Monday, April 27, 2009

April 2009 GPU Tech

OK, this is where things get fun: WAR! Nvidia and ATI have been at it for a while, as they are the two Processor Juggernauts on the market after destroying all the competition along the way. Some years Nvidia is the king of CPUs, until ATI develops a technological breakthrough and steals the thrown. Then, from out of nowhere Nvidia unleashes a devastating left hook jab and reclaims the crown.

Lets begin.

The 2009 Tech Guide Series of Articles are designed to be objective and unbiased. I will cover both Nvidia and ATI technology, both of their strengths and weaknesses, and post up where they stand on the benchmark scales from a variety of websites. I will also briefly mention the up coming Intel Larrabee GPU technology, but since not much concrete specs has been officially released by Intel we can only speculate.

Lets begin with Nvidia's Geforce GTX/GTS 200 series:

Current DirectX 10 GPUS beginning with the best performance to the least

GTX 295

Currently the King of the GPUs. This is actually a dual GPU card, meaning it has 2 GTX 260 GPUS on one board, along with twice the memory (2x896mb). This was a response to ATI's HD 4870x2, which will be covered later, that was at one time the King of the Cards. This card isn't just 2 GTX 260's, but also has the shader technology from the GTX 285 to augment the performance.

GTX 285

Basically an upgrade to the GTX 280, offering a smaller die size (55nm versus 65nm), faster shader core clock, and a little bit more memory over the GTX 280.

GTX 280

Was the flagship GPU when Nvidia first unveiled their GTX 200 series, offering the best graphics performance at the time.

GTX 275

Recently released in response to ATI's release of the HD 4890 GPU, matching performance and the price point. Basically it is a GTX 260 (216sp and 55nm version) only with higher reference clock rates which offers slightly better performance.

GTX 260 216 stream processors, 55nm fab

Third revision of the GTX 260 in a struggle to match performance of the HD 4870. This revision reduces the die size to 55nm, reducing power consumption, heat output, and maximizes over clocking potential.

GTX 260 216 stream processors, 65nm fab

Second revision of the GTX 260, which added additional stream processors to increase performance.

GTX 260

Was the "performance" card when the GTX 200 series was released, second only the the GTX 280. Has since been revised twice to match ATI HD 4870 performance and price.

GTS 250

Recently released, but technically not a 200 series card. The Nvidia 200 series uses G200 GPU models, whereas the GTS 250 uses the older G92 used in the previous 9000 series Nvidia cards. Many of these cards are actually rebranded Geforce 9800 GTX and GTX+ cards. Considered a "value" graphics card.

Now lets turn to see what ATI is offering these days with their Radeon HD 4000 series:

Current DirectX 10 GPUS beginning with the best performance to the least

HD 4870x2

ATI's Flagship card, like the GTX 295, is two 4870 GPUs on one board, along with twice the memory (2x1gb). Was the King of the Cards till Nvidia released the GTX 295, which was not just a dual GPU 260 board, but also had the GTX 285 shader technology.

HD 4890

Recently released to cover the $200-$250 price point that ATI didn't have. Has a slightly different core GPU (RV790, whereas the HD 4870 is the RV770). This core offers a more efficient die that allows for power efficiency when the system isn't playing games, and much more overclocking ability. Many users claimed they have broken the 1gz core clock speed on this card.

HD 4870

ATI's original Flagship card when the HD 4000 series was released. This card offered surprised everybody, particularly Nvidia, with its performance. ATI was lacking far behind Nvidia with their HD 3000 series. The HD 4000 series was also offered for less than Nvidia, forcing Nvidia to lower their prices on their entire GTX 200 series. First GPU to feature GDDR5 (Graphics Double Date Rate -5) memory technology.

HD 4850x2

Similar to the 4870x2 only with two 4850 GPUs.

HD 4850

The stop down from the 4870, with less stream processors, GDDR3 memory versus the 4870's GDDR5 memory.

HD 4830

Slightly slower core and memory clock speeds than the 4830, considered the lowest end "enthusiast" ATI card.

HD 4670

The fastest of the 4600 series cards, with GDDR2 memory and a 128mb memory interface versus the 4800 series' 256mb memory interface. Also has slower core and memory clock speeds. Only has 320 stream processors when compared to the 4800 series' 800 stream processors.

