Now, you may be a little perplexed, because VIA and SiS have had DDR333 for what seems like ages now. However, Intel isn’t one to jump the gun on memory standards, to say the least. They like to hang back, make sure all the standards are finalized and the interoperability is there before they do anything. Then they like to sip lemonade and kick back in the hammock for a few months. When Intel does move, however, the whole PC market moves with it. Put it this way: if you bought a Pentium 4-based system from Dell or HP before today, you probably had your choice of DDR266 or RDRAM. Now that these new chipsets has arrived, DDR333 memory will probably become the industry standard in a matter of weeks.
There’s not much more for me to tell you about the 845PE and GE chipsets that you can’t deduce from here. These chipsets add official support for DDR333 memory, though in the case of 845G, the chipset has been unofficially quite capable of running DDR333 memory since its launch. Beyond that, the features are familiar: AGP 4X support, an ATA/100 disk interface, a 533MHz bus, and nothing much new.
But I can tell you quite a bit about Pentium 4 chipsets overall, because we’ve lined up nine different configurations and tested them against one another to see who comes out on top. Can Intel’s new 845PE and GE chipsets match up with the latest chipsets out of VIA and SiS? Can DDR memory finally slay the RDRAM performance dragon once and for all? Would I really spend a perfectly good weekend testing core-logic chipsets and making graphs? Keep reading to find out.
Surveying Blue Mountain
The biggest surprise of this whole experience for me has been playing with Intel’s new 845PE-based motherboard. This board, dubbed “Blue Mountain” internally at Intel, looks like something out of Taiwan, not a typical mobo from Intel’s button-down desktop board division. Traditionally, Intel boards have compromised performance and tweakability for stability and simplicity. They have also been light on the fancy featuresperfect for a corporate desktop, but nothing an enthusiast would care to put into his system.
In fact, Intel boards have been something of a problem for us, because their relatively low performance has made chipset comparisons like this one difficult. For instance, we reviewed the original 845 chipset with DDR support using an Intel board and weren’t too impressed with the performance. Then, shortly thereafter, we were forced to reconsider once we got our hands on Abit’s 845 implementation.
Survey the Blue Mountain, and you know things have changed. This mobo comes with an array of on-board multimedia and I/O features unparalleled this side of Vegas, including Firewire, USB 2.0, Serial ATA RAID (courtesy of a Silicon Image controller chip), Ethernet, six analog sound ports, and optical and coaxial digital audio outputs. On a black PCB, for gosh sakes. There’s even an optional dancing midget.
(OK, I admit it: I made up the part about the Firewire ports. But you get the idea.)
Most importantly, perhaps, both Intel’s 845PE and GE boards now offer BIOS options for manual tuning of memory timings. With memory timings set aggressively, I’m happy to report that these boards perform quite well. That gives us the chance to see how the 845PE and GE chipsets really compare to the competition. Intel says these expanded BIOS options are here to stay, too, so Intel boards should be a little more adjustable from here out.
A fancy 845PE motherboard like this makes sense, because the 845PE chipset is, like its predecessor the 845E, Intel’s entry in the performance desktop market. Swanky mobos like Abit’s IT7-MAX are based on the 845E, but those boards have been hampered a bit by the 845E’s lack of support for DDR333 memory. The 845PE chipset addresses this shortcoming.
Intel’s 845GE board isn’t as flashy as the Blue Mountain board, but this one really is destined for truckloads full of corporate desktop PCs, because it packs Intel’s integrated graphics in the 845GE chipset. Intel’s “Extreme Graphics” are anything but extreme (unless you count “extremely cheap”), but they are adequate for most non-gaming tasks.
The 845GE’s officially faster memory will boost overall bandwidth available for the processor and built-in graphics from 2.1GB/s to 2.7GB/s. To best take advantage of this change, Intel has bumped up the speed of its graphics core from 200MHz in the 845G to 266MHz in the 845GE.
Like the 845G, the GE can support a discrete graphics card in an AGP slot, so if you have an 845G system, you’re not stuck with pokey 3D gaming forever. Appropriately, then, we’ve tested the 845GE with and without a discrete graphics card, so you can see what impact using the GE’s integrated graphics has on performance.
An RDRAM surprise
Beyond the DDR chipsets announcements, Intel has one other bit of news today: the company is finally validating PC1066 RDRAM for use with its 850E chipset. That makes me feel warm and fuzzy, because I’ve been using PC1066 memory for testing for a good while. Now, 850E system configs with PC1066 RDRAM are officially OK with Intel. To keep things interesting, we’ve tested Intel’s new chipsets against 850E systems with both PC800 and PC1066 memory.
