Okay, I got put in charge of a project my company is competing in. All of our national branches is setting up for a distributed "number crunching contest". Basically it is to advertise out company name by ranking high as we can get.
Each branch has been given a budget of $1200.00 to deign a system, and I'd really like my machine to be number one in the company. Now, I'm not asking anyone to deign a system for me, I got that covered. I've just been reading some mixed reviews, and would like some advice from other experts.
The basic process of the data to compute falls into two categories:
1) computing a single unique token as many times as possible, as fast as possible
2) computing a unique token, and comparing the results to dozens of individual predefined values, as fast as possible.
Stamina is important too! The machine must be going for one week at 100% CPU power. I plan on overclocking a slight bit, to squeeze every last CPU cycle. No worries, I've got cooling covered, since it will be in an environmentally controlled network operations center with dozens of critical boxes, and I'm experienced with overclocking cpu's and gpu's with reliable results.
One other consideration: Yes, I plan on taking advantage of the ability to use Cuda H.W. acceleration, as defined in the rules of the contest. Keep in mind, scores are given for CPU cycles as well as total CPU and GPU cycles. So, while Cuda acceleration is going to be big, I need to squeeze as many CPU cycles as possible. So just buying a dirt cheap CPU and spending the entire budget on the fastest Fermi GPU wouldn't help much. Just for the record, I'm overclocking SLI'd GTX 460'ss or GX2's (depnding on what ebay has to offer). I'm not asking for anyone to recommend a GPU.
Here are the candidate CPU's:
1) i5 2500 Sandy bridge - quad core
2) AMD 3.6 GHZ Phenom 4 Black Edition - quad core
3) Q9550 12MB L2 cache - quad core
I threw the Q9550 in the mix because I feel it would really take advantage of the 12MB cache when testing a computed token against the predefined results. Other than that pure, fast CPU cycles is what I need.
P.S. this is running on Windows 7 Pro 64-bit. So the O/S has been designed for optimization with multiple cores. Some of the operations in the distributed client application have been optimized for multi-core CPU's, but generically. So, an AMD six core, or i7 optimization is only built into the O/S. And I do not want to get an i7.
Oh, almost forgot; yes, I've considered an i7, but feel since the distributed agent is not optimized for multi-threaded cores, I'd be better off spending around $200.00 on the CPU, to put the money saved into GPU cycles.
TIA, for all recommendations, and advice!
Each branch has been given a budget of $1200.00 to deign a system, and I'd really like my machine to be number one in the company. Now, I'm not asking anyone to deign a system for me, I got that covered. I've just been reading some mixed reviews, and would like some advice from other experts.
The basic process of the data to compute falls into two categories:
1) computing a single unique token as many times as possible, as fast as possible
2) computing a unique token, and comparing the results to dozens of individual predefined values, as fast as possible.
Stamina is important too! The machine must be going for one week at 100% CPU power. I plan on overclocking a slight bit, to squeeze every last CPU cycle. No worries, I've got cooling covered, since it will be in an environmentally controlled network operations center with dozens of critical boxes, and I'm experienced with overclocking cpu's and gpu's with reliable results.
One other consideration: Yes, I plan on taking advantage of the ability to use Cuda H.W. acceleration, as defined in the rules of the contest. Keep in mind, scores are given for CPU cycles as well as total CPU and GPU cycles. So, while Cuda acceleration is going to be big, I need to squeeze as many CPU cycles as possible. So just buying a dirt cheap CPU and spending the entire budget on the fastest Fermi GPU wouldn't help much. Just for the record, I'm overclocking SLI'd GTX 460'ss or GX2's (depnding on what ebay has to offer). I'm not asking for anyone to recommend a GPU.
Here are the candidate CPU's:
1) i5 2500 Sandy bridge - quad core
2) AMD 3.6 GHZ Phenom 4 Black Edition - quad core
3) Q9550 12MB L2 cache - quad core
I threw the Q9550 in the mix because I feel it would really take advantage of the 12MB cache when testing a computed token against the predefined results. Other than that pure, fast CPU cycles is what I need.
P.S. this is running on Windows 7 Pro 64-bit. So the O/S has been designed for optimization with multiple cores. Some of the operations in the distributed client application have been optimized for multi-core CPU's, but generically. So, an AMD six core, or i7 optimization is only built into the O/S. And I do not want to get an i7.
Oh, almost forgot; yes, I've considered an i7, but feel since the distributed agent is not optimized for multi-threaded cores, I'd be better off spending around $200.00 on the CPU, to put the money saved into GPU cycles.
TIA, for all recommendations, and advice!