HD 4650

Slightly slower version of the 4670

HD 4550

Fastest of the two "value" cards. Only features 80 stream processors and GDDR2 memory.

HD 4350

Most basic ATI HD 4000 series card with slower clock speeds than the 4550.

First, lets discuss a few technical aspects of modern day graphics cards a little more ineptly than in the Basic Tech article series.


There is a lot more to graphics memory than size alone, the same as with RAM memory. Graphics cards have their own memory associated with them and an interface that connects the GPU to the memory. Most ATI cards are available in 512mb, 1gb, and 2gb cards. Nvidia G200 cards get a little confusing and range from 512bm, 896mb, and 1gb (sometimes labeled as 1024mb, which is the same thing).

Graphics memory is also available in a variety of speeds. ATI has the fastest Graphics memory available on their cards, GDDR5. The ATI performance level cards and Nvidia's G200 series cards use GDDR3.

The memory interface is usually either 256-bit, 128-bit, or 64-bit, and is the bandwidth for transferring date to and from the graphics memory.

Stream Processors

Stream processors are related to the programming of parallel processing. Essentially, these allow the GPU to more efficiently calculate all of the math, geometry, etc, associated with producing those intense graphics and physics in modern video games. I personally have a very basic understanding of this, but do know that the more the merrier when it comes to performance. The number of SP is also associated with the major differences between the enthusiast, performance, and value level cards, so its worth mentioning.

There are some other important factors to look at, such as die size, but that has already been discussed in a previous article. So as long as you have been reading these in order you should be ok.

Now lets discuss a few architectural differences. It isn't fair to compare an Nvidia and ATI card based on their architecture, because both brands manufacture their cards to process information differently. For example, the 4870 has 800 stream processors, while the GTX 260 revised model has 216, but both cards perform similarly. It is only fair to judge an ATI card with an ATI card and the same with Nvidia if you are comparing archicture features. For example, in the ATI 4000 series the 4870 with 800 stream processors will perform better than the 4350 with 80 stream processors.


Now lets get into the nitty gritty performance benchmarks! But first a disclaimer: benchmarks of CPUs and GPUs are performed on test systems, and tests are done to attempt to measure a non spurious relationship between the product being tested (GPUs in this case) and performance levels. It is ALWAYs important to look over their test system, see what other components they used, and more importantly possible differences between a test system for the different product. Sometimes spurious relationships occur where, for example, an ATI card might out perform a similar Nvidia card, not because the card is better, but because it might perform more efficiently with the CPU or Memory used in that test system.

Most test systems use Intel CPUs because they usually support both ATI and Nvidia cards equally, while ATI cards might run more efficiently on an AMD platform than Nvidia. These are all things to keep in mind while review benchmarks, and why I never suggest just one website or one review to look at, but several. There are a lot of websites that review hardware, and each website might get different results than another. My advice is to look at no less than 5 reviews of a product, look at their testing system and their test methods, and then read their conclusion to get their opinion on the product. This will give you a pretty good idea about whatever product your reviewing or comparing.

Another thing to keep in mind is the operating system and drivers used. You can find these under the test systems tab in a review. New GPU drivers are released almost monthly, and can greatly affect the performance of how a game runs on their hardware, thus meaning the benchmarks can be subjective after a few months.

Don't get disappointed because you ordered a graphics card based on some benchmark you read hoping to achieve the same frame rate. Well you install the card and run the game and you come to find that you aren't achieving nearly the same performance! More likely than not its because you didn't look up the test system for that benchmark, and the other system components such as the memory or CPU are more powerful than yours. Lesson learned.

The easiest way to find a benchmark for a product is the wicked cool Google machine. Simply input "[product x] review" or "[product x] benchmarks" and Walla! You have a long list of reviews to read through.

Here are a few websites that offer really great reviews on everything from CPUs and GPUs to Water Cooling Systems:

Toms Hardware


X-bit Labs

Overclockers Club

Trusted Reviews

Tech Spot


These websites are just a few out of hundreds. These websites are many (particularly Toms Hardware and Anandtech) that I read daily to keep up with tech news.

No comments:

Post a Comment