Original Link: https://www.anandtech.com/show/1279
nForce3-250 - Part 2: Taking Athlon 64 to the Next Level
by Wesley Fink on March 29, 2004 11:00 AM EST- Posted in
- CPUs
One of the nice things about a two-part review is that you get to address things you may have overlooked in Part 1. Since there has been so much discussion about Sound Storm in the comments for Part 1, a little more information about the sound capabilities of nForce3-250Gb is needed.
It was a bit surprising to see so much commentary about Sound Storm being absent from nF3-250Gb, since it was first removed from the previous generation nF3-150. There are several reasons Sound Storm is not a part of nForce3-250Gb:
- Customer surveys by nVidia found that most buyers did not use Sound Storm, and that Sound Storm did not enter heavily into the decision to buy nForce. So, the decision was made to choose the on-chip LAN, firewall, and much-expanded RAID capabilities, which benefit greatly from being moved off the bus.
- nVidia is committed to the one-chip chipset solution for Athlon 64. They are firmly convinced that the one-chip eliminates the potential bottlenecks of a north-south bridge communications bus. Even with the memory controller on the CPU, there is only so much real estate practically available on a single-chip chipset.
- There are new sound solutions in the works for nVidia. You may see them in a future chipset or on a sound card. Final decisions have not been made.
The other area that was questioned was nVidia's competence to deliver a decent RAID solution, given the past problems with IDE performance on earlier MCP platforms. We will provide a few disk benchmarks in the Performance tests. However, from a user's point of view, the memory testbed uses the same model SATA 10,000 rpm drives in an Intel RAID array that are being used in the nForce3-250Gb Reference Board RAID. We have not been a fan of past nVidia Disk solutions and normally skipped their IDE drivers, but this time around, the performance has been extremely stable and far less finicky than the Intel setup we use for testing. Keep in mind that this is a Reference Board; we will feel much more comfortable in reaching a conclusion about IDE/RAID stability and performance after testing production motherboards. One other point is that nF3-250Gb was shipped with prototype version 4.04 nForce Platform drivers, which have not yet been released.
There were also emails with questions about software that is part of the nForce3-250 package. This is also related to the Platform Drivers 4.04 and the BIOS of shipping motherboards, but the Reference Board has a very useful System Utility that allows basic overclocking (but no CPU multiplier option). The nVidia System Utility has been available since last fall, but only works if the board manufacturer enables it in BIOS. Also included is a working nVidia DVD Player à la ATI, a very flexible nVRAID Manager, and the usual nVidia Mixer for audio. Version 4.04 Platform drivers for Win2K/XP include:
- Audio driver version 4.09
- Audio utility version 4.09
- Win2K Ethernet driver version 4.16
- Win2K Ethernet NRM driver version 4.16
- Network management tools version 4.16
- GART driver version 3.77 (WHQL) with updated uninstaller files
- Memory controller driver version 3.38 (WHQL) with updated uninstaller files
- SMBus driver version 4.04 (WHQL) with updated uninstaller files
- Installer version 4.16
- Win2K IDE 2.5 driver version 4.15
nForce3-250Gb: IDE and RAID Benchmarks
Disk Performance Testing can provide extremely variable results, even when test conditions are well-controlled. For that reason, we were skeptical to drop IDE and RAID tests into our review of the nForce3-250Gb. However, it was certain, given nVidia's past lackluster drive performance, that HD tests should be included in nF3-250 benchmarking.Some of the most commonly used benchmarks for Hard Drive Testing are HDtach 2.7, PCMark 2004 HD Tests and SiSoft Sandra 2004 File System Benchmark. We have included all 3 test results here for both IDE and nVidia RAID. Tests in both cases were run on Hard Drives about 20% full, with the benchmark run from the IDE hard drive. The IDE hard drive was a 7200 RPM 80GB drive with a 50GB C: boot partition, and tests were run on C:. The RAID array was 2 Western Digital Raptor 10,000 RPM 36.7GB hard drives. Both the IDE and RAID were driven by nVidia IDE drivers in nForce Platform Driver 4.04.
Sandra 2004 shows a 38% increase in throughput with nVidia 10,000RPM RAID compared to a single 10,000RPM SATA drive. PCMark 2004 HDD Test Suite shows a smaller increase in performance with RAID, around 27%. Benchmarks for a single 7200RPM IDE are also included for comparison.
