Does anybody know of benchmarks for integrated vs discrete graphics in desktop systems?

[Bitterman]

First, a little background. As those of you who read this forum regularly doubtless know, I've been looking into contemporary computer hardware because I started playing Company of Heroes. Now, the rig I currently use isn't a gaming machine by any means. It has an older Core Duo processor, 1.5GB memory, and an Intel 950 GMA. By all rights it shouldn't be able to run anything more advanced than Nethack. Yet somehow it manages to run Company of Heroes at 20 FPS.

Now, conventional wisdom is that integrated graphics can't play games. But just how bad are they? Sure, my integrated graphics run Company of Heroes at low settings and only 20 FPS. But it runs and I can pretty much play it. That's got to be worth something, right? It got me thinking, just how good (or bad) are integrated graphics?

I looked online and searched a few hardware review sites, and read a few forums. The forums were almost completely unhelpful. Almost every post I read summed up as, "So bad you shouldn't even bother." Which is fair, I guess. I certainly believed it to be true. But now I have empirical evidence to show otherwise. Maybe we should bother.

The sites I've read are barely any more help than the forums. The few that do cover integrated graphics inevitably pit them against other integrated graphics solutions, or possibly against laptop versions of graphics cards. It's not really an apples-to-apples comparison for desktop performance. But again, the sites all seem to agree that integrated graphics simply cannot hope to run games, so they don't even bother testing the equipment.

Right now I see three possibilities that could adequately account for the data:

1) Company of Heroes plays unusually well on the Intel 950 GMA.

2) Current gaming and review culture is unreasonably biased towards high performance standards.

3) Conventional wisdom is wrong.

The third should be pretty easy to prove or disprove with current benchmark and review standards. Now, I'll grant that most games simply won't run on integrated graphics. The programs refuse to even try if they don't detect hardware shaders. But do the developers have it wrong, too? Should they allow the programs to at least start up and see what happens? It's an intriguing notion.

Have any of you seen any good benchmarks or comparison tests of integrated graphics on desktop systems?

[ADC]

I don't have a perfect link, but I did look at Macworld's comparison of two iMacs from the last generation, one with an Intel GMA 950 and the other an ATI Radeon X1600. They tested using Unreal Tournament 2004 and found that the Radeon cranked 65 fps, the Intel only 22. That's a significant difference. (And since the X1600 isn't exactly a smelly, Mt Dew swilling CS major's first choice, the real difference is even greater.)

As you noted, in order to get reasonable performance on the integrated hardware, you have to play in lower resolution with fewer effects, but it is reasonable performance. It's playable. It's just not ideal. And some of the others on the boards (who may bear some resemblance to my hypothetical CS major) will tell you that it could be the difference between fragging and getting fragged.

[stfram]

Have any of you seen any good benchmarks or comparison tests of integrated graphics on desktop systems?


None offhand, but you may want to check anandtech to see if they've looked into it.

That being said, I don't see why anyone would bother considering the solution would be to purchase a secondary video card that has the capabilities to run the games you want to play.

Even a regular PCI video card running a marginally decent ATI or Nvidia chipset, along with 128 or so megs on onboard ram would probably crush any Intel integrated video chipset.

Like this:
http://www.newegg.com/Product/Product.aspx?Item=N82E16814143032

$65 plus shipping and you've got yourself a marginally decent gaming computer.

[Bitterman]

None offhand, but you may want to check anandtech to see if they've looked into it.

I checked Anandtech, Tom's Hardware, Xbit, and a few other sites whose names I can't recall off the top of my head. Nobody had a good analysis, though a few had articles basically saying, "It'll suck, so we're not going to waste time with a real data."


That being said, I don't see why anyone would bother considering the solution would be to purchase a secondary video card that has the capabilities to run the games you want to play.

Even a regular PCI video card running a marginally decent ATI or Nvidia chipset, along with 128 or so megs on onboard ram would probably crush any Intel integrated video chipset.

