Rc5-72

[[XC] riptide]

Anyone still doing this? I know SkyCaptain is. I just fired up a client on CUDA again. The most recent client for CUDA is crap on Server 2008 R2. I'm getting like half of what I used to get. So I fired up an older client and it seems to be ok.

[Sky_Captain]

Read this, under the 27 Jul 2009 date as your completed work units from anything but the cuda 2.2 client will not be accepted and counted anymore.

http://n0cgi.distributed.net/cgi/dne...gi?user=bovine

So you have to use the cuda 2.2 client in order to continue crunching on rc5. I lost units from the cuda 2.1 client and learned the hard way. :-(

But like Bovine says, the cuda client still crunches work faster than the other clients, but at least the work they do now is correct.

[[XC] riptide]

Read this, under the 27 Jul 2009 date as your completed work units from anything but the cuda 2.2 client will not be accepted and counted anymore.

http://n0cgi.distributed.net/cgi/dne...gi?user=bovine

So you have to use the cuda 2.2 client in order to continue crunching on rc5. I lost units from the cuda 2.1 client and learned the hard way. :-(

But like Bovine says, the cuda client still crunches work faster than the other clients, but at least the work they do now is correct.

DAYUM!!!!! I did submit and old WU under the old client and got nothing for it. DAYUM! Sigh. So now I have to play with the new POS client. Great. Just ing great. And I was wondering why the blokcs were huge. They send out Dummy blocks.

[Sky_Captain]

You are supposed to be able to specify the size of the new blocks. I had experimented around with that but didn't have much luck. Under the cuda 2.2 client it has worked better. But I just decided to get one of the old 371 perproxy and run that on my net. It will only download 1 unit work units. And then I run my cuda clients against that. In the earlier days when the clients were running fast but not completing the multiple unit work units, the programmers were trying to cram the large ones down the gpu's throat saying that the setup time for getting ready to crunch the client took a lot of time, so that was the motivation behind the multi unit work units as it reduced the setup time overall as it didn't have to go through that on the multi unit work units. At the rate things run now, I don't see where a mili-second or two is going to matter that much cause it's taking a lot longer to crunch the units.

So all the folks that ran up the big points will be keeping them. But the points aren't going to run up so fast now. So while that's a bummer at least the project is getting good work out of the gpu's now.

What would be nice to know is how the ATI client is working in comparison to the cuda one. I'd go get one and give it a whirl now, but I'm a bit strapped with summer and vacations, etc. So it's going to be this late fall before I can try one of them out.

Which reminds me, I won't be submitting any work next week as I'm off on the road for the week and will be shutting my systems down as I don't have a computer literate babysitter around here. :-( But I'll be back after that. Actually I'll be shutting down Sunday morning.

Edit: And if you go to the nvida site to download the drivers, etc., they have the cuda 2.3 stuff out now. So I wouldn't be surprised to see distributed.net putting out a 2.3 client. I just hope they check that compiler out good before they put someting off the new libraries, etc., out to make sure that the completed work units are done correctly before they put the new client out.

[Sky_Captain]

DAYUM!!!!! I did submit and old WU under the old client and got nothing for it. DAYUM! Sigh. So now I have to play with the new POS client. Great. Just ing great. And I was wondering why the blokcs were huge. They send out Dummy blocks.

I saw your posts over at the Free-DC forum on the RC5 thread. :-) Yeah, the first thing I do with a new client is to run the -bench argument to see which core does the best since the default core selection has not been the best one for my gpu's for some time.

So you gonna stick with RC5 now or move on?

[[XC] riptide]

So you gonna stick with RC5 now or move on?

I dunno. There has been a lot of CUDA wastage... and I'm not worried about MY work.. but the other 1000'1 and thousands of CUDA blocks that have been done by others. Thats a lot of peoples electricity gone for nothing. I'll think about it.

[Sky_Captain]

I don't think that's gonna happen again though. The distributed.net programmers have undoubtedly learned that they are going to have to check the validity of and further client's completed work to make sure that there aren't any bogus issues creeping in. This issue was the fault of the nvida folks though as it was a compiler issue. Just one of those unfortunate things that happens, and this effort isn't the only one that has ever had tons of work having to be re-done again.

Well anyway, mull it over. I'm outta here for a week on another trip. Post whatever you are going to do. If you decide to hang it up on rc5, then I doubt I'll continue crunching it here as I'd be getting pretty lonesome.

[[XC] riptide]

RC5-72 just went ATI/Stream!!!. Anything above R600 (HD2000 series) works.

http://www.distributed.net/download/prerelease.php

[Sky_Captain]

Trying an ATI gpu out on it to see how it goes.

@riptide -- So have you decided whether or not you are going to put anymore effort into rc5?

