 |
 |
 |
||
|
|
||||
 |
|
|
| Introduction | ||||||||
When NVIDIA first resurrected the term "SLI" from the technology archives, it was welcomed back with open arms. Although the exact definition has changed from it's 3Dfx roots, it boils down to two graphics cards working on the same job. The obvious benefit is that two cards should be able to easily outperform even the fastest single card. The only real problem for some users was that the technology was limited to owners of nForce-based motherboards, and initially that meant having an AMD Athlon processor. NVIDIA eventually decided to address Intel's huge install-base as well, and nForce 4 SLI - Intel Edition motherboards arrived on the scene. It has only been six months since we first got a look at those boards, and there's already some room for improvement. In our initial reviews, the boards handled the benchmarks well, Occasionally outperforming Intel's lineup consisting of i915 and i925X chipset at the time. As we mentioned, Intel chipset based boards don't offer SLI support at all (althought the recently introduced i975X does support ATi Crossfire technology), making the nForce 4-SLI boards a top choice for hardware enthusiasts. A downside of the original nForce4-SLI chipset, however, was that the chipset's total number of PCIe lanes was limited to 20 and PEG slots had to share the 16 PCI Express lanes allocated to graphics. However NVIDIA's recently launched nForce 4 SLI X16 chipset also comes in both AMD and Intel flavors, with the "X16" standing for 16 devoted lanes for each PEG slot for 32 lanes total. As usual, Asus is one of the first to market, releasing the P5N32-SLI Deluxe, complimenting their P5ND2-SLI offering. Read on to find out whether it's always a good thing to be first in line.
|
| Layout and Features of the P5N32-SLI Deluxe | ||||
One topic that we hit on heavily when we evaluated Abit's AW8-MAX, was on the passive cooling methods which collectively were dubbed "Silent-OTES" technology. These methods consisted of a singular heatpipe connected from the Northbridge heatsink to a radiator placed near the I/O ports, specially placed heatsinks near the CPU, and "OC Strips" used to pull heat away from the PCB. Although we had already seen some of Asus' previous attempts at system cooling with the P5WD2, it appears that with the P5N32-SLI Deluxe Asus expands on similar techniques that Abit explored as well.
Rather than simply adding a single heatpipe to cool the chipsets, we've got two of them, one each running from the SPP Northbridge and MCP Southbridge to radiators placed around the CPU. While we appreciate the extra steps at cooling these chipsets (in fact, the SouthBridge still gets quite hot - enough so that we could not keep our finger on the heatsink for more than a few seconds), the placement of the radiators seemed somewhat puzzling. With the AW8-MAX, the radiator was placed at the edge, essentially bumping off legacy connections such as the parallel and serial ports. With the P5N32-SLI, the radiators are placed in a spot that seems to hinder heat dissipation. One is placed directly behind the IO ports, leaving little room between them and the CPU's heatsink and the other on the edge of the board, which would typically find itself placed right up against the PSU (although a properly place PSU exhaust fan sucking warm air in, could be very benefitial to this setup). This doesn't even take into account that the ATX power connection is placed in the middle of the two radiators, so we'll have cables snaked around here as well. It all adds up to a lack of airflow and increased heat build-up around the CPU, when we should be looking for the opposite.
To Asus' credit, we did receive a shipment a few weeks after we initially started testing the board that included two small fans that could be snapped onto the top of the heatsinks. Alas, these came with a warning that they should only be used in junction with passive or water cooling methods. Using them with a standard setup, the instructions warned, could interfere with CPU cooler airflow and thus should be avoided. A setup that would work perfectly here, in our opinion, would be installing one of the oversized heatsinks, such as those that Zalman markets. At least in this configuration, the large 120mm fan from the CPU cooler could cool down the CPU heatsink and blow air over these mini-heatsinks as well. In fact we've tested this exact setup in the HH labs with the Asus A8N32 SLI, which has a similar cooling configuration and it worked quite well. In direct contrast to the concerns we had so far concerning CPU cooling issues, the P5N32-SLI utilizes an 8 phase power desgin. The power array around the CPU socket is a neat and tidy design, that builds a steady clean power supply enviroment for the CPU. As such, the board should allow for more stability when operating at standard load as well as providing better overclocking potential. Additionally, as this board is one that can support more than one video card, Asus has placed additional space between the PCI-e x16 slots for improved thermal characteristics. Generally speaking, this allows for greater airflow over the heatsinks for each card, and specifically will allow users to install two of Asus' Extreme N7800GTX TOP graphic cards, or any cards that have dual-slot oversized coolers on them.
