Slashdot Mirror
Chipset Duel - VIA vs. Nvidia nForce
msolnik writes: "Tom's Hardware has put 13 motherboards to the ultimate test in their lab. The outcome? By and large, the VIA KT266A chipset knocks the stuffing out of the Nvidia nForce 420D. True bright spots were the candidates sent in by Soltek and Soyo."
I like the nForce still. I'm always scouting for cheap, fast linux-supported hardware. In a couple years these boards will be dirt cheap, compact slutions for nice linux boxes.
Moderation: Put your hand inside the puppet head!
It's quite possible, look at this pic.
Yes, the KT266A does show an approximate 5-10% lead on the nForce, however:
If you're looking for raw speed, over all other concerns, yes the KT266A is probably for you, and would go well with a Creative Audigy and GeForce 3 Ti500. However if you're looking for a less powerful system, with a still respectable specification, the nForce is likely to work out a lot cheaper.
Haven't got the time to read each motherboard in detail. In the end, KT266A looks to be a good choice if you already have a soundcard and graphics card, the nForce is a great first chipset and is great if you don't currently have a soundcard or graphics card - in fact the audio will be the best you can buy for under $100 at least.
And what was that about VIA taking the SiS735 memory controller? Eh? They are different companies, and SiS would certainly not give VIA their memory controller, that just doesn't make sense from a business or engineering point of view. The KT266A memory controller is taken from the P4X266 chipset.
I think that the nForce is being pitched improperly. What we need to know is - which board has the best TV out capabilities. With built in Dolby Digital, this thing could make a great set top box without making a huge investment. Hell, which motherboard has coax-in ports?
The PC-TV revolution is coming and the only one poised to take advantage is Microsoft. Are there any Linux or alternate OS based projects that have real funding behind them? What ever happened to Indrema? This would make a perfect hardware match for the platform.
Sigh...
Life is the leading cause of death in America.
The nForce is nVidia's first attempt at a chipset for AMD systems. How long has VIA been making chipsets?
I think this is a really good showing for nForce.
I also think that Tom is starting to lose focus when it comes to what people really want. With processors as cheap as they are, there's not much point to overclocking anymore. If a board doesn't make it easy to nuke your processor, that shouldn't be held against the manufacturer. Stability should be the priority, not how fast you can run the board out of spec.
VIA's stability / ease of setting up has been nothing short of awful in the past. The KT266A is suprisingly reliable, however. With VIA's 4in1 drivers 4.35 on Win 2000, or Windows XP (which includes VIA chipset drivers) users have found it to be pretty stable.
I agree with you they're comparing apples to oranges. When nForce was first announced, I questioned the decision to integrate an underpowered graphics chip, when most power users would want at least a GF3. The answer, as you say, is that this board isn't for power users. Sure our ears perked up when we heard nVidia was making a chipset, after all they revolutionized the 3D world. I'm sure future offerings from them will live up to their name. If you remember, it took them several tries with the Riva128, TNT, and TNT2 before they had a true performance winner in the Geforce256.
Did you actually fill all the banks? If so, did you encounter any problems? I'm shopping around and am looking for something that will be both decent performance and stable. I have heard all good things about the Shuttle board, but have also heard that the system can get flaky with all 4 banks filled due to memory timing issues.
XML is like violence. If it doesn't solve the problem, use more.
Not too sure about the 4 dimm slot issue, although, I do remember hearing the same as you about it. Something to do with the trace distance on the board being too great for 266Mhz DDR and having timing difficulties.
As to the raid, I recently purchased the Dragon+ reviewed by Tom and am overall quite pleased with it. It's onboard Raid is of the Promise FastTrak variety using a PDC20265R controller chip. There is a jumper on the board to select raid operation or to simply use it as additional IDE ports. When selected as Raid, upon bootup you get a menu allowing you to setup the array. It only supports raid0,1 but appears to work straight out of the box. I set it up and used my Win98 startup disk to install. When starting up FDISK, it sees the array as 1 drive of an (erroneously) large quantity. I created a partition and installed windows and all appeared fine. I think the win drivers are there to enable the UDMA features of the drives once you're inside windows. None of this mattered of course, since I needed everything to work in linux and the fasttrak module included with Mandrake 8.1 just isn't up to snuff yet. I was able to get so far as to start installing linux using the raid controller and suddenly the system rebooted itself. So, at the present time, I'm running in software raid0 mode still connected to the Promise IDE ports, since you can't boot off a CD-ROM connected to the Promise IDE ports. That's OK though. Still lets me keep all my IDE devices on separate ports and performance has been stellar.
No thanks. I don't smoke anymore.
As an owner, I take offense at THG's statements regarding Shuttle's AK31 v3.1 board. Clearly when there is a 5 FPS differential between boards running the same configuration, same chipset there is some problem with the test setup.
Shuttle's AK31 v3.1 has a known issue with detecting the proper memory timings, and I think Tom's hardware either forgot, or neglected to compensate for this. All you have to do is go into the BIOS and set the timings yourself. If you leave it on detect, it defaults to the slowest timings. They are using CL2 ram, and I wouldn't be surprised if the shuttle board was running at CL2.5. Yes, this is a problem with the board but it's a known problem and there is a workaround.
Tom's Hardware has cut plenty of corners in the past and this is just another example of their irresponsible reporting and benchmarking. They were one of the first hardware enthusiast sites but they've fallen off.
Phew, another article that focusses on overclocking potential and absolute performance. All well and good, but I'd like to see Tom's doing more comparisons on total component price and bang per buck and not try and match specifications without regard to the retail price. When I upgrade, I pick a budget first, then go shopping to see what I can get for that money. The price difference between a fully integrated nForce and a bare VIA + NIC + GeForce2 + sound means I can afford to put a significantly faster processor and a shedload more RAM in the nForce. There's a tradeoff in that it's harder to upgrade piecemeal, but I choose not to do that anyway as I find that it's cheaper and more rewarding to make infrequent larger upgrades, and easier to find a deserving home for the old hardware if it can form the substantial core of a box.
Informative article, but it's once again aimed at the geek who simply has to have the rootinist, tootinist fastest system west of the Pecos, with cost not an issue. Note to Tom's; for those of us who don't get free hardware, cost is always an issue. ;-)
If you were blocking sigs, you wouldn't have to read this.
Check out the November '01 archives at www.3dnow.net for many reviews of KT266A mainboards and some nForce stuff.
Some links from 3DNOW that I will highlight:
VIA KT-266A Motherboard 3-Way Shootout
VIA KT266A Initial Roundup - October 2001
I have filled all the banks -- I bought 4 DDR2100 256 meg dimms (CL 2.5) from Crucial and haven't had a single timing issue with them. I have just used the default SPD settings.
This is good since using the same memory, I was having issues with the MSI board.
The RAID controllers onboard are hardware RAID controllers, not software. Both the HighPoint and FastTrak RAID controllers pop up a BIOS screen before booting that let you configure the drives. They both support RAID 0 (striping), 1 (mirroring), and 0+1 (both i.e. 4 drives). They are presented to the operating system as a single SCSI drive attached to a SCSI controller. They are bootable.
This is not software RAID. This is not a 66MHz PCI Ultra 160 SCSI RAID controller with 128MB of battery-backed cache, but it is hardware RAID, and two striped 7200RPM ATA66/100 drives outperforms a single drive any day.
LOAD "SIG",8,1
LOADING...
READY.
RUN
Nota Bene, guys and girls - Tom didn't "do" the review. It was Frank Völkel and Bert Töpelt . Tomas Pabst is the site founder, and in reality the writing is done by staff. Credit where credit is due and all that.
http://futur.thednb.com
Wrong... the FastTrack series uses a BIOS implementation, true, but it offloads the striping calculations and such to the CPU. It's basically a BIOS trick that LOOKS like hardware RAID. For a true hardware RAID controller, you're looking at a few hundred even for IDE RAID, such as the Promise SuperTrack or, better yet, the 3ware Escalade series(which is supported in the 2.4 kernel series by default)
- Free tabletop fantasy gaming! Grey Lotus
Are you joking? Did you read the article? ALL 13 BOARDS were tested with identical GeForce3 cards. And furthermore, the tested NForce boards DO NOT have integrated graphics. If anything, the nForce board SHOULD have had the advantage, since it was employing the 128-bit memory bus, but it didn't.
You're correct on your other two points...however, you might take note of the fact that VIA chipsets sacrifice a tiny bit of stability for the sake of performance. With Intel and AMD chipsets, you lose a tiny bit of speed for the sake of stability. It's a give-and-take situation - you just have to decide which is most important to you.
Aces Hardware shows a different story. There, the nForce board from MSI seems to beat out the MSI K7T266 Pro 2 (KT266A chipset) in most of the benchmarks by a good margin. Not to say that I believe one over the other, but they do have widely different results.
All over the place, a board is marked as a clear winner, even if another board shows the *exact* same result. The archiving benchmark where MSI was top dog was interesting, what about the Epox which scored the exact same thing? Also, the other boards had less than 0.5% difference from the winner, I doubt that is statistically significant. How many runs were done? With the kernel compile, again, how many runs were done, in what order? The same config? I seriously doubt motherboards with the same chipset would have such a dramatic difference as 15 seconds on Kernel Compilation, given equivalent Disk I/O and CPU...
XML is like violence. If it doesn't solve the problem, use more.
It is an integrated chipset, and the performance is excellent considering that, but it is not as good as high-end non-integrated boards, and as a result of the extravagent design, it is very expensive.
The OEM market doesn't want it because it is more expensive than other integrated chipsets. They don't care that the performance is much better. If it is $5 more, they will ditch it in a minute.
So the other market to target is the performance market. Unfortunately, the GeForce2MX and regular DDR memory don't provide good enough performance for that segment, and the performance lags high-end boards sporting other chipsets even with an offboard video card.
By targetting two very different market segments with the same chipset, NVIDIA has put themselves in the position of having an excellent compromise product that is suitable for neither camp. I hope they get a P4 license, because the extra memory bandwidth might actually mean something there, whereas with the FSB limitations of Athlon processors it does nothing for them. That way, with an external video card, the NFORCE might actually roar, instead of being dead in the water as it is right now.
Even Slashdot wants to hide some things
(BTW, you should lose the <br> tags that break your post a third of the way across the window...they're hella annoying.)
20 January 2017: the End of an Error.
I'm bummed that the shuttle got such a lousy review. For $80 you can't knock that board. It's fast, and stable. I love mine, and so do many other people out there... Just take a look at the anandtech forums.
Um, this is my sig.
I'd like to see "we ran them in a burn-in oven at the maximum rated temperature for two weeks while running diagnostic programs that checked for correct functioning of everything". And "we put the machine on the shake table and ran the standard shake test profile". The reports should then show failure rates, like Consumer Reports. People notice when the product breaks. And then you'd find out which motherboard manufacturers actually build good PC boards.
It was a incomplete comparison
It really should have been a VIA KT266 vs VIA KT266A vs NVidia NForce vs AMD 760 vs SIS 735 vs SIS 745 vs ALI MAGiK 1 / M1647 (both revisions)
comparison.
I just picked up a Shuttle AK31 v3.1 board a week and a half ago along with an Athlon 1600+ XP (old motherboard just died, so it was upgrade time). It's good to see these kinds of review, but I've found that the performance difference is becoming truly negligable. The Shuttle board was one of the cheapest (I got it for $107) and one of the "lowest" performing in Tom's ratings, yet the difference between it and the best performing boards is generally about 2%. Similar KT266A boards at the recent computer show (where I bought mine) were $122-175 - much more than 2% higher cost. I think I found my price point.
What's even better is that it will overclock from 1.4GHz to 1.575GHz, effectively pushing it to an XP 1900+ with no stability issues. It's got 4 DDR slots, 6 PCI, and it's rock solid stable. I know all the true overclockers in the crowd will rant about the performance difference, but I think if a comparison was done, you'd find that an Athlon 1900+ XP chip on one of the slower boards (like the Shuttle) would probably both outperform and cost less than an Athlon 1800+ XP chip on the fastest board. Besides, at this point, the biggest bottleneck in most systems is still disk I/O.
It's the bus from the chipset to the CPU that's the problem; Nvidia's 128 bit memory interface is cool and all, but the CPU can't really use it. If it could, the system would simply scream. But it can't,
This results in a situation where the chipset can get more bandwidth from memory than it can deliver to the CPU. Nforce has a lot of potential that will only be realized as soon as the CPU can actually consume the bandwidth.
C//
What I can't believe is there were no memory benchmarks. How can you compare the nForce without doing memory benchmarks? The coolest thing about the chipset is the dual DDR channels.
