Milkyway@Home requires a GPU supporting Double Precision arithmetic.

NVIDIA:
Compute Capability 1.3 and Above
Example Products:
Geforce GTX 295
Geforce GTX 285
Geforce GTX 280
Geforce GTX 275 (credits to Bruce)
Geforce GTX 260
Tesla S1070
Tesla C1060
Quadro Plex 2200 D2
Quadro FX 5800
Quadro FX 4800
(Based on GT200 GPU)

ATI:
Example Products:
ATI HD Radeon 5970 (credits to kashi)
ATI HD Radeon 5870
ATI HD Radeon 5850
ATI HD Radeon 4890
ATI HD Radeon 4870
ATI HD Radeon 4850
ATI HD Radeon 4770
ATI HD Radeon 4830
ATI HD Radeon 38x0
(credits to cenit for the AMD documentation describing the products above)
ATI Firestream 9270
ATI Firestream 9250
ATI Firestream 9170
(credits to Cluster Physik)

Please post other working products and this list will be updated as time goes on.

you're forgetting fabulous ATi HD 4770.


edit: I'm still asking myself how ATi decided to not include DP on 57xx series...

edit2: this pdf from ATi
http://developer.amd.com/gpu/ATIStreamSDK/assets/ATI_Stream_SDK_Getting_Started_Guide_v2.0.pdf has all the data that you could need. Go to last page, section 4

GTX 275 works pretty good.
____________

Thanks for pointing out the 57XX cards.. looks like a 58XX card for me next!!

What about the current king of crunch?

ATI HD Radeon 5970

HD 57x0 series missing Double Precision is kind of shame.
I asked AMD about it and there is answer:

Double precision floating point operations are not required for any known or anticipated GPU compute applications for consumers (i.e. gaming, multimedia, and productivity), where single precision processing is sufficient. Implementing high performance DP FP processing in hardware carries significant costs in terms of die size and power consumption, and wouldn’t provide any value to the vast majority of our customers. This is the main reason why all of the companies who participated in the definition of the new industry standard OpenCL and DirectCompute APIs agreed to make this an optional feature.

That being said, there are some high performance computing applications in the scientific or industrial fields where double precision can be beneficial, so we chose to include that capability in our most powerful GPUs to provide an attractive solution for those markets as well.
- Antal Tungler

Please post other working products and this list will be updated as time goes on.

Geforce GT 240M

Getting the Message: Your NVIDIA GPU lacks the needed compute capability

No Mobile GPUs?
Only GTX?

Will there be an Update for those?

Presumably the message means that your card doesn't support Double Precision calculations - unfortunately, this project relies on that capability.

What is a normal (crunch)time for MW-wu using GTX295?

:/ pfff


17/02/2010 18:19:28 NVIDIA GPU 0: GeForce 8400 GS (driver version 19562, CUDA version 3000, compute capability 1.1, 256MB, 43 GFLOPS peak)
17/02/2010 19:15:06 NVIDIA GPU 0: GeForce 8400 GS (driver version 19621, CUDA version 3000, compute capability 1.1, 256MB, 43 GFLOPS peak)




17/02/2010 18:25:49 Milkyway@home Sending scheduler request: To fetch work.
17/02/2010 18:25:49 Milkyway@home Requesting new tasks for GPU
17/02/2010 18:25:54 Milkyway@home Scheduler request completed: got 0 new tasks
17/02/2010 18:25:54 Milkyway@home Message from server: No work sent
17/02/2010 18:25:54 Milkyway@home Message from server: Your computer has no ATI GPU
17/02/2010 18:25:54 Milkyway@home Message from server: Your NVIDIA GPU lacks the needed compute capability (1.3, required for double precision math

I'm getting 13 minutes per work unit on my gtx 295 (no overclocking or anything). ive turned in 10 GPU results for a 169 average per.
CPU is still waiting to finish even one WU per core :> 16% at an hour for an I7 920 slightly OC'd

12.5 to 13.25 minutes on a gtx260 no oc

Stock Performance stats:

GTX 260 Core 216 ~ 13 minute runtime
HD4850 ~ 4 minute runtime
HD5850 ~ 2 Minute runtime
____________
"Government big enough to supply everything you need is big enough to take everything you have."
-Thomas Jefferson
American Thinker

Hi Gang,
I'm running an ATI Radeon 9200, any reason why this is not up to specs ?



____________
Ivor
My Webpage

Hi Gang,
I'm running an ATI Radeon 9200, any reason why this is not up to specs ?

Hi Gang,
I'm running an ATI Radeon 9200, any reason why this is not up to specs ?

Too old. You need a double precision card.

Thanks for the info.
____________
Ivor
My Webpage

Anybody got WU run times for the new HD5830 (s/b on list I believe also)?

Trying to decide if I wanna upgrade a HD4850 with one of these.
____________
- da shu @ HeliOS,
"A child's exposure to technology should never be predicated on an ability to afford it."

Anybody got WU run times for the new HD5830 (s/b on list I believe also)?

Trying to decide if I wanna upgrade a HD4850 with one of these.

Anybody got WU run times for the new HD5830 (s/b on list I believe also)?

Trying to decide if I wanna upgrade a HD4850 with one of these.

Talk to me tomorrow as that is when it arrives at my house.

arkayn, what is your card running at? Mine's at 840/1100 and seems to be running around 2:21 to 2:23/WU.

-Dave
____________

Some 5870 performance numbers...

My OC'ed 5870 (900 core/1300 mem) runs normal wu in about 1:12. The recent shorties (yesterday) were generally 16 secs, some as fast as 10secs.

Manual fan speed ~3k rpm, gpu temp 75°C +/- 5°C, gpu load 99% as per GPU-Z,
ambient ~21°C, usually, stock cooler on the card.

Cheers!

arkayn, what is your card running at? Mine's at 840/1100 and seems to be running around 2:21 to 2:23/WU.

-Dave

My main ATI card is running MW WUs in 3 minutes 35 seconds (2 at a time)
____________
Go away, I was asleep

arkayn, what is your card running at? Mine's at 840/1100 and seems to be running around 2:21 to 2:23/WU.

-Dave

My card is running at stock speeds, but I do use the b-1 setting in the app_info.

Anybody got WU run times for the new HD5830 (s/b on list I believe also)?

