Better late than never! We decided to bring to a conclusion (this very article) our tests of integrated chipsets from NVIDIA. Another reason for writing this article is the new cut-down models, implemented in a single chip, that appeared last autumn (such motherboards have just started to appear in stores).
NVIDIA GeForce 6100/6150 + nForce 410/430
The chipset scheme hardly needs to be explained here, motherboards based on this chipset are already well known to users. Northbridge is responsible for PCI-Express bus — there is one graphical PCI Express x16 slot and one (GeForce 6100) or two (GeForce 6150) PCIEx1 slots. Other differences are provided by the graphics core. GPU of the top model operates at 475 MHz, the cheapest model works at 425 MHz. GeForce 6150 also supports DVI output, hardware acceleration of HD Video decoding. GeForce 6100 lacks these features, it got a simple TV-Out only in August 2006 in its new revision. The intriguing name NVIDIA Quadro NVS 210S is given to a complete copy of the GeForce 6150, which frequency is reduced to the level of GeForce 6100, certified according to AMD Commercial Stable Image Platform requirements.
Southbridge is connected via the standard bus — 16 bit HT bus operating at up to 1 GHz (maximum frequency is used as a rule, but as this bandwidth is excessive, it can be limited in BIOS by a motherboard manufacturer; you can often reduce the frequency and capacity manually). nForce 430 Southbridge supports four SATA-II, which can form RAID 0, 1, 0+1 and 5. nForce 410 supports only two ports, hence only RAID 0 and 1. nForce 430 has a MAC Gigabit Ethernet with widely known NVIDIA Active Armour (a firewall that can process TCP/IP traffic on the chipset level), nForce 410 offers Fast Ethernet without additional functions. Both versions support HD Audio. They initially supported AC’97 as well, but nForce 430 Southbridges (provided since August)support only HD Audio. That's good, because manufacturers of cheap motherboards don't have a chance to save on the audio codec. PATA support has been initially standard (two channels for four devices). But following the common trend, the updated revision of nForce 430 has only one channel.
NVIDIA GeForce 6100 + nForce 400/405
This family was supplemented last Autumn with a single-chip version of GeForce 6100. Functions of nForce 410 were transferred to Northbridge with minimum reductions in functionality (one PATA channel instead of two). GPU frequency remains the same, TV-Out is removed, and there are fewer PCI Express lanes. As a result, nForce 405 supports PCI Express x8 for a video card, while nForce 400 does not support a graphics port at all. These hybrids look good both in terms of their cost and economy (according to our tests of the first motherboards, based on the top single-chip model, heat dissipation is even lower than in a separate Northbridge in GeForce 6100; it must be the effect of removing some PCI Express lanes). Perhaps NVIDIA should have given them more logical names — GeForce 6000/6050, for example, as these chipsets do not have southbridges and PCI Express was supported by Northbridge even in earlier revisions. But these are criticules. On the whole, we should admit that this solution has almost no direct competitors. All popular integrated solutions for AMD64 were based on two chips, which had a negative effect on PCB design (more complex layout) and did not allow to reduce prices.
We should say a few words about GPU characteristics. It supports DirectX 9.0, PureVideo technology (hardware acceleration of decoding H.264, VC-1 and MPEG-2 video, post processing of HD video in up to 1080p resolution), it has 2 pixel and 1 vertex pipelines, and operates at 425 MHz (GeForce 6100) and 475 MHz (GeForce 6150). The integrated core and an external card can work together to build a multi-monitor configuration (up to 4 monitors, if a motherboard is based on GeForce 6150 and has two video outs). Interestingly, the competing solution (Radeon Xpress 1150) demonstrates a different approach to building a graphics core, there are 4 pixel pipelines (support for vertex shaders is emulated by the driver, but it's up to a CPU), operating at 333 MHz. It should be added that the real frequency may differ from the nominal. It depends on motherboard designers. The difference may be quite significant, already demonstrated by our tests.
In conclusion of the theoretical part, we publish the table of characteristics of existing models from this family.
