Do New Drivers Really Boost Performance?

If you believe the most emphatic gamers, a graphics drivers update is good for—what, 10 or 15 percent additional performance? As a result, running benchmarks on Catalyst 8.6 when 8.8 is out automatically invalidates those other numbers and clearly indicates bias. All kidding aside, how much truth is there behind the assumption that every new driver update delivers untold performance enhancements?

AMD brings out a new Catalyst driver every month and if there’s a sales push for a new graphics chip between releases, additional beta drivers are released to the press. At Nvidia, the cycle between the official WHQL driver versions is longer. However, beta drivers containing minor changes are available at irregular intervals.

Logically, it’s clear some of the performance claims must be exaggerated. AMD and Nvidia release numerous driver builds every year. If each of these drivers were to increase 3D speed by 10 percent, the graphics cards would double their performance in a few months. But that’s not the case, as even new graphics chips are struggling to outshine their predecessors by up to 50 percent. If the claim that every graphics driver brings more speed were true, no one would ever buy new 3D hardware again because the driver updates would deliver greater benefits.

In practice, the first beta drivers are optimized for speed for the purpose of 3D tests and aren’t necessarily perfect. The first “official” versions are more stable, which, for some games, costs a little in terms of performance. The greatest speed increases for new products generally come within the next three months, when more errors are found and corrected, and adaptations are made to meet current game requirements, ensuring the basic level of 3D performance. The subsequent three months see additional optimizations for benchmark games to achieve better test results. Then adaptations are made in favor of new games, with the drivers additionally optimized for CrossFire (AMD) and SLI (Nvidia).

Over the entire product life cycle of a graphics card, we might be able to expect as much as a 30 percent overall performance increase due to driver updates. Greater individual results are possible, but these usually result from errors in memory adaptation or poorly optimized games—for example, Flight Simulator X with Nvidia cards that have less than 512 MB of graphics memory or Crysis, which remains a constant source of changes as frame rates yo-yo depending on the driver version used.

In this evaluation we aim to determine how much of a performance increase could be seen between the Catalyst 8.6 and the Catalyst 8.8. Tests were conducted using a Radeon HD 4870.

There’s another rumor tied to the Radeon HD 4870 running in CrossFire mode: only beginning with Catalyst 8.8 is the performance of two cards running together properly supported.. The GeForce GTX 280 with the ForceWare 177.39 driver is apparently not very well optimized either—a recent test with the GeForce 177.92 beta driver showed up the differences.

All tests also were carried out with an overclocked CPU. This is the only way to determine how much potential could be seen in the fast graphics chips. The GeForce 9600 GT was tested as a control group with both the ForceWare 175.16 and the GeForce 177.92 drivers. As a series chip, it has the longest possible period of driver optimization and serves as the lower performance limit for the CPU power test.

This thread is closed for comments
    Your comment
  • pcgamer12
    Great article.
  • cangelini
    LedrosTo bad you used 8.8 drivers instead of the 8.9 ones. Real smart people.

    This is observed in the piece and should not affect performance.
  • curnel_D
  • neiroatopelcc
    SoulLance"Unfortunately for AMD and its customers, the PowerPlay optimization for energy-efficient 2D operation is not included in the Catalyst 8.8 driver for the Radeon HD 4870. Quake Wars refuses to start using the Catalyst 8.8, but it works fine with the Catalyst 8.6 and 8.7."But neither should comments like this appear when everyone on the planet reads a labels that clearly say "Use latest drivers from manufacturer".

    Can you please translate this into english?
    I have no idea what you mean to say with your comment, other than the obvious (that you disagree on some level with the article)
  • randomizer
    He's saying you shouldn't criticise newer drivers because most hardware and software packaging has a label, or a section in the manual, telling you to use the latest drivers. So latest = best as far as he is concerned.
  • In reference to overclocking graphics cards, I'm pretty sure that XFX card warranties cover overclocking to some degree, though you'd have to double check their warranty fine print.
  • neiroatopelcc
    SneppyIn reference to overclocking graphics cards, I'm pretty sure that XFX card warranties cover overclocking to some degree, though you'd have to double check their warranty fine print.