Trying to decide if I wanna upgrade a HD4850 with one of these.

ATI HD Radeon 4730 also supports double precision. This is my card and it is 100% works.

yesterday i replaced my 4850 with a 5830. Boinc message says: 4850 .. 1020 GFlops, 5830 .. 1792 GFlops peek. It produces less noise and draws less power (a bit less).

yesterday i replaced my 4850 with a 5830. Boinc message says: 4850 .. 1020 GFlops, 5830 .. 1792 GFlops peek. It produces less noise and draws less power (a bit less).

What's the utilization on the 5830 if it's drawing less power? ATI claims that card requires two 75W PCIe power connectors (vs. one for the 4850) and it peaks at 175W.

If you want to reduce your power draw you can do it quite easily by lowering your memory clock. For example I'm running my 4870 at 725mhz core(gets unstable if i push it to default 750 at lowest voltage setting), and 250mhz memory at 1.083 voltage (by using ATI Tray Tools). I tested higher and lower memory configs, and they didn't change anything in computation speed. Apparently 250mhz (still 32gb/s transfer rate) is more than enough for supplying the core with data. By doing this I was able to significantly reduce my load power draw, as well as load temperatures.

Yes Madhalo, you are quite right. Reducing GDDR frequencies in many tapes of graphic cards significantly reduces power draw and temperature. This is most usable for VGA wich has GDDR5. For example my VGA is Sapphire HD4730 512Mb 625MHz/3600MHz. Power consumption of my "stupid" VGA is more than 75W in idle!!! In idle state radiator is so hot that I can't touch it! Using ATI Try Tools v1.6.9.1451 I can easily change both GPU and DDR frequencies. This is the measurements of power usage of my PC (without monitor) in idle state/in boinc with active MilkyWay tasks/time to take for 1 task to complete:
1. 625MHz/3600MHz - 266W/340W/6m30s Default clocks
2. 625MHz/4000MHz - 275W/352W/6m35s
3. 625MHz/2000MHz - 232W/308W/6m15s
4. 625MHz/1200MHz - 220W/289W/6m01s
So we can see that reducing GDDR5 frequencies is important)) I can say the one thing. Sapphire and maybe even ATI are stupid brands! Because the power consumption in idle state vor my VGA is very big!! I can't find bios which can fix this problem. Because power consumption more than 50-70W in idle for non top VGA is a problem!

Just a quick note, the rare HD4830 works great also.

Just a quick note, the rare HD4830 works great also.

ATI Double Precision Floating Point Support as per Geeks3D.

hello,

er is this all true ? or some sort of joke ?
i have a 4890 overclocked at 900 by 4400 standard cooled so noisy
if i just reduced the mem frquancy by half default or something the same work would be done as is being done now ?
only with alot less electricity alot less heat therefore alot less noise

so am i correct in thinking the GPU clock speed pretty much determins your work throughput and only overclocking that makes any differance

i come at overclocking video cards as a n newbie as for years i only ever overclocked machines the vodeo just had to sort of keep up with the system

and i know the more you bump the frequancy of your cpu/system up the more work it gets through i.e. my i7 920 with a modest clock of 4 gig across all cores and threads does a ton of work ( top 20 in climate prediction )

so is it just the frquancy that gets brought down and not the clock speed
if this is all true i'm amazed that its not alot more known about on this project or have i just completely missed the boat on this subject ( may very well be )

if anybody has any technical answers that would be great as i'm Founder of team " Scotland " and know a couple of others thinking the way i do

at present time of writing using standard app the wu are taking 2 mins 44 seconds to complete

best regards
Ian

hi,

ok after some testing i found after dropping the mem speed to 2400 from 4400 the card worked fine but alot less noise and heat and electric i'm hoping

i left the gpu overclock as was i.e. 900

so my new question if the gpu clock is is left where it was and the mem clock is almost halved what performace has gone ?

i mean are manufacturers running these mem chips at speeds that are pointless ? or is it just this project which tends to show this up ?

best regards
Ian

or is it just this project which tends to show this up ?

hi,

thank you i will run some dvd's through it see if it stumbles if so will increase the mem freq until it stops
i dont use hd or blue ray disks so it shouldnt be to much of a problem
i'm also not a gamer so that helps

thank you for your quick response
Ian

i mean are manufacturers running these mem chips at speeds that are pointless ? or is it just this project which tends to show this up ?

I was suggesting you to add the new entry ATi FirePro V8800 when I found that V8750 and V8700 were never added to the list even if they were listed in the pdf I linked...

oh, and ATi Radeon HD 5830 is working on MW@Home

Wow, you guys with ATI cards are posting fierce times. A WU for my GTX 285s take over ten minutes! I'm just using BOINC with no special configurations. Is there something I could be doing to speed this up (besides OC'ing, that is)?

NVIDIA GPU 0: GeForce GTX 285 (driver version 19713, CUDA version 3000, compute capability 1.3, 1006MB, 702 GFLOPS peak)
NVIDIA GPU 1: GeForce GTX 285 (driver version 19713, CUDA version 3000, compute capability 1.3, 1006MB, 702 GFLOPS peak)

Wow, you guys with ATI cards are posting fierce times. A WU for my GTX 285s take over ten minutes! I'm just using BOINC with no special configurations. Is there something I could be doing to speed this up (besides OC'ing, that is)?

NVIDIA GPU 0: GeForce GTX 285 (driver version 19713, CUDA version 3000, compute capability 1.3, 1006MB, 702 GFLOPS peak)
NVIDIA GPU 1: GeForce GTX 285 (driver version 19713, CUDA version 3000, compute capability 1.3, 1006MB, 702 GFLOPS peak)

Wow, you guys with ATI cards are posting fierce times. A WU for my GTX 285s take over ten minutes!

Actually for Mw you can downclock. Oc doesn't speed up the wus.

Take the GPU memory down as far as it will go - even 150 would not be too low - keeping it high just creates a better space heater and wastes power.

GPU clocks can go as high as an indiviodual's setup/personal preferences allow, higher GPU clocks will speed up the WUs.

Regards
Zy

ATI Radeon HD 5970 need 1minute 43second to complet 2WU (no overclocked)

New cruncher here!

