The Tiger MP was a good board, but it didn’t quite light up the enthusiast’s mobo scene, despite packing twocount ’emfront-side busses. Now, AMD’s 760MPX chipset promises to boost dually Athlon performance while cutting costs, and Tyan is back with the Tiger MPX. Can this new board bring the creamy smoothness of dual processing to a larger audience? More importantly, will Damage be able to make it through this review without going completely overboard waxing rhapsodic about the joys of multiple CPUs? Read on…
We’ve spent a fair amount of time playing with the Tiger MPX, and we’ll walk you through those ups and downs. But first, let’s take a look at the specs, so you can see just what the Tiger MPX has to offer.
|CPU support||Socket 462-based CPUs, including AMD Duron and Athlon/XP/MP processors (only Athlon MPs are officially supported in dual configurations)|
|Chipset||AMD 760 MPX (762 North Bridge, 768 South Bridge, National Semi 83627 Super I/O ASIC)|
|Interconnect||64-bit/66MHz PCI (533MB/s)|
|PCI slots||6 (2 64-bit/66MHz; 4 32-bit/33MHz)|
|AGP slots||1, 2X/4X AGP|
|Memory||4 184-pin DIMM sockets for up to 4GB of PC1600/PC2100 Registered DDR SDRAM (w/ECC support)
Will support unbuffered DIMMs in slots 1 and 2
|Storage I/O||Floppy disk
2 channels ATA/100
|Ports||1 PS/2 keyboard, 1 PS/2 mouse,
2 serial, 1 parallel, 2 USB (non functional),
4 additional USB ports via included PCI card
1 3Com 10/100 Ethernet NIC (optional)
|Bus speeds||100/133MHz (200/266MHz DDR)|
|Monitoring||Voltage, fan status, and temperature monitoring|
That’s pretty much it for the spec sheet, but there are a number of things there that need explaining. We’ll discuss those below, after we take a look at the board’s layout.
The Tiger MPX is a relatively large mobo, if you’re comparing to competing single-processor boards. However, it’s not too big to fit into an average mid-tower ATX case with a little effort.
The Tiger MPX’s layout is similar to the Tiger MP’s, but it’s clearly a different board, not just a minor revision. Overall, the Tiger MPX’s design is very solid. There are quite a few capacitors packed in close near the two CPU sockets, but none that got in my way during frequent CPU swapping as I prepared this review. In fact, my only substantial gripes are about the placement of the IDE and floppy drive connectors. They’re all grouped near the “bottom” edge of the board, and folks with tall tower cases will face an all-too-familiar quandary trying to figure out how to get drive cables to reach.
Beyond that, the Tiger MPX has a few unique quirks worth mentioning. Right next to the standard ATX power connector are two supplemental power connectors, and you’ll need to use at least one of them. There’s a four-pin ATX12V connector like on Pentium 4 motherboards, and there’s a standard hard drive power connector. Tyan included both connector types so those of us without ATX12V power supplies wouldn’t have to pony up for a new PSU.
The Tiger MPX also includes one very useful goodie that the Tiger MP lacks: a built-in 3Com 3C920 Ethernet controller with a 10/100 Ethernet port.
The good stuff
Most of the Tiger MPX’s merits will be obvious to those of you familiar with the Tiger MP. This is one of a handful of motherboards capable of delivering AMD multiprocessing for a desktop system. As such, the Tiger MPX is endowed with all the goodness of AMD’s multiprocessor systems.
Many of those advantages are conferred by AMD’s 760MPX chipset, which consists of the AMD 762 north bridge chip and the 768 south bridge chip. The 760MPX differs from AMD’s original 760MP chipset only in the south bridge chip; the 762 north bridge is unchanged.
