Taking advantage of the 2nd CPU for gfx acceleration?

[Mr. Cold]

Will Matrox ever *TRULY* support SMP by utilizing _both_ processors under Windows 2000 Professional dual-processor ACPI systems?

If my idea is correct, then you could effectively get almost twice the graphics acceleration power from 2 CPU's as with 1. This would help a LOT for my current dual Celeron 366mhz setup which i do not have overclocked.

[C=64]

That's actually A LOT more difficult than it initially sounds. I don't believe any consumer level card's driver is actively utilizing multiple processors. The only thing they're worrying about is surviving in the hostile, multi-threaded environment that ensues from the application making use of the extra computing power.

First off, your computing power is not double that of a single processor system. It is greater than just a single processor, but there is overhead that has to be taken into account as the 2 processors have to communicate with each other and generally keep from stomping on other's toes.

Also, for 2 (or more) processors to be able to work on a single task, the task must be able to be broken up into 2 parts. For the processors to apply completely efficient computational power to the task, the task would have to be broken down into 2 pieces that are equally intensive and do not depend on each other so neither processor needs to pause and wait on the other.

That is a real difficult trick to pull off in most cases, but to do it in real-time with constantly fluctuating, interlocking data sets and computations that a graphic driver would have to handle... it's down right impossible.

Don't get me wrong; There are some drivers that do take advantage of multiple processors, but they are far from the norm, and from what I recall, only in the realm of "professional" graphics cards, and performance is far from doubled. But in that market, where users throw as many processors, as much ram, and as much cache as they can at the problem, this is a worthwhile optimization.


C=64

p.s. Not overclocked?! My goodness man! My 366's are cooking at 567! Show those celery sticks who's boss!

[This message has been edited by C=64 (edited 04 June 2000).]

[Mr. Cold]

Never say anything is impossible. That is what engineers mindsets are. There is always a way to do something!

Throw some guys w/ Ph.D.'s in Computer Engineering from MIT at the problem and they'll tackle it!

p.s. - I wonder how many Ph.D.'s actually work at Matrox?

[PaulS]

ATI and 3Dlabs advertise SMP optimised drivers. Many of the computations required by a 3D graphics card are inherantly parallel, so they are very suitable for multithreading. Maybe Matrox's recently announced emphasis on business users will see some resources allocated to developing excellent Windows 2000 drivers.

We live in hope!

Paul

[C=64]

I knew 3D Labs was doing it, but ATI too? So, if I understand this correctly, they can handle 2 processors that aren't on the video card, but can't quite handle 'm when they're on? I love irony...

I reread what I wrote, and I find that it does sound like I'm saying it ain't worth doing. Didn't mean for that; I'm just stating that we shouldn't expect double (or even near double) performance.

C=64

[PaulS]

C=64,

Yeah, it was a pretty pathetic effort on ATI's part, blaming W2k for not enumerating multiple video chips or something. About on par with Matrox's blaming W2k for their poor dual head implentation. It seems that video driver developers have become so reliant on the "fill in the blanks" generic driver models that MS provides that anything requiring them to code outside this framework is considered too hard. There is also a certain irony in the fact that MS wears the criticism from these manufacturers precisely as a result of their efforts to simplify development. They made it too easy, killing initiative and creativity in the process.

Paul