Yes, probably. Likely. Makes sense. But where's the science in that? I mean, does it make sense that time would dilate when bodies approach the speed of light? Of course not! Or that black holes would eventually evaporate? Not really. Or that a dimpled ball would fly farther than a perfect sphere after getting smacked with a golf club?

This isn't about what we think is going to happen, or what we all know is going to happen. It's about what really happens. I want to know what the real data are. If conventional wisdom is spot-on correct and a Riva TNT2 is superior to the 950 GMA, then so be it. But from what I've seen so far, I don't think it is.

[frijoles]

Is this what you are looking for? Link (http://www.extremetech.com/article2/0,1697,1821808,00.asp)

[Bitterman]

Is this what you are looking for? Link (http://www.extremetech.com/article2/0,1697,1821808,00.asp)
I already read that. I was hoping for something with a more contemporary processor in the test bed, given that vertex processing is dependent on the CPU for 950 GMA systems. So, something like that only with say a 950 GMA Core 2 Duo 4320 vs an nVidia 7300 or 8500.

[covnam]

It probably isn't worth comparing on a desktop. If you want to play something more than basic games you really need a dedicated graphics card for good performance. Most reviews on how cards perform have the graphics turned up to very high resolutions (1920x1200) with all the bells and whistles. If they had to scale it down so that it was reviewable on integrated graphics, the stand alone cards would trounce the competition. Even the low end cards are likely to perform better. For example, two DX10 cards under $50: nVidia - http://www.newegg.com/Product/Product.aspx?Item=N82E16814127296 or if you prefer ATI - http://www.newegg.com/Product/Product.aspx?Item=N82E16814102100

Why review something when for the price of a single game you can get better performance?

[Bitterman]

It probably isn't worth comparing on a desktop. If you want to play something more than basic games you really need a dedicated graphics card for good performance. Most reviews on how cards perform have the graphics turned up to very high resolutions (1920x1200) with all the bells and whistles. If they had to scale it down so that it was reviewable on integrated graphics, the stand alone cards would trounce the competition...
It's worth benchmarking integrated graphics specifically because most reviews focus on 1920x1200, 4x AA, high settings. Some even go so far as to define 45 FPS as somewhat playable.

Show of hands, how many people even have a monitor capable of displaying 1920x1200? I'm guessing the number is fairly low. The highest resolution monitor sold in most stores is about 1280. A lot of monitors can't do anything over 1024. And is 45 FPS really somewhat playable? Back in the day, we got 25 FPS on Quake 2 and called it good. 30 was great! God, 45 FPS was so silky smooth you didn't know what to do with all the frames.

Current review culture basically tells us, "You need to run games at 1920x1200, 4x AA, highest settings, and get about 60 FPS to play this game." Call that a minimum of $400 for the monitor, $500 for the video card, processor to go with it for about $220. To build a do-it-yourself machine of that caliber would clock in around $1500. To buy it retail would probably double the figure. Call it $2000-2500.

We should have benchmarks and comparisons of integrated graphics because they might be useful for real world conditions. If integrated graphics that come bundled into a $100 motherboard can sufficiently play a game at 1024x768, then there's no need to buy the 8800. It's a $300 source of heat and wasted wattage.

Maybe the problem is that we go about the problem and solution backwards; most hardware review sites say, "Get this hardware, and you can run Game ABC at settings XYZ."

But that's not what people really want to know. They want to know, "If I want to run Game ABC at setting XYZ, what do I need?" Based on what I've seen, the answer for at least some people is going to be integrated graphics. Or at least something lower-end than an nVidia 8800 GTS, though I've yet to see a review site willing to admit that.

[ADC]

At the end of the day, the issue isn't what the review says but what you can tolerate. Arguing that review methodology should be directed at the low end, however, is unrealistic. It's about what looks and plays best, and what you need to get the best looks and best gameplay. By choosing to use integrated graphics, you're giving up the looks. Sorry, chum, but it's true. And considering that graphics cards which will give you better looks and performance can be had for the price of a game, I think you'll have a tough row to hoe trying to convince anybody to change that position.