Performance analysis with an external video card can be performed only with archivers (or other applications sensitive to memory performance). Its only purpose is to make sure there are no problems in BIOS settings of motherboards taking part in our tests.
Which was to be proved. However, our motherboard reviews demonstrate
that even 5—7 sec deviations in this test are enough to
consider BIOS settings identical, because there are absolutely no
differences demonstrated in the other tests in this case. On this
basis, a performance drop demonstrated when we switch to integrated
video (due to an additional load on the memory controller in a processor
from a GPU) is also purely nominal. In practice (except for applications
that use 3D functions, of course), the effect of additional video
traffic is impossible to detect. Probably this very situation explains
why motherboard manufacturers are reluctant to install dedicated video
memory, when this option is provided by the chipset, like in Radeon
Xpress 1150. And the fact of a minimal performance drop can be logically
explained by characteristic features of Direct Connect architecture.
As is well known, modern AMD servers grant access to remote memory
via Hyper-Transport channel and memory controller built into the "end"
processor. It's only natural that if a memory controller (it's identical
in servers and desktops) is designed with regard to such a load, integrated
video requests are a piece of cake. It should be noted that a video
buffer may be useful in notebooks, where it may free a processor from
overheads (refreshing data in a frame buffer). It may come in handy
from the point of view of power saving and deeper CPU sleep modes
of an idle processor.
Another test sensitive to memory performance is XviD. In case of integrated video, it demonstrates insignificant differences between chipset results. But considering the previous test, we can still note that NVIDIA chipsets under review are a tad more economical in terms of memory exchange operations than Radeon Xpress 1150.
Let's proceed to much more interesting tests — games. Not to overcrowd diagrams with identical results, we are going to publish only integrated video results.
DOOM 3 is a serious load even in the minimal resolution, it's hardly
playable even for the top GeForce 6150.
But you can play FarCry with any integrated video version. Note
the difference in NVIDIA chipsets' performance between high and low
graphics quality modes.
The situation here is somewhere in between the two previous cases, it does not ruin the overall picture. So the general conclusion is as follows: significant (up to twofold) advantage of GeForce chipsets over Radeon Xpress 1150 in low resolution and graphics settings set to minimum (modern games actually look ugly in this case) disappears and becomes a disadvantage in medium and high resolutions, especially in case of GeForce 6100. The lag of GeForce 6100 with nForce 410 Southbridge from the single-chip solution (nForce 405), noticeable in all the three diagrams, does not necessarily mean that the problem is in chipsets. Tests of MSI K9NGM, which represents the "old" GeForce 6100 in this review, demonstrated that it was insignificantly outperformed by the Foxconn model on the same chipset (unfortunately, it was unavailable at the time we carried out "chipset" tests), while ECS GeForce 6100SM-M is currently the only available motherboard on GeForce 6100 + nForce 405. As the deviations are not too big, we can logically assume that motherboards based on both versions are generally similar, judging from gaming performance. If you want the fastest motherboard, you should analyze test results of motherboards themselves (according to our tests, the real GPU frequency differs in motherboards from different manufacturers).
From the practical point of view, NVIDIA chipsets look better for active network games, when reducing a resolution and graphics settings is the only way to get maximum response times. But ATI provides better support for "quiet" games — RTS, quests, etc. As Radeon Xpress 1150 is currently positioned even lower than GeForce 6100 (motherboards with this chipset are cheaper), we can note that all these chipsets demonstrate proper results, there are no outsiders.
Now let's proceed to even a less explored aspect of integrated video — support for decoding compressed video in MPEG2 and WMV formats. Support for "plain" MPEG2 playback appeared long ago, modern chipsets must work with the HD variety as well. But that's theoretically. In practice, the first question, especially after we analyzed test results of discrete video cards (top models), will be as follows — "will integrated video be powerful enough, and will it help modern CPUs?" The intuitive answer is "it won't make it worse", even insignificant help from a chipset will unload a processor for other tasks. But let's not jump at conclusions. In order to estimate CPU time dedicated to the video decoding process, we used standard information from Windows Task Manager — View/Select Columns/CPU Time. It's very easy to calculate the average CPU load in traditional percents (in our diagrams), if you know duration of each video file.