Perhaps even more important for critics of past nVidia IDE implementations is that nVidia and Intel throughput in IDE is now about the same with the nForce3-250Gb. Intel SATA RAID is a bit faster in Sandra, but the difference in measured benchmark performance was very small, at about 6%.
For those who hate the numbers from benchmarks like Sandra and PCMark, the HDTach screen captures provide a very interesting snapshot. The two graphs below, with blue title bars, represent nVidia's nForce3-250Gb drive controller performance.
We also tested the Intel RAID controller with another pair of 10,000 rpm Western Digital Raptors. Those results are shown with the green title bar below.
As you can see, performance of the Intel SATA RAID and nVidia SATA RAID are virtually identical.
The biggest surprise in the HD Tach benchmarks was the fact that SATA RAID requires half the overhead of basic IDE in nVidia's implementation. However, the CPU utilization, while greatly improved over numbers we have seen for nVidia in the past, is still very high compared to Intel's SATA RAID. Part of this has to do with the architectural differences between Athlon 64 and Intel processors and chipsets, but there is still a large difference in CPU overhead between the nVidia and Intel RAID solutions. Frankly, the real performance difference is likely to be nil on the typical desktop, but this will matter to some.
nForce3-250Gb: Stress Testing
We performed stress tests on the nForce3-250Gb Reference Board in these areas and configurations:- Chipset and motherboard stress testing was conducted by running the FSB at 249MHz at a multiplier of 9.5.
- Memory stress testing was conducted by running RAM at 400MHz with a DIMM slot filled and at 400MHz with 2 DIMM slots filled at the lowest memory timings possible.
Front Side Bus Stress Test Results:
As normally done in our testing of production motherboards, we ran a full range of stress tests and benchmarks on the nForce3-250Gb Reference Board to test stability at an overclocked speed. This included Prime95 torture tests, and the addition of other tasks - data compression, various DX8 and DX9 games, and apps like Word and Excel - while Prime95 was running in the background. Finally, we ran our benchmark suite, which includes ZD Winstone suite, Unreal Tournament 2003, SPECviewperf 7.0, and Gun Metal Benchmark 2. At default voltage, 249MHz was the highest overclock that we were able to achieve with the nForce3-250Gb and the HyperTransport setting reduced to 4X or 800 (Actual 996 at 249FSB) while running these tests.Unlike our experiences with some of the overclocks run on other A64 boards, the nForce3-250Gb was completely stable when overclocked. We did have stability problems at the highest 250 setting, which remains a puzzle, but 249 was completely stable. It would be impossible to run at 249FSB with SATA drives and our picky ATI 9800 PRO without a working PCI/AGP lock.
Memory Stress Test Results:
This memory stress test simply tests the ability of the nForce3-250Gb Reference Board to operate at its officially supported memory frequency (400MHz DDR) at the lowest supported memory timings that our OCZ PC3500 Platinum Ltd memory will support:Stable DDR400 Timings - 1 DIMM (1/2 DIMM populated) |
|
Clock Speed: | 200MHz |
Timing Mode: | N/A |
CAS Latency: | 2.0 |
Bank Interleave: | N/A |
RAS to CAS Delay: | 2T |
RAS Precharge: | 5T |
Precharge Delay: | 2T |
Command Rate: | N/A |
We had no problem running one DIMM of our standard OCZ PC3500 Platinum Ltd at the highest memory timings in the nForce3-250Gb Reference Board. We were also able to run the memory test suite with complete stability at 2-2-2-5 timings.
Filling all available memory banks is more strenuous on the memory subsystem than testing 1 DIMM, but 2 DS DIMMs worked just fine on the nForce3-250Gb. With 2 DIMMs, we could run the same aggressive timings used for one DIMM, which is excellent performance for an Athlon 64 motherboard.
Stable DDR400 Timings - 2 DIMMs (2/2 DIMMs populated) | |
Clock Speed: | 200MHz |
Timing Mode: | N/A |
CAS Latency: | 2.0 |
Bank Interleave: | N/A |
RAS to CAS Delay: | 2T |
RAS Precharge: | 5T |
Precharge Delay: | 2T |
Command Rate: | N/A |
We tested the memory timings with both banks filled using several stress tests and general applications to guarantee stability. Prime95 torture tests were run successfully at the timings listed in the above charts. We also ran ScienceMark (memory tests only) and Super Pi. None of the three stress tests created any stability problems for the nForce3-250Gb Reference Board at these memory timings.
