Why more memory is not always better on a graphic card

Views 2 Likes Comments Comment
Like if this guide is helpful
I'm going to describe here how does it happen that many (mainly mainstream or low-end) graphics cards with ridiculously large memory are actually slower than their "standard" counterparts.

In short, it is because the memory is only one of the potential bottlenecks of a graphic card. If the performance is limited by a slow GPU (graphic card's processor or "core") then no amount of memory will help. Besides, in the vast majority of cases, those large-memory cards are actually built at lower cost and on worse components than an original design.

The graphic chips producers (ie nVidia & ATI) always get a lot of dies which have not passed the quality control, BUT are still capable of stable operation with reduced clock speeds. Obviously, the chip company will happily sell those processors at or even below the producing cost to the actual card manufacturer (ie. like Xpertvision, Gigabyte, etc.). The card manufacturer then buys some cheap memory in large quantities (again it might be something that has not passed the tests at higher clock frequencies), puts it all together and hey presto, there's your dream card: the same core, twice the memory, lower price... It works at lower clocks, but who cares - it has this HUGE memory, that surely can't be wrong! Or can it be?...

Let's consider an example; I think drawing a frame of animation is a bit like moving homes:

- The number and speed of people packing stuff up is reflected in your GPU's architecture and clock speed.

- The speed of the car you are using is the card's memory speed.

- The size of your car is the memory size.

- The amount of stuff you need to pack is the quality of displayed graphics - resolution, antialiasing, HDR etc.

Ok, let's assume that the original, standard card design is _fairly_ well balanced (which is normally the case). Say, you've got 4 people to carry and pack items, and a small but nippy van, something like Renault Kangoo. People pack things, the car runs there and before it's back there's a good amount of stuff laying on the floor, prepared to be loaded and the whole operation finishes before long.

Now, let's imitate our cunning card manufacturers - let's kick one "packer" out, but hire a large, slow lorry instead of our Kangoo. Obviously, the lorry not only will take longer to travel, but also it will spend most of its time waiting for the packers!

This example is a bit exaggerated - but not as much as you might think. Your mileage will vary, of course, and, undeniably, THERE ARE profits from having a large graphic memory. It's just the fact that in order to put that memory into use, you will have to set very high level of details and resolution - which will invariably lead to unacceptably slow framerates, unless your card's PROCESSOR is actually fast enough to cope with the task.

So, to sum up:

- take time to compare the performance of similarly priced cards. Google is your friend.

- make sure that the particular card you like works at the reference clocks for the given design. Many sellers ommit this bit in their listings or give you the meaningless RAMDAC frequency instead (which is 400MHz no matter what the card). A good list of cards along with their clocks and performance charts can be found on the Tom's Hardware Guide

- 512MB or more of graphic memory (as of March 2008) only makes sense on high-end cards like nVidia 8800 series.

- 256MB memory only makes sense on cards like nVidia 7600 series or faster.

- look at the overall picture: Can the card you like be overclocked? Does it have all the outputs/adapters etc that you will require? Is it silent? Does it have any special requirements as to the computer's power supply unit?

I hope the above text will help you choose the right card. If you are still in doubt, there's many forums on the internet - if you ask politely (and your question is not outright stupid), you are very likely to get more or less proffesional (well, at least independent...) advice.

Good Luck!
Have something to share, create your own guide... Write a guide
Explore more guides