Standard 5870 with mild overclock @ 900 core/1300 mem runs a wu in about 1:26 minutes. (could reduce my mem speed for MW, but I am also encouraging all ATI gpu projects! The more the better! And very nice work getting this GPU app running this efficiantly.)

I put my fan speed at 65% with a GPU temp of +/- 60°C to 64°C, case temp +/- 27°C. Ambiant +/- 21°C. Looking to keep the GPU nice & cool for the long run.

I am also running other projects on my cpu (sorry MW, but the difference is just not worth running it on my CPU!), i7 920 @ 3.2 with 8 treads loaded at 100%. So I might be losing a few seconds per wu because of it, but I figure it's a fair trade-off.


M.

Try the 5870 @ 930/300 - it will run fine, and cooler with less power, memory speed is irrelevant here, higher mem speed just creates a better space heater, save settings as a profile in CCC for switching between apps if needed.

If you are looking for a CPU app, the FP's over at Aqua are good when they have a run going. The next run is due in a couple of days, work keeping an eye on:

News on next Aqua FP Run for CPUs

Regards
Zy

To all you people with ATI cards: I really hate you right now. I just built a new el cheapo boinc cruncher with an Nvidia GTX260 because there is CUDA support for the majority of the astrophysics and other boinc projects I like to participate in. But the times you guys are posting for your ATI cards are just ridiculous. I'm assuming the WU's themselves are the same quantity of data as the ones for CUDA? Or are there differences in size that would make them process so much faster?

Tasks are the same for all platforms both here and at Collatz Conjecture.

DNETC@HOME has shorter tasks for NVIDIA cards, I thought this was by design but it may be a bug. The work processed is the same, it just processes less packets per task on NVIDIA. There are also longer DNETC tasks designed to be processed only on ATI Cypress class cards, although I think they may also get sent to NVIDIA cards but the CUDA application only processes a small amount of the total task.

To all you people with ATI cards: I really hate you right now. I just built a new el cheapo boinc cruncher with an Nvidia GTX260 because there is CUDA support for the majority of the astrophysics and other boinc projects I like to participate in. But the times you guys are posting for your ATI cards are just ridiculous. I'm assuming the WU's themselves are the same quantity of data as the ones for CUDA? Or are there differences in size that would make them process so much faster?

To all you people with ATI cards: I really hate you right now. I just built a new el cheapo boinc cruncher with an Nvidia GTX260 because there is CUDA support for the majority of the astrophysics and other boinc projects I like to participate in. But the times you guys are posting for your ATI cards are just ridiculous. I'm assuming the WU's themselves are the same quantity of data as the ones for CUDA? Or are there differences in size that would make them process so much faster?

To all you people with ATI cards: I really hate you right now.

To make you hate us more, the cards typically draw less power as well ...

For all the guys who have too much money left:

SAPPHIRE Toxic Radeon HD 5970, Full Retail, 4096MB
ATI Radeon HD 5970, 4096 MB GDDR5, aktive Kühlung
PCI-E 2.1 16x, 2xDVI (Dual Link), Mini DisplayPort
Dual GPU,GPU Takt: 900MHz; Speichertakt: 4800MHz,
2 x 256bit, Stream-Prozessoren: 2 x 1600
DirectX 11, Eyefinity, CrossfireX ready
€ 1.111,00

I am running an ATI HD Radeon 4870 X 2
Why can I only get 6 WU's in my que at a time?

When I do network maintenance I run out of work real quick. What a shame.
____________

MW operates with a rule that only gives 6 wu's per CPU core. As much as we have asked for more, Travis indicates that more would cause database problems as the size would grow accordingly and the server just couldn't cope.

I am running an ATI HD Radeon 4870 X 2
Why can I only get 6 WU's in my que at a time?

When I do network maintenance I run out of work real quick. What a shame.

And to add to the GPU list, the 5830 is missing. Unless what I've read is wrong on it doesn't support double precision FP calculations?

And to add to the GPU list, the 5830 is missing. Unless what I've read is wrong on it doesn't support double precision FP calculations?

It does support double precision, the list has just not been updated in a long time.

6 WU Per core? Why not per GPU?

I put the ATI HD Radeon 4870 X 2 in a single CPU core system just to run the GPU work. I have another ATI HD Radeon 4870 (Single) to put in a system. Maybe I should think hard about what to put it in now.

I still think it is a shame to waste available resources.

____________

And to add to the GPU list, the 5830 is missing. Unless what I've read is wrong on it doesn't support double precision FP calculations?

It does support double precision, the list has just not been updated in a long time.

Good to know for sure. Thinking of getting one though more for gaming than crunching. If I were buying for crunching I think I'd go NV again since CUDA is more widely used. But I've always liked the ATI drivers better :)

CUDA might be more widely used, but the ATI cards are thrashing them in the WU completion category.

The top 200 hosts all run ATI cards here, pretty much the same thing at Collatz.

GTX 480 is taking about 200 seconds more than a 5870 and about 100 seconds more than my 5830.

are the GTX 470 and 480 supported yet. I am running Boinic Ver 6.10.56
and it just ran through compution errors on all 48 Wu

I have suspend project for now

I've just acquired an s/h pc with a PCI 16 express slot, but due to it being a small form factor case, standard length/height cards wont fit. Are there any double precision half height cards out there that could be used to crunch MW? I can get an HD4550 but I think its only single precision.

Thanks
____________
Don't drink water, that's the stuff that rusts pipes

If you leave thee side off the case, can you get the full height HD4870 card in?
____________
Go away, I was asleep

No John, its too long as well. Also I cant update the psu as a standard one wont fit. I might put it to one side till single precision projects come out.
____________
Don't drink water, that's the stuff that rusts pipes

17/02/2010 18:19:28 NVIDIA GPU 0: GeForce 8400 GS (driver version 19562, CUDA version 3000, compute capability 1.1, 256MB, 43 GFLOPS peak)
17/02/2010 19:15:06 NVIDIA GPU 0: GeForce 8400 GS (driver version 19621, CUDA version 3000, compute capability 1.1, 256MB, 43 GFLOPS peak)


15/06/2010 23:44:54 NVIDIA GPU 0: GeForce 8400 GS (driver version 25721, CUDA version 3010, compute capability 1.1, 244MB, 43 GFLOPS peak)

:)

Has anyone used the ATI HD 4850 card on MW?
That seems the lowest priced double percision card that ATI
has that will work with MW. The 4770 seems to be costing more and
has 640 streams to the 4850's 800 streams.