We suspect production nForce3-250Gb motherboards will have 3 or 4 DIMM slots, but as we have reported in the past, we have had some difficulties with every Athlon 64 motherboard so far when we tried to use more than 2 DIMMs. We will test memory slots again when production motherboards appear on the market in a few weeks.
Certainly, we will see 4 DIMMs, or two dual-channel pairs, on the nForce3-250Gb Ultra version for Socket 939.
Benchmarking nForce3-250GB
Before we benchmark nVidia's new nForce3-250Gb, we needed to decide exactly what we were going to test. There were several areas of particular interest:How does nForce3-250Gb compare to other chipsets that we have tested? In this case, the FX5950 video card was removed and our standard ATI 9800 PRO 128mb video card was installed. So that results could be truly compared to previous benchmarks published at AnandTech, the 3400+ was replaced with the 3200+, which we have used for virtually every Athlon 64 board that we have tested. In the interest of consistency, we also replaced the 512MB of 2-2-2-5 memory with 1GB of memory (2 x 512MB) that also ran fine at 2-2-2-5 timings.
How does nForce3-250Gb perform with top components? Is the nF3-250Gb a better performer with an nVidia FX5950 Ultra? Here, it would not have been comparing apples to run ATI benchmarks with the 9800 PRO, since the 9800 XT is the current top-of-the-line. So, full benchmarks were run with both the nVidia top-line 5950Ultra and the ATI top-line 9800 XT. To provide a sanity check, we also ran a full series of benchmarks with the SiS 755 chipset, a 3400+, and the ATI 9800 XT. Since we had found the SiS board to be one of the fastest that we have tested, we wanted to see if nF3-250 benchmarks with ATI were comparable to the best performance we had seen with ATI on Socket 754. If nForce3-250 only performs comparably to the best with a 5950U, the utility of the chipset is somewhat limited. However, if nF3-250 performs just as well as the best with ATI - and even better with 5950U - then the 5950U performance is a nice bonus.
Performance Test Configuration: Chipset Comparison
Performance Test Configuration | |
Processor(s): | AMD Athlon64 3200+ (2.0GHz) |
RAM: | 2 x 512MB Mushkin PC3500 Level II OR 2 x 512MB OCZ PC3500 Platinum Ltd |
Hard Drive(s): | Seagate 120GB 7200RPM SATA (8Mb buffer) |
Video AGP & IDE Bus Master Drivers: | SiS 1.17a AGP/IDE 2.04/RAID 1.03 VIA 4in1 Hyperion 4.51 NVIDIA nForce version 4.04 for nF3-250 |
Video Card(s): | ATI Radeon 9800 PRO 128MB (AGP 8X) |
Video Drivers: | ATI Catalyst 4.3 |
Operating System(s): | Windows XP Professional SP1 |
Motherboards: | nVidia nForce3-250Gb Reference Board ECS 755-A2 (SiS 755) MSI K8T Neo (VIA K8T800) |
Benchmarks used either Mushkin PC3500 Level II or OCZ PC3500 Platinum Ltd memory modules. Both DIMMs use Winbond BH5 chips and perform virtually the same in our tests.
All performance tests were run with the ATI 9800 PRO 128MB video card with AGP Aperture set to 128MB with Fast Write enabled. Resolution in all benchmarks is 1024x768x32 unless otherwise noted.
Game Benchmarks
We have added several new benchmarks to our standard Gaming tests. These include Halo, Microsoft's Direct X 9.0b game; Splinter Cell, a DX9 game; X2 Benchmark, a DX 8.1 game that includes Transform and Lighting effects; the DX9 Aquamark 3; and the DX 8.1 Comanche 4 benchmark. Since we have found that Comanche 4 can become video card limited at higher resolutions, we will only include benchmarks run with 4X anti-aliasing enabled to differentiate system performance better using our standard ATI Radeon 9800 PRO video card.Content Creation and General Usage - 3200+
With the memory controller for AMD chipsets on the CPU, we really did not expect much difference in Content Creation and General Usage benchmarks compared to other chipsets. This is particularly true in the case of Socket 754 chipsets where earlier chipsets from nVidia and VIA performed about the same and SiS performed a bit better. We tested with the fastest board in each category and found General Usage and Content Creation were virtually the same on all 3 boards.
Since nForce3-250Gb is basically a Socket 939 design, we may see greater performance differences due to the higher HyperTransport speed when 939 processors and boards are launched.
