Intel’s P35 Express chipset

INTEL HAS A LONG HISTORY of making good chipsets, the most storied of which is easily the 440BX from oh so many years ago. With few exceptions, Intel’s chipsets have been pretty solid since, even if many were largely ignored by enthusiasts during the Prescott era. But that era is long gone, and the Core 2 Duo has spawned a revival of enthusiast interest in Intel core logic. Over the last year, most of that interest has centered around the mid-range P965 Express, whose speedy performance, ample features, low power consumption, and availability on a slew of mid-range and high-end boards alike have made it a fast favorite among enthusiasts.

As much as we love the P965, its 1066MHz front-side bus is ill-equipped to handle the next generation of Core 2 processors. Those new chips will ride a 1333MHz front-side bus, and in preparation for their arrival, Intel has created the P35 Express chipset, otherwise known as Bearlake. The P35 Express features native support for 1333MHz front-side bus speeds, and it’s compatible with future processors built using 45nm process technology. More interestingly, it’s also the first core logic chipset to offer support for DDR3 memory.

Does Intel’s new P35 Express meet the lofty standards set by the legendary 440BX and the possible legend-in-the-making P965 Express? Is the transition to DDR3 one you’ll want to make? Keep reading to find out.

The skinny on DDR3
DDR2 memory has been around for close to three years now. Intel first brought it to PC desktops with the 925X chipset in June of 2004. At the time, DDR2 ran at a relatively pokey 533MHz, but in the years since it’s scaled all the way up to 1066MHz, though admittedly at a much slower pace than many would have liked. DDR2 effectively hit a wall at 1066MHz. You can find premium modules today that clock up to 1200MHz and beyond, but they require loads of extra voltage and carry hefty price premiums.

DDR3-1333 (top) and DDR3-1066 (bottom) modules from Corsair

DDR3 was created to push clock speeds for run-of-the-mill modules beyond 1066MHz, and to do so with lower power consumption, as well. The default voltage for DDR3 is just 1.5V—down from 1.8V in DDR2. That’s not a precipitous drop, but it’s rather impressive when you consider that DDR3 is expected to scale all the way up to 1600MHz. Initially, expect to see DDR3 modules running at 1066 and 1333MHz. 1.6GHz will have to wait.

Of course, there’s more to DDR3 than higher clock speeds and lower operating voltages. The size of DDR3’s prefetch buffer has been increased to eight bits, which is double the four bits you got with DDR2. This prefetch buffer acts like a cache for memory, allowing DDR3 to grab more data preemptively than its predecessor. (Interestingly, DDR2’s 4-bit prefetch buffer was double that of DDR memory, which had a 2-bit prefetch buffer.)

Increased densities are also on the menu for DDR3. Where its predecessor was limited to chip densities of 2Gb, DDR3 chips will be available in densities as high as 8Gb. Those denser chips will allow for higher capacity memory modules. DDR2 modules topped out at 4GB, but DDR3 modules will be capable of hosting up to 16GB of memory. That should be just enough to run the next version of Windows comfortably. We hope.

Despite its numerous advantages over DDR2, DDR3 memory modules share a similar 240-pin interface. The modules are keyed differently, so you can’t accidentally plug a DDR2 DIMM into a DDR3 slot or vice versa.

Latencies are also much higher for DDR3. The Corsair DDR3 modules pictured above are rated for 7-7-7-21 timings at 1066MHz, but the company’s DDR2 sticks can handle 5-5-5-15 timings at the same speed. It gets worse for DDR3-1333. Corsair’s CM3X1023-1333C9DHX modules are rated for latencies of 9-9-9-24. Real-world performance tends to be dictated by the balance of bandwidth and latency, so DDR3 will have to rely heavily on the former to realize any gains over DDR2—much like DDR2 had to rely heavily on its higher clock speeds to overcome its latency disadvantage relative to DDR.

The P35 Express chipset
Intel is set to release a whole line of new “3 series” chipsets, from low-end models with integrated graphics to a high-end X38 Express that’s set to replace the aging 975X in the third quarter of this year. The P35 Express sits in the middle of this new core logic range, replacing the P965.

Source: Intel

Busting out a block diagram of the P35 Express reveals relatively few surprises. Most of the action is at the north bridge, where the P35 offers native support for front-side bus speeds up to 1333MHz. Nvidia’s latest nForce 680i SLI (and that company’s more budget-oriented 650i SLI) also support 1333MHz front-side bus speeds, but as you can see in the north bridge feature comparison chart below, Intel’s P965 and 975X do not.

P35 Express P965 Express 975X Express nForce 680i SLI SPP
Front-side bus 1333/1066/800MHz 1066/800MHz 1066/800MHz 1333/1066/800MHz
Memory controller DDR2-800/DDR3-1066 DDR2-800 DDR2-667 DDR2-800
PCI Express lanes 16 16 16 18
Multi-GPU support CrossFire* CrossFire* CrossFire SLI
Chipset interconnect DMI DMI DMI HyperTransport
Peak interconnect bandwidth 2GB/s 2GB/s 2GB/s 8GB/s

Along with the P35’s support for a 1333MHz front-side bus comes compatibility with 45nm Penryn processors that are expected to make their debut in the second half of the year.

