The Inquirer-Home

Nvidia cheats on 3DMark with 177.39 drivers

Different workload, let's just pretend it's legit
Mon Jun 23 2008, 20:12

NVIDIA IS VOILATING Futuremark rules with the latest PhysX drivers, and doing it in the usual sleazy way. The rules are simple, violating them isn't a trick, and doing so in order to pump up your numbers is the height of unethical behavior.

Well no, it isn't the height, this is, but calling Nvidia on unethical behavior is what you might deem a target-rich environment. That said, the company's behavior this time could be done in an ethical way, but it chose not to. The explanation needs a little background though, so bear with us.

3DMark Vantage has four major components, two CPU and two GPU. One of the CPU subtests is a physics-based test. The physics test is based on the Ageia PhysX API, a fairly widespread API in use by a large number of games. Between the time that 3DMark Vantage development was started and the time it was released, Nvidia bought Ageia.

The problem is that the PhysX DLLs, and for that matter, the whole API is now owned by Nvidia. In and of itself, this is not a problem, especially if the company involved had a history of honesty, integrity, and fair play. Nvidia has none of these attributes, and has a proven history of cheating on 3DMark.

To be fair, ATI has been caught at the same thing as well, but nothing lately, and Intel compilers come with curious optimisation defaults as well. No one is clean, but only Nvidia seems to take dishonesty as a corporate mandate.

So, with the latest driver, Forceware 177.39 drivers, Nvidia put its now in-house PhysX APIs into the drivers. Instead of it running on the CPU or on the PhysX chip, it is running it on the GPU. It owns the GPU and it's drivers along with the physics API and all those drivers. This is a dangerous situation.

There are two problems with Nvidia doing this, it isn't a legal driver for 3DMark, and it isn't even running the same program as others who run 3DMark. Either one is enough to preclude people from using those drivers and calling the results 3DMark scores.

If you look at the 3DMark Vantage Driver Approval Policy, section 3.5 clearly states, "Based on the specification and design of the CPU tests, GPU make, type or driver version may not have a significant effect on the results of either of the CPU tests as indicated in Section 7.3 of the 3DMark Vantage specification and whitepaper." When you run a CPU test on a GPU, it clearly violates the rules.

The other problem is that when you install the drivers, they replace the PhysX DLLs with a completely different set of DLLs. If you look at the Nvidia PhysX reviewers guide, a PDF that NV hands out to help people write up their newest toys, they say the following for UT3 installation.

alt='nvidia_ppu_ut3'

That looks completely above board!

Note steps 4 and 5 that say "Uninstall the existing AGEIA PhysX v7.11.13 driver (installs with UT3 installation)." and "Install the new PhysX 8.06.12 driver." Same with 3DMark Vantage, and they offer the helpful hint of "GeForce PhysX is enabled in CPU Test 2. We recommend testing in Performance Preset for the best final score with GeForce PhysX. In Extreme Preset the score is mainly determined by the GPU score. A faster CPU Test 2 result will not make much difference."

This means two things, when you are running 3DMark Vantage with the 177.39 drivers, you are not doing the same work as every other driver running 3DMark Vantage. You are doing a completely different workload on 25 per cent of the tests. To rub salt into the wound, Nvidia then tells you that the Extreme preset, the one meant for high-end GPUs, doesn't show off the cheat sufficiently, so use one that weights it more heavily. What gall.

The end result is about a 10n per cent increase in scores, and a claimed 7.5x advantage on the physics subtest. You can see how they word it for yourself.

alt='nvidia_ppu_vantage'

If the glove doesn't fit, you must acquit.

If you look at it as a whole, Nvidia is doing two things, first of all doing a different workload than the rest of the world and claiming it is the same, and second not following the Futuremark rules. This is nothing less than blatant dishonesty.

We asked Futuremark for a comment, and they referred us to the above Driver Approval Policy, and then to the approved driver page here.

Nvidia has not submitted 177.39 for approval, and likely will never do so because the chance of it being approved are something between zero and having to buy Futuremark. Until they submit a bad driver, no harm, no foul. The Futuremark policy is that if it isn't up on ORB, it isn't a 3DMark score, and that is quite sensible.

Nvidia engineers know that there is no way they can get this driver approved, so they don't try. They know they are not running 3DMark, but they don't even try to hide it. They are however disingenuously doing a different workload and trying to cynically pass it off as the same old workload. There is a word for this behavior, cheating. µ

Note: At time of this writing, the ATI Catalyst 8.6 drivers are not approved yet, the latest valid set is 8.5. Nvidia's latest approved is 175.16, and Intel isn't even submitting a set, it seems they don't want the world to know something. In any case, this does not preclude all three companies from quoting scores liberally with unapproved drivers.

Bad industry, no cookies for you.

Reviewers take note.

 

Share this:

blog comments powered by Disqus
Advertisement
Subscribe to INQ newsletters

Sign up for INQbot – a weekly roundup of the best from the INQ

Advertisement
INQ Poll

Heartbleed bug discovered in OpenSSL

Have you reacted to Heartbleed?