Please keep in mind that all tests here were performed with an ATI video card. You will also want to see the impact that using an nVidia video card has on benchmark results in the second part of this review.
Gaming Performance - 3200+
Gaming performance also shows the ATI 9800 PRO performing about the same on nVidia, VIA, and SiS chipsets. nForce3-250Gb is generally at the top in the graphs, but the performance differences are small enough to be within the margin of error for most benchmarks. One exception is Aquamark 3, where nForce3-250Gb and the VIA-based MSI K8T Neo do outperform the SiS 755-based ECS 755-A2. This is likely the result of the SiS Gart, which still slightly lags VIA and nVidia, but continues to improve with each new release.
Workstation Performance - 3200+
Workstation Performance is another area where the 3 chipsets appear to switch leads in each successive Workstation benchmark. With the chipsets all running the same ATI 9800 PRO on a Socket 754 board, performance of the best boards from SiS, nVidia, and VIA is very close. When looking at Socket 754 boards, the feature set becomes more important if you are comparing the best from each chipset. We found a greater variation in Socket 940 chipset performance in the past. Socket 939, like 940, may show larger performance differences between chipsets and boards.
Performance Test Configuration: Graphics Card Performance
nVidia claims that the best performance with nForce3-250Gb is achieved with an nVidia graphics card. To test that, we compared performance of the nForce3-250Gb Reference Board with the top ATI and nVidia graphics cards and a 3400+ Athlon 64. We also included benchmarks from the same 9800XT 256MB video card on a similarly configured SiS 755 board. All benchmarks were run on a single IDE drive to remove RAID as a factor in the performance comparison.Performance Test Configuration | |
Processor(s): | AMD Athlon64 3400+ (2.2GHz) |
RAM: | 2 x 512MB Mushkin PC3500 Level II OR 2 x 512MB OCZ PC3500 Platinum Ltd |
Hard Drive(s): | Seagate 120GB 7200RPM SATA (8Mb buffer) |
Video AGP & IDE Bus Master Drivers: | SiS 1.17a AGP/IDE 2.04/RAID 1.03 VIA 4in1 Hyperion 4.51 NVIDIA nForce version 4.04 for nF3-250 |
Video Card(s): | nVidia FX5950 Ultra 256MB (Gainward) ATI Radeon 9800 XT 256MB |
Video Drivers: | nVidia Detonator 56.64 ATI Catalyst 4.3 |
Operating System(s): | Windows XP Professional SP1 |
Motherboards: | nVidia nForce3-250Gb Reference Board ECS 755-A2 (SiS 755) FIC K8T800 (VIA K8T800) |
Benchmarks used either Mushkin PC3500 Level II or OCZ PC3500 Platinum Ltd memory modules. Both DIMMs use Winbond BH5 chips and perform virtually the same in our tests.
All performance tests were run with either the 256MB ATI 9800 XT or the 256MB nVidia FX5950 Ultra video card. On both cards, AGP Aperture was set to 256MB with Fast Write enabled. Resolution in all benchmarks is 1024x768x32 unless otherwise noted.
Additions to Performance Tests
We have included PCMark 2004 Standard Score for General Usage comparison. This is in addition to Veritest Multimedia Content Creation Winstone 2004 and Veritest Business Winstone 2004 for system benchmarking.Content Creation and General Usage - 3400+
When we compare Content Creation and General Usage performance with the 5950U, the nForce3-250Gb breaks away a bit from the pack. Certainly, the 9800 XT performance on both the SiS 755 and NF3-250 are about the same, so 5950U just gives a bit more performance boost. We see the same pattern on Future Mark's PCMark 2004, with all scores close, but nF3-250/5950U slightly ahead.
You can compare these results to others that have been published to see the expected results with 5950U versus ATI 9800. There certainly does seem to be a synergy in the 5950U/nF3-250Gb combo that yields performance that is a bit better. The ATI 9800 XT performs just as well in nForce3-250 as it does in other Athlon 64 Socket 754 boards, but 5950U is a little faster.
Gaming Performance - 3400+
The 5950U with the nForce3-250Gb continues to pull away in the gaming arena - - outperforming both boards with the 9800 XT in Open GL (Quake 3) and DX 8.1 gaming, such as Comanche 4, X2, and Unreal Tournament 2002. In DX9 titles like Halo and Splinter Cell, the Radeon 9800 XT is equal or slightly faster than 5950U. However, the 9800 XT performs just as well or better on nF3-250Gb as it does on the SiS 755.