Things get a little more interesting when we look at the memory controller, where the P35 stands out as the only core logic chipset to support dual channels of DDR3 memory. Intel has only validated DDR3 speeds up to 1066MHz, but as we’ve seen with previous chipsets, motherboard makers are free to offer the necessary dividers to run memory at higher speeds. The Asus motherboard we’ll be looking at today, for example, supports DDR3 speeds up to 1333MHz, albeit only in tandem with a 1333MHz front-side bus.

If you’re not keen on upgrading to DDR3, the P35 Express also has a dual-channel DDR2 memory controller that’s been validated up to DDR2-800. Again, motherboard makers are free to support faster memory bus speeds. They’re even free to create motherboards with both DDR2 and DDR3 memory slots, although this is another solution Intel hasn’t officially condoned. Intel does concede it’s possible, though, just as long as you don’t try to run DDR2 and DDR3 at the same time.

We’ve actually seen Intel-based motherboards supporting dual memory types before. Those boards were based on the 915 Express chipset, which featured DDR and DDR2 memory compatibility just as the latter was debuting in the market.

Apart from the memory controller and front-side bus, the P35 Express looks a lot like the P965. You still get 16 lanes of PCI Express at the north bridge, but those lanes can’t be split up between a pair of graphics cards for CrossFire. Instead, the P35 supports multi-GPU configurations much like its predecessor did: primary graphics cards get 16 lanes of PCIe bandwidth from the north bridge, while secondary cards hang off a four-lane connection from the south bridge. This configuration puts additional stress on the 2GB/s DMI interconnect that runs between the chipset’s north and south bridge components, but as we saw with the P965 Express, that doesn’t seem to slow CrossFire performance.

We should note that although there’s been some talk of Bearlake’s support for second generation PCI Express, that’s only for the high-end X38 Express that’s scheduled for release later this year. The P35’s PCI Express lanes are all first-generation PCIe.

PCI Express lanes 6 6 6 28
Serial ATA ports 6 6 4 6
Peak SATA data rate 300MB/s 300MB/s 300MB/s 300MB/s
Native Command Queuing Y Y Y Y
RAID 0/1 Y Y Y Y
RAID 0+1/10 Y Y Y Y
Matrix RAID Y Y Y N
ATA channels 0 0 1 1
Max audio channels 8 8 8 10
Audio standard HDA HDA HDA HDA
Ethernet 10/100/1000 N N 2 x 10/100/1000
USB ports 12 10 8 10

At the south bridge, the ICH9R chip that will most commonly be paired with the P35 on enthusiast motherboards might as well be called the ICH8R Plus. This is essentially last year’s south bridge with a little extra kit, including two additional USB ports and a Gigabit Ethernet MAC.

The GigE MAC is the most notable new addition, and it’s one we’re happy to see. Gigabit Ethernet has become ubiquitous on even budget motherboards, and that functionality might as well be integrated into the chipset rather than farmed out to an auxiliary chip. Curiously, though, all three of the P35 Express-based motherboards that have arrived in our labs thus far feature third-party Gigabit Ethernet controllers. There may be a GigE MAC aboard the ICH9R, but mobo makers don’t appear to be taking advantage of it yet.

It’s a shame that more motherboard makers aren’t using the ICH9R’s integrated networking capabilities, because that’s really the only new feature. The chip sports six 300MB/s Serial ATA RAID ports just like its predecessor, the ICH8R, and it’s still missing an IDE port. At least by now there are enough affordable SATA optical drives on the market to make the lack of IDE support easy to swallow. Motherboard makers will probably continue to offer IDE ports through third-party storage controllers, as well.

Perhaps the most surprising attribute of the ICH9R is the fact that the chip’s being fabbed using relatively pedestrian 130nm process technology. With Intel apparently mastering 65nm production for its Core 2 line and already pushing 45nm for Penryn, we’d expected a process shrink for newer chipsets. The P35 north bridge is built on a 90nm process, at least.

Asus’ P5K Deluxe and P5K3 Deluxe motherboards
We’ve been testing the P35 Express in our labs for the past week on P5K3 Deluxe and P5K Deluxe motherboards courtesy of Asus. The former takes advantage of the P35’s support for DDR3 memory while the latter uses plain old DDR2.

Asus’ P5K3 Deluxe motherboard

Apart from differing on the memory front, the P5K3 and P5K Deluxe are virtually identical. Both are loaded with intricate heatpipe chipset cooling, eSATA, Firewire, and integrated Wi-Fi capabilities. We’ll be taking a closer look at these boards as well as others based on the P35 chipset in the coming weeks. Stay tuned.

Our testing methods
Today we’ll be looking at the P35’s performance from three angles. The first is with DDR2 memory on the P5K Deluxe. For that system, we’ve elected to use memory running at 800MHz with reasonably aggressive 4-4-4-12 timings. To explore the P35’s performance with DDR3 memory, we’ve tested the P5K Deluxe with a pair of DDR3-1066 sticks running at their default 7-7-7-21 timings. Corsair also hooked us with a pair of DDR3-1333 modules designed to run at 9-9-9-24 timings, which we naturally wanted to test, as well. However, the P5K3 Deluxe has a relatively limited number of memory bus dividers, so the only way to get a 1333MHz memory bus is to crank the front-side bus up to the same speed.

