Let's Talk Performance

This section is likely to generate a lot of flames if left unchecked. First, though, we want to make it abundantly clear that raw, theoretical performance numbers (which is what is listed here) rarely manage to match real world performance figures. There are numerous reasons for this discrepancy, for example the game or application in use may stress different parts of the architecture. A game that pushes a lot of polygons with low resolution textures is going to stress the geometry engine, while a game that uses high resolution textures with lower polygon counts is more likely to stress the memory bandwidth. Pixel and Vertex Shaders are even more difficult to judge, as both ATI and NVIDIA are relatively tight-lipped about the internal layout of their pipelines. These functions are the most like an actual CPU, but they're also highly proprietary and the companies feel a need to protect their technology (probably with good cause). So while we know that AMD Athlon 64 chips have a 12 stage Integer/ALU pipeline and 17 stage FPU/SSE pipeline, we really have no idea how many stages are in the pixel and vertex pipelines of ATI and NVIDIA cards. In fact, we really don't have much more than a simplistic functional overview.

So why even bother talking about performance without benchmarks? In part, by looking at the theoretical performance and comparing it to the real world performance (you'll have to find such real world figures in another article), we can get a better idea of what went wrong and what worked well. More importantly, though, most people referring to a GPU Guide are going to expect some sort of comparison and ranking of the parts. It is by no means definitive, and for some people choosing a graphics card is akin to joining a religion. So, take these numbers with a grain of salt and know that they are not intentionally meant to make one card look better than another. Where performance seriously fails to match expectations, it will be noted.

There are numerous factors that can affect performance, other than the application itself. Drivers are a major one, and it is not unheard of for the performance of a particular card to increase by as much as 50% over its lifetime due to driver enhancements. In light of such examples (i.e. both Radeon and GeForce cards in Quake 3 performance increased dramatically over time), it is somewhat difficult to say that theoretical performance numbers are really that much worse than changing real world numbers. With proper optimization, real world numbers can usually approach theoretical numbers, but this really only occurs for the most popular applications. Features also play a part, all other things being equal, so if two cards have the same theoretical performance but one card is DX9 based and the other is DX8 based, the DX9 card is should be faster.

Speaking of drivers, we would be remiss if we didn't at least mention OpenGL support. Brought into the consumer segment with GLQuake back in 1997, OpenGL is a different platform and requires different drivers. NVIDIA and ATI both have full OpenGL drivers, but all evidence indicates that NVIDIA's drivers are simply better at this point in time. Doom 3 is the latest example of this. However, OpenGL is also used in the professional world, and again NVIDIA tends to lead in performance, even with inferior hardware. Part of the problem is that very few games other than id Software titles and their licensees use OpenGL, so it often takes a back seat to DirectX. However, ATI has vowed to improve their OpenGL performance since the release of Doom 3, and hopefully they can close the gap between their DirectX and OpenGL drivers.

So, how is overall performance determined - in other words, how will the tables be sorted? The three main factors are fill rate, memory bandwidth, and processing power. Fill rate and bandwidth have been used for a long time, and they are well understood. Processing power, on the other hand, is somewhat more difficult to determine, especially with DX8 and later Pixel and Vertex Shaders. We will use the vertices/second rating as am estimate of processing power. For the charts, each section will be normalized relative to the theoretically fastest member of the group, and equal weight will be given to the fill rate, bandwidth, and vertex rate. That's not the best way of measuring performance, of course, but it's a start, and everything is theoretical at this point anyway. If you really want a suggestion on a specific card, the forums and past articles are a better place to search. Another option is to decide which games (or applications) you are most concerned about, and then go find an article that has benchmarks with that particular title.

To reiterate, this is more of a historical perspective on graphics chips and not a comparison of real world performance. And with that disclaimer, let's get on to the performance charts.

The Way It's Meant to be Played Number nine… Number nine…
POST A COMMENT

43 Comments

View All Comments

  • JarredWalton - Thursday, October 28, 2004 - link

    43 - It should be an option somewhere in the ATI Catalyst Control Center. I don't have an X800 of my own to verify this on, not to mention a lack of applications which use this feature. My comment was more tailored towards people that don't read hardware sites. Typical users really don't know much about their hardware or how to adjust advanced settings, so the default options are what they use. Reply
  • Thera - Tuesday, October 19, 2004 - link

    You say SM2.0b is disabled and consumers don't know how to turn it on. Can you tell us how to enable SM2.0b?

    Thank you.

    (cross posted from video forum)
    Reply
  • endrebjorsvik - Wednesday, September 15, 2004 - link

    WOW!! Very nice article!!

    does anyone have all these datas collected into an exel-file or something??
    Reply
  • JarredWalton - Sunday, September 12, 2004 - link

    Correction to my last post. KiB and MiB and such are meant to be used for size calculations, and then KB and MB can be used for bandwidth calculations. Now the first paragraph (and my gripe) should be a little more clear if you didn't understand it already. Basically, the *bandwidth* companies (hard drives, and to a lesser extent RAM companies advertising bandwidth) proposed that their incorrect calculations stand and that those who wanted to use the old computer calculations should change.

