Building a GPGPU System

Our simulations and visualisation work depends heavily upon having fast and efficient computers. Some of our recent experiments have used computer systems based upon hybrids of multi-core CPU processors with auxiliary GPUs.

The following photo history illustrates the construction of a typical GPU-based system. Click on the quarter-size image below to access full-size versions.

The collected new components, prior to assembly. This will eventually be a Quad-core Intel CPU, 3-GPU, 4GByte Memory system running our simulations under Linux.

Empty case ready to accomodate the new components. Although our prototypes do not need a lot of disk rack space, we have planned for the future and have chosen a case that will be big enough to accomodate the rather large GPU cards in the PCI-Express slots. More air space also aids air flow for cooling.

Dan fitting parts into the case. We have used a Silverstone mid-tower cases for our prototypes. It was a temptation to succumb to having see-through sides to the case, but this design seems to have good cooling properties with an extra fan at the top.

Close-up of the Motherboard (nFORCE 780iSLI - 3-way SLI). We have used an Intel Q8200 quad-core processor running at 2.66 GHz for this prototype system.

Fitting the first of the three GPUs into the motherboard. This is an NVIDIA GTX260 Black Edition GPU. The motherboard can accomodate three such GPUs and they can be connected with an "SLI" connector that supports fast direct data transfers between GPUs without going through the normal bus sub-systems on the motherboard.

The motherboard and Power supply in situ. In a previous experiment we found that GPUs easily trip out a "normal" power supply so this system has a 1000 watt supply in an attempt to future-proof it.

Ready to power up for first test - apart from the tangle of cables that are still to be pinned back.

Much relief when the new system exhibits a "heartbeat" and "it lives!"

We are running Ubuntu Linux on this system. Our simulations are generally written in C or C++ or D or Java, although for the purposes of utilizing the GPU capabilities as a general purpose computation engine we use NVIDIA's CUDA software system.

[ Complex Systems and Simulations Group (CSSG) | Massey University | IIMS ]

	The collected new components, prior to assembly. This will eventually be a Quad-core Intel CPU, 3-GPU, 4GByte Memory system running our simulations under Linux.
	Empty case ready to accomodate the new components. Although our prototypes do not need a lot of disk rack space, we have planned for the future and have chosen a case that will be big enough to accomodate the rather large GPU cards in the PCI-Express slots. More air space also aids air flow for cooling.
	Dan fitting parts into the case. We have used a Silverstone mid-tower cases for our prototypes. It was a temptation to succumb to having see-through sides to the case, but this design seems to have good cooling properties with an extra fan at the top.
	Close-up of the Motherboard (nFORCE 780iSLI - 3-way SLI). We have used an Intel Q8200 quad-core processor running at 2.66 GHz for this prototype system.
	Fitting the first of the three GPUs into the motherboard. This is an NVIDIA GTX260 Black Edition GPU. The motherboard can accomodate three such GPUs and they can be connected with an "SLI" connector that supports fast direct data transfers between GPUs without going through the normal bus sub-systems on the motherboard.
	The motherboard and Power supply in situ. In a previous experiment we found that GPUs easily trip out a "normal" power supply so this system has a 1000 watt supply in an attempt to future-proof it.
	Ready to power up for first test - apart from the tangle of cables that are still to be pinned back.
	Much relief when the new system exhibits a "heartbeat" and "it lives!" We are running Ubuntu Linux on this system. Our simulations are generally written in C or C++ or D or Java, although for the purposes of utilizing the GPU capabilities as a general purpose computation engine we use NVIDIA's CUDA software system.