|
| BIOS and Overclocking Results | ||||||||||||||
Leave it to Asus to try something different with the bios of the P5N32-SLI Deluxe. The AMI BIOS we found in control of things had a few screens with typical offerings that one would expect, but also a slew of settings that allow for some fine tuning. Initially, we were able to view the type and speed of the processor as well as the total usable amount of RAM installed. To keep track of which drives were installed, and where, we could take a look at the IDE and SATA devices on the first menu screen. While some newer boards are cutting back on IDE options, here we can see that a Secondary IDE channel is alive and kicking. Of course, with so many options available to users including multiple SATA RAID options, one also has to decide the boot sequence. In our case, the choice is made much more simple by assigning the single hard drive as the first boot option and the CD-ROM as the second.
Unfortunately, getting the system booted was not always that easy. In many cases, booting up the P5N32-SLI resulted in a blank monitor screen and we had to manually shut down the board using the kill-switch on the power supply. Originally, we thought this might have been a heat-related issue, as the system typically would not reboot after running benchmarks and the like. After checking fans and airflow, we were convinced that the setup, at least, was not at fault. We also noticed that the next boot after using the PSU kill-switch, would invariably halt during POST with the message that "overclocking failed or overvoltage failed". The problem with this is that in just about every case, we were neither overclocking nor overvolting any of the components. Investigating the issue, we found that in the Hardware Monitoring section of the BIOS, that the CPU was consistently undervolted, usually by a solid volt off of the expected value. That is, even if we raised the voltage elsewhere, the VCORE Voltage would be one volt lower. We managed to gain a bit of stability by manually raising the CPU voltage, but still found that the system became unresponsive from time to time. We also had what appeared to us to be a small discrepancy when installing a second 7800GT card for SLI testing. During POST, the PCI-E configuration was shown as 16-1-1-1-1, which was identical to the readout when only one card was installed. In our experience with other nForce4-SLI boards, a one-card setup might show 16-1-1-1-1, but installing a second card would show the division of the PCI-E graphic card lanes, giving us 8-8-1-1-1. We were able to test the board using SLI, but the results are debatable without knowing if the full 32 lanes were opened up on this board. After speaking with Asus concerning these matters, we were sent a few new BIOS versions, including a second board to try, but all issues persisted throughout. If stability is a top priority for you (and for who is it not?), then you might want to wait until Asus irons out some of the P5N32-SLI's BIOS wrinkles.
Overclocking the P5N32-SLI is straightforward; under the Advanced Options check into the CPU Configuration and turn off the AI Overclocking, which is Asus' built-in method of raising the speeds, by changing the setting to "Manual". This allows the user to choose from a new menu called Performance Options. It's on this screen that the most tinkering will be done. Curiously, instead of entering in the CPU Clockspeed, as it appears to suggest that you should, the FSB Clock is entered. For example, to manually enter the stock speed of the Pentium 4 550 CPU, you would enter in 800MHz (200MHz x 4). It's a little too abstract, in our opinion, and could throw some users off as they are typically used to a range of 200-450MHz instead of 800-2000MHz. If the clock mode is left as linked, the new memory speed will be displayed above the CPU speed, or this can be set to Manual which allows the user to manually enter both speeds, independent of each other. Memory timings can also be set on this page, depending on whether or not the user wants to manually enter the timings or simply leave them to be determined by SPD. Finally, the PCI-E frequency can be manually set to prevent the graphics card PCIe links from running too far out of specifications. Running this too far off of expected speeds can easily disrupt the operation of most motherboards, so it's best to leave it at 100MHz.
