d.net noob

[plonk420]

so, if it seems i'm not getting the performance i'm expecting, should i try running multiple instances? how do i do so? separate folders? (and set dnetc -config to allow multiple instances?) and to test it do i want to tell it just to crunch 1 unit?

[Brucifer]

you need to elaborate a bit more, like what are you running project wise, what o/s, and what hardware. What makes you think you aren't getting the correct performance? What versions of the client?

[plonk420]

RC5-72, win7x64, i7 920 (but it may get moved over to a PhII X3 705e) 5870. it uses no CPU, anyways. most recent client from site. i'm getting 400mKeys/sec, which SEEMS to be as many (or fewer) than what 4870 users are getting. maybe even 8800GT users, as well.

[NeoGen]

RC5-72, win7x64, i7 920 (but it may get moved over to a PhII X3 705e) 5870. it uses no CPU, anyways. most recent client from site. i'm getting 400mKeys/sec, which SEEMS to be as many (or fewer) than what 4870 users are getting. maybe even 8800GT users, as well.
Wait... You mean you're running dnet on one of the new ATI Radeon 5870's? :)

If so, do post a few screenshots for us please.:)

[plonk420]

Wait... You mean you're running dnet on one of the new ATI Radeon 5870's? :)

If so, do post a few screenshots for us please.:)

i may not entirely know what i'm doing, tho, but here we go:

http://forgottenwebsites.com/dnetc.png

note: they ARE running on sorta prerelease drivers, tho. "8.66RC6_Vista_Win7_Sep21"

edit: also, i still haven't gotten my email :(

edit2: it's definitely got SOME kind of black magic power (if not more shaders/tmus, etc): it whooped up on GTX285s on this nonstandard benchmark: 1080p (http://plastic-demo.nazwa.pl/wordpress/?page_id=266) 1080p8xAA (http://plastic-demo.nazwa.pl/wordpress/?page_id=266&prod=Into+The+Pink&res=Full+HD+%281920x1080%29&multisampling=MSAA+8x). note: benchmark isn't SLI/CFX friendly, if you try to run it.

mirror here if their site is slow: http://09-f9-11-02-9d-74-e3-5b-d8-41-56-c5-63-56-88-c0-forever.com/IntoThePink_final_by_Plastic_v1.04.zip

[NeoGen]

I was hoping you would post pics of the gfx card itself as it's the brand new 58xx series from ATI :icon_mrgreen:

But Brucifer might be the correct fellow there to answer your question about performance. The only thing I know is that with CUDA the dnet app takes 1 core of cpu usage for feeding the GPU, and the rest is the GPU working itself.
If the dnet app for ATI Stream does not use a cpu core for feeding input (you mentioned there was no CPU usage), then it's even better than the CUDA counterpart. :)

[plonk420]

haha .. maybe when i try to figure out if it will run on a ghetto/"budget" AM2+ board i'll take another pic. i'm a /b/tard (an (anonymous) occasional participant on the worst site on the internet) so i took a pic i'm not sure i feel like posting here ;)

[Brucifer]

Do not use a screen saver with the dnetc clients on windows as once the screen saver starts, the output will decrease to some degree. That is what I found as the reason why my output was lower in windows than what I had been getting under the linux systems. Once I disabled the screen saver and left the dnetc gpu client in the foreground the output went back up to what I expected it to be.

Yes, it will run on a ghetto budget AM2+ m/b. I've got one card running on that setup. I leave a cpu open for feeding the Stream client, but don't need to on the cuda client. I don't know if that issue has been fixed on the beta9 stream client. I guess I need to experiment with it. On the stream client (under beta8 anyway) you needed to set the priority to 2 I believe it was. I've got the stream client running by itself so I don't have issues for resource hogging.

In general, the cuda client seems to be more robust than the stream client. However the stream client is faster than the cuda clients. I've also noticed output differences between stream cards (4850) on different cpu/mobo combinations.

But if you are only running the distributed.net client once in a while, then I'd optimise for the project that you run on the bulk of the time.

I do not have the latest and greatest stream card to test run stuff on. And I doubt that I will get one for some time as there are other places in my budget that I need to put the money. Myself I like the 4850 cards as they don't eat me out of house and home and they run much much better than any cuda cards I have around here.

