Sean's Technological Journeys

Current vs. High-end. In short, this is a distinction between technology that is chronologically new to market – and thus supports the latest software facilities that programs rely on – and technology that is very capable compared to other hardware available at the time of its release.

So the degree to which hardware is “current” is only based on time. And the degree to which hardware is “high-end” is only based on how relatively powerful that hardware was when it first shipped.

These two axes are independent of one another because their basis is completely different. You can have old, high-end hardware, or you can have new, low-end hardware, and so on. I am going to apply these concepts to the arduous task of shopping for a new video card. I’m not doing this for myself, but as a guide to others who might be in the market for a new video card.

When shopping for a new video card, you have to answer one question along each of the axes. And you absolutely must go into this with a budget in mind, no matter how large or small it may be. Otherwise you can’t weigh reality against your needs.

First, ask how new are the games I want to play? When were they published? If you don’t game, but use other 3d applications, then the same question applies to those apps too. Almost without exception, new games will require many of the newest features of video cards. That means, to play new games, you need a current video card.

In general, you will need a video card that – at the time of the software’s release – was either the first, second, or third most-recent generation of video card from the vendor. The third generation cards are sometimes not supported, either, so choose wisely. For the sake of future-proofing your investment, I would never recommend that anyone buy a new video card older than the current second-generation, unless the price is extremely marked-down.

The second question to ask is, how high-end are the graphics being used in the game or app I want to use? Games these days are released with a broad spectrum of hardware expectations in terms of sheer texture fill rate. Texture fill rate is a general measure of how much complex scenery a video card can handle. Due to the natural evolution of technology, there is significant gain in fill rate on more current hardware, even the low-end stuff. But if the software you want to run is expecting the kind of fill rate you’d find on a $600 enthusiast video card, you can’t expect to buy a low-end video card that will play it with acceptable speed.

Similarly, there are some games that require only a small fill rate, but they require current hardware to support the game’s request for modern graphics APIs such as DirectX 10. Now for the tricky part. You have a vague idea of whether you need a low, mid or high-end card, and you know how current it needs to be. But now you have to size up your budget against the cost of a card to determine whether you can satisfy your needs under your current budget. If you can’t, you may find yourself going bargain hunting. But the financial optimization of purchases is a much more complex topic than I can cover here.

Now I will break into an overview of the top four graphics card manufacturers’ strengths and weaknesses, along with a “snapshot” of what their high-end, mid-grade and low-end video cards are today. This information should help you evaluate which vendor can best suit your requirements, as gathered above if you followed my post so far. The contenders are Nvidia, ATi (now a division of AMD), Intel, and S3. Other vendors out there might claim to manufacture 3d graphics cards, but I deliberately exclude them for their lack of competitiveness in the market.

But first, some terminology for the uninitiated:
Integrated: Often contrasted with discrete. Integrated means that the video “card” (i.e. the Graphics Processing Unit, GPU) is hard-wired to the motherboard, and can not be removed from it, nor upgraded. These are popular on laptops to save space and energy.
IGP: Integrated Graphics Processor. This is just an abbreviation for an integrated GPU.
Discrete: As opposed to integrated. A discrete graphics card is one that plugs into a slot on the motherboard. If you have discrete graphics, you can upgrade your video card later to a more powerful one (or run multiple video cards at once if your motherboard supports it). When upgrading, pay careful attention to the slots available on your motherboard, versus what slot is required for the card.
Texture memory: Very fast memory that is used to store graphics during 3d rendering. On some setups (IGPs), texture memory can be a piece of your main system memory that has been set aside for this purpose. All discrete cards have texture memory on the video card itself.
GDDR#: Graphics Double Data Rate RAM. The number after GDDR roughly translates to how quickly the GDDR can send and receive data. Higher numbers are much faster than lower numbers. GDDR is the modern terminology used to describe how fast and how much texture memory you have.
OpenGL, DirectX: Two different 3d graphics programming interfaces that software uses to interface with the GPU. These are always evolving to new versions that support more complex graphical operations; however, older hardware may be unable to support these newer revisions. That’s why maintaining a current GPU is important if you want to play 3d games.
Future-Proofing Guarantee (FPG): A term of my own invention that represents how long I think the specific video card in question will continue to work with newly released software. I always give a range for the FPG; the low number is the amount of time you can continue to use this card with the very latest high-end games; the high number is the amount of time this card will continue to work with new casual games.

Nvidia: The Evil Empire. Extreme performance, at all costs (even your paycheck).