We quickly ran through a number of speed checks, as our experience with this particular CPU has already limited our sights into the 230-240MHz range. Our Corsair XMS2 memory fares well with overclocking, so we left the speeds linked. We managed to quickly rise up to 225MHz (900MHz in the BIOS) before we had to bump the voltage for the CPU up to 1.55V, although as we mentioned earlier, this resulted in the CPU actually getting 1.45V. The RAM we left at 2.1V, which is the best voltage for the pair of Corsair sticks we used in testing. From there we continued on to 940MHz, or 235MHz, whichever works best for you, with the RAM operating at 391MHz on the memory divisor, which actually lists it as 783MHz (for sure this is an odd way to list memory timings). This appeared to be the ceiling for overclocking on the PN532-SLI with our particular CPU, as any further attempts to raise the speeds and/or voltage would give us an "overclocking failed" message at POST. As this fell into the range we expected, we decided to run a few benchmarks and call it a day. Raising the CPU to 3.99GHz provided about a 15% increase in CPU performance. |
| Testing Configuration and SANDRA Scores | ||||||||||||
How we configured our test systems: When configuring the test systems for this review, we first entered the system BIOS and set each board to their "Optimized" or "High-Performance Defaults" settings. The hard drive was then formatted, and Windows XP Professional with Service Pack 2 was installed. Then we installed all of the necessary drivers, and removed Windows Messenger from the system. Auto-Updating and System Restore were also disabled, and we setup a 1536MB permanent page file on the same partition as the Windows installation. Lastly, we set Windows XP's Visual Effects to "best performance", installed all of our benchmarking software, defragged the hard drives and ran all of the tests.
We began our testing with SiSoftware's SANDRA, the System ANalyzer, Diagnostic and Reporting Assistant. SANDRA consists of a set of information and diagnostic utilities that can provide a host of useful information about your hardware and operating system. We ran three of the built-in sub-system tests (CPU Arithmetic, CPU Multimedia, and Memory Bandwidth) that partially comprise the SANDRA 2005 suite of benchmarks. All of these tests were run with the Asus P5N32 powered by an Intel Pentium 4 550 3.4 GHz CPU with 1GB of Corsair XMS2 Pro DDR2 and compared against similar systems from SANDRA's database.
SANDRA's CPU Arithmetic and Multimedia benchmarks had the Asus P5N32 slightly behind the sample P4 550 scores from the database, and were also lower than what we saw with Abit's AW8-MAX or the Asus P5WD2. Memory bandwidth, as shown in the Memory Bandwidth Benchmark, was also off the mark. When compared to the i955X boards, the bandwidth was close to 200MB/s lower on the P5N32.
|
| PCMark05 Comparisons | |||||
For our next round of synthetic benchmarks, we ran the CPU and Memory performance modules built into Futuremark's brand new PCMark05. In the course of working with PCMark 05 we have found it to be even more robust in terms of test features than its predecessor. That said, the CPU and Memory test modules we use for comparison are very similar to the '04 version of the test suite. For those interested in more than just the graphs, we've got a couple of quotes directly from Futuremark that explain exactly what these tests do, and how they work:
"The CPU Test suite is a collection of tests that are run to isolate the performance of the CPU. The CPU Test Suite also includes multithreading: two of the test scenarios are run multithreaded; the other including two simultaneous tests and the other running four tests simultaneously. The remaining six tests are run single threaded. Operations include, File Compression/Decompression, Encryption/Decryption, Image Decompression, and Audio Compression" - Courtesy FutureMark Corp.
The PCMark05 CPU module's results were a close grouping of the Abit and Asus boards with the only notable distinction being the MSI P4N Diamond, which was a full 50 points off the pace. The overall lead, if you will, goes to the Asus P5N32-SLI. However, with such a small difference between it and the nearest board, the AW8-MAX, we aren't going to make any early calls on CPU performance.