    Having had experience with two failing xfx cards (both 8800gtx's actually), I've got to warn you though, that their warranty is a bit fictional. While they may actually cover broken products, you can easily risk waiting months for a replacement.
    Proof of my claim in case someone unexpectedly doubts it :
    That was the first failing card. The second I haven't even bothered sending in, and just bought an 4870 instead of waiting months.
  • cryogenic
    We all know there were cases in the past when "Newer drivers provided substantial performance improvements" in various applications or games. Analyzing performance with a few select driver sets can can't lead to the conclusion that drivers are irrelevant, or that from a "professional" stand point the latest drivers shouldn't be chosen.
  • randomizer
    CryogenicAnalyzing performance with a few select driver sets can can't lead to the conclusion that drivers are irrelevant

    But it can lead to the conclusion that current drivers are irrelevant, or rather less relevant.
  • cryogenic

    That's a good,true and useful conclusion, but it's not a general one, only specific to this particular set of drivers.
  • warezme
    according to that chart two 8800GTX's is faster than a single 280GTX and Two 4870's in CF.??? 18005 as opposed to 16422 for their fastest run on 3Dmark06 @1280x1024 default settings. I thought their game scores where kind of low to. Must be a dual core CPU making the difference.
  • When the Q3 08 Vga charts appeared, there where lots of critics to the fact that the ATI drivers used where 8.6. Maybe this article is some kind of response to those critics. Well, the results of the tests here "almost, "kinda", say that Q3 benchmarks are fair enough with new-gen ATI cards. But... profesionally speaking, did Tom's knew THEN -back when the Q3 benchmarks where made- that that will be the case? That drivers released before the new cards did, didn't keep those cards from showing some better numbers? If 8.7 or 8.8 show almost the same numbers 8.6 do, is irrelevant. We can know that only now, that there are avaiable tests that actually show those results. But how can we, nor Tom's, or anybody knew that THEN? So, you must admit at least that if you are benchmarking a card using drivers dated BEFORE the card was launched, that there is a "chance" that results may actually vary with recent drivers. And if its true that performance boost in newest drivers are not "that important" to compare with the gap between diferent card generations, is true also that you can't(and in fact you don't) use 6, 12 or 18 months old drivers on current hardware. So maybe not "that important", but "important" nonetheless.
  • neiroatopelcc
    CryogenicWe all know there were cases in the past when "Newer drivers provided substantial performance improvements" in various applications or games. Analyzing performance with a few select driver sets can can't lead to the conclusion that drivers are irrelevant, or that from a "professional" stand point the latest drivers shouldn't be chosen.

    Where's it say drivers are irrelevant?
    What the article states, is that you don't nessecarily gain performance by using current drivers. It doesn't mean the bug fixes and 'futureproofing' with regarding to unreleased games isn't relevant.

    The article merely deals with the urban legend claiming new drivers unconditionally improve hardware performance.
  • neiroatopelcc
    @ alfguy - the article does state that drivers released within roughly the first 3 months after the hardware is released is more likely to improve the hardware's peformance, but that it doesn't nessecarily improve older hardware's. That's pretty much the same point you're making I believe?

    ps. catalyst 8.6 is the version shipped with my 4870 - so it's not dated before the cards, it's merely the initial retail version.
  • JPForums
    Overall, this was a good article. I've seen driver updates on past hardware that netted more significant FPS improvements than is show here, but the time frame for optimizations is consistent with the article introduction. Also consistent with the article is the fact that the larger improvements don't usually occur on the high end cards.

    Disclaimer: This is not a fault in the article, just speculation on one of the results.

    Referring to the Crysis Very High setting on the Radeon 4870:

    In the Very High setting, fluctuations of the frame rates can also be seen. There does not appear to be any reason for these variations, although a performance increase is almost always seen.

    It appears 1280x1024 is cpu limited, as you see a fair FPS improvement at this resolution (and not others) when overclocking.
    Moving to 1680x1050, you see more significant improvements as the burden has shifted back towards the GPU, allowing the enhancements to be seen.
    At 1920x1200, we are simply reaching the limits of the hardware, so the enhancements are less relevant.

