are the dual DIMM sockets for two different types of SDRAMPC133 and DDR. Also notable are support for AGP 4X, ATA-100, integrated AC’97 audio, and bus speeds up to 166MHz. And pay attention to this one: you can’t run PC133 and DDR SDRAM modules in the motherboard at once. It’s either one or the other, folks.
The layout, and dawdling on the south bridge
Here’s how the board looks:
Notice a couple of interesting things about the AV32’s layout. First, there’s that funny connector next to the AGP slot that looks for all the world like a PCI slot that’s been turned around and offset just a bit. That’s an Advanced Communications Riser (ACR) slot, which is similar in function to the AMR and CNR slots that sit unused in thousands of motherboards around the globe. The idea behind both ACR and CNR is that cheap modems, network cards, and the like ought to go into those slots. Because CPUs and south bridge chips can now collaborate to deliver fuctions like modems and NICs, these little riser card slots were added to facilitate very inexpensive upgrades. After all, you need an RJ-11 socket for a modem, no matter how much of its logic is handled by the CPU.
The CNR standard is backed by Intel, but a range of companies have pushed ACR as an alternative, more broadly compatible standard. For more info on this stuff, see here. For now, let’s just say Shuttle’s choice of ACR over CNR makes sense in light of the Via chipset on this board. Also, ACR retains AMR compatability. You can grab an AMR modem to use in the AV32 for next to nothing online.
Next, the AV32’s AGP slot is adorned by a black brace, with a clip at the back that holds the AGP card in place. This brace is a nice touch, since video cards tend to get torqued out of their slots by video cables.
This mobo’s onboard sound comes courtesy of an Advance Logic ALC100P codec chip, which, in combination with the Via south bridge audio controller, delivers some pretty decent sound. Note that this is not a purely “soft” audio solutionthe Via south bridge acts as an AC’97 audio controller. (For more info on the ALC100P chip, see here.)
In casual use, I noticed no serious distortion or bus noise, though it was only casual use. In fact, using this mobo’s onboard sound made me wonder out loud whether PCI add-in cards are really necessary. If you’re building a new system based on the AV32, I’d at least give the built-in audio a shot before deciding to disable it and plunk down the cash for a PCI sound card.
Finally, the AV32’s layout is marred by one really annoying problem. Access to one of the two DDR DIMM slots is blocked by any standard-size AGP card in the AGP slot. Shown below is an Asus GeForce2 card blocking the tab. I had to remove the video card in order to put a DIMM into the DIMM slot.
There’s really no excuse for this sort of problem in this day and age. It’s not a huge problem once the system’s built, but having to pull your AGP card to swap out some RAM is just wrong.
The VIA Apollo Pro 266 chipset
Both of the chips in the Pro 266 chipset are substantially improved over their predecessors. Most prominent on the feature list, of course, is that “266”as in 266MHz access to memory. Double Date Rate (DDR) SDRAM sends data on the rising and falling edges of the clock, effectively doubling peak bandwidth. The fastest common variant of DDR SDRAM, PC2100, runs at 133MHz, or 266MHz DDR. Hence the name. The Via VT8633 north bridge chip provides this memory interface.
However, with Socket 370 processors stuck on Intel’s relatively slow GTL+ bus, the system isn’t likely to be able to take great advantage of that extra memory bandwidth. We’ll put that theory to the test below.
More interesting are some of the other enhancements Via has made to their north and south bridge chipsespecially V-Link. Most north and south bridge chips are connected by the PCI bus, which they share in common. Starting with its 8xx series chipsets, Intel put a much faster interconnect between these chips. Intel dubbed their new arrangement the Accelerated Hub Architecture. Via now follows suit with V-Link.
Like Intel’s Accelerated Hub, V-Link offers twice the bandwidth of PCI, or 266MB per second. And like Intel’s solution, V-Link is interesting because it doesn’t use a “wide” connection with high pin counts, like 64-bit or 128-bit interconnects, to provide all that bandwidth. Instead, it takes a more “serialized” approach, using an 8-bit, 66MHz bus that transfers data four times per clock cycle. V-Link doubles up on DDR, effectively offering a 266MHz transfer rate. (We’ve seen the same trick in Intel’s Pentium 4 bus, which runs at 100MHz, but is “quad pumped” for an effective 400MHz transfer rate.)
With V-Link in place, the PCI bus moves onto the VT8233 south bridge chip. Via’s Super South Bridge has long offered a whole slew of I/O options, and this latest revision offers more than ever. The chip now comes with ATA-100 support, a total of six AC’97 channels for AMR/ACR modems or audio, two PHY interfaces to support Ethernet on ACR, and an LPC interface.