Fortunately, the P5K3 Deluxe does offer control over the CPU multiplier. Running our E6700 processor with an 8X multiplier on a 1333MHz front-side bus gives us the same 2.67GHz clock speed as the default 10X multiplier on a 1066MHz FSB. Just keep in mind that the DDR3-1333 results in the following graphs are also running a faster system bus.

Since memory types and speeds shouldn’t have much of an impact on peripheral performance, we’ve limited peripheral testing to our P5K3 Deluxe with DDR3-1066 memory.

Before we dive into our test results, we should also note that we encountered an apparent compatibility conflict between Western Digital Caviar RE2 hard drives and the ICH9R running in AHCI mode. Running the south bridge in AHCI mode is necessary to enable Native Command Queuing, but doing so with Caviar RE2 hard drives gave us all sorts of blue screen errors in Windows XP and even during the XP installation process. We’ve notified Intel of the issue, but haven’t yet received a fix. Western Digital’s Raptor X and WD1500ADFD hard drives had no problem with AHCI mode on the ICH9R.

All tests were run at least twice, and their results were averaged, using the following test systems.

Processor Core 2 Duo E6700 2.67GHz
System bus 1333MHz (333MHz quad-pumped) 1066MHz (266MHz quad-pumped)
Motherboard Asus P5K3 Deluxe Asus P5K3 Deluxe Asus P5K Deluxe Asus P5B Deluxe Wifi-AP Edition Asus P5W DH Deluxe EVGA 122-CK-NF68
Bios revision 0201 0201 0201 1101 2004 P24
North bridge Intel P35 Intel P35 Intel P35 Intel P965 Intel 975X Nvidia nForce 680i SLI SPP
South bridge Intel ICH9R Intel ICH9R Intel ICH9R Intel ICH8R Intel ICH7R Nvidia nForce 680i SLI MCP
Chipset drivers Chipset
ForceWare 9.53
Memory size 2GB (2 DIMMs) 2GB (2 DIMMs) 2GB (2 DIMMs) 2GB (2 DIMMs) 2GB (2 DIMMs) 2GB (2 DIMMs)
Memory type Corsair CM3X1024-1333C9DHX DDR3 SDRAM at 1333MHz Corsair CM3X1024-1066C7 DDR3 SDRAM at 1066MHz Corsair TWIN2X2048-8500C5 DDR2 SDRAM at 800MHz
CAS latency (CL) 9 7 4 4 4 4
RAS to CAS delay (tRCD) 9 7 4 4 4 4
RAS precharge (tRP) 9 7 4 4 4 4
Cycle time (tRAS) 24 21 12 12 12 12
Audio codec Integrated ICH9R/AD1988B with drivers Integrated ICH9R/AD1988B with drivers Integrated ICH9R/AD1988B with drivers Integrated nForce 570 SLI MCP/AD1986A with drivers Integrated nForce 590 SLI MCP/AD1988B with drivers Integrated nForce 680i SLI MCP/ALC885 with Realtek HD 1.66 drivers
Graphics GeForce 7900 GTX 512MB PCI-E with ForceWare 93.71 drivers
Hard drive Western Raptor X 150GB
OS Windows XP Professional
OS updates Service Pack 2

Thanks to Corsair for providing us with memory for our testing. 2GB of RAM seems to be the new standard for most folks, and Corsair hooked us up with some of its 1GB DIMMs for testing.

Also, all of our test systems were powered by OCZ GameXStream 700W power supply units. Thanks to OCZ for providing these units for our use in testing.

We used the following versions of our test applications:

The test systems’ Windows desktop was set at 1280×1024 in 32-bit color at an 85Hz 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
Memory subsystems don’t always dictate performance in real-world applications, but these synthetic tests are the most direct way to explore the impact of the P35’s DDR3 memory controller.

With DDR2-800 memory, the P35 Express offers equivalent memory bandwidth and latency to its P965 predecessor. Working DDR3 into the mix yields more bandwidth in both Cachemem and Sandra, and with our DDR3-1066 sticks, latency improves, as well. However, our DDR3-1333 system’s latency is quite high, no doubt thanks to its conservative 9-9-9-24 memory timings. There’s nearly a 20 millisecond latency gap between the P35 Express with DDR3-1333 and the nForce 680i SLI with DDR2-800—a virtual eternity within the confines of a modern PC.

The following Cachemem latency graphs are a little indulgent, but they do a good job of highlighting access latency across various block and step sizes. The Core 2 Duo runs out of on-chip cache after a block size of 4096KB, so you’ll want to pay more attention to the memory access latencies that follow with larger block sizes.

I’ve arranged the following graphs in order of highest to lowest latency with a common Z-axis to aid comparison.

Amusingly, the DDR2 memory controller in Intel’s aging 975X Express chipset manages to offer much lower latency than the latest P35 Express. The P35 is really a replacement for the P965, though, and the two are evenly matched with DDR2 memory. DDR3 running at 1066MHz with 7-7-7-21 timings looks to be a reasonable option, as well, since it doesn’t suffer the latency penalty associated with our DDR3-1333 config.

