It was only a matter of time
- Thread starter Doomtrooper
- Start date
Bouncing Zabaglione Bros.
Legend
micron said:
Check the drivers the scores were taken with. If they contain cheats that invalidate the results, yank the scores from the ORB. The same should go for any manufacturer. It's the equivalent of not counting a spoiled ballot vote or a spoiled test score paper.
Myrmecophagavir
Newcomer
Isn't it the case that newer Detonator revisions are WHQL only for the FX series and not for the rest of the cards? Does that mean you can upload scores from FX cards using those drivers but not from other cards, or does the ORB just check the version number and not which card it was used with?
Bouncing Zabaglione Bros.
Legend
Myrmecophagavir said:
Public WHQL drivers are a requirement of the 3Dmark2003 licence, but WHQL doesn't stop any drivers that identify applications and then implement cheats.
I suppose the WHQL requirement was a way of making the scores repeatable, and to prevent any special developer drivers with the sole target of racking up huge ORB scores for marketing purposes. I don't think Futuremark expected such a massive effort from Nvidia to write and hide such cheats, even going to the extent of encrypting their drivers.
Do you happen to know which drivers the 5900Ultra sitting in the ORB's top position was using?Bouncing Zabaglione Bros. said:
In order for that card to grab the top spot and get published, the drivers had to have been aproved by FM in the first place. Futuremark is saying that it is ok for that card to hold the number one position.
We all know, including FM, that Nvidia is going to cheat in every driver release, but FM cannot survive without letting FX cards submit scores.
Bouncing Zabaglione Bros.
Legend
micron said:
No, sorry I don't know. However, IIRC soon after the 330 patch came out and dropped the GFFX scores, another Nvidia driver came out that put the scores back up to cheat levels. We've had no comment on that from Futuremark one way or the other.
Given what we know of the Futuremark climbdown, and the other Nvidia cheats still in place, I can only infer that GFFX could only have reached that score by cheating as per previous drivers. It's poor shader performance should have seen to that.
I'm not very happy about taking this "guilty until proven innocent" approach, but based on past performance and the lack of comment from Futuremark or Nvidia, I need to be convinced the cheats are not happening. Caveat emptor and all that. Nvidia has used up all their free passes.
As I said before, letting cheating drivers submit scores simply devalues the long term reputation of Futuremark and their products. What happens when ATI decides to do the same? The whole thing becomes useless.
The whole thing already is, actually... FutureMark is caught between not letting the FX line post scores (if the currently approved drivers on the ORB cheat, which may or may not be the case), which makes 3DM more or less useless if no comparison is possible, and letting the benchmark be cheated, which also makes comparisons meaningless.
And for most enthusiasts, Futuremark's reputation is already tarnished beyond repair : people are either turned against synthetic benchmarks, "thanks" to Nvidia's very efficient anti-3DMark smearing campaign using their "guys with webpages" as proxies, or don't trust FutureMark anymore because of the infamous "It's optimizations not cheats" FM/NV joint statement.
Bouncing Zabaglione Bros. said:
A problem I've had with looking at this issue 20-20 is that I've been spoiled rotten by ATi over the last year in getting an official driver release on average every 4-6 weeks. So, every time I hear something about the "latest" nVidia drivers I tend to think of them as company-sanctioned drivers which have been officially released. Just checked the nVidia site though and it looks like nVidia hasn't officially released a set of drivers since May 14, the 44.03's. Everything else apparently falls under the so-called "leaked" driver definition, and IMO, really doesn't count. That is, I wouldn't expect FM to release patches based on these "leaked" drivers. But I would and do expect them to release recompile patches every time nVidia releases an official driver, though. At least, if FM wants to preserve the tatters of its credibility, it should be prepared to do this automatically each time nVidia releases a new driver set officially. If they don't do this then IMO it will simply tell me that FM just no longer cares whether its benchmarks are manipulated. I'll support FM only so long as they defend their software--when they stop defending themselves so will I. But I really don't think it's fair to expect them to respond to anything other than officially released Detonators. I think they have a little more rope, yet.