• Market leader for mid-grade and high-end gaming. Trend-setter and forerunner.
• Both integrated and discrete GPUs using all the latest PCI-Express 2.0 technology.
• Gold-standard Windows drivers with comprehensive compatibility.
• Gold-standard Linux drivers with comprehensive OpenGL support, although their drivers are proprietary.
• They release each “generation” of video card as low-end, mid-grade, high-end, and enthusiast models for discrete GPUs. Less variety in the IGP product line, which targets casual gamers, composited desktops, and multimedia.
• All discrete chips have dedicated texture memory; currently ATi’s high-end cards have a slight edge in terms of memory bandwidth and capacity. Integrated chips may have dedicated memory or share system RAM, or both.
• All chips scale appreciably well to complex games; the low-end cards can run high-end games with detail settings turned down a bit.
• Decent value at the low end; rapidly decreasing value at the high-end. Their best cards cost way too much – a good $100 to $200 above the most expensive ATi.
• Power schmower! Ask your congressman to build you a new power plant; you’ll need it to run a discrete Nvidia GPU! Their integrated GPUs are among the most inefficient on the market, too, but recently they’ve started to compete in that regard.
• Nvidia makes motherboards, too, and it’s always a good idea to run an Nvidia GPU with either an Nvidia or Intel motherboard.
• No sane 3d application developer – games, high end computing, whatever – would dare release a product without extensive testing on the three most recent Nvidia series of GPUs. They’re the gold standard, after all.

For high-end gamers and enthusiasts, the GeForce 200 series is recommended. Unfortunately, Nvidia has yet to develop this generation into lower-end hardware. FPG is 2 - 4 years.

For gamers playing older 3d games or casual games, the GeForce 9 series is recommended. There is a lot of variety in this series of cards, so go with your price point. FPG is 1.5 - 3 years for the value cards; 2 - 4 years for the 9800 GX2.

If you don’t need all the latest features at all the latest expense, there are some affordable GeForce 8 series cards providing great gaming speeds. FPG is 0.5 - 1.5 years.

If you just want to run a game from 2004-2005 or earlier, and have no plans to play newer games, go with the GeForce 7 series. FPG is already expired; i.e., only future-proof for the next 6 months at most.

ATI/AMD: The New Republic. Good value all around, with competitive performance.

• In terms of market share, they’re a close second place for mid-grade and high-end gaming. Occasionally beats Nvidia to market with bragging-rights features (e.g., GDDR5, 1 TFlop.)
• Both integrated and discrete GPUs using all the latest PCI-Express 2.0 technology.
• Near-perfect Windows drivers with comprehensive compatibility for the latest DirectX.
• Quite good Linux drivers with comprehensive OpenGL support, although their drivers are proprietary. They have difficulty updating their drivers fast enough to keep up with the evolution of the Linux kernel and X server.
• They release each “generation” of video card as low-end, mid-grade, high-end, and enthusiast models for discrete GPUs; less variety in the IGP product line, which targets casual gamers, composited desktops, and multimedia.
• All discrete chips have dedicated texture memory; currently they have the best dedicated texture memory solutions available. Integrated chips may have dedicated memory or share system RAM, or both.
• All chips scale appreciably well to complex games; the low-end cards can run high-end games with detail settings turned down a bit.
• Very good value at the low end; decreasing value at the high-end. Their best cards are expensive, but cheaper than Nvidia.
• Power usage isn’t a big focus for ATi, because they are trying to keep up with Nvidia performance-wise, and power efficiencies often equate to performance drops.
• AMD makes motherboards and CPUs, too, and it’s always a good idea to run an ATi GPU with either an AMD or VIA motherboard.
• No sane 3d application developer – games, high end computing, whatever – would dare release software without extensive testing on recent ATi products, for risk of alienating many customers. The ATi user base is big enough that support for these GPUs is almost never neglected.

Cards with an “HD” in the model number came around the 2007-2008 timeframe. HD4x cards are pretty recent, and largely composed of enthusiast boards, with an FPG of 2 - 4 years. Luckily, the price of a current ATi video card can be very affordable compared to the price of a similar Nvidia card. You should be able to find an HD model GPU that is within your price range. This is one advantage ATi has over Nvidia: customers benefit from the innovation of their latest builds because ATi is more agile at paring down their enthusiast cards to the low-end range. Of course, FPG is negatively impacted by a low-end card, even if it’s current.

Intel: Just starting to realize that they’re in the graphics business.