I am looking for the lowest costing ATI card for GPU.
4670 is the lowest costing I would buy, but it does not work
with MW. So if I about double the cost to $100, then I can
go to 4850 card.

My GT220 card died a few weeks ago, so I was looking at getting a replacement
and for Collatz, ATI was the only choice for the price. I can get over 4 times
the credits going ATI for the same price as the GT220, so now I am looking for
the lowest costing ATI that works with the most projects.

I am running an AMD Phenom 9650 quad core @ 2.3 GHz, with only 1 PCIe x16 slot.
It does have the smaller x1 slot though. I have 4 gig ram and 1 TB internal and 1TB external. I run Ubuntu 64-bit Linux. I could run 64-bit XP, but need the free DVD authoring and video editing software, since on a fixed income I cannot afford the software prices. That is why Ubuntu Linux.

I am not a gamer and run GPU only for BOINC. My monitor max out at
1368 x 768 as a widescreen 16:9 monitor. I would love a bigger one, but
that was all I had the money for. I am really limited with spare money to
buy computer parts and such, so I need to get the best items for the lowest
prices I can. $100 is the most I can spare each month for non monthly bills
or food. Very limited.


ALSO
for anyone who is a programmer of C++ on Linux, please tell me what
is the easiest C++ system on Debian/Ubuntu to learn and use. Also
If there is an easier programming system, I would like to know. I
wish to relearn programming skills after 2 strokes took much of it away.
I used Pascal more than C/C++, but want to learn the easiest, most
powerful language "system" I can find "FREE". I miss creating things
like Mainframe Accounting systems and dedicated language editor, like I
did for RPG, all those years ago before I went PC only. Now I just want
to relearn it for my own fun, since it was fun for me or I would not
have become a programmer all those years ago.

Let me know OFF the lists at seti "at" krackedpress "dot" com.

____________

Has anyone used the ATI HD 4850 card on MW?

You should be able to do about 75k daily with a 4850, I got about 60k when I was using a 4830 on MW.


____________

If you have an AGP graphics slot PC, then a Radeon HD3850 AGP will also do double precision. This card will give an output of about 25K to 30K.
____________
Go away, I was asleep

THE NEW HD 5670 WILL ALSO PERFORM DP MATH.

Not from what I read, it will use the Juniper cores which are in the 57xx series.

The 57xx does not do double precision.
____________

Hi,

if somebody has the feeling that his system is a bit too slow, try this:

ARES/2DIS/4GD5
intel centrino2
Limited Edition , Unlimited Power

- Dual Radeon HD 5870 with 4GB GDDR5
- Performance King~~32% faster than generic Radeon HD 5970

http://uk.asus.com/product.aspx?P_ID=bGf2f0s2zeWhT0w8

Happy crunching!

Does the new Geforce 460 do double precision? What kind of performance should I expect from it?]

Bill

Yes it does, but there is not currently a fermi app for Milky way so you will need to wait a little while..

AFAIK they are working on one..

Does the new Geforce 460 do double precision? What kind of performance should I expect from it?]

Bill

"Yes it does, but there is not currently a fermi app for Milky way so you will need to wait a little while.."

So there is no way to tell the app that it's an older card and force it to work like the ATI 58XX cards, like you can with Seti@Home?

BTW, thanks to all for the prompt answers.

Bill

I've just acquired an s/h pc with a PCI 16 express slot, but due to it being a small form factor case, standard length/height cards wont fit. Are there any double precision half height cards out there that could be used to crunch MW? I can get an HD4550 but I think its only single precision.

Thanks

THE NEW HD 5670 WILL ALSO PERFORM DP MATH.

What about my ATI Radeon HD 4600? It's not DP? Does that mean MilkyWay@home can't use it at all?

What about my ATI Radeon HD 4600? It's not DP? Does that mean MilkyWay@home can't use it at all?

Just curious, what is the difficulty in recompiling the software to support newer GPUs like the GF104 or other cards that have double-precision floating point operation support? Why the big delay? I'd consider Fermi support to be important considering the massive performance increase it brings. That is the whole point of projects like this; to crunch a vast amount of information as quickly and efficiently as possible, is it not?

Even if on the back-end you guys are refactoring your programs to use OpenCL for better cross-hardware and cross-platform support, wouldn't it be worth it to update the old code to accept newer cards as a stop-gap solution? My GTX 460 can literally chew through anything I can currently throw at it without breaking a sweat. I'd love to see how it handles MW when it has WUs that don't just get discarded because the binary doesn't think it has a Compute Capability 1.3 or newer card (the GF104 actually has Compute Capability of 2.1).

It looks like they are currently working on the nbody-simulation, this requires their full attention. I started with them 7 hours ago and have ~40 wu's ended with errors.
As Travis posted a couple of days ago, they are working on an gpu-app for the simulation as well http://milkyway.cs.rpi.edu/milkyway/forum_thread.php?id=1875&nowrap=true#41548.

Alexander

As I've learned from gpugrid, openCL works fine on nVidia, but is not yet optimized for the ati-cards. And it runs only on 58xx cards.
Are these infos true for the mw-apps as well?

Regards,
Alexander

Even if on the back-end you guys are refactoring your programs to use OpenCL for better cross-hardware and cross-platform support, wouldn't it be worth it to update the old code to accept newer cards as a stop-gap solution?).

It would actually be less work to just finish the OpenCL. I haven't been able to get the existing CUDA to build since I started working on the project. I actually expect to have the OpenCL actually working later today, and ready to send out within a week or 2.

Here's a PDF describung what he's talking about.. (GPUGrid experiments with OpenCL... epic fail)
Experiences porting from CUDA to OpenCL -> http://www.cse.scitech.ac.uk/disco/mew20/presentations/GPU_MattHarvey.pdf

Hello!
I have ATI Radeon HD5570 & it's a great card!
It's a waste that boinc doesnt use it to calculate...!!!
Please advice how can I make Boinc use it as well.....

Only all 58XX and up cards have double precision.
To begin with the 5830 but i think not many will buy it since for a bit more you can buy a 5850 which performs much better, and has more streaming cores.

