PC Hardware forum

General discussion

Curious question: why are integrated graphics slow?

I was perusing some hardware sites and the question popped into my head - why are integrated graphics so significantly slower than 'graphics-card' graphics? More specifically, I cannot see the advantage of having an interface card.

Consider GPU manufacturers having a standard socket that is build into motherboards. It has its own DDR channel independent from the CPU (like on interface card architecture) but it is routed into the board. A standard socket, motherboard mounted architecture offers some clear advantages:

Bus speed: 32x PCIe, rapidIO, HyperTransport, whatever the designers want, can be implemented. No interface card physical restrictions.

System upgrading: with DDR on the mainboard, users can upgrade graphics card memory by swapping dimms. They can upgrade the GPU by swapping out a new GPU so long as it retains the socket standard.

Cooling, power: The GPU can potentially share cooling with CPU, and this really makes water cooling far more attractive. delivering power is a design constraint on the mainboard now instead of a design constraint on the interface card connection.

Discussion is locked
You are posting a reply to: Curious question: why are integrated graphics slow?
The posting of advertisements, profanity, or personal attacks is prohibited. Please refer to our CNET Forums policies for details. All submitted content is subject to our Terms of Use.
Track this discussion and email me when there are updates

If you're asking for technical help, please be sure to include all your system info, including operating system, model number, and any other specifics related to the problem. Also please exercise your best judgment when posting in the forums--revealing personal information such as your e-mail address, telephone number, and address is not recommended.

You are reporting the following post: Curious question: why are integrated graphics slow?
This post has been flagged and will be reviewed by our staff. Thank you for helping us maintain CNET's great community.
Sorry, there was a problem flagging this post. Please try again now or at a later time.
If you believe this post is offensive or violates the CNET Forums' Usage policies, you can report it below (this will not automatically remove the post). Once reported, our moderators will be notified and the post will be reviewed.
Collapse -
Price. You get what you pay for.

In reply to: Curious question: why are integrated graphics slow?

It's basic.


Collapse -

In reply to: Curious question: why are integrated graphics slow?

An embedded video port uses the systems own ram to run its graphic needs which in turn reduces system ram and causes a slower system. That's the main difference to exclude any other h/w based issue. The benefit is you have a ready to go w/o adding the video card as yet another expense. The next benefit(if), should you decide to upgrade the video, then embedded video can be disabled when a new video card installed having its own ram and any better gpu provided. For most embedded video its usually basic and thus the cost of the mtrbd. remains low. The overall system cost remains low as a result if brought complete or via some OEM, like Dell, etc.

tada -----Willy Happy

Collapse -
Okay, but

In reply to: Curious question: why are integrated graphics slow?

Well, the price point was something I thought about but there are a lot of motherboards specifically marketed to the enthusiast (which is the main market for component-wise upgradable graphics system) that have very expensive options that nobody else would pay for. Like DIMMs for DDR2 and DDR3. Or 2 PCIEx16 expansion slots AND 2 PCIE2.0x16 expansion slots.

I was investigating and found this. It looks like AMD is actually thinking about doing something like this. It's definitely not for the enthusiast but the 780G chipset has a GPU on it that can be OC'd to equal the performance of a HD3450.


Integrated graphics are even more common among laptops and portables. So especially in the performance portable market, powerful integrated graphics seem like a very viable solution. MXM has failed repeatedly because the demands for cooling and form factor change with every generation. Building it into the mainboard gives the whole thing much lower profile, too.

Collapse -
consider this

In reply to: Okay, but

a modern ATX or BTX mainboard is something like 6 or 7 layers and has something like 8 primary ASICs and at least a dozen busses on it, not to mention it has to have power control circuitry for a few hundred watts of juice at fairly high current/low voltage coming in, to be corrected and adjusted as the devices need it, with correct phase and everything

a modern discrete graphics card is also 6 or 7 layers, and has its own proprietary power circuitry, etc

mating these together would create a mainboard at least 10-12 layers thick, which means insane production costs, and it would be madly expensive for the consumer, not to mention you'd give up A LOT of resources (since you'd have to give up space for more power circuitry)

then we enter the memory interfaces on modern discrete graphics, take the 8800 Ultra (current king pin), it has a 384-bit wide memory interface using GDDR3, modern DDR2 dual-channel solutions are only 128-bits wide, not to mention they're running MUCH SLOWER (max of around 800MHZ, vs somewhere around 2000MHZ) along with the timings not being ideal for the GPU, and DDR not being compatable with any modern GPU (GDDR2/GDDR3 are NOT DDR2/DDR3)

then add to this the cooling equation, modern GPUs can pull down more power than some CPUs, which would mean more heat on the board, and even more heat from your power circuitry, those are both problems

then we also realize that all modern GPU packages use contact balls, instead of pins or pads, and no two GPUs have the same ball-out in terms of contact orientation, or # of contacts, so there'd be that to contend with

then we also must realize that neither nVidia no AMD packages their GPUs in a way that would allow them to be really handled in this manner (they're just not built the same way as CPUs)

so ultimately, this is impossible to do on a basis of power/thermal/electromechanical limitations

in theory, yes, it would be an amazing advancement for desktop systems, although it is much more pragmatic to provide a common interface for discrete cards, that offers a large amount of bandwidth and decent electrical connections (such as AGP or PCI Express)

as far as using HyperTransport of Flex IO ("or whatever else") consider that a GPU needs two connections: 1 to its RAM and 1 to the system, the one to the system isn't really bandwidth intensive, its latency intensive, most modern backplane interfaces (like Flex IO) are worried more about bandwidth, and the connection to the RAM shouldn't be held up by any means (again, latency)

as far as mainboards with 4 PCIe x16 slots (it isn't to provide 2.0 support or non-2.0 support, its that one element of the chipset provides 2.0 compliant output, and the other element doesn't) is to provide for multiple PCIe expansion cards, such as graphics cards, RAID controllers, sound cards, etc

yes, x16 slots are generally used for graphics cards, but any PCIe device can be used in them, generally speaking

as far as the "performance portable market", there are powerful integrated solutions, however like most portable devices, they're highly proprietary and rather expensive relative to their performance (although nVidia has managed to cram 7900's in SLI into a laptop, as well as 8800's in SLI into a laptop (recently))

not to mention that the Molex standard for external PCI Express was supposed to provide for external gfx cards (Asus even built and demo'd a few models), although none of that ever came to fruition

Popular Forums

Computer Newbies 10,686 discussions
Computer Help 54,365 discussions
Laptops 21,181 discussions
Networking & Wireless 16,313 discussions
Phones 17,137 discussions
Security 31,287 discussions
TVs & Home Theaters 22,101 discussions
Windows 7 8,164 discussions
Windows 10 2,657 discussions


This one tip will help you sleep better tonight

A few seconds are all you need to get a better night's rest.