I would prefer it if FM would cease posting scores from non-official driver sets, regardless of IHV. With reference to "leaked" drivers no one really knows "who" is doing the "leaking" or what the drivers actually entail--they could easily be older sets, a mixture of older sets, renamed, etc. With official Detonators coming 3-4 months apart as they sometimes do I can understand people with nVidia products becoming enthusiastic about such "leaks", but nonetheless websites should not be evaluating products with such drivers. It's issues like these that make me really glad I can view the nVidia predicament from a distance these days...
Bouncing Zabaglione Bros.
Legend
WaltC said:
Agreed. I thought it was made pretty clear that the 3DMark2003 licence agreement *requires* a publicly released WHQL driver (I speculated on the reasons why in an earlier post). If scores are submitted to the ORB that do not comply to this, they should be yanked from the ORB.
There's no point in having a score in the ORB if it was made with a hacked driver that renders a black screen at 500 fps for a 30,000+ final score.
Can somebody clarify which NVIDIA drivers after the 44.03s are definitely WHQL? I was under the impression that there aren't any - so that will preclude all ORB entries, whether published or not, from being included in the data sample for the HoF. I have the funny feeling that another revision is though...
Also note that the positioning problem of certain cards due to their chip names is a driver/IHV problem and not FMs; once everybody gets that sorted out and enough entries are submitted to the ORB, that particular issue should disappear. For example, the 44.03 drivers recognise a NV31 rev1 as a 5600 (no "Ultra) but labels a rev2 as being a 5600 Ultra.
Also note that the positioning problem of certain cards due to their chip names is a driver/IHV problem and not FMs; once everybody gets that sorted out and enough entries are submitted to the ORB, that particular issue should disappear. For example, the 44.03 drivers recognise a NV31 rev1 as a 5600 (no "Ultra) but labels a rev2 as being a 5600 Ultra.
Actually the licence doesn't include entries to the ORB and as far as I can tell, the licence for the 330 version doesn't say anything about WHQL drivers at all.
From guru3d the only higher than 44.03 drivers I can see that are WHQL are the popular EVGA 44.67:
http://download.guru3d.com/pafiledb.php?action=file&id=640
EDIT: thus these can distort the results of the HOF asuming they still cheat.
Nick, there are board vendor variants that are WHQL. Yours may not be WHQL, but other boards may be bundled with them post the WHQL process. I think there is at least one other set thr AIB's are bundling that are WHQL, but aren't available as an official download from NVIDIA.
DaveBaumann said:
Dave, do you have any ideas on why this is done? I mean, it seems to me that if nVidia's going to WHQL a set of drivers for one of its OEMs there should be no problem releasing a concurrent WHQL version of its reference drivers officially. I'm also assuming that eVGA, like most OEMs, is 100% dependent on nVidia to supply it with its drivers.
If memory serves, I've also seen some mention of "WHQL" drivers which are WHQL for nv3x only (and not for other cards supported in the driver set) or else WHQL only for the older cards and not for the nv3x series.
I can understand cards shipped with non-WHQL drivers prior to the drivers completing the WHQL process, but am not clear on why nVidia would supply reference-design board drivers which are WHQL'ed to an OEM but fail to make them available as official WHQL reference drivers from its web site. I'm wondering now if nVidia isn't mixing in elements from older WHQL driver sets along with some newer elements, and shipping drivers with a newer version number to OEMs as "new WHQL drivers."
I don't have the time to read the whole thread ( or at least not now ) , but just one quick comment...
The cheat causing the artifact shown in the screenshot posted in the original post of this thread IS gone in some leaked drivers already, and will also be gone in future official releases. The same is true for the "no-back-buffer-clearing" optimization.
All shading optimizations remain and will continue to exist in future driver releases, at least in the near future.
Uttar
The cheat causing the artifact shown in the screenshot posted in the original post of this thread IS gone in some leaked drivers already, and will also be gone in future official releases. The same is true for the "no-back-buffer-clearing" optimization.
All shading optimizations remain and will continue to exist in future driver releases, at least in the near future.
Uttar
They did miss one though.
Which one?DaveBaumann said: