Intel excuse for no GMA900 WDDM driver: no "HW Scheduler" no driver

[bzo]

People have been rightfully (but childishly) complaining about the lack of Vista WDDM drivers for the GMA900 on the Intel forums. The latest reason mentioned is that a WDDM driver is impossible because the GMA900 does not have a "HW Scheduler". Does this seem plausibile?

http://softwareforums.intel.com/ISN/Community/en-US/forums/permalink/30224347/30225947/ShowThread.aspx#30225947

 

[Richteralan]

I would guess no.
Because GMA900 and GMA950 has no significant difference besides GMA950 added PS3.0 support.

 

[DavidC]

You guys may not agree with me, but I think what Intel says is plausible.

People still believe 865G=845G in architecture and at same bandwidth, it would perform same, only difference is the graphics clock speed

http://www.beyond3d.com/reviews/intel/d865gbf/index.php?p=14

Despite the same clock speed(845GE is at 266MHz, 845G=200MHz) as the 845GE, 865G manages to outperform it greatly, even with single channel memory. There are other gaming benchmarks I have seen where it can reach 2x the performance.

Intel doesn't explicitely mention it, but 865G has greater size buffers for the zone rendering and the on-chip cache to improve performance.

I would guess no.
Because GMA900 and GMA950 has no significant difference besides GMA950 added PS3.0 support.

(BTW, they both have PS2.0, not PS3.0, the one with PS3.0 is the GMA X3000 with the latest drivers.
The confusion for support between both is the software VS part, which I believed was GMA900 supporting version 2.0 and GMA950 supporting version 3.0)

Now I am not sure anymore, while the pages for GMA900 and GMA950 shows different VS support, Intel Software Network article(http://www.intel.com/cd/ids/developer/asmo-na/eng/201974.htm?prn=Y) says GMA900 also support VS3.0 like GMA950, and even 845G/865G!!

But, then I can't explain why GMA950 is significantly faster than 3dmark05: http://www.computerbase.de/artikel/...st_turbocache_hypermemory/6/#abschnitt_3dmark

There's also this:

GMA900: OpenGL* 1.4 support(http://www.intel.com/support/graphics/intel915g/sb/CS-012579.htm?iid=graphics+915main&)
GMA950: OpenGL* 1.4 support plus ARB_vertex_buffer and EXT_shadow_funcs extensions and TexEnv shader caching(http://www.intel.com/products/chipsets/gma950/index.htm)

Until people finds it out I suggest accepting what they said is better idea.

(Just like Intel said 865G is a new core, while not more advanced in terms of 3D, it is indeed different performance-wise, and is not a exact replica of 845G)

 

[Richteralan]

You guys may not agree with me, but I think what Intel says is plausible.

People still believe 865G=845G in architecture and at same bandwidth, it would perform same, only difference is the graphics clock speed

http://www.beyond3d.com/reviews/intel/d865gbf/index.php?p=14

Despite the same clock speed(845GE is at 266MHz, 845G=200MHz) as the 845GE, 865G manages to outperform it greatly, even with single channel memory. There are other gaming benchmarks I have seen where it can reach 2x the performance.

Intel doesn't explicitely mention it, but 865G has greater size buffers for the zone rendering and the on-chip cache to improve performance.



(BTW, they both have PS2.0, not PS3.0, the one with PS3.0 is the GMA X3000 with the latest drivers.
The confusion for support between both is the software VS part, which I believed was GMA900 supporting version 2.0 and GMA950 supporting version 3.0)

Now I am not sure anymore, while the pages for GMA900 and GMA950 shows different VS support, Intel Software Network article(http://www.intel.com/cd/ids/developer/asmo-na/eng/201974.htm?prn=Y) says GMA900 also support VS3.0 like GMA950, and even 845G/865G!!

But, then I can't explain why GMA950 is significantly faster than 3dmark05: http://www.computerbase.de/artikel/...st_turbocache_hypermemory/6/#abschnitt_3dmark

There's also this:

GMA900: OpenGL* 1.4 support(http://www.intel.com/support/graphics/intel915g/sb/CS-012579.htm?iid=graphics+915main&)
GMA950: OpenGL* 1.4 support plus ARB_vertex_buffer and EXT_shadow_funcs extensions and TexEnv shader caching(http://www.intel.com/products/chipsets/gma950/index.htm)

Until people finds it out I suggest accepting what they said is better idea.

(Just like Intel said 865G is a new core, while not more advanced in terms of 3D, it is indeed different performance-wise, and is not a exact replica of 845G)

So..you mean you can't explain it and want people just simply to believe that and don't ask questions?

uh, ok.

 

[Geo]

Well, this would be the first time I've heard someone insist that it was impossible to write a DX9L WDDM driver for an existing DX9c XP GPU, so I'd like to hear about that in more detail before I'm willing to believe it. It's certainly not my understanding as of today that that would be all that likely.

Btw, are we saying here that the vast ocean of Intel GMA900 owners can't run Vista?

Not being familiar with Intel forums, what's the street cred of mad\cvdesylv? He's got 35 posts, so I'm guessing he's not Intel's official rep/staff on that board?

Edit: Hrrm, now I'm really confused. http://www.intel.com/products/chipsets/915g/index.htm It says "Ready for Microsoft Windows Vista" on Intel's own page?

