nvidia's early crashing on vista... DX10 connection?

[taltamir]

So, I was seeing nemesis1 spouting his conspiracy theories about MS + nVidia changing DX10 to screw AMD... When I remembered two things.
1. The counter argument I have used against his theory. (which he admitted he made up completely by himself without it ever being mentioned anywhere on the web, based on his own "hunches")
2. That nvidia had 2-3x the amount of driver crashes in vista early on...

The two connected in an unusual manner...
Firstly, a timeline:
1. Nvidia releases the first G80 GPU, the first DX10 card, the same day (and I think the same hour too) MS releases windows vista.
2. 6 months afterwards AMD releases their first DX10 card.
3. 5 months afterwards AMD releases their first DX10.1 card.
4. 2.5 months afterwards MS releases SP1 (with DX10.1, and a variety of graphics related bug fixes that were NOT made available on windows update).

so for the first 6 months or so. nvidia's drivers, the #1 crash cause in vista. Had DX10 and DX9 support. While AMD and Intel drivers both had only DX9 support.

Also, a variety of hot fixes for windows vista were made available on an individual basis (ie, people file a support ticket), for various video card related crashes. Those fixes have only been made available via windows update when SP1 was released (they were rolled into it).

Basically the bottom line is. I just realized that nvidia didn't just happen to have a "slip up" period of horrible drivers out of nowhere. They happened to be the ONLY ones dealing with a brand new API that massively restructures everything, and OS bugs as well.

 

[Aberforth]

DX10 was in development for a very long time, Many companies including nv, amd and other software companies were involved in it's development. The first goals of dx10 was to standardize the platform so it shouldn't really matter whether it's a PC, console or a mobile, so naturally the API required a new design, vista driver model changed and so on. But in the middle something went badly wrong, prototypes sent by nv to partners were having problems, it was already delayed- nv couldn't stand the pressure and rushed it to stay ahead of AMD.

Dx10 implementation is very friendly and developers sort of took it for granted by adding too many dynamic elements into the games. Almost 60% of the effects were dynamic...completely dependent on Shader Processors, the old dx10 architecture wasn't designed to handle this sort of scenario.

And developing drivers for a new driver model is very excruciating, you need people with experience, train them about new MS DDK and DX10. NV failed here...almost 30% of vista bugs are due to nv products....but that doesn't mean they were trying hard. NV accepted that (after G80 release) they didn't expect vista driver model to be so hard, bottom line is they just didn't spend enough time and resource in developing driver...they were too busy with hardware and marketing.

 

[MarcVenice]

What exactly is the use of this topic? Are you trying to let Nvidia of the hook?

 

[BFG10K]

DX10 wasn't much of factor given DX10 games took months to arrive on Vista.

 

[taltamir]

@BFG: just because DX10 games weren't available doesn't mean that DX10 was not a factor, they were still integrating DX10 into their drivers, which could destabilize the drivers as a whole, or maybe draw away from their resources as they implement both APIs at once. (while intel and AMD had to only implement DX9, with DX10 following 6 months later for AMD, and not yet for intel). Ofcourse it could all be a big conincidence.

@Marc: What is the purpose of any topic? I am not trying to hook or unhook anyone. I just realized there is a mitigating factor here that did not occur to me before, and wished to discuss it.

@Aber: so basically, a few statements that have nothing to do with what I said, and then the conclusion that nvidia just didn't care about drivers at that time, and wasn't trying enough? maybe they took their driver writers (the majority of their staff) and put them to work engineering cards instead (note, software engineers are not electrical engineers, and nvidia is known for its lack of firing rounds)

 

[Aberforth]

Originally posted by: taltamir

@Aber: so basically, a few statements that have nothing to do with what I said, and then the conclusion that nvidia just didn't care about drivers at that time, and wasn't trying enough? maybe they took their driver writers (the majority of their staff) and put them to work engineering cards instead (note, software engineers are not electrical engineers, and nvidia is known for its lack of firing rounds)

Yes, nv didn't care about the drivers, they said they didn't expect the architecture to be so vastly different so didn't bother to look into the new driver sdk, this proves how pathetic the company really is- give the experience they have in the industry...they didn't bother to take drivers as a priority. When people started complaining all they did was release a statement that they were working on drivers.

Also Device driver development is not like software development, you need a separate group of programmers to do it.

 

[ViRGE]

I'm with BFG, I think you're stretching things a bit thin taltamir. I would look at their implementations revolving around the new WDDM; everyone with Vista certified drivers had to implement it. Keep in mind too that to Vista, DX9 isn't very different from DX10 - the former is implemented largely as a subset of the latter. While there are certainly things you have to do specifically for DX10 beyond the bare minimum for DX9/SM2.0 + WDDM, it's not necessarily a great deal of things. Most of the "work" for DX10 is getting WDDM compliant.

In short, I highly doubt that the problem was in the OS (although the memory allocation bug was pretty nasty) because WDDM is the deciding factor, not DX10.

 

[taltamir]

Originally posted by: ViRGE
I'm with BFG, I think you're stretching things a bit thin taltamir. I would look at their implementations revolving around the new WDDM; everyone with Vista certified drivers had to implement it. Keep in mind too that to Vista, DX9 isn't very different from DX10 - the former is implemented largely as a subset of the latter. While there are certainly things you have to do specifically for DX10 beyond the bare minimum for DX9/SM2.0 + WDDM, it's not necessarily a great deal of things. Most of the "work" for DX10 is getting WDDM compliant.

