Does Chipset-to-GPU Matching Matter?

Experienced users might already know that Nvidia and ATI chipsets support competing-brand graphics cards, but is there a performance penalty to such mismatched combinations? We put the latest chipsets and graphics cards to the test to find out.

Indeed

Haven't got a chance to read the article other than the intro and conclusion, but haven't we known already for quite some time that there is no need or real gain to matching the chipset to the GPU?

I've never heard definitively that there was not a correlation between chipset and GPU. I appreciated this article that, in my mind at least, put to bed an urban legend with clear empirical data. Cheers to THG and Patrick for a job well done! As an aside, I've never known another hardware site with so many pissy readers! If any of us had to write an article, under a deadline, that would be scrutinized by thousands or millions of readers that could analyze every word at their own leisure, I'm sure we would all make mistakes too! Give THG a break, please.

GREAT Article. I am one of those matching noobs, just seems logical in my head. This article assured me that it doesn't matter.

Its ashame to see that the ATI chipset consistently performed quite a bit under the competition.

Shockingly good article by THG.

There seems to be a general performance issue for the ATI solution but given its relatively short existence it may very well be driver issues. Seems like they spent most of the driver development getting the IGP performance numbers up and need to spend a little work on increasing the numbers for card based solutions.

Anyway, great work THG

I am surprised to say the least... i think if ATI or NVIDIA gains enough of a market share advantage over the other... then it would only be smart business for the chipset manu. to optimize brand specific boards to these high performing cards.

Thus forcing their own market share to follow the coat tails of the winning GPU.

I think it makes too much sense for this to happen... i fear the day when someone with a suit and a fancy pen wakes up to such a realization and we are all forced to lay down our hard earned cash to play in their monopoly.

But as it stands right now, the consumer wins (relatively speaking).

although people have been telling potential buyers it doesn't matter for quite some time now i still think these articles are good as it allows solid evidence to present to anyone who may challenge such a view.

Wont have any problems with flaming at least this time, nice article. Regardless of how many times this topic is answered in the forums, people really like a "professional" opinion. Now all we have to do when this topic comes up again (which it will), we can just link this article.

Makes one wonder if he was instead trying to refer to the original version of the EVGA motherboard...

Firstly, thank-you Thomas Soderstrom and THG for producing an article with some meat in it. It is a welcome and refreshing change.

Next,

I have to agree with Multiplectic. As this is intended as a comparison between GPU manufacturers and chipsets makers, could we see an update with the gaming tests done at higher resolutions, such as those found in 19" and 21" LCD panels? This would fall in line with the more prevalent upgrade philosophy outlined in the conclusion to the article, and as Multiplectic said, this would stress the throughput of data from chipset to GPU more. Afterall, it is the communication between chipset and GPU that is really being tested here.

Finally, this subject needs ongoing monitoring as the manufacturers will read this article and will also likely react in the form of driver enhancements and eventually board revisions to address shortfalls and gain advantages.

Q

Perhaps Thomas was emphasising that ECS was working in both high-end and budget markets?

Just a thought, as thats how I read it.

Q

Agreed. They shouldve included 975X chipset as it is intended for high-end market; even though it's been out for a while. The price range for a 975X chipset falls in the same range with 680i; while the P965 chipset are generally cheaper which are in the same marketing segment as the 650i. And in general, the benchmarks around the web have consistently shown that 975X is a better performer than 965 at stock settings.

I'd want to point out that they should make the effort compare chipset that are competing in the "same" marketing segment and price range. Also drive higher resolution as we all know that performance advantage does not always shows up in low resolution settings. However, I do believe that Nvidia's claim of matching Memory (EPP Profiles), Mobo, and Graphics card is a marketing tool.

Despite what I just said, I'd recommend 975X or 965 over any nvidia chipset for anyone who don't give a crap about dual graphics setup (In other words, SLI). Better RAID and data performance, cooler chipset, lower power consumption, good OCers.

exactly why would modern games be better. a game is a game is it not. what exactly are the chipsets going to change. they will stay the same.

i fail to see this logic as well as the resolution ones?

what would changing the games and resolutions achieve?

Using a game that is known to stretch the limits of GPU and mainboard, especially at higher resolutions, will show a more stark difference between the platform/GPU matchings referred to in the article. Where differences are greater, more solid conclusions can be drawn with better confidence from the results.

This is a fairly basic premis for empiricism; test conditions that will give meaningful results.

