This week I'm taking a request. I'll be talking about the release of the SETI@home/BOINC CUDA applications and some of the backlash it has caused. Let's do this interview style, for clarity's sake.

Q. Well that didn't go quite the way you planned, did it?

A. Well, no, it didn't. But I have to admit I've seen worse software rollouts. The initial release of SETI@home comes to mind. I still hear complaints about that one.

Q. Then why did you do it?

A. For years we've been under pressure from various sources to get out a GPU version of SETI@home. When NVIDIA offered to do the port we jumped at the chance. The timing could have been better, though. We felt pressure to release when we did, partially because we had prearranged the date to coincide with the press release.

Q. NVIDIA gave you a big pile of cash to do it during the holiday season, right?

A. No. NVIDIA's total donations to SETI@home in the last year have been two graphics cards, with total value less than $500. Even if they had, my understanding of the University rules is that we cannot agree to provide specific services in exchange for a donation. If we did, it would stop being a donation and become a contract, which is subject to entirely different rules. The big thing NVIDIA did was to develop the application. If someone from another graphics card manufacturer were to do the same for their cards, we would work with them.

To be completely honest, ATI donated a graphics card in 2007, but couldn't donate any resources to the porting effort. So an ATI port didn't happen at that time. If they've got a programmer with the time now ....

Q. Did you ask NVIDIA to make a donation?

A. Not in exchange for any work we did. Dan did query a contact of his at NVIDIA (unassociated with the porting effors) and was informed that current economic conditions made a donation unlikely.

NVIDIA has donated to SETI@home in the past. (IIRC, the last donation was in about 2003, although I would have to do a lot of digging to find out exactly when). It was around the time we had a student working with porting the preliminary version of Astropulse to GPUs. At that time we decided the limited graphics card memory and the asymmetric nature of the AGP bus limited the performance of GPUs to be less than that of CPUs. Things have changed since then.

Q. How do I know that NVIDIA didn't pay you under the table?

A. That's really a ridiculous accusation. They wouldn't be dumb enough to offer it, and we wouldn't be dumb enough to take it. They could have made an above-board donation, gotten a tax deduction for it, and additional accolades for being a good corporate citizen.

What it comes down to is that people at NVIDIA thought this release and the effort they put into it was valuable to NVIDIA. David felt that it was valuable to BOINC and the BOINC community. And I felt that, after years of being asked to develop a GPU application, it wasn't a bad thing for SETI@home. In hindsight I could have pushed back on the release date. It's too late to do that now. I could have also pushed David to have CUDA be an "opt-in" system, with CPU apps being the default. That can be fixed, and I'll probably push on it in the future.

Q. Are you going to fix the bugs?

A. NVIDIA's developer (and Jorden and Raistmer (thanks guys)) have been working on it since the release. If seti_cuda 6.08 works out in beta overnight, we'll probably release it late tomorrow. It should fix a lot of the problems.

Q. Can I ask more questions?

A. Feel free. But it's late, and I have to take Angela to a conference in Burlingame in the morning. I'll answer some of them tomorrow.

In support of full disclosure, I own a small number of NVIDIA shares, which I purchased in 1999. The total dollar value is far below the threshold that the University and the State of California have for reporting a conflict of interest. In fact when we did do a matching grant proposal for the NVIDIA donations (also circa 2003), I got laughed at for reporting it as a potential conflict of interest. Somehow I don't think SETI@home affects NVIDIA's stock price much.

---

@SETIEric@qoto.org (Mastodon)

Thank you very much for this Eric. It is nice to know that the issues we are experiencing are being looked at, and even nicer to know that I am not alone in experiencing them. It is nicer to know that a potential fixed client is on it's way.

To all my wingmen that suffered at the hands of the malfunctionning CUDA client and my possibly insane GPU, I am deeply sorry. We'll give it another go with teh new client, shall we?

Eric........

Thank you for the courtesy of your reply.

As I trust you implicitly, I will take your responses at face value and will assume that any other information I have been given is false.

I will only say that if the rollout had been suspended until the proper testing had been given time in Beta, the effort might have been greeted with very different attitudes by myself and many others.....

I hope that the difficulties that most have experienced while trying to make it work will be sorted soon.

And I hope, my friend, that this shall be a lesson learned when trying new additions to your project in the future...please keep it in Beta until it is ready, and don't release something because it is expedient...