The LPC interface replaces the PCI-to-ISA bridge you’d find in previous chipsets. The LPC (Low Pin Count) interface comes royalty free from Intel, who first implemented LPC in its 8xx-series chipsets. LPC handles legacy I/O functions like keyboards, floppy disk drives, and parallel ports. To software, it looks just like an ISA bus, but it runs at 33MHz off the PCI clock, instead of at 8MHz like ISA. This move explains the AV32’s lack of ISA slots, and in the grand scheme, it’s definitely a good thing. (For more on LPC, start here.)
All in all, the Pro 266 is the most advanced chipset available for Pentium III-based systems.
The AV32’s overclocking options are pretty decent, and Shuttle’s obviously trying here to appeal to the enthusiast market with a little more gusto than they have in the past. The AV32 does have BIOS-based overclocking options, as you can see below…
However, using the AV32’s “jumperless” front-side bus speed settings can require the use of a jumper. For instance, I used a PIII 500E to test the AV32, and I wanted to crank it way past its default 100MHz bus speed, to well over 133MHz. With the AV32’s FSB jumper set to the default setting, I was only able to choose bus speeds near 100MHz in the BIOS menu. In order to hit 133MHz and beyond, I had to set the jumper to 133MHz. This “safety measure” may appeal to OEMs and folks building systems for Aunt Erma, but I found it mildly annoying.
Also, the AV32 doesn’t allow the fine-grain control most of the real overclocker’s boards do. For example, one may choose speeds of 133, 136, 140, 145, 150, 160, and 166MHz, but nothing in between. Believe me, there’s a heckuva a lot of difference between 150 and 160MHz when you’re trying to push that CPU to the edge. With the competiton offering speed adjustments in 1MHz increments, this arrangement seems a bit anemic.
The AV32 redeems itself by offering a 1/5 PCI divider at 160 and 166MHz. This means that if you have high-quality RAM and a willing CPU, you could run your system with a 166MHz bus without making clock-sensitive peripherials like UDMA hard drives lose their cookies. Very nice.
Personally, I don’t have any such highfallutin’ RAM, but the AV32 was able to push this 500E to 750MHz on a 150MHz bus. At that speed, things were a bit flaky, but the system was quite stable at 725/145MHz, so long as I gave it a little extra voltage. Not a bad overclocking feat, all told.
I should also note that the AV32 has very clear BIOS-based options for memory tweaking, including 2- and 4-way bank interleave. Such tweaks tend to matter more on recent Via-based mobos, and not every board exposes these options.
Now, let’s see how this puppy performed…
As ever, we did our best to deliver clean benchmark numbers. All tests were run at least twice, and the results were averaged.
The AV32 test system was built with the following parts:
Processor: Intel Pentium III 500E processor at 667MHz on a 133MHz bus
Memory: 256MB PC133 SDRAM memory in two 128MB DIMMs, 256MB PC2100 DDR SDRAM memory in two 128MB DIMMs
Video: NVIDIA GeForce 2 GTS 64MB (Detonator 3 version 6.50 drivers)
Storage: Western Digital WDC AC41 8000D 18GB ATA-66 hard drive
OS: Microsoft Windows 2000 Professional SP1
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 2001
- Quake III Arena 1.17
- Vulpine GLMark 1.1
In the Quake III Arena timedemo tests, we used the game defaults for “Normal” and “High Quality” rendering.
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.
I’ve included scores here for both PC133 SDRAM and PC2100 DDR SDRAM. However, the AV32 didn’t get along well with the Micron DDR DIMMs we used for testing. These particular DIMMs came out of an AMD 760-based eval box from AMD (see our review here), and they work perfectly in the Gigabyte GA-7DX mobo in that system.
But they’re clearly marked as engineering samples, not final products. The AV32 wasn’t entirely stable while using them. When I asked Shuttle about this problem, they said they hadn’t yet qualified the AV32 with Micron DDR memory, and recommended Mushkin, Samsing, or TwinMOS memory, instead. Given the fact that Micron has just today, as I write this, started selling PC2100 DIMMs online through its Crucial subsidiary, I find it hard to fault Shuttle for this problem. Still, caveat emptor: it’s not clear this mobo works with Micron DDR memory.
And yes, mom, in the future I will stock multiple varieties of DDR DIMMs for testing.
We’ll kick things off with memory tests, to see how DDR memory stacks up against the older stuff. I’ve also included scores from a variety of other systems we’ve tested, just to provide a reference. First up is SiSoft Sandra’s modified Stream benchmark, which measures memory bandwidth.
The AV32’s performance doesn’t look too impressive next to that long, orange bar representing the Pentium 4 with PC800 RDRAM. But look again. The AV32 manged to beat out an 800MHz Pentium III on a 133MHz bus; that system used an older Via-based Asus P3V4X motherboard. The KT266-based AV32 beats it soundly, even with a 667MHz CPU. Via claims to have improved the memory controller in the Pro 266 chipset, and this performance bears that out.