"The Memory test suite is a collection of tests that isolate the performance of the memory subsystem. The memory subsystem consists of various devices on the PC. This includes the main memory, the CPU internal cache (known as the L1 cache) and the external cache (known as the L2 cache). As it is difficult to find applications that only stress the memory, we explicitly developed a set of tests geared for this purpose. The tests are written in C++ and assembly. They include: Reading data blocks from memory, Writing data blocks to memory performing copy operations on data blocks, random access to data items and latency testing." - Courtesy FutureMark Corp.
The memory performance module was also quite close, although once again the Asus P5N32-SLI barely held on a scant lead over the other boards. Only 36 points separated the top board from the bottom, measuring out to a margin of less than one percent. Synthetically speaking, CPU and Memory performance is on par for both the i955X and nForce4 SLI boards.
|
| World Bench 5.0 - Photoshop 7 & Office XP | ||||
In the WorldBench 5.0 tests, we found that all but one of the boards were operating at nearly the same level. For whatever reason, the Asus P5WD2 was consistently about 15 seconds slower than the other three. On the other hand, the Asus P5N32-SLI finished the Photoshop module in the shortest amount of time (tied with the MSI P4N Diamond) and was the second quickest in the Office XP Module. Of course, we're only talking averages in numbers of seconds here, so we won't get too excited yet. |
| Encoding Speed Tests | ||||||||
We continued testing with another module from World Bench 5, this time based on Windows Media Encoder 9. PC WorldBench 5's Windows Media Encoding test reports encoding times in seconds, and like the tests on the previous page, lower times indicate better performance here.
Five seconds. That's the spread in this test with both Asus boards at the "extremes". So far in our testing it appears that the nForce4-SLI boards are faring just a bit better than the i955x duo. Abit's AW8-MAX is right in the mix, but the P5WD2 has usually brought up the rear. Being one of the first retail i955X motherboards, it might not be as fully optimized as the later released AW8-MAX and certainly it's brethren, the P5N32-SLI, has a tiny bit more performance under the hood.
In our custom Lame MP3 encoding tests, we convert a large WAV file to the MP3 format, which is a very popular scenario that many end users work with on a regular basis, to provide portability and storage of their digital audio content. In this test, we chose a large 223MB WAV file (a never-ending Grateful Dead jam) and converted it to the MP3 format. Processing times are recorded below. Once again, shorter times equate to better performance.
Not much can be said for the MP3 encoding tests. Each board completed the conversion in either 2 minutes 25 seconds or 2 minutes 26 seconds. Given some room for error, we may as well call this benchmark a four-way tie.
|
| Kribibench 1.1 Benchmarks | ||||
Next up, we ran Kribibench, a 3D rendering benchmark produced by the folks at Adept Development. Kribibench is an SSE aware software renderer. A 3D model is rendered and animated by the host CPU, and the average frame rate is reported. We used two of the included models with this benchmark: a "Sponge Explode" model consisting of over 19.2 million polygons and a gargantuan "Ultra" model that is comprised of over 16 billion polys.
The overall differences amongst the boards hasn't changed much when testing with Kribibench, although the relative order has. Whereas much of the testing so far had the nForce4-SLI boards as the top performers, we find that the frame rates in Kribibench are leaning ever so slightly in Intel's favor. Abit's AW8-MAX led both tests, followed by either of the two Asus boards, and the MSI P4N Diamond was found bringing up the rear. |
| Cinebench 2003 and 3DMark05 Results | ||||||||
The Cinebench 2003 benchmark is an OpenGL 3D rendering performance test, based on the commercially available Cinema 4D application. This is a multi-threaded, multi-processor aware benchmark that renders a single 3D scene and tracks the length of the entire process. The time it took each test system to render the entire scene is represented in the graph below (listed in seconds).
Our Cinebench 2003 chart mirrors exactly what we saw in the LAME MP3 encoding. That is, the Asus P5N32-SLI and Abit AW8-MAX boards are tied for the lead with the fastest scene rendering. On the other hand, the Asus P5WD2 and MSI P4N Diamond are also tied, this time for last. It's almost eerie how similar this graph is.