The key features of AMD multiprocessor systemsmost of which are provided by the 762 north bridgemake this platform very potent, so we’ll review ’em:
- Dual front-side busses Unlike Intel’s multiprocessor systems, AMD’s multiprocessor systems give each CPU its own, dedicated front-side bus. A shared front-side bus has been a bottleneck in traditional desktop SMP systems, especially in Pentium III systems where the FSB only offers 1.06GB/s of bandwidth. The 762’s dual busses each offer 2.1GB/s of bandwidth per processor. Not only that, but Athlon systems use the EV6 bus protocol, borrowed from the (DEC-then-Compaq-then-Intel) Alpha, which is a bit more advanced in some respects than the GTL+ bus used in the Pentium III and Pentium 4. (Not that Pentium 4-based Xeons, with their 400MHz FSBs, are exactly hurting for bus bandwidth.)
- Cache coherency Athlon MP systems use a protocol called MOESI to manage data in the processors’ data caches. This mechanism, combined with the 762 north bridge, allows either processor in the system to fetch data stored in the other processor’s cache without first transferring that data into main memory. Instead, if CPU 1 wants to grab some data stored in CPU 2’s cache, it will request the data, and CPU2 will pass this data through the 762 memory controller and into CPU 1’s cache. Managing cached data in this way ought to offer much improved performance.
- DDR SDRAM AMD first brought DDR to the mainstream with the 760 chipset, and a dual-processor system is the perfect place to take advantage of the extra memory bandwidth DDR brings.
Don’t underestimate the potency of this combination of features. Implementing these things in any platform is by no means trivial, and the fact AMD and Tyan can deliver them on the desktop (or in low-to-mid-range server and workstations) is exceptional.
Face it, SMP on the desktop hasn’t exactly set the world on fire. Enthusiasts toyed with SMP Celeron rigs back in the day with BP6 motherboards and the like, but generally, the price premium hasn’t brought enough performance along with it. Folks soon discovered that, for a variety of reasons, two processors don’t usually perform anywhere near twice as fast as one. But with dual FSBs, superior cache management, and more memory bandwidth, AMD’s dually systems eliminate some of the key roadblocks to SMP performance.
The 768 south bridge chip in the 760MPX chipset adds a few new features to this mix. Among them:
- A 64-bit, 66MHz PCI bus The 760MPX now has a 66MHz PCI bus, so it can accommodate high-end PCI cards for server applications where regular old PCI won’t cut itthings like SCSI RAID controllers and Gigabit Ethernet NICs. With 533MB/s of bandwidth, 64-bit/66MHz PCI quadruples the available bandwidth over the usual 32-bit/33MHz standard. The 760MP, by contrast, offered 64-bit PCI, but only at 33MHz for peak bandwidth of 266MB/s.
- A faster north-south bridge interconnect Like many previous-gen desktop chipsets (think KT133 or Intel BX), the 760MP used the shared 32-bit/33MHz PCI bus for communication between north and south bridge chips. The 760MPX still uses the shared PCI bus, but it does so at 64 bits and 66MHz. As a consequence, the north-south bridge link jumps from 133MB/s to 533MB/s.
The 762 chip has always been capable of sitting on a faster PCI bus, but the 766 south bridge chip in the 760MP chipset wasn’t. The 768 chip enables the faster interconnect.
- A secondary 32-bit, 33MHz PCI bus Since 66MHz PCI slots aren’t compatible with the vast majority of expansion cards out there, the 768 chip also integrates a traditional 32-bit/33MHz PCI bus. On the Tiger MPX, four of the six PCI slots are 32-bit/33MHz slots.
- AC’97 audio Although the Tiger MPX doesn’t take advantage of it, the 768 chip now incorporates an AC’97-compatible audio controller capable of powering on-board audio.
The not-so-good stuff
You’re probably all ready to sign up for the SMP brigades now that you’ve read about all the good things the Tiger MPX can do. Simmer down, though, and get a load of the downsides first. The 760MPX comes with a host of “gotchas.” They’re not necessarily dealbreakers, but they are annoying.