[[XC] riptide]

For now, I'm doing this. http://www.xtremesystems.org/forums/...d.php?t=232049

[Sky_Captain]

okay, have fun.
I think I'll look around at other stuff.

[Sky_Captain]

I pulled my account out of the team so that I could retire two testing accounts into it. Then this evening after the stats run, I'll put it back in the team. Seems to be a little quirk in the system that the unassigned points don't show up in the stats until the new account is moved into the team so that those new points are then transfered into the account. Doesn't make sense I know.... :-) but it's what happens. :-)

Been messing around with the ATI cards. I have to say that I am really impressed with them. I picked up a HD 4850 to play with. That little puppy puts out some serious points a day, like in the neighborhood of 8,000 which is a teensy bit better than I used to get out of a cuda GTX 260. And the price is much less, and the current draw is much less. And the HD 4890's really rock on, and there is a dual one of those out there. It has me drooling, but I don't have the bux at the moment to invest in one of those yet. But that doesn't stop me from dreaming about it! :-)

So I think I'll keep tinkering around with rc5 on and off.

[MentholMoose]

Been messing around with the ATI cards. I have to say that I am really impressed with them. I picked up a HD 4850 to play with. That little puppy puts out some serious points a day, like in the neighborhood of 8,000 which is a teensy bit better than I used to get out of a cuda GTX 260. And the price is much less, and the current draw is much less. And the HD 4890's really rock on, and there is a dual one of those out there. It has me drooling, but I don't have the bux at the moment to invest in one of those yet. But that doesn't stop me from dreaming about it! :-)

What kind of power usage are you seeing with a client running on an HD 4850?

So I think I'll keep tinkering around with rc5 on and off.

Keep at it and eventually you'll match my 1.7M blocks, unless I add some GPU clients...

The thing I like about the RC5-72 project is that there is a clearly defined goal... eventually the key will be found (hopefully before the entire keyspace is tested) and the project will end.

[Sky_Captain]

Alpha at Free-DC posted the below:
Originally Posted by alpha View Post
Well, from this thread alone we can summarise:

HD 3650 - ..57.3 Mkeys/sec: 194? watts: 120 shaders
HD 4650 - 174.9 Mkeys/sec : ..55 watts : 320 shaders
HD 4850 - 489.1 Mkeys/sec : 110 watts : 800 shaders
HD 4870 - 589.5 Mkeys/sec : 150 watts : 800 shaders
HD 4890 - 680.0 Mkeys/sec : 190 watts : 800 shaders
HD 4890 - 740.0 Mkeys/sec : 200 watts : 800 shaders : Core Clock 950MHz : Memory Clock 1050MHz

So to answer your question a 4850 is running roughly 110watts, for 489.1 Mkeys/sec, which is nothing to sneeze at. And the card can be had for under $150 bucks. You can't put a cruncher together for that price, and get anywhere near the output on 110 watts... Now of course you have to add to that the power the host cpu system sucks down to support the gpu. As you can see there are hotter systems to run, but I prefer the HD 4850's. They don't cost an arm and a leg. You can run multiple cards and if one craps out you still have respectable output from the other cards. So I don't put all my eggs in one basket so to speak.

I crunch for multiple rc5 accounts, currently running 30,000 units a day. Presently sitting at 3.17M units crunched. The bulk of the units have come along since the linux cuda beta clients started last fall, and then I picked up some stream cards when they released the windows stream beta client this August. At present it would take me 54 days to catch you assuming I didn't get any more cards. :-) I doubt I will though as stream is still in the early beta. As for why I don't crunch more here, it's primarily because I just get bored crunching with only one or two people.

I too really enjoy the rc5 project. Even with the stream cards coming online, it is still a huge huge task to crunch even 1/3rd of the rc5 effort. But like you mention, who knows when the magic answer will be proven! :-) Crunch forth and prosper!!!

[MentholMoose]

Cool, thx for the info. I have a HD 4850 on the way, so I'll dedicate some crunching time on it!

[BatteryOperated]

Is there a team for xtremesystems? I am itching for a project to join up with.

[Snow Crash]

We have teams on lots of different projects ... currently we could use some help at the Einstein project.
What type of project are you looking for? CPU, ATI/ nVidia, bio, astronomy, physics?
Let us know if you need any details on signing up

[Paladin]

Yes, we have RC5 and OGR teams. Stats are down at the moment, but here's the team page link:

http://stats.distributed.net/team/tm...=27&team=31529

[Sky_Captain]

So they finally got the rc5 database back to people can edit their user info now. :-) So back to adding some points. :-)

[Sky_Captain]

Just slowly crunching away yet. Slowly coming after the #1 rc5 slot. <cheesy grin> I think I'll start putting some effort into the OGR effort too as I see some large dumps coming in from another soul!!!

[Paladin]

I had to hold off Skycrane's attack. :P