Advertisement
If you have a new account but are having problems posting or verifying your account, please email us on hello@boards.ie for help. Thanks :)
Hello all! Please ensure that you are posting a new thread or question in the appropriate forum. The Feedback forum is overwhelmed with questions that are having to be moved elsewhere. If you need help to verify your account contact hello@boards.ie
help with high end mobo choice
Options
-
28-05-2006 10:36am#1
Comments
-
-
-
-
-
-
Advertisement
-
-
-
-
-
-
Advertisement
-
-
-
-
-
-
-
-
-
-
-
Advertisement
-
-
-
-
-
-
-
-
conzymaher wrote:Its not used yet though and the Physx PPU is going to be PCI express 1X
Releasing it as PCI was a mistake IMO.....
OFFTOPIC:
You're right in saying it should've been on PCIe 1x+ from launch, but I'm not convinced that would negate the latency issue. It would appear (from the recent benchmarks and investigation) that it's more of a combination of physical integration and how the PPU APIs place the functionality in the rendering process.
Tbh, PPUs could[should?] be integrated into the GPU or as a copper (co-processor) in either a dedicated/custom socket on the mobo or as a socket AM2/940-ish HT node like the Opteron FPGA coppers.
While this *may* not be a whole lot different to just having it as a discreet PCIe component, it may open possibilities of synching all or some of the PPU work with either the GPU or CPU operations. This may require redesign of the way things are calculated in games, but somethings simply cannot be computed until others have finished. Quantum computing coul be useful here
Of course the counter-argument is that we may simply have to accept that having physics acceleration will produce massive increases in physics-accelerated objects in games at the cost of a hit in rendering efficiency. The gpu essentially has to wait around for PPU/CPU to finish calculating the scene to be rendered. This is where tighter integration of the scene-setup and rendering process is needed. Graphical physics rendering is helping to build the bridge from the other side, as are things like MSIs 2x MXM socketed GPU cards.
THe traditional concept of how things fit together is slowly exploding, there is alot of flux in the tech industry at the moment as people try different ideas and it'll be another 4 years or so before things get sorted. The sheer size of cooling solutions required today (and needing 2 full cards for SLI!) is a major issue. I think people are accepting that as par-for-course far too easily. Clock-speed in GPUs isn't necessarily as beneficial as having lots and lots of parallelism. 2 gpu cores (or 2x the pipelines) at 300Mhz would consume (*waste*) far less energy while delivering similar performance to one core running at 600Mhz. Both solutions *can* fit on one card/package, but the wider core would consume less power at the lower frequency. Therefore, the PCB real-estate required for excess power circuitry could be reduced, allowing *more* cores to be integrated.
Likewise, reducing the voltage will also result in an exponential drop in power consumption.
This is kicking off now in multicore cpus (Thank you Alpha EV8) on the desktop as they started hitting frequency walls too often. It's more a matter of a change in CPU design (TLP Vs ILP) and programming style (Single Vs Multi-threaded) and fitting as many cores on dies as you can and running them at moderate clocks. The Cell is an example of this, though the SPIs are not general purpose procs.
Jesus. That's a post and a half.
On-Topic:
If you're going SLI, there's no reason you *have* to have dual-slot cards. Watercooling is still an option - if even just for the GPUs, and there will always be single-slot solutions available.0 -
Nice post syxpak
First of all it was crazy to release the PPU on PCI.
*example taken from forums.amd.com*
Who is more likely to buy it?
- Granny, who has a PCI only motherboard
- Jimmy, who is on s754 with his AGP 9500 and can barely run Doom 3
- Xavier, who just got done buying 2 7900gtxs and already has his two PCI slots filled with a sound card and TV tuner:D
This is the epox SLI board, you can still use 2 PCI slots and the PCI-Express 1X when running sli.......
And I think CPU/GPU/PPU should all be installed in sockets on the mainboard and cooled one large cooler with a 140mm fan
Excuse my paint diagram, but PCs should be designed like this:
P.s SLI is a waste on money If you dont already have one card..............
A single high end card is always cheaper, quiter, cooler and has the option to add a second one later........ 0 -
Advertisement
-
Advertisement