I do not expect your project to be static....change can be a good thing, and I am sure as time goes on, there may be other approaches that you may want to try from a scientific perspective...and that is good.

But when you are enlisting the donated time of many thousands to assist you in your research.......keep the experimentation in Beta, where failure is an accepted part of participating there, rather than causing much dismay among your faithful followers by presenting an application that is not ready for them.

I hope you will accept my apologies for 'fanning the flames' so to speak, but I had to express my dismay when it was obvious that this Cuda app was NOT working when released for almost anybody....and was also causing disruption for those who were not involved with it, such as myself.

When the storm passes, the rest of the development work is done, and things are working properly, I may even play with Cudacizing a rig or two myself.

I still respect you and your project...

Best Regards,
Mark

---

"Time is simply the mechanism that keeps everything from happening all at once."

Q: Why CUDA App was pushed to S@H Main so fast (before cleaning most of bugs)? It has too little time for testing in Beta...
Mark was faster than me :)

---

@ Eric Korpela

Thanks for the FAQ.. :-)


It would be nice to have a BOINC-client which can manage simultanously Enhanced and Astropulse on CPU and Enhanced (maybe in future Astropulse also?) on GPU.
Or combined applications.


I'm a 'hardcore'-member.. ;-D
And I would like to buy a nVIDIA-GPU.. GTX 2xx series.. but I'm afraid because of the GPU if she's crunching 24/7.
So I can relax or you hear maybe somewhere that it could be 'dangerous'?

I thought about to buy an overclocked GPU from a manufacturer.

If the GPU (OCed or not) will after using not calculate correct, I can change it under warranty?


@ Astlor.ca.SETI@home

BTW.
BOINC is the client.
SETI@home-Enhanced and Astropulse are the applications.

---

We felt pressure to release when we did, partially because we had prearranged the date to coincide with the press release.

OK. Let's take Eric's post at face value, as Mark suggests, and ignore the rumours involving David Anderson's occupational opportunities at NVIDIA (which, by the way, would have made another nice Q for the Q&A).

Still, the statement I've quoted from Eric's post alerts me. The release date of an application supposed to generate vaild scientific data was determined by a forthcoming press release by a third-party company (*)? We know the science has been compromised by the rushed release because of the false -9s from the CUDA app, not to speak of the severe problems some people have experienced (BSODs and the like). All this for the sake of the schedule of a press release, of NVIDIA PR? Science has not only been compromised within the master database, the scientific approach of the whole project has become questionable.

(*) one week before Christmas - time enough for the most determined crunchers to put a high-end graphics card from said company on their last-minute wish list. But that's speculation.

(*) one week before Christmas - time enough for the most determined crunchers to put a high-end graphics card from said company on their last-minute wish list. But that's speculation.

OK........
The kittyman has posted his response to Eric's reply.......

Let's quit speculating....as much as it might make us wonder......

Let's get on with it.......
Either make Cuda work properly, or shut it down.

I think they shall work with the nvidia people to make it work.

Let's drop the reasons why it was a bad mistake to release it before maturity.....

---

"Time is simply the mechanism that keeps everything from happening all at once."

Learning from past errors is smart, repeating past errors is dumb.

Let's learn, get smarter and move on.

As the BORG would say:

CUDA will be repaired and assimilated, speculation is futile !!!

Eric,
Thank You for your posting and answering some questions. It is nice to hear from someone such as yourself who is looked upon with great respect from a lot of people here.

Thanks again!

. . . as usaul - Thank You Dr. Eric Korpela for Your Informative Posting - Much Appreciated Sir!

---

BOINC Wiki . . .

Science Status Page . . .

My thanks to you for taking time to answer the many questions. My thanks also to Raistmer and all that came out with the apps that helped me get this thing going on my computer. I'm looking forward to 608 and hope it overcomes the problems we are having so far.

Now if only someone can get the AP validator running again!

---

PROUD MEMBER OF Team Starfire World BOINC

Excellent post Dr Korpela, many thanks. Looking forward to seeing the app stable.

---

Nice.. Valuable input indeed.

Wonder what the new cuda version fixes..
I sincerly hope that crashes disappear :)

Kind Regards Vyper

---

_________________________________________________________________________
Addicted to SETI crunching!
Founder of GPU Users Group

Dr. K...thanks for the info...and I like your avatar...