In short, I highly doubt that the problem was in the OS (although the memory allocation bug was pretty nasty) because WDDM is the deciding factor, not DX10.

Now that is a very interesting response... what I was looking for too. Technical info corroborating, or in this case, countering, a theory.

You make excellent points, according to which, it shouldn't have placed an extra burden justifying that level of issues. So I guess they really did just goof, not because of DX10.

@Aber: How exactly do they not make the drivers a priority? they did not fire any driver writers, or hire any, they COULD NOT reassign them to other tasks (except maybe menial labor, which makes no sense), and they probably weren't paying them to sit around and do nothing all day.

Now if they fired a bunch of driver writers, and then hired more later on, that is a different story. Or if their driver department was significantly smaller then the competition and they would not upgrade it. But AFAIK that is not the case.

Or are you saying that they insisted their driver department focus on increasing FPS instead of solving crashes? controversial approach, but doable, and not necessarily malicious. If it ever occurred. I would prefer stability though.

 

[Aberforth]

Originally posted by: taltamir

@Aber: How exactly do they not make the drivers a priority? they did not fire any driver writers, or hire any, they COULD NOT reassign them to other tasks (except maybe menial labor, which makes no sense), and they probably weren't paying them to sit around and do nothing all day.

Now if they fired a bunch of driver writers, and then hired more later on, that is a different story. Or if their driver department was significantly smaller then the competition and they would not upgrade it. But AFAIK that is not the case.

Or are you saying that they insisted their driver department focus on increasing FPS instead of solving crashes? controversial approach, but doable, and not necessarily malicious. If it ever occurred. I would prefer stability though.

What am saying is, the driver developers didn't focus on WDDM- you need special training for WDDM and DX10, I guess nv didn't do this- they were trying to port XP driver code to vista and when the crashes started to occur they must have realized that a complete reworking of drivers is needed...so maybe at that point they started to re-write the entire thing from scratch but it's only a guess I dunno exactly. You might notice they suddenly jumped from v96 to v164.

btw they are never going to make the drivers open-source (even though it helps) because everyone will know what how bloated they really are and what other patch up work they are doing since g80 to boost fps.

 

[taltamir]

mmm... yea, AMD wouldnt open source theirs either, even though they created an initiate to create an open source drivers, they insist they start from scratch

 

[chizow]

The majority of TDR errors being attributed to NV weren't directly tied to DX10 or DX10 games/apps, but they were absolutely a result of NV dominating high-end market share when Vista first released without any DX10 compliant hardware available from ATI.

This follows the basic premise that people who upgrade to cutting edge hardware will do the same and upgrade to cutting edge software. I've already shown without a doubt NV dominates ATI with DX10 gaming machines in the Steam survey corroborated by a few other sources. I believe the ratio of NV to ATI cards was 4:1 which is much higher than its 3:1 ratio of Vista errors and also higher than its 2:1 market dominance over ATI.

Also, the original WDDM Vista driver had a much greater impact and footprint than just the video card and its driver as there was a copy of the frame buffer virtualized in system memory which greatly increased the likelihood for any other system instability to cause a TDR error. Chipset problems, BIOS instability, bad RAM could all result in a TDR error which would ultimately be reported as a driver crash. Much of these problems were fixed once MS hot fixed Vista's WDDM and removed virtualization of the frame buffer from system memory.

There's no doubt NV's Vista drivers were a mess at Vista launch, but there's also no doubt MS was at least equally at fault with all the hot fixes needed to stabilize the new WDDM. Its amazing how many TDR errors you see now from ATI parts now that they're actually selling and people are using their cards. Some people would've had you believe the problem was isolated to NV.

 

[40sTheme]

I HATE nVidia and DX10. I think almost all DX10 games crash no matter what drivers I install; even third party ones with supposed fixes. YES I DO USE DRIVER CLEANER. Before that becomes and issue. DX9 games do just fine. The biggest offender for DX10 crashes is Company of Heroes, but that is partly Relic's fault and they have been unable to find the cause.

 

[thilanliyan]

It's not like nVidia first saw Vista when it was released. They had obviously gotten a hold of it long before it came out so their drivers should have been a bit better than they were.

 

[taltamir]

Originally posted by: 40sTheme
I HATE nVidia and DX10. I think almost all DX10 games crash no matter what drivers I install; even third party ones with supposed fixes. YES I DO USE DRIVER CLEANER. Before that becomes and issue. DX9 games do just fine. The biggest offender for DX10 crashes is Company of Heroes, but that is partly Relic's fault and they have been unable to find the cause.

maybe your card is defective. I had an 8800GTS 512, a 4850, and now a GTX260. And the only crashing I had were isolating game related bugs, and once a leaked unofficial driver beta version that made City of heroes crash all the time on the 4850, and was solved when i reverted to WHQL drivers.
But hate nvidia? and hate DX10? and these really have nothing to do with this thread, please keep to the subject.

@chizow: so you think I really was unto something then? interesting.
anyone else with a technical view on why this is or isn't an issue?