3DMark05's built-in CPU test is a multi-threaded "gaming related" DirectX metric that's useful for comparing relative performance between similarly equipped systems. This test consists of two different 3D scenes that are generated with a software renderer, which is dependant on the host CPU's performance. This means that the calculations normally reserved for your 3D accelerator are instead sent to the central processor. The number of frames generated per second in each test is used to determine the final score.
3DMark05 showed us a wider disparity between the two chipsets. Basically, it appears that the i955X chipset is handling CPU calls more efficiently than the nForce4-SLI boards. There's a definite 100-point gap that separates the two sets of board. Further testing will determine how this will affect actual gameplay results. |
| Testing with UT2004 and DOOM3 | ||||||||
To start our in-game testing, we did some low-resolution benchmarking with Unreal Tournament 2004. When testing with UT 2004, we use a specific set of game engine initialization settings that ensure all of the systems are being benchmarked with the exact same in-game settings and graphical options. In the following two game tests, we used "Low-Quality" graphical settings and low screen resolutions which isolates CPU and Memory performance.
There's quite a close grouping in UT2004 testing, with the Asus boards bookmarking the top three boards. Less than a quarter of a frame per second separated them. The AW8-MAX was a full frame behind, although this equals less than one percent of a difference.
For our next game test, we benchmarked all of the test systems using a custom multi-player Doom 3 timedemo. We cranked the resolution down to 640 x 480, and configured the game to run at its "Low-Quality" graphics setting. Although Doom 3 typically taxes today's high-end GPUs, when it's configured at these minimal settings it too is more CPU and memory-bound than anything else.
Again we're looking at some frame rates that for all purposes should be considered equal. The list from top to bottom doesn't remain the same, however, with the AW8-MAX moving towards the front and the PN532-SLI slipping to the rear. Still, we're talking fractions of a frame per second at already-blistering rates.
|
| SLI Testing - Asus P5N32 vs. MSI P4N Diamond | ||||
From the low-quality gaming benchmarks, we're now going to move on to some more demanding tests using a pair of 7800GTs in SLI mode. For these tests, we will only be using the MSI P4N Diamond with the original n4Force-SLI chipset and the Asus P5N32-SLI using the updated n4Force-SLI X16. Basically, what we're looking for here is some evidence that upgrading from 16 to 32 PCI-e lanes improves performance. We will stick with two of the more demanding graphic engines; Doom3 and Far Cry.
Raising the resolution and enabling driver optimizations such as anti-aliasing places much more strain on the graphics cards than our earlier tests. As the games get much more advanced, using more powerful shaders and real-time physics, access to available CPU and Memory Bandwidth becomes much more vital. When the AGP bus became the bottleneck, manufacturers turned to PCI-e to help alleviate this issue. Using two GeForce cards in SLI mode further improves performance by having two cards tackle the job of one. However, original nForce4-SLI boards didn't offer any increase in the number of PCI-e lanes over other Intel and AMD-based boards, splitting up the 16 lanes that were available. Asus' P5N32-SLI comes with 32 lanes thanks to the updated nForce4-SLI X16 chipset. In our testing of two popular games using two 7800GTs in SLI mode, we can see a small, yet favorable increase in frame rates with the P5N32-SLI. Typically, we were looking at about a 5-10 fps difference. It might not be enough for current nForce4-SLI owners to make the upgrade, but new system builders, especially those who bleed benchmark numbers, would probably find reason to start here.
|
| SLI - 8xAA and 16xAA Testing | ||||
Using the Coolbits tweak, the new SLI anti-aliasing rendering modes are available from within the advanced global driver options. After selecting "SLI antialiasing" from within the "SLI Rendering Mode" drop-down menu, two new AA modes are made available, SLI 8X and SLI 16X. We should note that these two modes replace the traditional AA options available within NVIDIA's Forceware drivers when enabled. Essentially, each GPU renders a frame and applies the appropriate filtering, then the two frames are blended, and finally the blended frame is sent out to the display. In the SLI8X AA mode, NVIDIA's 4X multi-sample AA is applied; in SLI16X AA mode, 8xS (4X multi-sample / 2X super-sample) AA is applied.