Still, the nForce 680i SLI’s DDR2 memory controller delivers by far the lowest access latencies of the lot.

Cinebench rendering

The chipset has little impact on Cinebench scores, which are virtually tied across the board. Our DDR3-1333 configuration is consistently faster than the rest of the field, but only by a hair.

Sphinx speech recognition

Sphinx favors faster memory subsystems, and with a little help from a 1333MHz front-side bus and similarly clocked memory, the P35 Express takes the lead. Even then, though, it’s just a little bit quicker than the 975X Express.

As we saw in our memory subsystem tests, the P35 is slightly faster with DDR3-1066 than it is with DDR2-800. Performance with the latter is virtually identical to that of the P965.

We should also note that despite having a clear advantage in our memory subsystem tests, the nForce 680i SLI trails all the Intel chipsets here by a substantial margin. Raw bandwidth and latency scores aren’t everything.


WorldBench overall performance
WorldBench uses scripting to step through a series of tasks in common Windows applications. It then produces an overall score. WorldBench also spits out individual results for its component application tests, allowing us to compare performance in each. We’ll look at the overall score, and then we’ll show individual application results alongside the results from some of our own application tests.

WorldBench spit out a score of 155 for each of our test systems, even after multiple runs. At least they can all agree on something. Let’s dig deeper to see if WorldBench’s individual test results tell a more interesting story.

Multimedia editing and encoding

MusicMatch Jukebox

Windows Media Encoder

Adobe Premiere

VideoWave Movie Creator

Scores are close through WorldBench’s multimedia editing and encoding tests, although all the Intel chipsets have an edge over the nForce 680i SLI in Premiere.

3D rendering

3ds Max

WorldBench’s 3ds max tests don’t really favor any of the chipsets. The P35 with DDR3-1333 does come out slightly ahead, but then it does enjoy a much faster system and memory bus.

Image processing

Adobe Photoshop

ACDSee PowerPack

Scores in WorldBench’s image processing tests are close, although as we saw with Premiere, the Intel chipsets have an edge over the nForce 680i SLI in ACDSee.

Multitasking and office applications

Microsoft Office


Mozilla and Windows Media Encoder

Intel core logic sprints out to a lead in WorldBench’s Office XP tests, too. There, the P35 Express leads the way with DDR3 memory, and is notably faster than the 975X. But the honeymoon doesn’t last; in WorldBench’s Mozilla and multitasking tests, the P35 is sent to the back of the class. Running DDR3-1333 on a 1033MHz front-side bus is the fastest P35 configuration, but even it can’t manage better than third place.

Other applications



WinZip and Nero don’t give these chipsets much opportunity to stretch their legs. It’s interesting to note that the P35 Express with DDR3-1333 is slightly slower in Nero, though.


Only F.E.A.R. and Quake 4 manage to coax higher frame rates from some of the field. In the former, the P35 wraps up second through fourth place and is fastest with our DDR3-1333 config. That setup also takes top honors in Quake 4, where the P35 is again faster with DDR3-1066 than it is with DDR2-800.

The fact that DDR3 at 1066MHz turns in a better performance than DDR2 at 800MHz isn’t as much of a foregone conclusion as one might imagine. After all, our DDR2-800 sticks are running with tight 4-4-4-12 timings while the DDR3-1066 is stuck at 7-7-7-21.

Multi-GPU gaming performance
Normally this is where we bust out CrossFire and SLI configurations to see how performance scales when a second graphics card is installed. However, we couldn’t get a pair of Radeon X1900 graphics cards running in CrossFire on the P35 Express. According to the latest Catalyst 7.4 graphics drivers, there are no communication lanes between PCI Express slots on the P35, presumably because they’re split between the chipset’s north and south bridge components. This configuration works just fine with the P965 Express, so it’s likely something that will be corrected with future drivers.

Despite not being able to test the P35’s CrossFire performance, we did run the chipsets through a wave of gaming tests at high resolutions and detail levels. This is how most folks actually play games, but with the graphics subsystem as the bottleneck, don’t expect to see much difference in performance between the chipsets.

Apart from Quake 4, where a couple of configurations are a little slower, there’s little variance in scores for these high resolution gaming tests. For what it’s worth, though, the P35 Express with DDR3-1333 ekes out the highest frame rate in Quake 4.

Serial ATA performance
The Serial ATA disk controller is one of the most important components of a modern core logic chipset, so we threw each platform a selection of I/O-intensive storage tests using a Western Digital Raptor WD1500ADFD.

We’ll begin our storage tests with IOMeter, which subjects our systems to increasing multi-user loads. Testing was restricted to IOMeter’s workstation and database test patterns, since those are more appropriate for desktop systems than the file or web server test patterns.

For whatever reason, the nForce 680i SLI won’t run IOMeter with a load of 128 or 256 outstanding I/O requests. We’ve seen this behavior before from other chipset and RAID controllers—including older Intel designs—so it’s nothing to get too worked up about. Even 64 outstanding I/Os is hitting a desktop chipset pretty hard.

The Intel Express crowd leads the nForce 680i SLI when we look at IOMeter transaction rates, but not by much overall. Note that there’s virtually no difference in performance between the P35, P965, and 975X Express.

