Asus P6T6 WS Revolution Core i7 Motherboard

We started our high-resolution 3D game testing with the recently released Far Cry 2.  This time we have the resolution dialed up in the game to 1920X1200, the native resolution for our 24" test LCDs.  We also turned up all quality settings and anisotropic filtering for textures to maximum levels.  Benchmarks were then run with super-crisp 8X AA with Ultra High Quality enabled as well.  This is a torture test that will bring any graphics subsystem to its knees, unless you're running at least a pair of high end cards in tandem--though we were testing a trio of GPUs at once, in a 3-Way GeForce GTX 285 SLI configuration. 

Testing Three Full X16 Gen2 PCI Express For Graphics:
The object of our test methodology in this benchmark run and the ET: Quake Wars test on the following page, was to try and stress system bandwidth over PCI Express with as much GPU traffic as we could throw at it, at the maximum resolution of our test bench monitor, which is 1920X1200.  Running 3-Way SLI with the Far Cry 2 game engine at these extreme settings should accomplish that goal nicely.  Also, this time we put both the MSI Eclipse and Gigabyte GA-EX58 Extreme motherboards up against the Asus P6T6 WS Revolution.  Here we're going to look at just what three full X16 Gen2 PCI Express links can do for multi-GPU rendering (X16, X16, X16), versus a X16, X16, X4 setup on the MSI board and a X16, X8, X8 setup, which is more common for motherboards that don't incorporate NVIDIA's NF200 PCIe switch chip.  

However, remember we're talking about Gen2 PCI Express here, which offers a ton of bandwidth. Specifically, a single X16 connection offers 8GB/s (4GB bi-directional) of concurrent bandwith, while a X8 offers 4GB/s and a X4 offers 2GB/s.  So, in reality, the Asus P6T6 WS Revolution has 24GB/s of bandwidth available over PCIe for graphics transfers, the Gigabyte board has 20GB/s total and the MSI Eclipse, with its standard X58 Express chipset configuration, also has 20GB/s available to it.  Let's have a look...

The first thing to note here is that processing horsepower does affect performance in this test setup, for a couple of reasons.  First, there is a fair degree of inter-GPU transaction workload for the host processor to contend with and also, with all this GPU bandwidth available, even at this high resolution and these ultra high image quality settings, the test becomes slightly more CPU-bound.  As such, going from 2.8GHz to 3.2GHz on the Core i7 actually yielded a 6-7% performance gain on the Asus P6T6 WS Revolution and Gigabyte boards and a 9% performance gain on the MSI Eclipse.  Apparently, with the MSI board's lower available PCIe bandwidth in one of its slots, there are a higher number of PCIe transfers taking place that puts more of a strain on the processor, hence the enhanced benefit of a faster CPU.  For the Asus P6T6 WS and Gigabyte boards, a faster CPU offers slightly less advantage, since they can afford a larger transfer size with each transaction.  The net result is that both the Asus P6T6 WS Revolution and Gigabyte GA-EX58 Extreme, with a 2.8GHz Core i7 processor clock, actually perform as fast as a 3.2GHz processor on the MSI board, under this specific benchmark scenario.

The rest of the data here speaks for itself, under these high workload conditions, there is no real tangible benefit for 3-Way SLI with three full X16 PCIe links versus a boad that can support a X16,X8,X8 series of connections.  In fact, likely due to slightly higher latency going through the NF200 chip, the Asus board is a frame or two per second slower than that Gigbyte board.  However, it's perfectly clear that 3-Way SLI on a motherboard that supports something less than a X16, X8, X8 connection, is giving up some available performance to be sure.  Incidentally, MSI is reportedly going to offer the MSI Eclipse Plus motherboard, which will come equipped with the NF200 chip for full X16 3-Way SLI as well.