Nvidia, Apple, Comdex?
Over the summer the CEO of Nvidia commented about them an Apple having something to announce around this time...well nothing has been announced so far...the NV30 is due to be announced at comdex...could the NV30 have anything to do with a future Apple product? The only thing that comes to mind is an Xserve renderfarm which would contain more than one NV30 ie dual AGP slots...or perhaps whatever they had to announce got postponed or cancelled because of the economy
Comments
<strong>The only thing that comes to mind is an Xserve renderfarm which would contain more than one NV30 ie dual AGP slots...</strong><hr></blockquote>
Why would you want dual graphix cards in a renderfarm? <img src="confused.gif" border="0">
And why on earth hasn't anyone tried using GPUs for final rendering to disk before?
The main question about the Xserve and nv30 is not whether it makes sense, but whether the nv30 runs too hot for the Xserve... even one of them. Two (or more) of them will really push the envelope.
When the rendering programs are updated to use the new GPUs (and OpenGL 1.4 in Jaguar supports this) we ought to see an tremendous leap in Mac 3D rendering performance. The PC will leap too, but since both use the same GPUs this should benefit the Mac far more, largely making up for the G4's short comings. Even better, Apple is being aggressive about optimizing MacOS X's OpenGL implementation in terms of performance and features.
There'll be the usual bunch of naysayers complaining about the lack of double precision floating point in the GPUs but frankly, in the vast majority of cases, its not actually needed -- they just haven't done the work to figure out why their code is squandering precision.
<strong>Could be that Apple is going to use nVidia's mobo chipset.</strong><hr></blockquote>
No.
<strong>The Radeon9700 and nv30 are the first GPUs (at the consumer level, at least) which support floating point frame buffer formats.</strong><hr></blockquote>
What about rendering bezier paths?
Also, what happened to OpenGL 2.0?
<strong>The main question about the Xserve and nv30 is not whether it makes sense, but whether the nv30 runs too hot for the Xserve... even one of them. Two (or more) of them will really push the envelope. </strong><hr></blockquote>
Well, I have an Xserve (dual gig) and the blowers are quite capable at cooling the system. The blower over the PCI/AGP slot is doing almost nothing on mine based on how stable the temperature is on my system.
Compared to the 1U Xeon systems, there's a *lot* less heat being moved from the Xserve. Assuming it can tolerate the same steady-state internal temps as the Xeons (not necessarily a safe assumption), I'd say the system could move about 3x the heat that it does now.
The two 1GHz G4s push, what, about 50 watts? Would a single nv30 dissipate more than that - or is it one of those plug-in-the-wall beasts?
Makes me wonder when we'll see a card that handles (or I should say, actually uses) 8X AGP bandwidth, and how long before MoBo tech has to move on to the next new internal bus, PCI-EXpress or whatever ???
From what I've been told is that they are only at the beginning of the optimizations. Supposedly there is far more that they can do.
[[[in the vast majority of cases, its not actually needed -- they just haven't done the work to figure out why their code is squandering precision. ]]]
Ahhhh! I've posted this before on these forums. (Just do a search for my past posts)
Essentially Programmer is right. Double precision is not NEEDED EVERYWHERE. It's likely that it isn't even needed at all. Besides, we wouldn't want to have this feature in our SIMD unit UNLESS VMX/AltiVec was bumped to 256-bit. Again, this was also mentioned in one of my prior posts. The early G5 discussions perhaps. The point is that double precision is just used to cover up a lot of lazy/bad code that just happens to work, but only because double precision allows for more error.. Most programmers are depending on the hardware to makeup for the shortcomings of their code. It's likely that there are some excellent single-precision algorithms that produce results as well as double-precision clacs. However, the story might be a little different for CAD applications where the precision is actually needed. It make you stop and think though.
On another note... I wonder if Apple has found a way to harness the power of "multiple" graphics/video cards on the same system the same way multiple processors are currently utilized. Anyway, It's just another thought. Could something like this be done?
--
Ed\tM.
<strong>
On another note... I wonder if Apple has found a way to harness the power of "multiple" graphics/video cards on the same system the same way multiple processors are currently utilized. Anyway, It's just another thought. Could something like this be done?
</strong><hr></blockquote>
This is definately possible...Back in the day I had a Monster II 3D card in my PC. It was a 3D only add-in card with something like 8MB RAM. It hooked into the regular vid card with an external VGA cable. Anyway, point being, you could buy a second Monster card and tie them together with another VGA cable and get, effectively, double the 3D performance...if only it could be done through an 8x AGP BUS.
:cool:
<strong>... I wonder if Apple has found a way to harness the power of "multiple" graphics/video cards on the same system the same way multiple processors are currently utilized. Anyway, It's just another thought. Could something like this be done?
--
Ed\tM.</strong><hr></blockquote>
I would imagine that it would be better to use 1 video card with multiple GPU's on board. That way you would only be using the PCI bus to communicate with the system (remember that PCI only supports 1 AGP slot) and the GPU's can communicate with each other on the same card.