---

Clk2HlpSetiCty:::PayIt4ward

Q. Is David Anderson trying to get a job at NVIDIA?

A. I highly doubt it. It's not really his area of expertise, so I don't know what he would do there. So far he's got enough grant money to pay his salary here. I don't see a reason for him to leave.

On a related note, someone mentioned that we wouldn't want to jeopardize our "tenured" positions here. Nobody on the SETI@home or BOINC staffs are tenured. We're research scientists, not teaching faculty. We jump through the same hoops that the faculty does to get tenure, but at the end of that process we don't get a guaranteed job/paycheck.

The process I went through to go from "Assistant Research Astronomer" to "Associate Research Astronomer" was similar to that an "Assistant Professor" goes through to become a tenured "Associate Professor." But at the end of it my paycheck is smaller than a Professors and I'm required to bring in enough grant and donation money to fund my own salary. If I don't I won't get fired, but I also won't get paid.

Q. How badly was the science damaged by the problems with the CUDA app?

A. About 3% of our results are generated by machines running CUDA and about 0.1% were from cuda results validated against cuda results. About 20% (by my survey of CUDA a fraction of CUDA results) of those were bad overflows (although they come in bunches that make the problem look worse than it is.), so we've lost about 0.02% [edited because of typo] of our science. Compared to our problems with radar leaking through, it's minor. But we may decide to reanalyze all of the overflow workunits from these dates if we start catching up the the rate data is coming in from the telescope. We also may want to do that for a lot of tapes when/if we are able to improve our radar rejection.

Q. If my GPU gets bad results or locks up my system, will it be replaced under warranty?

That's up to the company that made the graphics board. If it were my GPU I would certainly try to get it replaced. If it gets bad results in SETI@home, it might hang when you are doing something important like pwning n00bz.

Q. Was the release rushed through beta?

Yes. I thought I made that point in the original post. I underestimated the problems that were encountered in beta. Hopefully most of those are fixed in 6.08.

---

@SETIEric@qoto.org (Mastodon)

Thanks to Eric for his openness. I've worked on many technology development projects over the last 35 years, and the balance between waiting to get it right, and getting good stuff out as soon as possible, is a delicate one. Erring in either direction can have undesirable results. Just look at at all the Shuttle "replacement" projects over the last 20 years, for example. The possible similarity between all the dead end replacement projects and something like the Cuda project is pretty obvious to me - if you keep waiting to get the new project perfect, some new technology (another GPU, or whatever) will come along and make it all obsolete before you deliver the product.

Or, as an old boss of mine used to say, if you don't royally @#$%-up once in awhile, you're not trying hard enough. He had a lot of other saying that applied to this situation, but the message board rules won't allow me to quote them.

---

@ Eric Korpela

Thanks for reply! :-)

I asked because of the GPU-warranty.. because normally the GPUs are for gaming and not for crunching.
So if one point of the screen make funny things it's maybe not so problematic and it's not a warranty-topic, but you can't crunch anymore with it.
I'll contact some manufacturer because of crunching-damage and warranty..


Because of simultanously MB/AP on CPU and MB (AP?*) on GPU..
The devs are on this, that it'll be possible in future?
(also with optimized apps?)

I'm a 100% SETI@home-Fan and don't want to crunch for other projects.. ;-)


[* We'll see in future an AP-CUDA-app also?]

---

As the BORG would say:

CUDA will be repaired and assimilated, speculation is futile !!!

I would never say that.

---

You will be assimilated...bunghole!

Q. Are you going to fix the bugs?

A. NVIDIA's developer (and Jorden and Raistmer (thanks guys)) have been working on it since the release. If seti_cuda 6.08 works out in beta overnight, we'll probably release it late tomorrow. It should fix a lot of the problems.



Will this be a general release or on GPU ratings memory size and board attributes.

i.e 8300GS and 8500GT at present auto excluded to download 6.08 On beta but after downloading files and with the help of claggy and richard I am now able to run.

Michael

Because of simultanously MB/AP on CPU and MB (AP?*) on GPU..
The devs are on this, that it'll be possible in future?
(also with optimized apps?)

That's the plan...

[* We'll see in future an AP-CUDA-app also?]

That is definitely the plan. At least one volunteer is working on it.

---

@SETIEric@qoto.org (Mastodon)