IOMeter response times don’t vary much across these four chipsets.

CPU utilization is quite reasonable, as well. The chipsets are all clustered around half a percent.

iPEAK multitasking
We developed a series of disk-intensive multitasking tests to highlight the impact of command queuing on hard drive performance. You can get the low-down on these iPEAK-based tests here. The mean service time of each drive is reported in milliseconds, with lower values representing better performance.

The P35 Express is all over the map in iPEAK, but there are a couple of trends worth noting. With the exception of a workload that combines compressed file extraction and a Virtualdub import, the P35 is never significantly slower than the fastest system. That’s a better result than with the nForce 680i SLI, which steals a number of wins but also trails the pack in several workloads.

Overall, all the chipsets are evenly matched here. If you average the response times across all nine workloads, the P35, P965, and 975X all tie with a score of 2.06, while the nForce 680i SLI turns in a slightly slower 2.08.

SATA performance
We used HD Tach 3.01’s 8MB zone test to measure basic SATA throughput and latency.

HD Tach’s write speed test makes the Intel chipsets look particularly fast, but it’s not entirely accurate. Intel has confirmed that a bug with Western Digital’s Caviar RE2 hard drives can affect performance in this test when combined with ICHxR south bridge chips. We’re using Western Digital Raptor drives here, but this is the same behavior we observed with the Caviar RE2, so it’s likely the same problem. You can read more about the issue here.

ATA performance
ATA performance was tested with a Seagate Barracuda 7200.7 ATA/133 hard drive using HD Tach 3.01’s 8MB zone setting.

The P35’s lack of an IDE controller forces motherboard makers to use auxiliary storage controllers to support older ATA devices. More often than not, those auxiliary chips come from JMicron, and performance is pretty decent. The nForce 680i SLI’s integrated ATA controller is faster, though.

USB performance
Our USB transfer speed tests were conducted with a USB 2.0/Firewire external hard drive enclosure connected to a 7200RPM Seagate Barracuda 7200.7 hard drive. We tested with HD Tach 3.01’s 8MB zone setting.

USB write performance has always been a little slow with Intel’s south bridge chips, and the P35 is no exception. Read performance is fine, and even CPU utilization isn’t bad if you take HD Tach’s +/- 2% margin of error for CPU utilization into account.

3D Audio performance

Rather than being handled by the chipset, positional audio calculations are performed by codec chips and their associated drivers. Our P35 board features an Analog Devices codec, and they’ve traditionally had lower CPU utilization with 3D audio than codec chips from Realtek. Analog Devices codecs also handle EAX occlusion and obstruction effects properly, unlike their counterparts from Realtek. In fact, Realtek’s EAX implementation is so poor that it renders games like Battlefield 2 unplayable with EAX effects enabled.

Ethernet performance
We evaluated Ethernet performance using the NTttcp tool from Microsoft’s Windows DDK. The docs say this program “provides the customer with a multi-threaded, asynchronous performance benchmark for measuring achievable data transfer rate.”

We used the following command line options on the server machine:

ntttcps -m 4,0, -a

..and the same basic thing on each of our test systems acting as clients:

ntttcpr -m 4,0, -a

Our server was a Windows XP Pro system based on Asus’s P5WD2 Premium motherboard with a Pentium 4 3.4GHz Extreme Edition (800MHz front-side bus, Hyper-Threading enabled) and PCI Express-attached Gigabit Ethernet. A crossover CAT6 cable was used to connect the server to each system. The boards were tested with jumbo frames disabled.

Unfortunately, we’ve yet to see a P35 Express motherboard take advantage of the chipset’s integrated GigE MAC, so our Ethernet results are from the P5K3 Deluxe’s auxiliary Ethernet controllers. And there’s certainly room for improvement there. The PCI-based Realtek RTL8187 offers much lower throughput and much higher CPU utilization than competing solutions, and all of the Marvell chips consume more CPU cycles than the nForce 680i SLI’s integrated GigE MACs. PCI Express performance
We used the same ntttcp test methods from our Ethernet tests to examine PCI Express throughput using a Marvell 88E8052-based PCI Express x1 Gigabit Ethernet card.

The P35’s CPU utilization is a little higher than the rest in our PCIe performance test, but throughput is spot on. PCI performance
To test PCI performance, we used the same ntttcp test methods and a PCI VIA Velocity GigE NIC.

PCI performance isn’t a problem for the P35 Express.

Power consumption
We measured system power consumption, sans monitor and speakers, at the wall outlet using a Watts Up power meter. Power consumption was measured at idle and under a load consisting of a multi-threaded Cinebench 9.5 render running in parallel with the “rthdribl” high dynamic range lighting demo.

Now that’s unexpected. Despite lower DIMM operating voltages, the P35’s DDR3 configurations consume more power at idle and under load than when the chipset is running DDR2 memory. The power consumption associated with running a faster memory bus may be offsetting the lower per-DIMM voltage some. However, even with DDR2 memory, our P35 system consumes more power than the nForce 680i SLI reference board and our P965 and 975X systems. Our P35-based boards do have loads of onboard peripherals, including integrated Wi-Fi, but then so does the Asus P5B Deluxe Wifi-AP Edition motherboard we used to test the P965.

