Whilst in the past CUDA support would get my vote I think instead of a company specific language which blocks ATI users the best thing to work on would be OpenCL support which works on both ATI & Nvidia Cards, though I suggested in some time before the seti dev's didnt think the GPU could do much work for them whilst I disagreed nowdays it does the majority of it amusingly enough (as evidenced by the oldest post on such things).

Obviously the benefits vary due to the limited nature of calculations possible on the GPU vs a CPU however given how rapidly GPU's can do what calculations they can do if any part of your project can work out how to use it I'm sure it would benefit, and given the whole point of @home style projects is to leverage maximum burning potential it's an aweful shame you have no GPU support of any kind yet.

Especially seeings as how with SETI unable to provide any work units nowadays due to frankly mismanagement, poor planning not enough coordination, incompetence, lack of caring about their userbase or a combination therein, I'm looking to consider swapping my crunching project, but alas I have 3-5 nvidia gpu's in 3-4(or 5 with ati) (depending on when I have the 4th on(3 run 24/7) pc's(and an ATI's but they've not been in use) but how can I swap to a project that would ignore a fair bit of my crunching potential, otherwise known as under utilised and I'm swapping to avoid that kinda thing.

---

I think that the SETI@home problems are caused by people using GPUs. They process the WUs so quickly that they overload the system's capabilities, which were not designed for such processing speed. Other projects like AQUA, well run and well financed by a commercial firm, have renounced to the GPUs for this reason. I have no GPU and am running 6 projects (CPDN included) on my Opteron 1210 with Linux, and also SETI on a virtual machine running Solaris on top of Linux. I am never out of work.
Tullio

---

I don't know of any climate modelling project anywhere (CPDN isn't the only one) that uses GPUs intead of CPUs yet. Whether GPUs can be used to process work seems to depend on the type of calculations that have to be done, and I don't think there's any likelihood in the near future that climate models will be able to use them.

Just small differences in how the models are compiled for the three OSs can have large effects on the stability of the models and the validity of the results they produce. At CPDN the researchers pore over the data generated by alpha and beta model versions and decide what can be used and what can't. They don't care one bit, to tell you the truth, about the speed of the processing. All they care about is receiving valid data. In consequence, almost all the data CPDN models produce can be used by the researchers.

Sometimes when CPDN models are 'optimised' it entails slowing them down to reduce the occurrence of processing errors. This had to be done with the HadCM models (not being used at the moment) and again with the FAMOUS models we're processing now.

As I see it, the speed of processing doesn't necessarily correlate with the value of the results produced. Some of the most valuable results in medical research have emerged from longitudinal studies where data is collected and collated for years or even decades. The most important factor is often the design of the studies.

You will probably need to look at the possibility of using your CUDA cards on the projects that can make use of them, the CAL cards on other projects, and the CPU cores still available after that on projects like CPDN.

CPDN is more than happy for members to spread their crunching capability over several projects so that it can all be used to the full.

---

Cpdn news