[plonk420]

mine's (maybe) more ghetto than YOURS! ;)

the PCI-E 16x slot blocks the 1x if a 2slot GPU is used (or if i'm misremembering, there may even be 0 adtl PCI-E slots). the mobo has 100mbit onboard, and my only gigabit card was PCI-E, so i had to take out the GPU with the big HSF to use gigabit to encode some video... in fact my two low end DX10 cards have 3rd party coolers, so i had to take out the GPU altogether. thankfully my friend is sending me some PCI gigabit cards. (or at least one).

even worse, there was a sticker covering the cpu socket when i opened the box "65 Watt TDP Max". the box said nothing of this. newegg said nothing of this (except for a few reviews after digging a bit).

ah well. you get what you pay for.

[plonk420]

oh weird, you DO need a free core, even tho the app uses 0% cpu (at least according to task manager. kinda wishing i had another i7 as it's now on my X3 lan gaming pc...

was getting ~40 million keys. shut down a core on boinc .. 300 million (with a newer client that last i used)

[Jason1478963]

I am running my 4850 on a 5000+ with WCG running on the extra CPU cycles. I'm not sure if this is a good way to do it but I just upped the priority of the dnet task to real time priority.

[plonk420]

any ideas as to why i still haven't gotten an email? i tried another account, too, a gmail one. still no email. spam folder is empty, too.

[AMDave]

I checked your email address and it appears to be ok.
I have initiated an email for you.
It should arrive in the next 24 hours (sometimes the stats have to go through a full cycle before their server kicks out the email for you after you press the button - depends on your timing - you could even get it in the next 10 minutes)

[plonk420]

thanks!

now to see how many points i'm getting. it's been really cold so i've been running it last few nights X)

on distributed.net, do GPUs do the same work that CPUs do? (within the RC-72 project)

[Jason1478963]

I am curious to see what you get compared to my ASUS eah4850. I was getting about 9000 blocks/points a day with it running 24/7 and an athlon x2 5000+ feeding it. I look forward to seeing what your GPU does.

[plonk420]

143rd place, 4,369 blocks, 18,764,712,116,224 keys yesterday.

but i'm not sure if i ran it all day

[plonk420]

33,118,492,819,456 keys yesterday (ran all day)
383,315,889 keys/sec
7,711 blocks
80th place for the day :D

[Jason1478963]

That's not to bad. I guess you only ran half a day before. Is your CPU able to keep the GPU activity pretty high or is it starving for a bit more work? I imagine its hard not to pull it away here and there for some test games and other activities. I've been looking forward to more AMD/ATI GPU based projects and the launch of a BOINC client that supports it.

[plonk420]

i'm actually wasting a core on it, sadly, even tho it's using probably not even 1% cpu... my experience is that unless you actually STOP a core, performance suffers (with other apps)


I've been looking forward to more AMD/ATI GPU based projects and the launch of a BOINC client that supports it.

i wouldn't mind a burst of seti on stream :0

[Jason1478963]

Have you set the affinity to one core for dnet and BOINC using the other 1 or 3? I wanted to keep both my cores running WCG so i sit the priority of dnet to realtime and this kept my GPU in the 98% activity range. This worked well for me as its just a cruncher and i don't use the desktop much. It did cause my GPU to run around 80C though. I've been watching for a koolance vid485 for it, but may have to look into other options.

I also would like to see WCG do something with stream. I'm not sure if that would work well for the type of research they are trying to do.

[plonk420]

7 seems pretty good at managing affinity. i've even heard claims that it'll manage an i7's affinity, too (where cpu 0/2/4/6 in HT mode will achieve the most work opposed to cpu 0/1/2/3), but i haven't seen any tests based on that. i'd ASSume Vista (and linux) are good at the former, too. not sure about xp. i've dumped xp like a bad date ;)

(unless modern (fast) 8gb SSDs come out at the $20-30 price point sometime soon)

edit: an i7 (with HT on) would probably be optimal for at least d.net... you'd probably only lose ~2-5% efficiency using 7/8 threads.

edit2: yesterday:
63rd place,
9,319 blocks
40,024,800,231,424 keys
463,250,003 keys/sec

[Jason1478963]

It does work on XP as I learned this trying to run UDagent with two instances on my dual core athlon x2. I have also used it with folding@home a few times on XP

[Jason1478963]

I think we should be able to squeeze some more out of that card if we can figure out whats going on. If your still interested in getting more out of the card on this project I think its possible.