[tonguesomewhatincheek]

OKAY! Who's the that decided CUSTOMER SERVICE has to be an OXYMORON!!! It's NOT an adversarial contest. It's not like I expect to be nor am I into I just want some actual service that's right.

[/tonguesonewhatincheek]

You wouldn't believe the latest on my laptop. I think this is going to be one for a letter to an executive. We'll see how it plays out. I had to chuckle earlier today when I called the phone company about my DSL. After asking for my phone number, they actually told me that there was a known outage in my area, all without me talking to a human. Hmm....that puts them up to about the 1970's service-wise I think. The part that made me laugh was the part of the recording that mentioned that all of their call centers are in the US. Guess their off-shoring them prior to that didn't work quite as well as expected!

My USD .02 ...
The white paper is simple enough, but they gloss over why they want to solve QUBO problems and how that applies to what their stated goal is (machine learning). I didn't chase down all the references. Maybe it was described better elsewhere. The AQUA simulations will have to be validated by the hardware, which, if they get it working (properly), won't take long.

This all sounded somewhat familiar. Superconducting computer hardware was researched in the late '70s and simple working prototypes were built. IBM was trying to build and commercialize a Josephson junction processor, but ran into a large number of practical problems. A lot has changed since then and D-Wave may be more successful.

As far as multithreading ... it doesn't look like the BOINC support is quite ready for prime time, judging from the amount of activity in that area. My quick read of what's going on leads me to the conclusion that they will let a multithreaded application monopolize a system if it wants to because BOINC doesn't really recognize multiple threads. This could cause CPU thrashing if you tell BOINC to use all cores (real and HT) and one or more of the WUs is multithreaded. If you don't want to run any other BOINC project then maybe it isn't a problem. However, if you have just a single core and no HT, it'll run slower, not faster. I think AQUA has jumped before they looked. Bottom line: I don't like the idea, at least not yet and maybe not for some time.

- D

And conversly they are taking 80% of the time on my 8500GT than a 9800GT.



Because the GPU usage is not constant and fluxuates widely over the length of the wu.


Thanks ... a very good description. Multithreading aside, I've noticed something on the GPU side. Wu's built after 06/01 are now coming into the queue as 1.44 hours. Where I use to get 1, I now get 3. When one starts the estimate jumps to 30 hrs or longer. It may end up running more or less but now I have 3 long-running jobs queued even with the 0.01 day queue setting. If you add multithreaded CPUs, this could take everything and hang on to it for a long time. I agree with you, BOINC does not handle the long-term debt very well now. I don't think it will handle this to the benefit of other projects.

Anyway .... I'll refrain from conspiricy theories about this progression of events not being accidental.

Bottom line is all my cards are back at GPUGrid except the 8500GT. It can't do anything anywhere else right now.

What I don't like about GPUGrid is their really short deadlines, and that I can't use older cards. Other than that they're ok. :)

I just saw we're 198 on the boincstats?

Dr. Mike