Before we move on, let’s take a quick look at how the 845PE and GE stack up to other Intel chipsets and to the Taiwanese competition. Below is a soul-sucking table full of core-logic chipset features. Feel free to poke through it until you drool on your keyboard. I think I’ve got everything right.
|Intel 845E||Intel 845PE||Intel 845G||Intel 845GE||Intel 850E||SiS 648||VIA P4X400|
(DDR only listed)
|PC1600 PC2100 DDR SDRAM||PC1600 PC2100 PC2700 DDR SDRAM||PC1600 PC2100 (PC2700) DDR SDRAM||PC1600 PC2100 PC2700 DDR SDRAM||PC800 PC1600 RDRAM||PC1600 PC2100 PC2700 DDR SDRAM||PC1600 PC2100 PC2700 DDR SDRAM|
|Max. effective memory clock speed||266MHz||333MHz||266MHz (333MHz unofficial)||333MHz||1066MHz||333MHz
|Memory bus width||64 bits||64 bits||64 bits||64 bits||2 x 16 bits||64 bits||64 bits|
|Peak theoretical memory bandwidth||2.1GB/s||2.7GB/s||2.1GB/s
|Maximum addressable RAM||2GB||2GB||2GB||2GB||2GB||3GB||3GB|
|Max. AGP mode||4X||4X||4X||4X||4X||8X||8X|
|North/south bridge interconnect||Accelerated Hub||Accelerated Hub||Accelerated Hub||Accelerated Hub||Accelerated Hub||MuTIOL||V-Link|
|Interconnect clock speed||266MHz||266MHz||266MHz||266MHz||266MHz||533MHz||533MHz|
|Interconnect bus width||8 bits||8 bits||8 bits||8 bits||8 bits||16 bits||8 bits|
|Peak theoretical interconnect bandwidth||266MB/s||266MB/s||266MB/s||266MB/s||266MB/s||1.06GB/s||533MB/s|
|Disk interface support up to…||ATA/100||ATA/100||ATA/100||ATA/100||ATA/100||ATA/133||ATA/133|
|USB mode support||2.0||2.0||2.0||2.0||1.1||2.0||2.0|
|IEEE 1394 controllers/ports||0/0||0/0||0/0||0/0||0/0||2/6||0/0|
|AC97 audio channels||6||6||6||6||6||6||6|
|Network interfaces||10/100Mbps Ethernet||10/100Mbps Ethernet||10/100Mbps Ethernet||10/100Mbps Ethernet||10/100Mbps Ethernet||10/100Mbps Ethernet
If you’re still with us, you can probably see that even these newest Intel chipsets don’t have all the latest features supported by the VIA P4X400 and SiS 648 chipsets. Those chipsets both have support for AGP 8X, faster interconnects between the north and south bridge chips, ATA/133 support, and unofficial support for DDR400 memory. Remember what I was saying about Intel being more deliberate about supporting new memory standards? The same goes double for other things in core logic. Not that any of these new features matter greatly now. In the case of AGP 8X and DDR400, the current implementations are somewhat imperfect.
I should follow up that statement by saying that VIA’s AGP 8X implementation worked beautifully with a Radeon 9700 graphics card in our testing. The SiS 648 wasn’t quite as smooth. We saw repeated crashes in 3D applications, and one of them, Unreal Tournament 2003, simply locked up every time we tried to benchmark it. Our attempts to turn off AGP 8X mode in the board’s BIOS didn’t help, either.
Wisely, both VIA and SiS have backed off of supporting DDR400 memory officially. This new memory speed has not been easy to implement successfully, and even with good Corsair XMS3200 memory, we’ve not been able to make it go. Our VIA P4PB 400 board wouldn’t boot with DDR400 memory. The SiS-based Abit SR7-8X would run DDR400 stable enough to produce benchmark results, but the conservative memory timings required for stability dragged on performance. DDR400 scores were consistently slower than DDR333.
So Intel’s conservative approach to chipsets isn’t all bad. Effectively, the VIA and SiS products offer a few extra features, but as you’ll see shortly, those features don’t always add up to better performance.