• A modest market share overall; massive success at business-class integrated graphics.
• Current chips are only integrated (with discrete graphics planned for next year).
• Well-supported Windows drivers and good DirectX 9 compatibility, with some chips now supporting DirectX 10.
• Enthusiasts, Red Hat employees, and Intel employees make a coordinated effort at rapidly evolving the Linux graphics stack by developing open source drivers for Intel cards, while evolving the DRI framework at large. Their work benefits S3 and ATi customers too.
• No real sense of low-end or high-end within the Intel product line. Newer is better; only desktop and notebook variants of their GPUs exist. However, on the market at large, Intel chipsets appear at the entry level (low-end).
• These IGPs use system memory as texture memory, reducing available RAM for programs. However, the amount of RAM used can afford lots of texture memory to the IGP.
• GPU design does not scale well, as texture fill rates are modest.
• Intel IGPs sport outstanding value because the IGP is included with a motherboard or laptop purchase.
• Power usage of Intel IGPs is much lower than a comparable Nvidia or ATi product.
• Intel motherboards integrate Intel GPUs, opening the possibility for great optimizations.
• These cards are on the radar for most game developers, but moreso for casual games than high-end games. ATi and Nvidia are the market leaders, so no game developer would dare release a title incompatible with one of those; but they might also work in Intel support as an afterthought.

Go with the X3100 or, preferably, the X4500. Prior to that the cards weren’t really a contender at all for 3d gaming. FPG of 0 - 2 years.

S3 Graphics: The Unknown, the Unusual, the Underdog.

• Very small market share overall; niche market success at the ultra low-end.
• Current chips are either integrated, or discrete cards that need a modern PCI-Express x16 slot.
• Well-supported Windows drivers and outstanding DirectX 10 compatibility.
• A rapidly-growing community of enthusiasts on the Linux/Open Source front are making modern S3 chipsets usable on Linux for 3d.
• Cheapest chipsets target the ultra low-end market; slightly more expensive are the entry-level chips that actually compete with current low-end ATi/Nvidia products.
• Outstanding texture memory value for the price.
• GPU design does not scale well to extremely complex rendering environments, such as the Valve Source engine, or the Gamebryo engine (that’d be Half-Life 2, Oblivion, and Fallout 3, notably.))
• $49.99 price tag on the Chrome 530 GT undercuts ATi and Nvidia’s entry-level chips by a long shot.
• S3 claims to excel at power management. Less watts = lower electricity bill!
• Joint venture with VIA, an experienced motherboard manufacturer, leads to possible performance wins for S3/VIA combo.
• Barely, if at all, on the radar of game developers for QA/testing. Any hardware-dependent problems will probably go unanswered.

For S3 chip selection, just pick whatever their latest model is – their cards retail around $50 and have been pretty consistent in staying that way.

In Conclusion, here is how I sum up the current state of the market for these graphics card vendors:

Value (bang for buck): S3 and Intel are close on this one, but I’d have to call it in favor of S3, because their discrete graphics are more capable than Intel’s best IGPs, yet still extremely cheap. AMD is slightly more value-driven than Nvidia for those needing performance.

Performance: In general, Nvidia. AMD does have its moments where its offerings leapfrog Nvidia on some key performance metrics, but the “safe” bet is on Nvidia. Intel’s recent offerings pull ahead of S3 in terms of performance, but not by much.

Compatibility: Best bets are on Nvidia and ATi, a dead tie. Depending on the software APIs used, and the generation of video card, S3 and Intel may be better or worse in terms of compatibility. S3 compatibility may suffer due to S3’s low market share.

Market Share: S3, Intel, ATi, Nvidia – least to greatest. Not much else to say about that.

Power Consumption: Nvidia and ATi love to chew up power; Intel and S3 are more miserly about it.

Well-Testedness/QA Factor: Nvidia gets preferential treatment. ATi is a first-class citizen, but in line behind Nvidia. Intel and S3 are often ignored during development, and if they work, it’s by sheer luck.

Future-Proofness: Nvidia. Cards as old as the GeForce 6 Series – launched in 2004 – can still run the latest games. ATi cards do not fare quite as well, and older value-priced cards from Intel and S3 have no hope whatsoever after 2+ years of product age.

Stability: Dead tie. These are all professionally manufactured, high-reliability parts. Graphics cards have no moving parts, so their lifespan of stable use is quite long (10+ years). As long as you maintain a healthy software environment, there’s no reason that graphics hardware should cause any instability for you, unless you physically damage it.

If you are in the market for a video card, I would suggest that you choose among the above factors your most important item, and start from there. For example, if you value power consumption the most, you would want to look at an Intel or S3 chip.

I hope this has been helpful. Even if it’s not, I at least got to record a braindump of my thoughts on this matter for future reference