First, AMD didn’t design the 762 memory controller to work with regular old unbuffered DIMMs. Instead, the 762 usually requires registered DIMMs. If you’re like me, that means you can’t just transplant your DDR DIMMs from your previous motherboard when upgrading. Frustrating.
Fortunately, Tyan has made substantial improvements to the Tiger MPX’s memory handling. Now, unlike the Tiger MP, the Tiger MPX can make use of unbuffered DIMMs in memory slots one and two. If you want to use more than two DIMM slots, though, Tyan recommends registered DIMMs still.
Next, there’s the Tiger MPX’s tweaking options. The juiciest screen in the whole BIOS looks like this:
..and that’s about it. You can’t tweak the RAM timings, can’t juice up the CPU voltage, can’t change a single default clock frequency or multiplier.
If you look at it wrong, a big, electronic hand will reach out of the screen and grab you by the collar while the motherboard chastises you for even thinking about overclocking. This is a serious computer, fool!
Even switching from a 200MHz to 266MHz front-side bus doesn’t happen in the BIOS or via CPU auto detection. Instead, you’ve gotta move a total of four different jumpers on the motherboard in order to make this change. Other FSB frequencies are not offered.
What’s more, Tyan won’t officially sanction running a pair of Durons or Athlon XPs; the board only supports Athlon MPs in dual configurations. (No doubt AMD had some say in this one.)
Then there’s the USB problem. Thanks to a bug in AMD’s 768 south bridge chip, all AMD 760MPX motherboards currently have a broken USB controller. Tyan says it will make a version of the Tiger MPX with working USB ports just as soon as AMD fixes the 768 chip. However, I’ve seen no indication that current Tiger MPX owners will be eligible for replacement boards. To compensate, Tyan includes a four-port USB controller card with the Tiger MPX. Unfortunately, the card is a USB 1.1 controller, not a USB 2.0 card like those included with some of the Tiger MPX’s competitors.
All of which leads to…
The USB problem is worse than it seems at first, because the USB card will occupy one of the Tiger MPX’s four 32-bit/33MHz PCI slots. The two 66MHz/64-bit PCI slots on the Tiger MPX, meanwhile, aren’t compatible with the vast majority of PCI expansion cards out there. (The older Tiger MP could accommodate regular old 32-bit PCI cards in its 64-bit slots, but the Tiger MPX’s 66MHz PCI bus makes those slots incompatible.) Now, if you need 66MHz/64-bit PCI for a high-end SCSI RAID controller card or something like that, those 66MHz PCI slots are great. But for most folks, at the end of the day, what you’re getting with the Tiger MPX is three open PCI slots.
Without on-board RAID or built-in sound, those three open PCI slots are gonna get filled up quick. If it were my own system, for instance, I’d probably drop in a sound card, a Firewire card, and an IDE RAID controller. Assuming that IDE RAID controller isn’t compatible with 66MHz PCI, that’s the end of the road. If you want to add a TV tuner card or most anything else, you’re pretty much out of luck. Thank goodness the Tiger MPX includes a decent on-board NIC, or the PCI slot situation would be nearly impossible for most of us.
This mess makes the original Tiger MP look mighty attractive by comparison. If you don’t mind buying registered DIMMs, the Tiger MP will give you working USB and six PCI slots. (Yep, those 64-bit/33MHz slots on the Tiger MP will accept 32-bit/33MHz cards, as well.) Depending on your needs, the Tiger MP may be a better option than the MPX.
Doing things we ought not to do
Naturally, some of the Tiger MPX’s spec-sheet restrictions looked like challenges to us. So we tried a bunch of messed up configurations to see if they would really work. Among them:
- Running a pair of 1GHz Durons Running Durons in SMP is an attractive alternative to building a single-processor system based on a more expensive Athlon XP, so there’s some real interest in this one. Like the Tiger MP before it, the Tiger MPX had no trouble at all running a pair of “Morgan” Durons (the kind derived from the Athlon MP’s Palomino core) in a dually config. In fact, we’ve got a full set of benchmarks with dually Durons.