Also, I would have liked to have seen was a GeForce 2 MX on a KT266A compared. I think anything that was using lots of memory bandwidth would have shown different numbers. Did you guys notice there was no 32-bit color depth for quake3? Oh well, little things aside it was still a good review though.
JOhn
Campaign for Liberty
Though I trust Tom's Hardware with their benchmarks and recommendations, I have to question the details when it comes to their writing.
Unless I'm missing something, what are these DIE devices and cables they talk about here and here? Could they possibly mean IDE cables? Though the article is in English, perhaps IDE is the same as DIE in another language.
They are also confusing Windows 2000 and Windows XP, and I really hope they aren't compiling Linux under Windows 2000, as someone mentioned earlier. All the benchmarks are run under Windows 2000, yet they state that Lame MP3 Encoder under Windows XP was used. When describing the benchmarks, they state that they chose not to use Sysmark 2001 because values can fluctuate under Windows XP. Am I supposed to assume they don't trust it under Windows 2000? They need to state that, not leave us to assume, if they meant to say XP at all.
I'm just picking nits, but I think if you are going to write a technical review, you must be accurate and specific.
Best is 100.5, NVidia is 100, worst is 99.5.
Considering that for pretty much every benchmark except 2 or 3 of them, the NVidia chipsets are in the middle of the back of VIA chipsets, and the difference is around half a percent, which is less than the variation amount the VIA chipsets, it is pretty clear that these benchmarks don't show that the VIA chipsets are faster.
Other differences in the motherboards are more significant than the chipset differences. Note that within the VIA motherboards, it tends to be the same ones near the top and the bottom of the results.
All of these boards appear to be performing so close as to not make ANY difference in real world use to the end user. They're all fast.
None of the boards really shine over another in performance (there are a few that lag in some of the tests). I think now you really have to look at the total package (the mobo, included accessories and software), and the prices to make a fair comparison anymore. Tom did say a few things about what the boards came with, but neglected the average retail prices of the boards.
Overclockability needs to become a side note after the conclusion, or part of a different review. Stability, integrated stuff (like IDE RAID, SCSI, ethernet, USB 1/2, sound, slots, etc), BIOS features and cost are pretty much all I (and a lot of people I know) care about in a hardware review.
Where the hell did you get this from?
Let's see: nForce solution:
- MSI nForce 420D - $129
- GeForce2 MX (at 8x AGP, using system DDR RAM)
- Dolby Digital 5.1 audio controller
- integrated 10/100 NIC
- 6 USB ports
- integrated ATA100 IDE controller
- Athlon XP 1600 - $114
- 512 MB PC2100 Micron DDR RAM (2 x 256 MB) - $72
Total: $315Includes:
Now, let's try the "superior" alternative (and I'm trying to pick out the lowest prices I can):
- Shuttle AK31 (KT266A) - $77
- integrated ATA100
- VIA AC97 sound
- 2+2 USB ports
- Athlon XP 1600 - $114
- 512 MB PC2100 Micron DDR RAM (2 x 256 MB) - $72
- Abit Siluro Nvidia Geforce2 Mx200 (but at 4x AGP w/ less memory bandwidth) - $43 --OR-- MSI G3TI200 PRO-TD GEFORCE3 TI200 $162
- cheap-ass Realtek NIC - $10
Total for GF2MX solution - $316 (for GF3 - $435)Includes:
(Prices from Pricewatch... I skipped the "house brand" GF2 MX cards when finding that bank-breaking $43 Abit to spare myself much embarassment when the Shuttle combo came in $10 cheaper). I also neglect shipping - where I'll let you do the math as to which is cheaper: 3 boxes vs. 5.
So for one dollar more, you would recommend to your customers that they get infinitely crappier sound, a far worse NIC, undoubtedly a crappier video card (by nature of the external AGP interface), less USB ports and a mobo with no good reputation for reliability?
Wow. Way to do your homework.
Or maybe you would instead recommend that they get the GeForce3 for an added $119. While they're at it, why not throw in an OEM Creative Audigy for an extra $50?
Don't even try that lame argument that it's not "high-performance" enough. Do you mean to tell me that Johnny Necktie will notice the difference between 200 fps and 210? I've got an Asus A7V, Athlon TB 750 and Diamond TNT2... I play all the games I want. I still type my Word documents with no impediments. Honestly, the only thing I find might need changing is the four-generation-old TNT2.