Notice also that the difference between PC133 and DDR memory ain’t muchnowhere close to twice the speed. Our skepticism about the PIII’s ability to capitalize on DDR’s extra bandwidth may be confirmed.
Now for a really cool Linpack graph:
Linpack graphs can be a little confusing if you don’t know what you’re getting, but this one is easy. Because we’re not concerned with measuring the performance of the various processors’ on chip caches, we want to focus on the right half of the graph, where the data matrices are over about 320K or sotoo large to fit into cache. The lines represent data throughput, so higher is better.
On that front, we see something really odd. Yes, we see the Pentium 4 with RDRAM trouncing everything else, and the Athlon DDR rig doing pretty well, but those things are expected. What’s odd is that the AV32 is actually slower with DDR SDRAM than with PC133.
In technical terms, the little gold line is above the little red line.
That’s not what we’d expect here, and it may be the result of our memory tweaks. In order to get the AV32 running stable enough to complete our benchmarks with our Micron DDR DIMMs, we had to use very conservative memory settings in the system BIOS. It’s possible our tweaks harmed performance.
Nonetheless, DDR memory is supposed to offer twice the bandwidth of PC133. Clearly this mobo/CPU combination isn’t taking advantage. That’s especially strange because DDR won handily in the Sandra test.
Content Creation Winstone 2001
Maybe a few application benchmarks will help sort things out. CC Winstone is a decent mix of high-end apps, all of which are run together, like in real life.
Oddly enough, PC133 wins again. What can I say? The difference is slight, but it favors, the older, cheaper, and ostenstibly slower memory.
Note that the AV32 was only able to complete this test with a single DDR DIMM slot populated. To make things fair, I ran both the DDR and PC133 tests with a single, 128MB DIMM on the board. Keep that in mind if you try to compare these scores to a comparably-specced system.
Quake III Arena
Quake is known for being bandwidth hungry. Let’s see how the RAM types handle it…
And again, it’s PC133 with a very, very slight lead. Makes you wonder.
Finally, we’ll plug in a shiny, new 3D benchmark, Vulpine GLMark. This is a pretty intensive graphic test that pushes a lot of info through a system, creating some stunning visuals along the way.
DDR comes out ahead here, evening up the score a bit. Again, though, the difference is slight.
Obviously, the Pentium III isn’t ideally suited to take advantage of DDR SDRAM’s extra bandwidthand especially not at the 667MHz speeds we tested. This is an odd finding, because every lecture about cache hierarchies I’ve ever heard or read has talked quite a bit about how much of a performance penalty going out to main RAM imposes, and how processors are much faster than the rest of the systems in which they reside. In some ways, that may well be true, but I have to wonder if said penalty isn’t more about latency than bandwidth in the case of the Pentium III.
Of course, there’s also the big, bad bottleneck known as the Pentium III front-side bus. At 133MHz, this bus can transfer up to 1066MB/sec, or half the theoretical peak bandwidth of PC2100 DDR SDRAM. Other devices in the system will also be accessing memory via the north bridge chip (think UDMA hard drives, AGP cards, and the like), but the processor’s the main thing. At the end of the day, the PIII just isn’t well equipped to take advantage of DDR memory.
That said, the AV32 did an admirable job of showing us its greatest virtue: those PC133 DIMM slots. PC133 memory is cheap, and those two slots will hold a pair of very inexpensive 256MB DIMMs. If, for whatever reason, you are intent on building a Pentium III-based system soon, you may want to consider using the most advanced PIII chipset, the Pro 266, with the best value in PIII memory, PC133 SDRAM. Few other motherboards will give you this option.
Still, on balance, the AV32 isn’t the end-all, be-all of enthusiast mobos. For one thing, I can’t fathom why anyone would build a PIII-based single processor system right now. The AMD Duron and Athlon and Intel’s Pentium 4 are all more compelling choices, and at least one of those three choices is cheaper at present, if not two.
If you are intent on building a PIII system, other motherboards will offer better overclocking options, built-in ATA-100 RAID controllers, more PCI slots, and more DIMM slots of a single type. All of these things are probably important enough to enthusiasts to make other choices more compelling.
With the exception of the Micron DDR problems we mentioned above (for which we can’t really fault the AV32), this board was impressively stable in our testing, even though we consistently overclocked our PIII during the tests. The AV32 offers flexibility in RAM choices, decent built-in sound, a 1/5 PCI divider for mad overclocking, and better memory performance with PC133 SDRAM than motherboards based on Via’s older Apollo Pro 133A chipset. And, by virtue of its Pro 266 chipset, it has a leading-edge feature set. Not bad.