I have a machine with a decent mobo, an i7 3770k and 8Gb ram @1600. Graphics are quite good with the integrated card.

I am about getting a dedicated graphics card but I don't know how to compare the IntelGraphics 4000 with the cards out there.

Just as example:

  • The Sapphire 4850 is clocked ~700Mhz with 1GbDDR3 and 800 stream processors whilst
  • The GT 630 is clocked at 810Mhz with 2GbDDR3 and 96 cuda cores

Using Intel's Extreme Tuning Utility I've found that IntelGraphics reaches 650Mhz while doing stuff.

Now there is the cores - stream processors issue. How is this compared to this chip?

Is there a way to compare my integrated graphics with other cards? Even in terms of performance, if not in terms of specs.


As a very rough comparison, have a look here: http://www.videocardbenchmark.net/video_lookup.php?gpu=Intel+HD+4000

According to that, a GTX 680 (for instance), is about 10x faster than your integrated graphics.

Your individual observed performance improvement will depend on many things. It's probably not 10x "better" for everyday use (since the benchmark is a bit synthetic) and you can probably do with a smaller card than the top-of-the-line one - but the chart should give you some idea on where the sweet spot is, comparing performance vs $.

I think it is safe to say that you will observe an improvement on graphics performance with a mid- or top-range graphics card, especially because the graphics performance is likely to be the bottleneck in your system.


Comparing an integrated gfx card (any brand/kind) is a little more complicated than simply looking at some benchmarks. An integrated gfx card uses the same RAM and sometimes can even offload some of the gfx processing onto the CPU. Anything that offloads onto the system has an immediate performance hit since you are having to take away resources from the system that could be used instead to process other information (like parsing a web site or calculating the bullets trajectory in your 3D FPS). A dedicated card will take over that extra processing (and in some cases even allow you to use it as a second CPU), immediately gaining performance for your system (even if marginal).

Comparing the different brands of cards is also a little more complicated (i.e. Nvidia cards vs. AMD/ATI cards vs. Intel cards) because of how they work at a hardware level (CUDE cores vs. stream processors vs. standard GPU). The more apparent hardware at your disposal is not always analogous to a better/faster machine, this is also due to how the hardware works, but also how the drivers of the individual cards work.

Here's a good 'basics' to video cards: http://www.enthusiastpc.net/articles/00001/

It should give some clarity as to why comparing them isn't as simple as some benchmark testing (though benchmarks can be a good visual indicator).

The real question to ask is what is the driving force of your decision to get a new dedicated card? Is it gaming, gfx editing, 3D rending, or just for some slight system performance increase?

Answering that will help to push in the right direction of which brand/model/etc is a better option for you.

It's a lot like choosing a car; a Porche Boxster 911 is a blazing sports car, but extreme overkill to pick up the groceries, just like a Honda Civic is an economical car but wouldn't stand a chance in an uphill race against the former.

Hope that helps.


The most important factor to answering your question is "What do you want to use it for?" Since you don't indicate whether you are simply browsing web pages, editing video, or playing the most demanding video games, it is nearly impossible to tell you what you will gain in terms of performance. That said, Tom's Hardware, in my humble opinion, has the best information available on what you can expect from your hardware and what will be the best value in multiple dollar ranges and workloads. Here is their graphics hierarchy chart, which includes dedicated and integrated cards. Tom's Hardware Graphics Hierarchy

Here is the list based upon the amount you want to spend.

Best Graphics Cards for the Money

Unlike previous posters, I will tell you that when you want to compare across the spectrum, benchmarks are what you want to look at in terms of performance, rather than the technical specifications. Tom's covers a broad range of tasks and you can see where one card may do well in one area, but be outperformed in another. Choose the one that benchmarks best in your targeted area. That means the most important piece of information for you to determine is the question that I asked first, "What do you want to use it for?"


It really depends on what you want to do. And there are so many variables to include that comparing raw numbers simply does not make sense: efficiency of a graphics card depends on many more factors than just only on clock rates and amount of on-card RAM. For example, chip design plays a major role, and it cannot be quantified (think what difference is between your i7 3770k and, let say, core 2 duo with the same clock rate, amount of memory and cache).

You say that the integrated card is pretty good. Does it then fulfill your needs? Why buying an additional piece of hardware if the current one might be enough for you?

If you want to play games—just look at the benchmarks of frame rate in these games. Long time ago Quake 1-3 were most often used for such benchmarks, recently I heard that games like Crysis are used. Just googling around might give you enough information.

If you want to use GPU for calculations, then that's totally different matter! Bitcoin mining works three times as good on AMD cards than on nVidia cards because of chip design. But nVidia triumphs AMD on more complex calculations. Also while nVidia shines when running CUDA code, it's not as good at OpenCL code, whereas AMD cannot run CUDA, and their OpenCL implementation is the best one available.

Also look at power consumption. Your current integrated graphics will be much cheaper in energy usage than a hefty GPU designed for heavy gaming or computing.

As a summary: each use case is different. Measure for yourself or look for benchmarks that test your specific use case. There is no other way.

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Not the answer you're looking for? Browse other questions tagged or ask your own question.