    It seems one of the enhancements affect AA performance in Crysis as it seems to be improved across the board. I'll leave you to figure out the rest, as I'm short on time, but I'd wager the enhancements involve trading more efficient processing of data on the video card for more driver overhead. This would explain why more CPU limit settings are seeing lower rates, while more GPU intense settings are seeing benefits. The only real oddball is 1680x1050 at high, but I'll bet with a little time, it could be explained as well.

    On another note: Did the Catalyst 8.8 improve performance over Catalyst 8.6 in CF mode when both were overclocked. I suspect the answer is no, but it seems there may be more CPU overhead in the newer drivers. Given more CPU cycles, an insignificant, but measurable improvement may be measured. (My interest in this is purely academic as I don't own a newer ATI card at the moment)

    That said, I'd be interested in a similar article with mid/high end video cards and a mid/high end processor. What kind of real performance improvement (or lack there of) can someone with a GeForce 8800GT or Radeon 4850 expect with paired with an Athlon X2 6400+ or a Core 2 Quad Q6600 (without overclocking).
  • CanadaKen
    I wish someone could explain why ATI/AMD cards can't run Microsoft FlighSim X as well as Nvidia cards. The specs are there but the performance sucks?!
  • neiroatopelcc
    @ CanadaKen : I can only guess, but flightsim is very dependant on the cpu. I'm guessing the ati drivers need more cpu time to do well in comparison to the nvidia one, thus the nvidia card works better with limited cpu ressources. It could also be possible that microsoft has been utilizing some features that run better on nvidia cards as opposed to amd/ati.

    As the monthly graphics roundup has been telling us for the last past months, which highend card you should choose (4870, 260 or the dual gpu nvidia one) depends on the gaming titles you expect to use it on, as they're not delivering the same performance in all games. Some are sponsored by nvidia and more often than not those games run better on nvidia hardware, even if it's slower in a raw shader processing power value. The opposite is probably true with games where ati has had their hands on the graphics engine though I can't at the top of my head remember any.
  • zodiacfml
    newer drivers is important to solve issues. but it becomes irrelevant to improve a game performing more than 30 or 60 frames per second.
    of course, exception is crysis wherein additional of, like 5 frames can help in all resolutions and little fps difference there can be worth debating.
    anyways, this an article many of us waiting for. high end crossfire/sli does have the least value. crysis is the only game could benefit from high end hardware video card to processor.
  • dagger
    The myth that new drivers drastically improve performance is made up by fanboys to discredit benchmarks that show their favorite cards don't perform as well as competition. It's nothing more than an excuse, against valid benchmark results.

    Some drivers do improve performance, just not nearly enough to make a noticeable difference.
  • Aragorn
    Since 8.7 did so well in the few tests it was used in I'd love to see the results of running it through the whole test suite.
  • wh3resmycar
    so this article is basically a slap in the face to the ati fan boys who cried foul when the 8.8 drivers werent used on the GPU charts.

    i've been using videocards for like 7 years already and i'm not expecting new drivers to give me performance boosts, they provide more "stability" instead (like fixing corrupt textures).

    so for the apes who thought they were cheated, and thought theres a conspiracy between nvidia and tom's, here what i can say to you.. the aliens did land on roswell!
  • Dmeza
    Great Article, but i wonder why dont you guys include the Radeon 4870X2... i'm thinking of buying one and i would like to see how it performs...
  • kamkal
    alex57340141Any ideas why they don't test at 1080P resolutions, I have a nice 1080P tv that i use to play games on. Which resolution would be closest performance to it when looking for a new video card?

    look at the 1920x1200 resolutions in the article for comparison

    your 1080p tv has a similar resolution of 1920x1080
  • kamkal
    DmezaGreat Article, but i wonder why dont you guys include the Radeon 4870X2... i'm thinking of buying one and i would like to see how it performs...

    performance i believe would be similar to 4870 in CF

    the 4870x2 would probably be a bit better since its CF on one card