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 845E||Intel 845PE||Intel 845G||Intel 845GE||Intel 850E||SiS 648||VIA P4X400|
|Processor||Athlon XP 2800+ 2.25GHz||Pentium 4 2.8GHz||Pentium 4 2.8GHz||Pentium 4 2.8GHz||Pentium 4 2.8GHz||Pentium 4 2.8GHz||Pentium 4 2.8GHz||Pentium 4 2.8GHz|
|Front-side bus||333MHz (166MHz DDR)||533MHz (133MHz quad-pumped)||533MHz (133MHz quad-pumped)||533MHz (133MHz quad-pumped)||533MHz (133MHz quad-pumped)||533MHz (133MHz quad-pumped)||533MHz (133MHz quad-pumped)||533MHz (133MHz quad-pumped)|
|Motherboard||Asus A7N-8X (pre-release sample)||Abit IT7-MAX2||Intel D845PEBT2||Abit BG7||Intel D845GEBV2||Asus P4T533C||Abit SR7-8X||VIA P4PB 400|
|North bridge||nForce2 SPP||82845E MCH||82845PE MCH||82845G MCH||82845GE MCH||82850E MCH||648||VT8754|
|South bridge||nForce2 MCP-T||82801DB ICH4||82801DB ICH4||82801DB ICH4||82801DB ICH4||82801BA ICH2||963||VT8235|
|Chipset drivers||2.77||Intel Application Accelerator 6.22||Intel Application Accelerator 6.22||Intel Application Accelerator 6.22||Intel Application Accelerator 6.22||Intel Application Accelerator 6.22||SiS AGP 1.10||4-in-1 4.43|
|Memory size||512MB (2 DIMMs)||512MB (1 DIMM)||512MB (1 DIMM)||512MB (1 DIMM)||512MB (1 DIMM)||512MB (4 RIMMs)||512MB (1 DIMM)||512MB (1 DIMM)|
|Memory type||Corsair XMS3200 PC2700 DDR SDRAM||Corsair XMS2400 PC2100 DDR SDRAM||Corsair XMS3200 PC2700 DDR SDRAM||Corsair XMS2400 PC2100 DDR SDRAM||Corsair XMS3200 PC2700 DDR SDRAM||Samsung PC800/PC1066 Rambus DRAM||Corsair XMS3200 PC2700 DDR SDRAM||Corsair XMS3200 PC2700 DDR SDRAM|
|Graphics||ATI Radeon 9700 Pro 128MB (Catalyst 7.76 drivers)|
|Sound||Creative SoundBlaster Live!|
|Storage||Maxtor DiamondMax Plus D740X 7200RPM ATA/133 hard drive|
|OS||Microsoft Windows XP Professional|
|OS updates||Service Pack 1|
Thanks to Corsair for providing us with memory for our testing. If you’re looking to tweak out your system to the max and maybe overclock it a little, Corsair’s RAM is definitely worth considering. Using it makes life easier for us as we’re dealing with brand-new chipsets and pre-production motherboards, because we don’t have to worry so much about stability and compatibility.
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:
- Cachemem 2.6
- SiSoft Sandra Standard 2002 SP1
- ZD Media Business Winstone 2001 1.0.3
- ZD Media Content Creation Winstone 2002 1.0.1
- Sphinx 3.3
- MadOnion 3DMark 2001 SE Build 330
- Unreal Tournament 2003 demo benchmark
- Comanche 4 demo benchmark
- Quake III Arena v1.31
- Serious Sam SE v1.07
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.
Sandra’s modified version of the Stream benchmark really cranks, because it uses SSE, MMX, buffering, and the like to move as much data as possible back and forth. As a result, all of the DDR333 chipsets produce similar results, approaching DDR333 memory’s theoretical peak of 2.7GB/s bandwidth. The PC800 and PC1066 RDRAM systems are even faster here, although they’re well off their theoretical peaks of 3.2 and 4.2GB/s, respectively. You also can see how sharing memory with the graphics controller on the 845G and GE chipsets steals bandwidth from the CPU.
Cachemem’s bandwidth tests are probably a little more indicative of real-world performance. RDRAM systems are pulled back to the pack here, and we see especially strong read/write performances out of the P4X400 and 845GE memory controllers. The 845PE, strangely enough, comes out quite a bit slower than the GE in the memory write tests.
Memory access latency is every bit as important to overall performance as those Sandra numbers above, if not more so, and the Intel chipsets beat all comers here. (Our comparison Athlon XP system does quite well, but it’s running its memory synchronously on a 333MHz bus and talking to dual banks of DDR memory in the process.) Here RDRAM shows its Achilles’ heel: high access latencies. However, higher frequencies with PC1066 memory help reduce latencies considerably.
The two Winstone tests are important because they measure overall system performance apart from synthetic memory tests, and because they lean heavily on a system’s disk controller. In Business Winstone, the 845PE and GE chipsets outrun even our PC1066 RDRAM rig. Business Winstone has traditionally favored chipsets with low memory access latencies over chipsets with higher latencies and more bandwidth.