- Running an Athlon MP 1800+ and an Athlon XP 1800+ Again, no problems here. If you want to save a buck, you might be able to get away with running a pair of Athlon XPs on a Tiger MPX. Heck, other than the markings on the CPU cores and the fact the MP’s L1 bridges are not severed, the Athlon MP and XP look identical. They’ve just gotta be the same chip.
- Running a Duron 1GHz and a Duron 1.1GHz We put a Duron 1.1GHz chip into the CPU 1 socket and a Duron 1GHz into the second socket. The system booted into Windows and ran stable. Both WCPUID and SiSoft Sandra reported that both processors were running at 1.1GHz. Apparently, the system picked up the multiplier from CPU 1 and enforced it for CPU 2, as well.
- Running a pair of Durons on a 266MHz bus We tried this one inadvertently the first time we installed the Durons into the board. Unfortunately, our Durons weren’t up to the task. The system refused to POST.
- Running a full load of unbuffered DIMMs So if you slap four DIMMs full of high-quality DDR memory into the Tiger MPX’s DIMM slots, what happens? We put in a gig’s worth of Crucial and Kingmax (all based on Micron chips) and gave it a go. We were able to get the system to boot and run a solid mix of Windows apps, like so:
However, after a little while, the system locked up. We cut down to three DIMMs, but the results were the same. On the other hand, the system was stable with a pair of unbuffered DIMMs or with registered ECC memory. That’s disappointing, because we were so close to being able to get away with unbuffered memory modules. But there’s a reason Tyan’s specs call for registered DIMMs.
All told, the Tiger MPX took our abuse very well. Now, let’s see how it performs.
What to watch for in the test results
What follows is a series of benchmarks intended to show off the advantages (or non-advantages) of a dual-processor system in everyday use. It’s kind of like a cross between watching paint dry and a car wreck. On the one hand, we’ve got a series of graphs presenting test data. On the other hand, the SMP systems are going to be demonstrating vividly why traditional benchmarks just can’t do them justice.
Remember: adding a second processor doesn’t mean a 100% performance gain. Depending on kinds of software and the sort of tasks you throw at an SMP system, the performance gains can range from, heck, negative 10% (due to SMP overhead) to nearly 100% (in extreme, usually contrived scenarios). But most of the time, you’ll see something in between those two extremes.
Unfortunately, only some of our benchmarks are multithreaded, so only some of our tests can break out the work onto multiple processors in order to improve performance. Don’t think badly of SMP systems because 3DMark doesn’t run any faster on them. That’s 3DMark’s fault (and maybe Microsoft’s for Direct3D and maybe just a teensy bit NVIDIA’s for their drivers). The fact single applications are not multithreaded doesn’t mean SMP will have no value for you in everyday use, because nearly all modern OSes are SMP-aware, including Windows NT/2000/XP Pro, Linux, BeOS, and nearly every flavor of Unix. More on this after the test results.
Our testing methods
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:
|Tyan Tiger MP||Tyan Tiger MPX|
|Processor||2 x AMD Athlon MP 1800+
||2 x AMD Athlon MP 1800+
2 x AMD Duron 1GHz
AMD Athlon MP 1800+
|Front-side bus||266MHz (133MHz double-pumped) for Athlon MPs
200MHz (100MHz double-pumped) for Durons
|Chipset||AMD 760MP||AMD 760MPX|
|Memory size||256MB (1 DIMM)||256MB (1 DIMM)|
|Memory type||Micron PC2100 DDR Registered ECC SDRAM (CAS 2.5)|
|Graphics||NVIDIA GeForce3 Ti 500 64MB (Detonator XP 21.83 video drivers)|
|Sound||Creative SoundBlaster Live!|
|Storage||Maxtor DiamondMax Plus D740X 40GB 7200RPM ATA/133 hard drive|
|OS||Microsoft Windows XP Professional|
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:
- ZD Media Business Winstone 2001 1.0.2
- ZD Media Content Creation Winstone 2002 1.0
- SiSoft Sandra Standard 2002
- MadOnion 3DMark 2001 SE
- Quake III Arena 1.30
- picCOLOR NT Image Analysis 1.00
- Cinebench 2000 1.0
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.