 

[Demirug]

I am not sure what Intel mean witch “missing HW Schedulerâ€￾ therefore I could not say if this is a needed part to implement a WDDM driver. But if we take a look in the WDK we see that the example driver is for the good old R200 chip. It’s primary only a tweaked XP driver and I could not see anything there that let me believe that you need something like a “HW Schedulerâ€￾ for a WDDM driver.

I have the feeling that Intel don’t want to write a WDDM driver for <= 915 because they know that these chips would not match the performance requirements to run Aero. Therefore they don’t want to waste money.

geo, Vista supports XP drivers. You can’t use AERO and D3D9EX but everything else will work.

 

[KillerB]

Normally I just lurk, but this tickled me enough to register and reply.

Chuck is shifty alright, but he also happens to have been working on Intel graphics since before they were integrated !

Check the first reference he gave, (link below), foil 19, top bullet (and 3 sub-bullets). Capability to support that is specifically what's missing.
http://download.microsoft.com/download/5/b/9/5b97017b-e28a-4bae-ba48-174cf47d23cd/MED134_WH06.ppt#19

And from the second reference he gave:

A Windows Vista Capable PC includes at least:

A modern processor (at least 800MHz1).
512 MB of system memory.
A graphics processor that is DirectX 9 capable.​

That is really all it takes to be "Vista Capable". Both of those directly from MS.

And nope you shouldn't trust me if you don't trust mad\cvdesylv. But if you wanted to answers to questions there you go. And the only reason I bothered replying is that I also know first hand the scheduler is needed, and is not in the 900. Honestly I'd be surprised if anyone would even be happy running Aero on the 900 anyway (or 950 for that matter). And no, I don't know anything more than that anymore, and I don't work with anyone who would either (now) .

 

[Geo]

Ah, I think I see where you're going. "Vista capable" does not require Vista drivers by definition (rather than common sense). Yes, sometimes I've wanted to bitch-slap MS for some of the bait-and-switch I see going on with Vista terminology. Tho having said that, one presumes that the whining/crying/wailing/threatening is coming from the hardware people that ends us up here.

 

[KillerB]

Yup, MS wants to ship their OS to as many people as they can, even if they don't have HW that is "premium ready". And there was a WHOLE lot of complaining done when MS decided on final HW requirements for Vista 'native' drivers that excluded relatively recent products. But they make the rules and they get what they want . And like I said, if you really want the 'premium' (aka aero) experience, I'd be surprised if many found it worth the performance hit on entry level (aka low end freeD) gfx solutions...

 

[bzo]

KillerB, perhaps you are correct, but my reading of the slide points you mention doesn't confirm what was claimed about the hw scheduler:

Glitch resilience
* Improved queuing and latency control
* HW should implement Vsync reporting, deep queues
* Multimedia class scheduler

The only HW mention I see is point 2. Besides, this presentation is about the new video playback architecture in Vista, so I'm not sure why he is referencing this as reference on requirements for WDDM drivers?

 

[Farid]

If the original poster doesn't mind, I did updated the thread title a bit to make the relevance to the GMA900 architecture more visible. As it was, at a first glance, one could had expected the thread to be software only, while the the fact the thread actually involves talk about the IC architecture.

On topic, color me surprised with regards to the explanation given by Intel in this case. Needs some investigation, I guess.

 

[bzo]

Vysez, better thread title, thanks.

Assuming that this "HW Scheduler" is required for a fully compliant WDDM driver as the Intel rep cvdesylv claims, does anyone find it ironic that this is in the context of the GMA9xx which doesn't even have hardware T&L? The absence of this more significant hardware feature didn't seem to stop Intel from creating a GMA950 WDDM driver.

 

[KillerB]

A wee bit of hw support is needed to enable all 3 of those sub bullets, and this bit is called "hw scheduler". It really isn't huge, that's why it was able to get into 950 relatively late in the design cycle. Trying to put it into 900 would have caused a hard schedule slip, missing the refresh window and making OEM's very VERY unhappy. Not having this support means your product will not be WDDM compliant, and you won't be put in the stack. If you understand what that means, then you know the ball is all in MS' court (and they don't want to let the GMA900 play).

It's just that simple.

Yes you could argue over the importance of that wee little 'feature'. But it's Bill you have to take that up with, not Paul...

 

[Fox5]

I am not sure what Intel mean witch “missing HW Schedulerâ€￾ therefore I could not say if this is a needed part to implement a WDDM driver. But if we take a look in the WDK we see that the example driver is for the good old R200 chip. It’s primary only a tweaked XP driver and I could not see anything there that let me believe that you need something like a “HW Schedulerâ€￾ for a WDDM driver.

I have the feeling that Intel don’t want to write a WDDM driver for <= 915 because they know that these chips would not match the performance requirements to run Aero. Therefore they don’t want to waste money.

geo, Vista supports XP drivers. You can’t use AERO and D3D9EX but everything else will work.

Supposedly in earlier beta versions of vista, people were running Aero on the GMA900 and it performed fine.

 

[KillerB]

Exactly why you can argue the importance of forcing HW to support that one 'feature' in order to get on the stack...

(note just like in the early betas of 2k and xp, it was easy hack drivers into the OS for development - for some components you could even copy dll's over on the fly without a reboot!)