With native support for 1333MHz front-side bus speeds, the P35 Express’ exudes overclocking potential, especially considering that its P965 predecessor was no slouch in that department. So, with an eye towards pushing the chipset to its limits, we fitted our P5K3 Deluxe with a retail Core 2 Duo E6300 processor, dropped the CPU multiplier to 6X, lowered the memory bus divider to take it out of the equation, and started turning the screws on the front-side bus. And then we resumed breathing.

The P35’s front-side bus sailed past 333MHz with ease and glided all the way up to 490MHz—close to 2000MHz quad-pumped—without a fuss or even a hint of extra voltage. The board refused to POST at 500MHz, though, regardless of how much voltage or extra cooling we applied to the system.

490MHz ties the fastest front-side bus we’ve had working with this particular CPU, and it’s the only chipset to hit that speed without requiring extra voltage. Remember that this is with passive chipset cooling, as well. Given those factors, I suspect the P35 is capable of even higher speeds with the right processor.

CPU-Z hasn’t been updated to report memory bus speeds with the P35 Express, but we decided to push Corsair’s DDR3-1333 modules to see just how fast they’d run. For this round of overclocking tests, we used a 1:1 FSB-to-DRAM ratio and kept the stock 9-9-9-24 timings at the default 1.5V. These DDR3-1333 modules are fresh off the line, and we were able to get them up to a perfectly stable 760MHz—or DDR3-1520—with little effort. That certainly bodes well for DDR3’s clock scaling potential this early in the game.

The P35 Express marks the debut of DDR3 memory on the desktop, and the performance we’ve seen today hints at the new memory type’s potential. Despite running at much looser timings, our DDR3 configs kept pace with or offered better performance than low-latency DDR2, which is pretty good for the first wave of modules. Combine that with our stunning DDR3-1333 overclocking results, which had our modules running at DDR3-1520 speeds with default timings and voltage, and DDR3 looks to have a bright future indeed. We certainly expect it to become an attractive alternative to DDR2 much quicker than DDR2 became our preferred choice over DDR.

That said, initial DDR3 modules won’t be cheap—expect to see the Corsair DDR3-1066 sticks we tested today running at least $400 for a 2GB kit, with the DDR3-1333 carrying an even heftier price premium. The rare and minimal performance gains aren’t worth the extra scratch, so you’re better off sticking with more affordable DDR2 unless you’re looking to enhance your street cred or set records with extreme memory overclocking.

Fortunately, the P35 Express is well equipped either way. Our P5K3 Deluxe motherboard overclocked like a dream, more easily than any other Core 2 motherboard we’ve tested. Any motherboard based on the P35 chipset should work with a 1333MHz front-side bus, too, so there’s “free” overclocking headroom built-in if you have a current Core 2 processor designed for 1066MHz front-side bus. If you don’t want to pay the premium associated with DDR3, the P35 works with DDR2, too.

With DDR2 memory, the P35 performs very much like the P965 Express. The new ICH9R south bridge looks like a minor update to the ICH8R, too. That’s not terribly exciting, but it’s not entirely disappointing. The addition of a GigE MAC is a positive one, provided its performance is competitive, and plenty of folks could use the additional USB ports. But we probably don’t need more than six Serial ATA RAID ports, and with SATA optical drives readily available at affordable prices, IDE support doesn’t need to be in the chipset either. I’m not even sure second-generation PCI Express would have been appropriate for this chipset given its mid-range roots and the consistent dearth of PCIe peripherals.

That said, the P35 Express isn’t perfect. To start, compatibility problems with Western Digital Caviar RE2 hard drives and the ICH9R’s AHCI mode really need to be resolved. The power consumption of our Asus P5K and P5K3 Deluxe motherboards isn’t encouraging, either, although we won’t pass judgment on the P35’s power consumption as a chipset until we see what it can do on other motherboards. Expect detailed results on that front soon.

Overall, the P35 looks like a solid successor to the P965. Only time will tell if this is another BX in the making, but with a solid array of integrated peripherals, apparently ample overclocking potential, and flexibility on the memory front, it has more than a fighting chance. In fact, if I were building a Core 2-based system for myself today, it would have a P35 Express under the hood—albeit with DDR2 memory.

Comments closed
    • Ihmemies
    • 14 years ago

    #25: Wikipedia says: “Because partition tables on master boot record (MBR) disks only support partition sizes up to 2 TiB, you must use dynamic volumes to create bootable NTFS volumes over 2 TiB.”

    • MadManOriginal
    • 14 years ago

    Test oc’ing with a quad-core, it seems to be where the P35 becomes worthwhile.

    As for Penryn support…is VRM 11.0 correct for Penryn? I’ve been trying to find out this info for a while and haven’t seen anything, including vague rumors. Is there any solid evidence of which VRM is needed and which current P965 mobos or revisions support it?

      • Prototyped
      • 14 years ago

      It’s pretty unclear. I’d read that Bloomfield (of the Nehalem generation) would require VRM 11.1, indicating that perhaps the Penryn generation (Wolfdale/Yorkfield) wouldn’t need VRM 11.0 support:

      §[<<]§ but then I came across this: §[<<]§ which suggests that Wolfdale and Yorkfield will indeed need motherboards that supported VRM 11.1. I'm guessing things won't get clearer until Intel releases spec sheets for Wolfdale. (Conroe-capable boards use VRM 11.0.)

    • Prototyped
    • 14 years ago

    Edit: oops, should’ve been under #18.

    • somegeek
    • 14 years ago

    I have the P35 Asus P5K (regular).

    FSB: 375
    C2D 6400 @ 3000 (1.325V)
    2x512MB Corsair DDR2 5400 675 @ 750 (4-4-4-12 1.9V)
    GeForce 7100GS
    X-Fi ExtremeMusic
    Vista Basic 32-bit

    Idle – 103W (92W at stock speeds) no SpeedStep
    CPU Load – 143W

    Power consumption is higher than expected but it’s fine. Everything works and it’s fast.

      • d0g_p00p
      • 14 years ago

      The C2D 6400 multiplier can go as low as X2? Interesting…..

        • Prototyped
        • 14 years ago

        375 x 8 = 3000 MHz . . . it’s an 8x multiplier (remember, 266 x 8 = 2133, its stock speed).

        The 1500 MT/s number is the number of transfers per second, not the actual FSB clock rate.

    • blitzy
    • 14 years ago

    well it doesn’t stir any envy from me, pretty happy with my p965 here. one of those long term things.

      • PetMiceRnice
      • 14 years ago

      I hear ya! I have a 965 Express with an E6400 and plan to hang on to both until the end of the decade regardless of what comes out. I hunkered down with an 865PE, 2.8GHz P4C and a 9800 Pro from the summer of 2003 until November of 2006 and survived, so hanging onto my current system for the same length of time should be a piece of cake!

    • Bensam123
    • 14 years ago

    Do core logic chipset makers strive for higher performance anymore or just to break even?

    • gratuitous
    • 14 years ago
      • Jigar
      • 14 years ago


    • carlemueller
    • 14 years ago


    I would like to know if the 2TB maximum size for a RAID array is still limiting the ICH in this chipset. The memory controller bump is nice, but DDR3 and PCIX2 are things I don’t care about for building a system right now.

    I want to know ICH9R performance and limitations relevant to the ICH8R. Screw all the lame gaming benchmarks.

      • Krogoth
      • 14 years ago

      IIRC, 2TB is actually a limit of NTFS 5.0 😉

        • ew
        • 14 years ago

        Mm… ZFS

        • GreatGooglyMoogly
        • 14 years ago

        Really? It’s supposed to support 16 exabyte partitions. Maybe it’s yet another Explorer limitation though?

      • Forge
      • 14 years ago

      PCIe 2.0, I assume you mean, since PCI-X 2.0 was out years ago.

    • indeego
    • 14 years ago

    Power draw is interesting. That is somewhat significant increase in power at idle with nary a benefit in real-worldg{<.<}g

    • krazyredboy
    • 14 years ago

    Will Nvidia’s 680i boards support 45nm chips?

    I don’t know, but I may sway toward the Nvidia chipset, in hopes of running SLI, as it seems it would be better suited for doing so. Everything else is pretty much on par with each other otherwise.

    What do you guys think? Is it better to go that route if I’m looking into using SLI with higher end cards? Also, maybe even the possibility of running a PhysX card? I want my next build to be a big one, as I haven’t upgraded very much in the last few years.

    • Forge
    • 14 years ago

    DDR3 = expensive, not faster, and not interesting in general. I’m guessing we’ll have to wait for five or six speed grades and chipset revisions before it becomes interesting, much like DDR2 did.

    Yawnfest. Makes me happy with my 680i, though. I was leery about jumping on DDR2 this late in it’s life, just to get the C2D love, but it looks like it’ll be performance competitive and price desirable for a while to come.


    • Prototyped
    • 14 years ago

    You guys are being unfair to the DDR2 setups, running the DDR2 at 800 MT/s while the FSB is at 1066 MT/s. You should be using a 1:1 clock ratio and run it at 533 MT/s speeds, dual-channel. It’s not like the FSB can make use of any more bandwidth:

    FSB: 1066 MT/s FSB, 64 bits per transfer, comes to 8,533 MB/s.
    DDR2: 533 MT/s per channel, 64 bits per transfer, comes to . . . 8,533 MB/s.

    I suspect the DDR3-1066 is faster because *[

    • SnowboardingTobi
    • 14 years ago

    The Serial ATA CPU utilization graph on page 11 is wrong. Either it’s mislabled or you guys colored the wrong bar.

      • Dissonance
      • 14 years ago


    • IntelMole
    • 14 years ago

    Hopefully DDR3 will have a better start than DDR2. That’d be nice – seeing some sort of performance gain (or at least, not having it hold back scaling of some other component) from the get-go.

    • Peffse
    • 14 years ago

    It’s nice to know things are coming together pretty well… with luck, I’ll be able to get this with a 45nm Penryn and DDR3 memory sometime before the end of the year. It’ll be a nice jump from a D865PERL with 90nm Prescott and DDR1.

    Now if only I could choose, which arm and leg should I give?

    • elmopuddy
    • 14 years ago

    815 wasn’t all that bad, the 512 memory cap sucked though.. I still have a P3-700@933..

    Nice boards, not much reason to upgrade from P965 though.


      • Krogoth
      • 14 years ago

      815 was effectively a 440BX designed to work with Socket 370. 😉

        • Chryx
        • 14 years ago

        no?, it was based on i810?

        • Forge
        • 14 years ago

        Not even sort-of.

        i815 was i810 with an exernal AGP interface and the internal graphics core torn out.

        I still have an i815EP running my ancient P3/800EB and love them both dearly.

          • Prototyped
          • 14 years ago

          The graphics wasn’t even necessarily torn out, cf. i815G.

          The i815s added 133 MHz FSB support and the external AGP interface, and that’s more or less it.

          If you wanted gobs of RAM you had to get the i840 chipset. i840 boards were godawfully expensive and usually needed dual-channel RDRAM (not that the extra bandwidth helped, the FSB being the bottleneck). You could use PC133 SDRAM with boards that came with an MTH, but that posed a latency deficit.

          • Chryx
          • 14 years ago

          it didn’t necessarily have the internal graphics removed, I had an Abit SH6 which was i815 with AGP AND built in graphics

    • Lord.Blue
    • 14 years ago

    I remember (and have) one of the bad ones. Anyone remember the 810, 812/815 & 820 sets?

      • Krogoth
      • 14 years ago

      815 and 820 weren’t that bad. 820’s problem was RDRAM not the chipset itself. 815 was just damage control when P3/820 were getting hammered by the cheaper and faster Athlon platforms of the day.

      810/812 had quite a bit of bugs and problems that Intel did a recall on motherboards that feature them. I know somebody who still has a board (not in use) with the defective 810! 😉

    • Krogoth
    • 14 years ago

    P965/975X to P35 = 865/P875 to 925X all over again!

    There is almost no reason to get a P35 chipset. P965 always could handle 1333Mhz from the start. The only reason it had an official FSB of 1066FSB is purely marketing reasons. Intel didn’t want to marginalized their high-end chipset platform the 975X.

    In reality, the P965 is just a die-shrink “975XE”!

    I suspect that some enthusiast class P965 boards just need a BIOS update in order to support upcoming Penyrn due to changes in VRM standards or do board revisions that update the VRM standards.

    • marvelous
    • 14 years ago

    This p35 doesn’t really do anything. No point to upgrade if you already have 965 chipset.

      • jhtrico1850
      • 14 years ago

      45nm support?

        • Prototyped
        • 14 years ago

        That has more to do with the motherboard than anything else, for the same reason that Conroe worked in boards with 945, 975 and even 865 chipsets — they have to support the VRM 11.1 scheme. (It’s not clear it’s necessary, but according to §[<<]§ the Wolfdale and Yorkfield processors will need VRM 11.1 based boards.)

      • pedro
      • 14 years ago

      I’ll be sticking with P965 it seems.

    • krazyredboy
    • 14 years ago

    You know…I’m still using a couple of 440bx boards…now that I think of it…Crazy!

      • Krogoth
      • 14 years ago

      Sorry, the only thing recent that matches the BX would have to be the 975X.

      975X has similar longevity, stability and a solid set of features.

        • slaimus
        • 14 years ago

        I have a socket 370 BX board (Abit BM6) still running. The crazy Intel VRM changes made it only compatible with Celerons under 600Mhz with cB0 stepping. I hope the older 965/975 boards have a better fate.

        • Pettytheft
        • 14 years ago

        It took about 3 generations of boards to catch up with the i850E. Yeah it used Rambus but it topped the performance charts longer than any board I can think of. It was just as stable as anything I’ve ever used.

          • Krogoth
          • 14 years ago

          i850E could only use Socket 478 chips with a 400/533 FSB. 😉

          The reason it took Intel so long is that they were stuck with the one agreement with RAMBUS and SDRAM was woefully inadequate for Netburst. When that agreement expired Intel dropped RDRAM support like a very bad habit in favor of DDR1.

          • Prototyped
          • 14 years ago

          Yeah, but even then with the i845E/i845G finally with DDR support, the processor was starved of bandwidth:

          FSB 400 MT/s, 64 bits per transfer: 3,200 MB/s
          FSB 533 MT/s, 64 bits per transfer: 4,266 MB/s
          The max bandwidth i845E/i845G could provide was:

          DDR333: 333 MT/s, 64 bits per transfer: 2,667 MB/s

          It wasn’t until the Granite Bay E7205 workstation chipset that the bandwidth shortfall was satisfied, with dual-channel DDR support:

          DC DDR266: 266 MT/s, 64 bits per transfer: 2,133 MB/s x 2 = 4,266 MB/s.

          Of course, /[<]§ §[<<]§ Of course, the fact that it and its high-end version, Canterwood (i875P) rocked so much is also why Intel crippled its successors i915P/G and i925X by disabling support for more than one core -- motherboard manufacturers used the i875P instead of the much more expensive E75xx chipsets for multi-socket Xeon boards, and Intel couldn't have that! (That's likely also why they had to release the i945P/G and i955X when the dual-core Pentium D came out.)

Pin It on Pinterest

Share This