We’ll omit single-processor results for these memory performance tests, because Sandra’s memory benchmarks are multiprocessor-aware. Comparing single and dual-processor memory bandwidth results may not be valid here.
The Tiger MPX is just a tad bit faster than the original Tiger MP, but it’s nothing to write home about.
Content Creation Winstone 2002
The newest version of Content Creation Winstone ought to give us a good start, because it runs multiple tasks at once and ought to exercise that second processor quite a bit.
As you can see, the Tiger MPX is again just a little faster than the Tiger MP. And yep, two processors are better than one.
Quake III Arena
We tested Quake III both with and without its SMP support enabled. Quake III is one of the few games that makes use of multiple processors, but its SMP support is far from robust. We had to start Q3A with a “realtime” process priority in Windows XP in order to make it work.
For the dually systems, we tested both with and without Q3A’s SMP support enabled. The results below marked ‘r_smp 1’ are the ones with SMP support in use.
Quake III does benefit measurably with both processors pitching in. Once more, the Tiger MPX is just an eyelash faster than the Tiger MP. Also, notice how the dual-processor results without ‘r_smp’ enabled are slower than the single-CPU results. The overhead from keeping a second processor around without putting it to use causes the duallies turn in slightly lower scores.
Without any multithreading, 3DMark doesn’t benefit at all from a second processor.
picCOLOR image processing
The picCOLOR image processing tests were sent to us by Dr. Reinert H. G. Mueller of the FIBUS institute. He wanted to see how his test would perform on a dual-processor Athlon system, and we were glad to oblige. Not all of the tests were multithreaded in the version of picCOLOR we used, but by the time we finished testing, Dr. Mueller had already revised picCOLOR so more of its operations could benefit from SMP. By nature, image processing work is easily done in parallel, so dually systems can show great performance gains. We hope to use those newer versions of picCOLOR next time around.
For now, it will be easy to see which of the tests below were multithreaded.
AddressMem, Fixed interpolation, Convert, and Convert with video all make good use of SMP. For this sort of work, a dually Duron can outrun a single Athlon XP/MP 1800+.
We’ll wrap up our SMP tests with Cinebench, a benchmark based on Maxon’s Cinema 4D XL rendering software. Specifically, we’re testing raytracing, one of the most intensive and realistic 3D rendering methods. Cinema 4D can take advantage of multiple processors; we’ll test it both with and without dual-processor support turned on.
Like image processing, rendering is easily parallelized. We’re gaining in the neighborhood of 70% more performance with SMP. Here again, for serious work, a dually Duron outperforms a single-processor Athlon XP/MP 1800+ system.
Chipset benchmark results
The next set of tests is just something we did to satisfy our own curiosity, and we thought you might enjoy seeing the results, too. Our recent review of VIA’s KT333 chipset provided us a chance to do a massive, seven-way Socket A chipset shootout. We kind of wondered how the Tiger MPX and AMD’s 760MPX chipset would compare to the single-processor Socket A chipsets of the world, so we decided to test it against its single-CPU competition.
We used the same testing methods we employed in the KT333 review. You can see a full accounting of the system configs and benchmark versions here. Basically, what we did was put a single, unbuffered PC2100 DIMM on the Tiger MPX and install a single Athlon XP 2000+ processor. Then we ran it through the same set of tests we used for the KT333, nForce, KT266A, SiS 745, and the like. The resulting test scores may not prove much, because hey, the MPX can support two processors, but they are interesting.
I’ll go light on the commentary for these scores. If you want to know more about what these scores mean, see our KT333 review.