The screens may look better, but the performance hit using SLI-AA is enormous. We doubt many readers would expect to see framerates of 20-30fps in a typical SLI review. More importantly perhaps, is that we're not seeing any differences between the two boards. While we knew that SLI-AA would heavily tax systems when using the newer engines, we had hoped that the extra bandwidth afforded by the SLI-X16 chipset would have helped out here. Unfortunately, regardless of whether 16 or 32 PCI-e lanes were available, the framerates were almost exactly the same.
|
| Benchmark Analysis and Our Rating | ||||
Benchmark Analysis: To determine the overall performance of the P5N32-SLI, we ran a series of benchmarks against competing high-end boards such as Abit AW8-MAX and Asus P5WD2, both using the Intel 955X chipset and an earlier nForce4-SLI board from MSI, the P4N Diamond. For the most part, all four boards were performing on-par with each other, which makes it hard to pick out a "leader" from the group. We could just leave it like that, that the two chipsets are nearly equal, but for the PC enthusiast there's something the nForce4-SLI boards have that Intel doesn't, yet. nForce4-SLI boards support two of NVIDIA's GeForce cards, which can greatly improve frame rates in most games. It's also here where part of our concern for the P5N32-SLI developed. When placing the P5N32-SLI in direct competition with the MSI P4N Diamond, the performance difference was less than we had hoped for. Doubling the number of dedicated PCI-E lanes for the graphics cards led to minimal gains in gaming benchmarks, and no gain whatsoever with SLI-AA enabled. Perhaps future game engines and usage models will further exploit the additional bandwidth the nForce 4 SLI X16 chipset has to offer but for now it seems the chipset's full potential is not being realized.
They say the numbers don't lie. As such, we're quite content with the overall performance of the Asus P5N32-SLI. It blasted its way through the benchmarks with similarly configured high-end boards, with barely a few percentage points separating them. However, when we got to the SLI testing, we didn't get the results we had been hoping for. The whole reasoning behind the nForce4-SLI X16 update was the doubling of the number of PCI-E lanes for graphics applications from 16 to 32. Seen here, this doesn't appear to offer too much of an improvement over the original nForce4-SLI. For the new system builder, that might not be a problem. It makes perfect sense to start with the latest board than search out an older model. For upgraders, on the other hand, there's just not enough of an improvement to make the switch. Of course, gaming benchmarks aren't everything. Stability and ease of operation are another. And it's these areas that the Asus P5N32-SLI really failed to impress us. Voltage problems surfaced during testing, continually showing that the CPU was getting 1V less than what was specified. While this could be easily remedied by adding an extra 1V in the BIOS, we were continually plagued with booting issues that could not be solved either with newer BIOS revisions or another board. We're also a little bit concerned on heat dissipation issues with the P5N32-SLI. The heatpipe strategy was appreciated, possibly taking a cue from Abit's OTES technology. It's quiet and should be moving heat away from sensitive components. Even so, we noticed that the SouthBridge heatsink was still quite hot to touch, and whisking the heat away only to place it directly around the CPU seems to be a foolish endeavor. Additional fans can be used to cool down these extra heatsinks, but can be installed in only a few cases, making their value nominal. In light of these issues, we're left wondering if being first to market is always such a great thing. It's great to see that Asus is willing to take a new chipset and get a board out there. But, if the headaches outweigh the benefits, is it really worth the added cost? Current street price for the P5N32 SLI Deluxe is $200 - $220, about $20 - $30 more than current nForce 4 SLI Intel Edition boards based on the SLI X8 version of the chipset. For now, we're giving the Asus P5N32-SLI a 7.5 on the Heat Meter, and hoping that later BIOS or board revisions can fix some of these issues.
|
;)
;)
;)
;)
;)
;)
;)
;)
;)
;)
;)
;)
;)
;)
;)
;)
;)
;)
;)
;)
;)
;)
;)
;)
;)
;)
;)
;)
;)
;)
;)
;)
;)
;)
;)