5830 - 1120
5850 - 1440
5870 - 1600
5970 - 3200

Its sad that Ati decided not to implement double precision on the 5700 serie cards but i guess it has todo with the cost reduction for manufacture.

My good old 4770 still does well at processing its most of the time faster then my old 4830
____________
Its new, its relative fast... my new bicycle

Only all 58XX and up cards have double precision.

I still dont get why this project doesnt support Single Precision arithmetic as well..

Maybe there are other pointless

Mapping THE Milkyway Galaxy requires a significant level of precision, to, i think, more than 16 places of decimals.

The way computer single precision is implemented will only give around 11 or 12 decimal places. this means the science cannot be carried out, unless the hardware is capable of double precision.

So, the GPU implementation of this project is only for ATI and nVidia cards capable of double precision, and the ATI GPUs have had this facility for several years. Even the Fermi nVidia GPU implementation has the double precision capability limited (crippled).

I am sure I will be heavily corrected by more knowledgeable posters, but I think this is the gist of the case?
____________
Go away, I was asleep

Mapping THE Milkyway Galaxy requires a significant level of precision, to, i think, more than 16 places of decimals.

The way computer single precision is implemented will only give around 11 or 12 decimal places. this means the science cannot be carried out, unless the hardware is capable of double precision.

So, the GPU implementation of this project is only for ATI and nVidia cards capable of double precision, and the ATI GPUs have had this facility for several years. Even the Fermi nVidia GPU implementation has the double precision capability limited (crippled).

I am sure I will be heavily corrected by more knowledgeable posters, but I think this is the gist of the case?

Double precision binary floating-point format
Double precision binary floating-point is a commonly used format on PCs, due to its wider range over single precision floating point, even if at a performance and bandwidth cost. As with single precision floating point format, it lacks precision on integer numbers when compared with integer formats. It is commonly known simply as double. The IEEE 754 standard defines a double as:

Sign bit: 1 bit
Exponent width: 11 bits
Significand precision: 52 bits (53 implicit)
The format is written with the significand having an implicit integer bit of value 1, unless the written exponent is all zeros. With the 52 bits of the fraction significand appearing in the memory format, the total precision is therefore 53 bits (approximately 16 decimal digits, ).
Source:
http://en.wikipedia.org/wiki/Double_precision

Maybe there are other pointless

Actually, Fermi's solution to double precision is much less crippled than ATI's approach to it. One thing Nvidia did cripple somewhat was its capability in the consumer line of Fermi cards due to the lack of ECC by reducing the double-precision to 1/8 that of their pure Tesla GPGPU Fermi cards. The performance in benchmarks shows that even that it's stated that the performance has been cut to 1/8, benchmarks show that is really only around 1/3 of the amount of performance loss. It all has to do with memory.

Only all 58XX and up cards have double precision.

I find this an unfortunately incomplete statement.

All ATI HD48xx and HD38xx series cards have double precision implemented and can successfully crunch Milkyway work quite readily.

When the server runs constantly, I have had an HD3850 crunching Milkyway work for over a year.

Bad news for Laptop owners with ATI graphics:
most of the ATI Mobility HD cards do not support doubles. This is even true for the new Mobility HD 58xx models.

(See ATI Knowledge base)

The only exception seems to be the "Mobility Radeon HD 4800" Series.

Both my ATI Radian 5830's work.
Just depends on what type of Motherboard one has.
The xfx 630i motherboard with a Hd 5830 does a milky way in 1Min53sec with a mild overclock.
The Intel 31dg with a Hd 5830 does a Milky way in 2Min00sec with a Mild overclock.
Both running at 68-69c.

Should I crank it up a little more on the overclock???

I am not sure but it seems to me that the cypress models all overclock well and easy
The way i overclock is to load furmark let it run at 1024x768 windowed in test mode not as benchmark and then watch the average/max fps
And i look for artifacts or errors, while i crank the card up with msi afterburner, there are other tools but i hate them ( traytools/atitool/riva tuner )
Now the core overclock is pretty straight forward you crank it up till you see either artifacts or when the machine freezes or reboots, when it does lower it like 10mhz from the point it crashed and try again, if again happens you found your limit on the core.
The memory on the newer cards is somewhat different since these cards correct themselfs you watch the fps of furmark.
Everytime i crank the speed up with 10 mhz and keep my eye on the fps if they keep going up your on the right track but if they become equal or lower then you found the limit. again lower 10mhz and try again if same result your found the limit. Now on most cases i test after i set these settings i start try major benchmarks for a day or 2 to see if the machine stays stable.
If it does remain stable then most overclockers do a step back 10 to 20 mhz and keep that as the final setting.
There seem to be ways to crank up voltage on these cards also, but thats not my thing (since there is a big change of blowing up your card)

Overclocking does give some minor increase of performance, to be honest they are really so small it hardly worth doing except for bragging to friends about the good syntetic benchmarks
____________
Its new, its relative fast... my new bicycle

Both my ATI Radian 5830's work.
Just depends on what type of Motherboard one has.

For the first time I have a gpu that can run this......time to go get a mil.

As far as I can see there is no Linux client for ATI GPUs. Is there any chance of getting that? I am currently running Collatz, but would love to do something more meaningful...

Hi, got a Juniper HD5770, Single Precision, though, also an 4850 which is
capable of doing Double Precision

It sure is a pitty, MW has no S.P. app.'s, but the 4850 does well, probably replace it for a 5850/5870 or 5900 series.
How much power they need to operate as I had some issues getting the GTX480 going, taking 15A on it's 8 pin, 15A on it's 6 pin connector and 5 or more through the Mobo. And they run quite hot, 85-95C (!)
According to spec's, that shouldn't be a problem and it isn't, only needed a 850W, instead of 650Watt.
____________

Knight Who says Ni

Those FERMI cards, just installed a GTX480 & GTX470 on an ASUS P5E with an QX9650
CPU, they are 'Double Precision', since it isn't mentioned in the spec's?

Anyway, they are fast, very fast.


____________

Knight Who says Ni

Those FERMI cards, just installed a GTX480 & GTX470 on an ASUS P5E with an QX9650
CPU, they are 'Double Precision', since it isn't mentioned in the spec's?

Anyway, they are fast, very fast.