In both of the important basic measurements of memory performance, the 760MPX is at or near the bottom of the pack. But does this lackluster memory performance translate into real-world slowdowns? Let’s find out.
Content Creation Winstone 2001
Here the Tiger MPX is at the bottom of the pack. There’s no candy coating this one. If you’re not going to put two processors into your system, this motherboard’s not for you.
Quake III Arena
The MPX again is behind the pack in Q3A. Note that the MPX is faster here, with a single Athlon XP 2100+ and unbuffered CAS 2 RAM, than it is with registered DIMMs and a pair of Athlon MP 1800+ processors, even when SMP support is enabled165.5 fps here versus 162.1 in our dually tests. And the fastest single-CPU solution just blows it away.
No surprises here, just more of the same.
Sphinx speech recognition
Speech recognition can be very memory intensive, and the MPX again doesn’t fare well.
So we’ve seen the Tiger MPX’s performance examined twelve ways from Sunday. What to make of it? The results are mixed, to put it mildly. We do know the Tiger MPX is consistently just a tad bit faster than the Tiger MP. Beyond that, we’ve seen some casesspecific ones like 3D rendering or image processing, and more general ones like Content Creation Winbenchwhere doing the dually thing will net some big performance gains.
However, the chipset tests have shown us the downside of desktop SMP. Quite simply, the pumped-up, pimped-out single-CPU hobbyists’ mobos with extensive performance tuning plus memory tweaking features make the conservatively tuned Tiger MPX look like a pussycat. They deliver gobs more memory bandwidth, which even a single CPU can put to great use. For absolute maximum performance in single-task applications like games, there’s no denying that single-processor rigs are the way to go. The Tiger MPX might have a beast lurking within, but Tyan hasn’t given us the keys to unlock it.
And the Tiger MPX is still a fussy cat. You’ll need to plan your RAM and PCI card purchases carefully make sure everything fits well and works right. Even if you do, the PCI slot quandary might bite you down the road. Meanwhile, the non-dually competition is bristling with open PCI slots, IDE RAID ports, and optical audio outputs.
The Tiger MPX is a solid, stable (when properly configured) motherboard that deserves to make its way into loads of low-end workstations and servers. There’s no denying it. The Tiger MPX brings some of the most advanced SMP technology anywhere into the realm of us mortals. But this mobo isn’t quite as good as what I want to see in desktop SMP. I’d like to see all the goodness of a single-processor enthusiast’s board like a Soyo DRAGON combined with the SMP smoothness of the 760MPX, and this ain’t it. Consequently, I’m going to have to give this board an 8 on our world-famous 11-point scale.
Before you write off the Tiger MPX, though, hear me out. If you’ve read this far, seen all the benchmarks, and wondered why anyone would bother with the Tiger MPX, you’re just not getting it. None of the benchmarks and none of our niggling complaints can change the reality of the user experience on a dually Athlon system.
While fast single-processor rigs are screamers at many things, they can slow down. Fire off an install script for a Windows program, and even an Athlon XP or Pentium 4 will drop to its knees. Open an Explorer window after inserting a new CD and your expensive system will stop everything and wait. Watch an errant application go haywire and chew up all the CPU time, and that killer box won’t even respond to keyboard input.
A system fortified with the creamy smoothness of SMP, on the other hand, simply will not slow down. It’s friction-free computing. There’s nothing else like it. Yes, a good OS can multitask well with one CPU, but as always, some things are just better done in hardware. With SMP, you can run a few game servers in the background while you compute unaffected in the foreground. Or rip and compress MP3swhatever. If you’re going to spend eight or more hours a day in front of the PC you’re about to build, and if you’re going to use it like I doumpteen different windows open running a zillion things with your hair on firethen there’s no substitute for some creamy smooth SMP action. That’s why I’d recommend investigating a dually Duron rig before dropping the cash on an Athlon XP 2000+ or the like. That’s also why this is one motherboard with an 8 rating that I’d really like to have in my own PC.