3DMark2001 Performance, XP 1600+ CPU, onboard GF2 video
Benchmarking -
Memory Speed
One of the most interesting things about this new chipset is its
use of dual 64-bit memory controllers, allowing a higher memory
throughput when there are two sticks of RAM present. I experimented
with this board under Win98 for a while with SiSoft Sandra but
was getting pretty inconsistent results. 3DMark also kept crashing
under Win98 which I couldn't find a solution for, so eventually
I gave up and installed Win2K and decided to use 3DMark2001 as
the primary benchmark. Hopefully later chipset drivers for Win98
will sort out those kinds of problems.
So, the testbed as it stands is:
MSI K7N420 Pro motherboard (BIOS
v2.0 dated 31-Oct-2001)
nForce chipset driver version: 2.03 (dated 23-Nov-2001)
AMD AthlonXP 1600+ CPU (1.4GHz clock speed, 133x10.5)
IBM 60GXP 40GB ATA-100 HDD (primary master)
On-board GF2 video (Detonator 23.11)
DirectX 8.1
Win2K SP1
Anyway, I was first thinking that, given TwinBank is kinda like RAID for RAM, the two sticks of memory would have to be identical. SecretNet kindly sent over 2 identical 256MB sticks of NANYA-chipped Apacer PC2100 CAS2 for testing. I later discovered this not to be the case and experimented with some different sticks. So anyway, in the graph below from left to right we have a single Apacer stick, then identical Apacer sticks in slots 1 + 2, the same sticks in slots 1 + 3, those sticks in 1 + 2 with a Corsair XMS2400 stick in slot 3 and finally a stick of Apacer in slot1 with a stick of KingMax DDR266 in slot 2. All sticks were 256MB except the KingMax which was deliberately chosen as a different size to the Apacer for the last test. The BIOS was set to force the CAS latency to 2 for all tests and the on-board video adapter was used.
Interesting results! As I was trying to wrap my head around them, someone by the name of SW emailed me this link which notes exactly the same results I had - use slot 2 and things slow down. If you take slot 2 out of the equation, you can clearly see that performance with even non-identical sticks in slots 1 and 2 makes for higher performance than a single stick in slot 1. In case you're thinking it's just Win2K liking 512MB of RAM more than 256MB, note that with 3 sticks (768MB), performance dropped to only slightly more than the single-stick score and much less than the slot 1 + 2 scores. I'm not sure what the Slot 2 issue is, hopefully something fixed via a BIOS update or similar soon.
Onboard Video
The next element to isolate for performance measuring is the integrated
GF2MX core in the North Bridge or IGP. This is simple enough to
do - I ran a few benchmarks using the onboard video then switched
to an Asus GeForce2 Ultra card I have here. It would have been
nice to compare to a cutting-edge GF3 Ti500 or even a GF2MX400
to get a more exact idea of where in the lineup the performance
of the integrated GF2MX fits, but this will do for now. One cool
thing about the integrated core is that you can use the normal
DetonatorXP drivers (from a certain version on, I'm not sure what
version that is, but at the time of writing the 23.11's are the
latest and support it and that's what I'm using). The testbed
was identical to as used above, memory configuration was two Apacer
PC2100 C2 sticks in slots 1 and 3, of course.
So, the onboard chipset is being pretty squarely beaten by the GF2 Ultra, but given it's a GF2MX that's not too surprising. It's certainly faring a lot better than any other onboard video I've seen. It's hard to quantify the usability of a video card using benchmarks so in a completely unscientific bout of testing I sat down and played some games for a couple of hours. Colin McRae Rally 2 is very playable at 1024x768x32 with all the detail settings on High and draw distance set to maximum. In fact, I've previously found CMR2 very jerky under Win2k, unplayably so on my KT133-based board - that problem was not present on the nForce board at all. Half-Life was also very playable at 1024x768. If you're going to go for higher resolutions then ultimately you're going to need a more high-end card, this is also true if you need the special new features of the GF3 core or another card like ATI's Radeon line. However from my observations at LAN parties lately, almost everyone is using some variety of GF2MX because they're simply the best bang for the buck at the moment. They're cheap and they work well enough for most people - which is how I feel about the onboard GF2MX in this chipset.
Overall Chipset
For this test I built another
testbed using the AMD760-based Gigabte GA-7DXR motherboard that
fared so well in our AMD760 roundup. Trainspotters will note that
the southbridge on this board is actually the VIA 686B unit not
the AMD part - but, given we're primarily testing CPU, memory
and AGP performance here which are all northbridge functions,
I think we'll be ok. We saw in the roundup that the performance
between chipsets using the different southbridges was pretty much
identical anyway. I levelled the playing field by using the GF2U
card in each machine. This should give us an idea of the comparative
overall speed of the chipsets. Maybe I'll look at southbridge
stuff like IDE speed in a later review.
So anyway, second testbed:
Gigabyte GA-7DXR Motherboard (BIOS F7, dated Sep.19, 2001)
AMD760 + 686B Chipset (latest drivers from here)
Other details same as above.
Again, this is using an Athlon XP1600+ CPU and 2 sticks of Apacer PC2100 DDR. It looks like the nForce board keeps slightly ahead in the lower resolutions. When we get to 1280x1024 the nForce is actually starting to move away from the AMD760 board.
For another more general look at things, I rebuilt the testbeds using a Thunderbird-cored Athlon 950MHz. This is a 100MHz-FSB (DDR, so 200MHz) which means on the AMD760 board you can't run the memory at 133MHz DDR, only 100MHz. To keep it fair, I ran the memory at 100MHz on the nForce board too. In reality you would run the RAM at 133MHz DDR on the nForce board which would definitely help your scores. I ran a Quake3 timedemo - to make sure the different sound hardware on each board didn't affect the scores I set Quake3 to disable sound using the seta s_initsound "0" command and restarting.
This time again the GF2U scores are very similar. The AMD760 machine pulls slightly ahead but a FPS or two here and there is not a big deal. Also, bear in mind that in reality you would run the memory at 133MHz DDR on the nForce board because you can. On the older AMD760 board, lacking asychronous memory timing, you are stuck at 100MHz memory speed with this CPU. Interesting how the nForce's onboard video was fine in the low-res stuff but when the required fill-rate went up, its scores dropped bigtime. No doubt the slower system memory, which the on-board card uses for a framebuffer, is dragging it down there.
Anyway, it's pretty clear that the chipset is in the same ballpark, performance-wise, as AMD760 which is no slouch itself. Given how early the BIOS and drivers are for this motherboard and chipset, and that you can run the memory higher than AMD760 in some situations, I think in the real world you wouldn't notice a difference performance-wise between these boards.
Onboard Audio
I'm not going to go into this in any depth because frankly I don't
have the right equipment or enough understanding of the technology
to perform a proper analysis - I'm simply not an audiophile. Those
of you who do know a lot about audio hardware will decide whether
or not it excites you, based on the feature list. Certainly there
was some excitement when the audio features of this chipset were
announced. I can say that, using the normal Line-Out connector
through my fairly expensive Koss headphones, listening to MP3's
and playing games is indistinguishable from my main system which
uses an SBLive! card. For my kind of usage, this already scores
it above the implementation of the AC97 codec in i815E for example, which has a noticeably duller sound in my experience.
On the next page, I'll cover installation, general usage notes, overclocking the board and wrap up the review.
 DIY Parallel-Cable Remote LCD! |
 CPU Radiator Zen |
 TwinMOS PC2700 DDR |