The MPEG2 stream (about 0.7 MB/s in this case) is evidently
no problem to decode even if the entire task is up to a processor.
Chipset support makes it even easier — CyberLink decoder results
illustrate this situation (we should note that all modern decoders
use hardware acceleration functions, not only "proprietary" functions
from NVIDIA and ATI). But what do we see in case of the proprietary
decoder from NVIDIA? Even if 10% load is not a critical value and
has little effect on performance of background tasks, we are testing
the easiest compression format here. What shall we see in more complex
cases? We have no rational explanation, even if we assume that the
NVIDIA decoder relies on chipset functions more than the others (and
the chipset is too weak to cope with this task), then why maximum
load on the processor is in case with the most powerful chipset (GeForce
6150)? As the real difference is not big in the latter case, we suppose
that it's the effect of background processes and measurement errors
(but they cannot explain the lag from CyberLink decoder). Unfortunately,
NVIDIA PureVideo continues to surprise us. The author of this article
has accumulated lots of gripes with it, because he had to retest it
several times with several motherboards, Windows and program versions,
drivers, and various versions of this decoder itself in order to make
sure that the test procedure is correct and the problem is in the
The data stream grew to 2.4 MB/s. While Cyberlink coped well
with it, the NVIDIA decoder was probably also engaged into some internal
computations (a working hypothesis — it calculated the pi value).
Note that in one case CPU load exceeded 50%. It means that if our
processor hadn't been a dual-core model, it wouldn't have coped with
the task (our tests with a weaker Sempron prove it). GeForce 6150
finally demonstrated a noticeable advantage over GeForce 6100. A leader
of the test (in case of the CyberLink decoder, of course) is Radeon
Xpress 1150. Results are even more surprising, considering that the
layout was practically inverted in case of discrete
video cards. In order to smooth over this categorical conclusion
(dictated by the test results!), we can note that the decoder from
NVIDIA might use more thorough and resource-intensive post processing.
In case of sufficient resources, it provides sharper image. But firstly,
the visual difference in quality cannot be seen (analysis with a magnifier
is good, when other parameters are similar; but here there is an absolutely
inadequate difference in resource requirements). Secondly, even if
we assume that all CPU resources are indeed used to increase video
quality, NVIDIA just has to know capacity of its own hardware solutions
and probably to let the decoder scale the load depending on a PC configuration.
Decoding Windows Media Video (using the WMV9 decoder built into Windows Media Player) showed no surprises. These chipsets are similar, because 3—5% deviations in this case are within the measurement error. Drawing a bottom line under video decoding tests, we can note that Cyberlink demonstrates the most stable results (it makes no sense to buy a proprietary video encoder from NVIDIA for a motherboard with integrated video, this is proved by the tests), GeForce 6150 has an advantage over GeForce 6100 only in a single test. Note the unexpected leading positions of Radeon Xpress 1150. It's a better choice from the point of view of video decoding.
By definition, integrated video solutions are more conservative than discrete ones. If the GPU level is initially rather high, they can exist on the market for several years. However, the chipset should also support a state-of-the-art set of peripheral interfaces. NVIDIA GeForce 6100/6150 chipsets are very good in both criteria. With the launch of single-chip versions, nothing stops this family from expanding to the segment of the cheapest motherboards, which requirements to functionality and performance are minimal and where VIA chipsets rule. However, our tests demonstrated that Radeon Xpress 1150, which career under another name started a year prior to its official launch, does not look weaker. It performs even better than we expected. Combined with the modern Southbridge (SB600), the only integrated chipset from AMD (ATI) is a strong competitor to GeForce 6100. And finally our conclusion from video decoding results — we are obviously not through with our tests. We want to carry out more tests with an expanded set of decoders and videos in various formats. The next article will most likely follow the launch of AMD 690 and the appearance of such motherboards in our testlab.
Write a comment below. No registration needed!