They natively do double-precision FP operations and the GTX460 is rated at 907 GFLOPS. That's at the 675 MHz clock speed. I am currently running mine at 850MHz completely rock solid crunching Collatz WUs in the ~600 seconds/WU range. And it doesn't even phase Aero performance on Windows 7. Temp is currently at 66C with fan speed at around 70%.

I would argue that they currently blow anything ATI has out there away barring the dual-GPU cards. they haven't even introduced a "full" Fermi card that has all 512 CUDA cores enabled. They are working with TSMC to get the wafer yield up and power consumption. The GF104 was a giant leap forward in thermal efficiency, power usage, and overall performance. Let me put it this way:

"When this baby hits 88 miles per hour, you're gonna see some serious shit."

With regards to the ati/nvidia duel. My 5970 will do a Collatz on each core in 330 to 360 seconds. That's almost twice as fast as the 460. Yes I have mine sitting at 900MHz (stock is 750) without trying. I suppose I'm glad it's winter as it sits at 61/71°C (GPU0/1) with the fan at about 60% on Collatz. Temps go up by 4°C to 5°C with MW (84 sec/wu/gpu) and another 2°C to 3°C on dnetc.

My old 4870 does a Collatz in 800 seconds.

So I think nvidia have a ways to go before they blow ati away.

Sorry to disappoint you Mutiny32, but I agree with TGG on the current position with ATI cards and nVidia GPUs that can run any ot the following projects - Milkyway, Collatz and DNETC.

When I run Collatz on my current ATI HD5850 GPU the individual WUs are seen off in under 410 seconds. This GPU will see off a Milkyway WU in <110 seconds.

This card is not the top ATI card currently - with the HD5870 and the dual HD5970 able to crunch faster again. Again, come November the ATI band wagon will roll on to it's new architecture with the release of the HD68xx cards soon. Rumour has it that these will be in the order of 70% faster than the equivalent HD58xx series cards. But this is all rumour. Mind you, rumour also has it these will be seriously expensive as well.
____________
Go away, I was asleep

460 is a nice improvement on the nightmare NVidia are battling, but its a way way away from the claim being made.

A simplistic benchmark, but common one says it all:

Ultra 460 Benchmarking V ATI Cards

The card benchmarked is the fastest 460 on the market today, but still cant compete at the required level. A look at the chart for "standard" 460's shows the norm for the 460 class - which has a hard time against the low end 5830, let alone its direct competitor the 5850. Given that the 460 is in the new NVidia Fermi technology offering Suite, and the 58xx class is 2 + years old, there should be a much closer fight - there isnt, Fermi was leapfrogged by 58xx before it was even off the drawing board. As was said above the AMD (ATI)6xxx are due release 4th Quarter this year. NVidia are already behind, 6xxx will leave them for dead.

I hope Fermi2 - son of fermi - whatever they call it, improves markedly, else competition out there will be a one way street to AMD (ATI brand name has been dropped), such a lead is bad news for pricing. Therefore despite returning to the AMD (ATI) fold for the first time in 20 years, I furvently hope NVidia get their act together, but on current track record I am not holding my breath.

Regards
Zy

One area that nvidia does have on ati is the stability of their drivers when gaming. I don't have an issue when crunching but on my now superseded 4850 it would regularly freeze or cause a system reboot when my son was playing Star Wars Battle Front 2 or when I was playing Need For Speed. Haven't played these on the PC since I got the 5970 and updated the driver, but that was a frustrating period.

Well I don't know about the drivers for games but the ATI OpenCL seems poor to me. Doesn't work on the 2nd core of a HD 5970, it's their top product and it's still not supported yet after a few different versions of the drivers and SDK software.

I was going to give the PrimeGrid PPSieve ATI/OpenCL test version a go but no point if it only works on one core.

I have a Ati 5700 series (Juniper) and it doesn't support ? double precision math's? How come? And what's the solution? My 4800 series went down yesterday so I had to change. Any Ideas? Greetz. Meteor

to us when calculating for gtx 4xx????
____________
Team Alliance francophone, boinc: 7.0.18

GA-P55-UD5, i7 860, Win 7 64 bits, 8g DDR3, GTX 470

I have a Ati 5700 series (Juniper) and it doesn't support ? double precision math's? How come? And what's the solution? My 4800 series went down yesterday so I had to change. Any Ideas? Greetz. Meteor

Or DNETC
____________
Go away, I was asleep

Or DNETC

Very true.

Server software upgrades and other issues. Looks like no DNETC this week end or early next week.
____________
Go away, I was asleep

Thanks for that list of compatable gpu's, i am running stock milkyway app cannot seem to see is there anything else i can be doing with milkyway and a gtx470 or should i just use stock app.I will have to shop around for another gpu. Would appreciate any advice...Thank You
____________

lol lets not get onto the rumours about the new series 6xxx i would seriously not expect anything more then say 7 to 20% gain if they get released.( anything above this will be nice bonus)

I know for a fact that all manufactors do not give the full potential ever, since we want our pricy cards to run longer then the warranty says ;)

I have seen some test with cpu's from DEC which where simply destroyed by going to their limits, so they will allways stay well below that risk.

Nevertheless here we see an overclocked GTX480 doing about 110 seconds over a unit, while a normal clocked AMd 5870 takes about 85 seconds..

So even though its not bad for a nvidia you must consider that any 5870 does that cooler, cheaper, more silent and last but not least needs less power todo this then those nvidia cards

On Collatz we see the same the GTX 480 OC can show in the top list but has a huge list of AMD cards above them.
To be exact a 5870 does one unit in 5 to 6 minutes and a GTX480 OC needs 6,6 to 8 minutes for the same unit.

So i can safely say nvidia made a huge improvement on the processing side on these projects but still is not able to beat a simple 5870 at all
____________
Its new, its relative fast... my new bicycle

How to deal with MilkyWay@home server diagnostics:
"Result de_16_3s_2_835115_1279816531_0 is no longer usable"
"Result de_13_3x_2_1254273_128284363_0 is no longer usable" ?

Since January, 2010, I've processed almost 42 million on Milkyway@home, using two ATI 4870s running on an ASUS M4A79 Deluxe motherboard.
In Late August I converted from Vista64 to 64-bit Win7. I then started noticing the above diagnostics. I'm using the latest ATI and BOINC download software.
The message seems to be telling me that my GPUs are malfunctioning. However, Collatz Conjecture continues to process normally using the GPUs.
Question 2: Do we need a GPU diagnostic/validation utility?