    There are problems, however. HDD and RAM both continue to use both calculations. RAM uses the simplified KB and MB for bandwidth, but the accepted KB and MB (KiB and MiB now) for size. HDD uses the simplified KB and MB for size, but then they use the other KB and MB for sustained transfer rates. So, the proposed change not only failed to address the problem, but the proposers basically continue in the same way as before.
    Reply
  • JarredWalton - Saturday, September 11, 2004 - link

    #38 - there are quite a few cards/chips that were only available in very limited quantities.

    39 - Actually, that is only partially true. KibiBytes and MibiBytes are a *proposed* change as far as I am aware, and they basically allow the HDD and RAM people to continue with their simplified calculations. I believe that KiB and MiB are meant for bandwidths, however, and not memory sizes. The problem is that MB and KB were in existence long before KiB and MiB were proposed. Early computers with 8 KB of RAM (over 40 years ago) had 8192 bytes of RAM, not 8000 bytes. When you buy a 512 MB DIMM, it is 512 * 1048576 bytes, not 512 * 1000000 bytes.

    If a new standard is to be adopted for abbreviations, it is my personal opinion that the parties who did not conform to the old standard are the ones that should change. Since I often look at the low level details of processors and GPUs and such, I do not want to have two different meanings of the same thing, which is what we currently have. Heck, there was even a class action lawsuit against hard drive manufacturers a while back about this "lie". That was the solution: the HDD people basically said, "We're right and in the future 2^10 = KiB, 2^20 = MiB, 2^30 = GiB, etc." Talk about not taking responsibility for your acttions....

    It *IS* a minor point for most people, and relative performance is still the same. Basically, this is one of my pet peeves. It would be like saying, "You know what, 5280 feet per mile is inconvenient Even though it has been this way for ages, let's just call it 5000 feet per mile." I have yet to see any hardware manufacturers actually use KiB or MiB as an abbreviation, and software that has been around for decades still thinks that a KB is 1024 bytes and a MB is 1048576.
    Reply
  • Bonta - Saturday, September 11, 2004 - link

    Jarred, you were wrong about the abbreviation MB.
    1 MB is 1 mega Byte is (1000*1000) Bytes is 1000000 Bytes is 1 million Bytes.
    1 MiB is (1024*1024) Bytes is 1048576 Bytes.

    So the vid card makers (and the hard drive makers) actually have it right, and can keep smiling. It is the people that think 1MB is 1048576 Bytes that have it wrong. I can't pronounce or spell 1 MiB correctly, but it is something like 1 mibiBytes.
    Reply
  • viggen - Friday, September 10, 2004 - link

    Nice article but what's up with the 9200 Pro running at 300mhz for core & memory? I dun remember ATI having such a card.
    Reply
  • JarredWalton - Wednesday, September 8, 2004 - link

    Oops... I forgot the link from Quon. Here it is:

    http://www.appliedmaterials.com/HTMAC/index.html

    It's somewhat basic, but at the same time, it covers several things my article left out.
    Reply
  • JarredWalton - Wednesday, September 8, 2004 - link

    I received a link from Matthew Quon containing a recent presentation on the whole chip fabrication process. It includes details that I omitted, but in general it supports my abbreviated description of the process.

    #34: Yes, there are errors that are bound to slip through. This is especially true on older parts. However, as you point out, several of the older chips were offered in various speed grades, which only makes it more difficult. Several of the as-yet unreleased parts may vary, but on the X700 and 6800LE, that's the best info we have right now. The vertex pipelines are *not* tied directly to the pixel quads, so disabling 1/4 or 1/2 of the pixel pipelines does not mean they *have* to disable 1/4 or 1/2 of the vertex pipelines. According to T8000, though, the 6800LE is a 4 vertex pipeline card.

    Last, you might want to take note of the fact that I have written precisely 3 articles for Anandtech. I live in Washington, while many of the other AT people are back east. So, don't count on everything being reviewed by every single AT editor - we're only human. :)

    (I'm working on some updates and corrections, which will hopefully be posted in the next 24 hours.)
    Reply
  • T8000 - Wednesday, September 8, 2004 - link

    I think it is very good to put the facts together in such a review.

    I did notice three things, however:

    1: I have a GF6800LE and it has 4 enabled vertex pipes instead of 5 and comes with a 300/700 gpu/mem clock.

    2: Since gpu clock speeds did not increase much, they had to add more features (like pipelines) to increase performance.

    3: Gpu defects are less of an issue then cpu defects, since a lot of large gpu's offered the luxory of disabling parts, so that most defective gpu's can still be sold. As far as I know, this feature has never made it into the cpu market.
    Reply

Log in

Don't have an account? Sign up now