Content Creation Winstone
Content Creation Winstone is intentionally very memory bandwidth intensive, and it shows in the results. Incidentally, the low scores for VIA’s P4X400 in both of these tests are unexpectedVIA’s chipsets have been good all-around performers in recent timesbut they are not a fluke. We saw the same thing when in our review of the still-born P4X333 chipset, which has since been renamed (and perhaps slightly revised) to P4X400. For whatever reason, VIA’s performance in these tests is relatively poor.
The 845PE and GE chipsets tie to lead the pack of DDR333-based systems. Quake laps up the extra bandwidth of RDRAM. 3DMark2001 SE
Despite their AGP 8X support, the P4X400 and 648 can’t keep up in 3DMark. Impressively, the 845PE and GE both beat out the 850E with PC800 RDRAM, too. Serious Sam SE
The story is much the same in Serious Sam, where the PE/GE chipsets run just ahead of the VIA and SiS competition.
Comanche 4 is mostly limited by graphics card performance, but there are some minor performance differences here. Once again, the Intel DDR333 chipsets perform quite well. Unreal Tournament 2003
As we said before, the SiS 648 system would not complete UT2003’s benchmarking sequence, so we weren’t able to include 648 results.
There’s nothing much new to say here. Again, the AGP 8X chipsets finish below Intel’s new AGP 4X chipsets.
Sphinx is a high-quality speech recognition routine that needs the latest computer hardware to run at speeds close to real-time processing. We use two different versions, built with two different compilers, in an attempt to ensure we’re getting the best possible performance.
There are two goals with Sphinx. The first is to run it faster than real time, so real-time speech recognition is possible. The second, more ambitious goal is to run it at about 0.8 times real time, where additional CPU overhead is available for other sorts of processing, enabling Sphinx-driven real-time applications.
Intel’s DDR333 chipsets make it in just below the 0.8 mark, while the competition can’t quite cut it. Even in this test, which we’ve always considered very friendly to lots of memory bandwidth, most of the DDR333 rigs beat out the PC800 RDRAM system.
Now we’ll pause to consider how the 845GE’s extra memory bandwidth and 66MHz higher graphics clock speed boosts performance over the 845G chipset. We didn’t flinch from running Intel’s Extreme Graphics though our extreme set of game benchmarks, and for the most part, it worked. Comanche 4 simply refused to run without hardware T&L, and UT2003 showed a few visual anomalies, but that’s all.
The 845GE does have measurably better performance than the 845G, but I wouldn’t take either to a deathmatch. For business use, the 845GE’s graphics are probably fine, but then again, I didn’t appreciate the ghosting I saw on the Windows desktop. I shouldn’t compare the 845GE in any way to our Radeon 9700 Pro test card, because it’s just not fair, but I couldn’t help notice how much sharper the 9700’s video output is.
The 845PE and GE chipsets put Intel firmly in the lead for Pentium 4 core logic chipsets, despite the feature deficits with SiS and VIA. Across a range of tests intended to stress memory performance, disk I/O, AGP performance, and all-around speed, the 845PE/GE chipsets ran consistently ahead of the competition. Given Intel’s dominance of the chipset market for its flagship processor, and given the fact Intel isn’t shy about muscling out competitors like VIA through intellectual property disputes, the 845PE and GE should become the default choices in Pentium 4 chipsets. Generally when something like this happens, Intel’s chipset competitors can rely on their nimbleness to regain the edge. Given the trouble VIA and SiS have been having with validating DDR400 memory, there may be tough times ahead for third-party chipset makers. DDR400 is slow to ramp, and DDR-II just isn’t going to be ready for some time yet. The next big push for these companies will probably be dual-bank DDR solutions, which we may see on the market before Christmas, if the rumors are true. But dual-DDR solutions aren’t likely to be especially cheap or easy to produce, at least initially. VIA’s QBM initiative is also very promisingessentially “dual-channel DDR on a DIMM”and potentially easier to implement, but from what I hear, QBM probably won’t reach the market until well into next year.
Don’t forget, also, that Intel has its own dual-DDR solution for desktops in the works.
Intel is widely rumored to be preparing to bring its Hyper-Threading technology to the desktop in a new version of the Pentium 4 coming soon. Intel says all of its chipsets except for one older revision of the 845 will support Hyper-Threading. I’m eager to see how the VIA and SiS chipsets handle this feature. Hyper-Threading performance may add a new wrinkle to the whole chipset performance picture, so we’ll keep an eye on it.