Best Regards,
Rich
____________

I have a very unique card EVGA GTX275 w/ integrated GTS250. This is a card that has two separate physical double precision capable processors on the same card. Normally the GTS250 (GPU1) will not support double precision if on a card by itself. But in this instance because it is intregrated with the GTX275 (GPU0) it is. You may want to add this card to your list for those who have this type of card but have not designated the GTS250 to crunch GPU WUs.

Normally, I only crunch SETI, but I was doing a test by downloading some MW tasks for timing purposes. I noticed that all of the downloaded tasks were GPUs and when they started crunching is when I discovered that both GPUs were crunching on MW; instead of one on MW and the other on SETI.

I am wondering what is wrong since there has been no announcement of any change. I have been using ATI 5870 and 5890 cards with no problems, crunching with GPU only. Suddenly all my cache ran dry and I am not getting any new work, yet the Server status indicate there is plenty of work.

Yeah, my 5850's not getting any work either.

It seems to be not just ATI cards. My NVIDIA card isn't getting work either.

Funny, my NVIDIA card is recognized by BOINC and MW and my client keeps requesting work for it with no error messages. For reference, my GPU is a notebook card (GeForce GT 300M) that does not support DP and only has 256MB anyway.

It works for PrimeGrid and supposedly for Collatz, although Collatz gives computation errors. Einstein won't run because of insufficient memory...

I'm looking into a Quadro 600 (brand new as of, like, yesterday) for my home desktop...Not only do I want to crunch and help more, but I'm sick of games made past 2004 crashing CONSTANTLY.

That said...wonder if I can somehow make my ATi RADEON X800 SE crunch...something.

Riddle me this?
I have a ASUS G73JH with a ATI HD5870 card that BOINC reads as a 5700 series card.
ATI GPU 0: ATI Radeon HD5700 series (Juniper) (CAL version 1.4.515, 1024MB, 1120 GFLOPS peak)

MW gives me the
Milkyway@home Message from server: No work sent
Milkyway@home Message from server: An ATI GPU supporting double precision math is required
Milkyway@home Message from server: Your computer has no NVIDIA GPU

Running Catalyst Ver. 2009.1217.1632.29627 and after I tried to update Catalyst to the most current version I locked so I am stuck with the version I have.
Downloaded the 64 bit .23 mW apps for 64 bit W7 and copied them after stopping (completely) BOINC. When run I produced WU in 2-4 seconds that were all errors. This happened to around 10 WU then I stopped it and retried the installation with the same results on another 10 WU's. deleted all and am just using the CPU at the moment.
Any ideas?

Anybody else getting this message. I just attached this host to DNETC and now I see this.
"10/12/2010 7:23:06 PM Milkyway@home Message from server: No work sent
10/12/2010 7:23:06 PM Milkyway@home Message from server: Your app_info.xml file doesn't have a version of MilkyWay@Home N-Body Simulation.
"
Mo idea what's going on here. Please help.
____________

4870 GPU
4870 GPU

When will nvidia cards be suported by this project.
I have been unable to run since my upgrade from a 275 to Dual 470's sli

Thanks

I get a message now that i cannot get more work units from Milkyway as my system is not supported. i run a MAC, power PC G4 1.6 Ghz and 1 MB memory. The graphics card is a Nvida GEOFORCE2 MX

This started a few days ago. Does this mean i am out of luck with future work units?

I want to know so i can remove Milkwway and dedicate it's space to some other application.

Right now i run SETI, Einstien and World Community work units. At one time i did Rosetta but it's requirement went beyond my systems capabilities

gfcisme
____________

Thanks for the info. I did not want to take it off too hastily. I will leave it on for a few weeks. Does not create a problem for me. My other work units will just have more time for themselves.

Hi Anthony,
You can update your list:
GTX480
GTX470
GTX460
are working with app MW_0.24_CUDA
Thanks to Crunch3r, who compiled it.
heinz
____________

Further to my problem. I was getting a message for all my work units that the cpu was overloaded and so it would halt and keep trying again. Reloaded my OS again as i had another process error, but turns probably not related to my problem as i see the same thing with the previous version of boinc that i reverted to. (was screenreaderd not responding)

Then i went backwards from boinc 6.10.58 back to 6.10.21 which i had been using.

I do not get the CPU error with all the work units and they are running normal. I had not tried down loading 6.10.58, just reloading the original copy.

I think i will hold off updating until they bring out something official that's newer than 6.10.58

many thanks for your help

Gerry

As said before Milky way needs Double Precision cards. I've also found that its the number of shaders or stream processors that define the crunching power of the card, not necessarily the card model number. It also seems to matter more than the type of memory or clock and memory speeds.

3850/3870 - 320 shaders (£50-£60)
4850/4870/4890 - 800 shaders (£75-£120)
5830 - 1120 shaders (£140)
5850 - 1440 shaders (£180)
5870 - 1600 shaders (£250)
5950 - 2 x 1440 shaders
5970 - 2 x 1600 shaders (£420)
6950 - 1760 shaders
6970 - 1920 shaders
6990 - 2 x 1280 shaders

The above info from Card info

*** late edit added some prices - you pay for what you get!
____________
Don't drink water, that's the stuff that rusts pipes

As said before Milky way needs Double Precision cards. I've also found that its the number of shaders or stream processors that define the crunching power of the card, not necessarily the card model number. It also seems to matter more than the type of memory or clock and memory speeds.

3850/3870 - 320 shaders (£50-£60)
4850/4870/4890 - 800 shaders (£75-£120)
5830 - 1120 shaders (£140)
5850 - 1440 shaders (£180)
5870 - 1600 shaders (£250)
5950 - 2 x 1440 shaders
5970 - 2 x 1600 shaders (£420)
6950 - 1760 shaders
6970 - 1920 shaders
6990 - 2 x 1280 shaders

The above info from Card info

*** late edit added some prices - you pay for what you get!

First of all I'd better confirm that I'm no expert here, just an ordinary end user! From the table on the website I quoted from,

5970 - clock 725, memory 1000
6990 - clock 850, memory 1200

Historically the xx70 and xx90 cards just had higher clock/memory speeds than the xx50 cards, the shaders staying the same. This seems to have changed with the 5000 and 6000 series cards. It will be interesting to compare real life results of the 5970 against the 6990, with relevance to the price.