That said, the stores you shop at haven't had a product refresh since about 1998 if their monitors top out at 1280. (I bought a 17-inch CRT that year that topped out at 1600x1200, which was glorious — $500, which was considered inexpensive at the time.) The Worst Buys and Circuit Shittys in my area all stock 1920x1200 and 1680x1050 LCDs. Don't know about the CRTs, but I guess people still buy those dinosaurs, so …

[Bitterman]

At the end of the day, the issue isn't what the review says but what you can tolerate. Arguing that review methodology should be directed at the low end, however, is unrealistic. It's about what looks and plays best, and what you need to get the best looks and best gameplay. By choosing to use integrated graphics, you're giving up the looks. Sorry, chum, but it's true. And considering that graphics cards which will give you better looks and performance can be had for the price of a game, I think you'll have a tough row to hoe trying to convince anybody to change that position.

I don't intend to argue that review methodology should be directed towards the low end; rather, I argue that it should include the low end. This isn't a zero-sum game. And nor am I arguing that integrated graphics run games as well as discrete solutions. What I said is if an integrated solution runs the game sufficiently (and I'm perfectly willing to allow that "sufficient" is in the mind of each individual user), then discrete solutions would be unnecessary. Obviously, if an integrated grapics solution isn't sufficient then a discrete solution is needed.

But, my main point still remains. Where are the data? Without the data, how can we even determine which side of the "if" we're on? I can sit here and say, "Well, I think it's possible that integrated graphics can get the job done," and whomever wants can say, "Well, I don't think it can," but with actual data we could maybe... I dunno... Settle the argument?

Alas, I suppose that is not what the Internet is for.

[stfram]

Show of hands, how many people even have a monitor capable of displaying 1920x1200?


The nearly three year old Inspiron XPS laptop that I'm using right now is running at that resolution. :)

Would I play games at that resolution? Hell no. But I did play FEAR: Extraction Point at 1024x768 with no real slowdown using the integrated ATI Radeon video.

[ADC]

Show of hands, how many people even have a monitor capable of displaying 1920x1200?

The nearly three year old Inspiron XPS laptop that I'm using right now is running at that resolution. :)
Forgot to answer this earlier; my MBP runs at that resolution. That, in fact, is why I bought the 17-inch model instead of the 15. Not so much for the gaming, really, considering the last Mac game I bought was Warcraft III, but I guess I should see how it does at 1920x1200.

[Sensuifu]

But, my main point still remains. Where are the data? Without the data, how can we even determine which side of the "if" we're on?


Desktop IGP solutions are mainly used in value-computers, primarily where the user is only a casual gamer. You won't see data you want because nobody wants to bother with the obvious. Sure the game might run, but at the bare minimum. Some games might run better than others, though I would consider the overall setup before assuming the IGP is working wonders. The type of CPU, the chipset, all the way to the RAM can affect the gameplay. Game developers make games that take advantage of the latest hardware, but they also consider the lowest common denominator of users. The gameplay you're seeing could just be borderline of the minimum requirements, but not necessarily the setup they'd recommend. To casual gamers, it's fine, but for a more active gamer, the recommended hardware might not even be adequate for the performance they're seeking. It's a matter of matching the game with the right hardware. The lack to benchmark hardware that was not intended to perform above the bare minimum is understandable. In the cases where there is a shootout, assume they'll perform below their discrete cousins.

To quote an article that basically sums up why IGPs are of little value to make direct comparisons with their discrete counterparts:

For comparison's sake, we ran the same tests on a midrange (under-$200) nVidia GeForce 7600 GT PCI Express graphics card. In most of our tests, it turned in silky smooth frame rates of 59fps or more, with only the 1,024x768 Quake 4 test turning in a score as low as 41.9fps. But even that's more than seven times the frame rate of the fastest motherboard-integrated graphics chip. (http://computershopper.com/feature/200704_the_right_gpu_for_you_4)

[Chacranajxy]

All I know is that the onboard graphics on my laptop are so shitty that they can't even run Counter Strike Source. And that runs on like... anything. Good thing I have a legitimate gaming PC for getting shit done.