With the launch of Athlon 64 processors AMD has stricken a blow at the chipsets market nearly stronger than if it would have started to distribute its own chipset. Though this launch didn't add competitors to the products from [ALi,] NVIDIA, SiS and VIA, but on the other hand it almost killed the challenge between the chipsets - in good old times speed was always the first comparison point. Now the memory operation speed depends solely on the processor, and chipset manufacturers have to recollect the long forgotten performance tests for on-board peripheral controllers, which are not so critical in the eyes of customers.

In this case almost the only trump card is obviously the functionality of a chipset, though we have always noted that the functionality will come with time. Indeed, the lack of on-board support for, say, SATA will not stop a mainboard manufacturer from installing a corresponding PCI-controller. As a result this model can set up SATA RAID outscoring competitor mainboards even based on more functional chipsets. We have omitted such factors as wiring complication, PCB upsizing, cost and performance of the resulted solution. Thus, as the old testing methods for the AMD64 platform eventually play out, in future we'll try to benchmark the systems by new parameters.

NVIDIA nForce3 250 and 250Gb

We shall base our evaluation of the new nForce3 generation on the comparison with old chipsets of this series and its competitors. nForce3 250/Gb functionality:

• Bidirectional HyperTransport 16/16bit bus (to a processor) with its clock operating at up to 1 GHz
• AGP 4x/8x for video accelerators
• Six PCI slots
• Two IDE channels for four UATA133 devices
• Two Serial ATA ports for two SATA150 devices
• Interface for another two Serial ATA ports (external PHY controller is required)
• RAID 0, 1, 0+1 for the devices connected to any of the on-board ATA controllers
• Network MAC controller: 10/100 Mbit/sec (Fast Ethernet) in nForce3 250 and 10/100/1000 Mbit/sec (Gigabit Ethernet) in nForce3 250Gb
• On-board firewall in nForce3 250Gb
• Eight USB 2.0 ports
• AC'97 compliant interface supporting 7.1 surround sound audio
• Low-speed periphery support, etc

Thus, two new nForce3 models of the 250 series feature a gigabit network [MAC-]controller (in the model explicitly called 250Gb). It also contains some firewall - to the order of the day. Now let's analyze their special features in more detail.

  

Frequency and capacity of the HT bus are maximized in comparison with competitors. We have already written that it makes no practical sense, at least for common desktops: even professional 3D-applications do not differentiate between 600 and 800 MHz (even 400 MHz is almost on the same level). We mean here the "down" channel (from CPU and memory to chipset), which is really loaded by AGP video cards, and the reverse traffic (remember that AGP is asymmetric) in the "up" channel is hardly noticeable. Nevertheless, some chipset manufacturers boosted the HT speed (for the lack of better solutions), so they have to repair the discrepancy of nForce3 150. Formally speaking, 1 GHz is a privilege of the top nForce3 Ultra model, being its only difference from 800MHz 250Gb. But it is positioned as a solution for "semi-professional" systems based on Athlon 64 FX (however we know that it's impossible to "tie" a chipset to a certain AMD64 CPU series). Meanwhile our nForce3 250Gb mainboard under review allows to easily eliminate this little inequality.

The area, where the 150 series from NVIDIA really loses to its competitors, is the disk storage support, so in this respect the new chipsets are considerably enhanced. nForce3 150 had three channels in the IDE controller (for six devices), but it lacked Serial ATA ports. Now, Parallel ATA support is limited to two channels, but four SATA ports are added (in maximum configuration). We wouldn't have mentioned the second pair of SATA ports requiring an additional PHY controller, because VIA also offers similar interface (SATAlite) in its south bridges, but we haven't seen mainboards with this feature implemented yet. However, we'll give you a real-world example a little below, so that you understand that it's all different with NVIDIA.

nForce3 and VIA VT8237 are on the level regarding the regular storage connection, but the RAID support puts the motherboard from NVIDIA ahead. It supports the same modes 0, 1 and 0+1 (and, of course, JBOD), but these arrays can be assembled from disks connected to any on-board controllers (e.g. 1 PATA and 3 SATA). Besides, backup disk(s) can be added to RAID 1 and 0+1, which will automatically replace a failed disk (when a failure is detected) and the information will be duplicated in the background mode, without user's interference. I also want to note the hot plug support for SATA disks.

The gigabit network controller interface in nForce3 250Gb is to a certain extent a break-through. Of course it's no problem to install a Gigabit Ethernet adapter on any mainboard, but it will be connected via the common PCI bus, which will cramp it even in the unidirectional mode with all other PCI devices inactive. To say nothing of the full duplex mode. Besides, the NVIDIA chipset also has an on-board MAC controller, that is mainboard manufacturers only have to provide a relatively cheap PHY controller. Such solutions (remember the Intel CSA bus) have started to appear only recently together with gigabit networks, but they will have a short life - PCI Express will rid us of the bottleneck problems in a natural way. Nevertheless, for now nForce3 250Gb is the only leader in this sphere, and nForce3 250 possesses a common to all rival chipsets functionality of an on-board Fast Ethernet controller.

Another marketing advantage of the "gigabit" NVIDIA novice is an embedded semi-hardware firewall. We tactfully call it semi-hardware, because we have absolutely no information about how some of the firewall functions are executed (or at least "accelerated") on the hardware level - the company confidently avoids details in this issue. Therefore we cannot possibly know whether it has hardware support or not. But a free and probably good (even if it is purely software) firewall in nForce3 250Gb is only a plus, though you can find a lot of similar freeware products of the acknowledged quality in Internet.