StrangeStranger, for someone with such a high post count as yourself and obviously a great involvement in THG forum, it is a suprise to read such questions from you. I can only assume that this momentary lapse is due to you concentrating your full attention on something of greater import - work perhaps?

sorry but i do not agree, this is not a gfx test and as such should not stress the gfx card. the whole point of a system test should be to put as much stress on the system and as little stress on the individual components. yes in games they use fps but there is much more than games on test here.

higher reolutions or other games may have stressed the gfx card but to what purpose they would have served given how many other apps were tested i do not know.

you may disagree and you may be right but i do not understand your reasons. please enlighten me further.

My understanding is that the channel between chipset and GPU is what needs to be stressed as that is what this article is actually trying to test. However, one part of the conclusion was aimed at piecemeal upgraders who are looking for whether the chipset/GPR match will have an effect on their choice of upgrade. Those people need the answers that will come from just such further testing of higher resolutions on more modern games.

Q

Why wouldn't the card need to be stressed? The test was set up perfectly for this. If one component (the card) shows an obvious advantage, then the next thing to look at would be the board... which is what he does in several examples. Say the GTX owned the X1900XTX... if it was significantly higher on the nVidia chipset, then we might call shenanigans. However, if it was the same ownage on both the nVidia and the ATI, then we could claim that neither was cheating... thus being a usefull result for the experiment in question.

Individual parts can be stressed as much as they want because the same parts are used in all systems, therefore one being better than the other still produces valid results.

Whats wrong with THG?

Putting an 8800 GTX against an X1950XTX at 1024x768 and saying that Call of Duty 2 works better on the ATI card because it gets 194 instead of 162 FPS..

I mean, come on!

This is not a graphics test? Why the heck are they benchmarking games? And using some gaming benchmark to form a so called "conclusion". To my understand, the goal of this article/experiment is to test the hypothesis that matching GPU to their respective chipset will yield performance enhancements, such as nvidia claiming a "performance" boost when you insert a Geforce video card into a Nforce chipset motherboard, etc.

The Quake 4 benchmark, which uses an aging Doom engine, you barely see a big difference in performance. Now if I was looking at this benchmark, I'd say, why waste an extra 200-300 bucks on the 8800 GTX and the 680i chipset for a mere 10-20 fps, AND when it's well above 100fps with a 965 and a 1950XTX. It is clear that they HAVENT stress the bandwidth between the chipset and the video card, which is by the way what this article is mainly aiming to acclomplish. Dude, this is like holding a competition for math scholars while giving them 5th grade math.

AND plus they should be using an 8800GTS which is in the same price range as the x1950XTX.

*sigh*

Track, they used a low resolution to shift the bottleneck to the platform.
They're testing the chipsets, not the video cards.

They never said COD2 works better on either card, did they?
This is not an 8800 vs. X1950 review.

finally some voices of reason i.e you agree with me

Blah.

Anyway. I never said it is a comparison of 1950XTX to a GTX. In this experiment the graphics card is a test variable. So are you saying the benchmarks would not change if they decide to use a 6800GT video card? Or say a X1650?

A good experiment would try to eliminate as much "bias" as possible. And this article is NOT A TEST OF CHIPSETs. It is an experiment to determine if there are BENEFITS of MATCHING a chipset to a VIDEO CARD of the same manufacturer. The way the tests are ran and set up, it is impossible to see the benefit of matching video card with chipset, hence the title of the article. And of course Nvidia claims performance boost by matching Geforce with Nforce. And it is about the communication/bandwidth between the GPU and Northbridge that is what they are really after.


Let me ask you this, does this article faithfully support that claim or any other claims such as "matching video card to chipset yields best performance?"

I'm just trying to point out some testings procedures or setup that may or may not bring bias to the test.

And honestly, anybody buying a 680i + 8800GTX to run resolution at 1078x768 is wasting their money.

Blah.

The conclusion would not change. The numbers themselves would change with a 6800GT/x1650, of course, but I expect the conclusion would remain the same. If one chipset performed best at a task with the nVidia graphics card, it would also perform best with the ATI graphics card, and vice versa. The x1650 results would presumably be better in all cases, but that doesn't have any effect on the conclusion.

No No, would the benchmarks be screwed up if a X1650XTX was pitted against a 8800GTX? If that's the case, then the video card comparing a X1950XTX to a 8800GTX is screwed up as well. In your argument, you're saying that the choice of video card is a variable that does not have a significant impact.