One DX9 standout for 5950U is Aquamark. Here, the combination of nForce3-250Gb with the nVidia FX5950 Ultra gives the highest score that we have seen for stock performance of a 3400+ at over 52,000.
Workstation Performance - 3400+
If there is a surprise area in the 5950U tests, it is certainly workstation performance. The last time that we looked at nVidia graphics compared to ATI in Workstation performance, we had to consider the results something of a toss-up. This time, with a 5950U on the nForce3-250Gb, nVidia dominates most of the SPECviewperf benchmarks.
We see the usual nVidia/ATI architecture patterns in the last benchmarks with 5950U dominating ProE and ATI dominating UGS scores. It is worth mentioning that the ATI 9800 XT once again performs as well or better in the nForce3-250Gb than it does in the SiS755, which had been our fastest board.
Final Words
nVidia prides themselves as the graphics card manufacturer and chipset maker that caters to gamers and computer enthusiasts. It must have been a bitter pill to swallow to see ATI making moves to the forefront in the graphics card business, and to see nVidia's hard-won market share in the AMD chipset business get lost in Athlon 64 to VIA. nVidia would tell us that nForce3-150 was interim, and that now the market gets more serious. Regardless of the words, it does look like nVidia has really concentrated on making the nF3-250 family all that the nF3-150 was not.The feature set of the nForce3-250Gb is excellent and well-balanced. We finally see a working PCI/AGP lock on an Athlon 64 board, which is good news for overclockers. The on-chip Gigabit LAN and Firewall are also welcome features that will make today's LAN gamers very happy once they have taken a test drive. The easy set-up of the Firewall for an avid LAN gamer will have them smiling in no time. No, there isn't Sound Storm or premium audio, but we think it's a fair tradeoff for nVidia, since they remain the only chipset vendor with a single-chip solution for Athlon 64.
nVidia also has demonstrated conclusively that their rumored problems with 1000 HyperTransport for socket 939 are a thing of the past. We were particularly excited to see that the nForce3-250Gb Ultra could also be used on Socket 754 boards, since it means that we may see a couple of nF3-250 overclocking dynamos in the near future. Dropping HT from 1000 to 800, dropping a multiplier, fixing the AGP lock, and cranking the bus means an easy route to the Athlon 64 overclocks that have seemed so elusive up to this point for the average overclocker.
The "any drive" SATA/IDE RAID is slick as we've seen and answers many enthusiast's worst nightmares. Finally, performance from an nVidia IDE solution is also up to the best available. However, this certainly does not mean that nForce3-250Gb is perfect, because it isn't. The downside is that CPU overhead is still higher than we would like. Perhaps with driver updates, we will see this area continue to improve. It would be a mistake, though, to place too much on the CPU overhead and overlook the wonderfully flexible setup, 8-drive RAID potential, and hot-spare mirroring that nVidia introduces with the full-blown versions of nForce3-250 family.
Performance testing showed nForce3-250 to perform about the same as the best Socket 754 boards that we have tested - no better and no worse with everything the same. This is pretty much what we have come to expect with the maturing Athlon 64 chipsets and the on-chip memory controller of the A64. There is, however, a little unexpected boost. While the best ATI graphics cards perform about the same in NF3-250 as they do in other Athlon 64 boards, the combination of nForce3-250 and nVidia graphics yields a nice performance boost. In this case, the sum is a little more than the parts, which is undoubtedly much easier to do when you manufacture both the chipset and the graphics card.
We are certainly impressed with nForce3-250Gb right now, but the real test comes with the introduction of the chipset with Socket 939. This chipset was clearly meant for that AMD socket, but the competition becomes stiffer with updated chipsets from both VIA and SiS for the new platform. If we were in the market for a top Athlon 64 today, we would search for a high-end nForce3-250Gb board - which you should be able to buy in a couple of weeks. The feature set and performance make it a great choice in today's market, but the main standout here is features because performance of all the chipsets is very similar. If you are an overclocker, then nForce3-250 may be your only choice for a working AGP lock - assuming production motherboards follow through. This is not to minimize the fact that there are a few VIA boards with additional multipliers for FSB and lower processor multipliers, which make some decent overclocking possible, but that solution is not nearly as flexible as a working AGP lock.
What about 939 and dual-channel? We suspect good things from the nF3-250 family, but until we see chips, boards, and the competitors' updates, the jury is still out.