With its audio solution NVIDIA also attempted to take a step forward by enhancing the standard functions of AC'97 to 8-channel audio and 20bit output (the latter is provided in version 2.3, but has no practical sense). In combination with several codecs in the market (e.g. Realtek ALC850 - four stereo DACs instead of the usual three) you can really get the 7.1 audio, but you must understand that it is not High Definition Audio and that you will not be able to listen to DVD-Audio adequately (without downsampling). There is no nForce APU again, and the company is not planning to enhance functionality with a full value audio controller embedded into the chipset. The same concerns the FireWire support - the chip area is limited, its cost leaves much to be desired, it's difficult for mainboard manufacturers to sell models with exotic functions... Once revolutional NVIDIA chipsets are moving from a rebellious to the respectable niche of high-quality mainstream solutions.

Performance tests

Testbed configuration:

• CPU: AMD Athlon 64 3400+ (2,2 GHz), Socket 754
• Mainboards:
  • Gigabyte K8NNXP (BIOS v.F12) on NVIDIA nForce3 150
  • Gigabyte K8NSNXP (BIOS v.F2) on NVIDIA nForce3 250
  • MSI K8N Neo Platinum (BIOS v.1.1) on NVIDIA nForce3 250Gb
  • MSI K8T Neo-FIS2R (BIOS v.1.6) on VIA K8T800

• Memory: 2x256 MB PC4000(DDR500) DDR SDRAM DIMM OCZ
• Video card: Manli ATI Radeon 9800 Pro 256 MB
• HDD: Western Digital WD360 (SATA), 10000 rpm

Software:

• OS and drivers:
  • Windows XP Professional SP1
  • DirectX 9.0b
  • VIA Hyperion 4.51
  • VIA SATA Driver 2.10a
  • NVIDIA GART 3.77
  • NVIDIA SMBUS 4.04
  • NVIDIA IDE 4.12
  • ATI Catalyst 4.5

• Test applications:
  • 7-Zip 3.13
  • WinRAR 3.30
  • Canopus ProCoder (Demo v1.25)
  • discreet 3ds max 5.1 + Brazil Rendering System 1.0
  • Gray Matter Studios & Nerve Software Return to Castle Wolfenstein v1.1
  • Croteam/GodGames Serious Sam: The Second Encounter v1.07
  • Digital Extremes/Epic Games/Atari Unreal Tournament 2003 v2225

Brief comparison characteristics of all mainboards under review are presented in this table:

Test results

Thus, the memory controller is the same in all our contenders. What is the difference then? Mainboard capabilities to start with minimal memory timings. MSI models had no problems with that (2-3-3-5), and Gigabyte K8NNXP (NVIDIA nForce3 150) restricted itself to 2,5-3-3-5. Moreover, though Gigabyte K8NSNXP (NVIDIA nForce3 250) allowed minimal timings in its BIOS, in fact it operated (according to the CPU-Z data) at 2-4-3-5. And don't forget that one of the MSI mainboards represent the VIA chipset. Let's see the difference in real tests.

Embedded memory controller Athlon 64 is up to the mark, and even latency-critical archivers almost don't detect the difference between the contenders. Interestingly, the worst result (up to 3% lag) is demonstrated by VIA K8T800. Out of the two cases of extended latencies the archivers prefer an increased by half cycle CAS Latency (Gigabyte K8NNXP) of the increased by a cycle RAS to CAS Delay (Gigabyte K8NSNXP).

Video encoding speed (measured according to our open technique) demonstrates the same feebly marked tendency: MSI K8T Neo-FIS2R on VIA K8T800 is a tad slower than the rest (up to 4%), the mainboard on nForce3 150 is a little faster than the one on nForce3 250. Note, that we have given results of a single test from the video encoding set, because in the other tests the difference is still smaller (in most cases it does not exceed 1%).

All the more we couldn't expect an explicit advantage of one contender over the others at rendering. As a rule, several percents of difference demonstrate the difference in CPU clocks of contenders but not the strengths and weaknesses of memory controllers - in our case we have nothing like it.

In games this difference is confidently reduced to 2% or more, and the only difference from the above results is the leadership of the mainboard on the VIA chipset. I repeat, the reason is not in a faster HT bus to processor (read, to the memory), but rather in a specific character of chipset AGP operation (I really mean chipsets).

Conclusions

Performance tests proved once more that the chipset for 64bit AMD processors has almost no impact on the operation speed. At any case, on the memory operation speed - the most important characteristic for common customers. You should also take into account that despite the single chipset chip, nForce3 mainboards are not cheaper than their competitors. Conclusion: they have to win by functionality. Will the new NVIDIA chipsets manage that? In general, the answer is positive.

Ideologically correctly designed gigabit network adapter with a free additional firewall, the best support for disk storage devices and RAID, and the parity with competing chipsets in the other parameters - all this distinguishes nForce3 250 (especially 250Gb). It's quite another matter that the AMD64 platform is in a period of transition now, and so it's very difficult to recommend a Socket 754 mainboard. Considering the general revolutional changes in this field and the appearance of PCI Express, perhaps you should wait a little not to be whipping the cat later.

Sergei Pikalov (peek@ixbt.com) Dmitry Majorov (destrax@ixbt.com) 29 July 2004

Write a comment below. No registration needed!