____________
Don't drink water, that's the stuff that rusts pipes

Nope...

It's been pointed out to me that I missed off two cards on the compatible list

4770 - 640 shaders (£80)
4830 - 640 shaders (£70)

Plus also this is a 6870 thread, and some of my last posts might be better placed in the GPU Requirements thread. If a mod wants to move them there that is fine with me. Thanks.

____________
Don't drink water, that's the stuff that rusts pipes

Hi Anthony,
You can update your list:
GTX480
GTX470
GTX460
are working with app MW_0.24_CUDA
Thanks to Crunch3r, who compiled it.
heinz

It is not the same, Crunch3r made a fermi app that does work, but you will have to use the anonymous platform instead.

http://milkyway.cs.rpi.edu/milkyway/forum_thread.php?id=1987&nowrap=true#42940
____________

It Appears that my GPU is not up to the job of crunching for Milkyway@home as it does not seem to be capable of Double Precision arithmetic.I just had a message to that effect. ( ATI Radion 4570 XT.)

This is a shame as I had wished to continue running the application on my machine.

I think it is rather ironic that projects that depend on home computer users for their existence and sucessful fuctioning, start to get fussy over what computers can run their precious projects.

Why don't you have some WU's available for running on systems that do not have
Double Precision arithmetic capabilities.

I might as well abort my existing WU and bid Milkyway@Home farewell, after all there are still plenty of projects that I can run.


____________

They did try to build a single precision app, it either too 4 times as long or did not validate.

They decided that double-precision was the best way to go for the project.
____________

.... Why don't you have some WU's available for running on systems that do not have Double Precision arithmetic capabilities.

It is not the same, Crunch3r made a fermi app that does work, but you will have to use the anonymous platform instead.

http://milkyway.cs.rpi.edu/milkyway/forum_thread.php?id=1987&nowrap=true#42940

Will the 6800 series be put on the supported list soon?

Also I see a lot of talk about CUDA development in another thread. Would it not make more sense to develop using OpenAL or DirectCompute so that the card used is not really an issue as much as proper driver support?

Matt,
Even tho I primarily am cruching with ATI GPU's I do have a PNY GTX480 laying here.
It is yours for your R&D if you want it!
All I need is an address to send it too.

Seriously,

David

Work is being done on OpenCL one, which is ready to replace the CUDA one for Nvidia cards, but I'm still working on speed for ATI. After that, there's the small issue that I have a fairly limited set of different hardware to test on; for example I don't have a Fermi GTX 4XX series around to test it.

It's actually OpenCL. OpenAL has something to do with sound. DirectCompute is limited to Windows only, and isn't a good solution. Work is being done on OpenCL one, which is ready to replace the CUDA one for Nvidia cards, but I'm still working on speed for ATI. After that, there's the small issue that I have a fairly limited set of different hardware to test on; for example I don't have a Fermi GTX 4XX series around to test it.

Found an
ERROR with an GTX480. No NOT this one, when I 'clicked'the link, I got :Unable to handle request?!

Since I have one myself, now crunching GPUgrid and Einstein and SETI waits to
install and run the new servers, soon I hope. I haven't tried a GTX480 & 470 on Milky Way. Curious if it beats an ATI 5870 and 4850, which I use now, with an Q6600, 650Watt PSU, is just enough to feed both cards and CPU, 550Watt, when used @ 100%, use (cold)air from outside to feed the intake fans.
Only useable if it's colder outside, most of the time, like the Netherlands :), can even make/replace a heater, only the fan noise, can be too 'present'.



It seems like the cards Compute Capabillity is wrong, it has to be 2.0, not 1.3.(OK for GTX260 and up)
____________

Knight Who says Ni

I tried my 460 on MW, it was coming in at 12:30 a unit, way slower than my 5830 and slower than the 4830 that it replaced.
____________

Is Cluster Physik involved in the ATI app development?

No. He disappeared before I got here, and never gave anyone the source. I don't understand why it was ever deployed without the source, but in any case we don't have it, and Cluster Physik has disappeared and stopped replying to messages so it needs to be replaced.

I'm one of the donors and one of the user in the community always asking for some news and updates from scientists behind the projects.
Anyway, this is a great article, but it doesn't stress enough on the merit of the community. If we had not been so active, milkyway wouldn't be second fastest distributed computing project.
Nowhere in the article is cited the great work by Andreas Przystawik (PhD), aka Gipsel/Cluster Physik in the community. Without his work, GPGPU on milkyway would be a dream and 1 petaflop would be far far away, so the researches at Rensselaer institute.

I have a new NVIDIA GTS 450 w/1024MB DDR5. BOINC reports it as follows:
'Tue 16 Nov 2010 11:19:53 AM EST NVIDIA GPU 0: GeForce GTS 450 (driver version unknown, CUDA version 3020, compute capability 2.1, 1023MB, 421 GFLOPS peak)'
This card doesn't appear to be recognised although according to NVIDIA, it has double precision capability. Is it an application issue?

Thank you

Hi, will there be support in the future for the new ati hd 6850 and 6870 series?

Hi, will there be support in the future for the new ati hd 6850 and 6870 series?

Hi, will there be support in the future for the new ati hd 6850 and 6870 series?

Hi, IU was just wondering if the FireGL V7700 would work with Milyway?

Thanks!

This pages says it does support double precision which means that it will work on MW.
http://www.beyond3d.com/content/news/613
____________

I see, thanks a lot.

Why are there not GPU WU's being dragged down right now?

Just wanted to know why my FirePro v5800 isn't use by Milkyway. It seem to be recognize as ATI Radeon HD5700.

(Windows 7 64 bits, Last driver version 8.773)

Is it normal ?

Thanks
____________

The FirePro V5800 is based on a Radeon HD 5770 (Juniper XT RV840) clocked at 700MHz so as to remain under the 75 watt PCI Express bus limit and thus not require a power connector.

As far as I know Juniper class cards do not have the double precision capability necessary for MilkyWay.

I just had the wrong drivers installed.

Needed the FireGL CCC and not..."other" CCC.

So the V5800 would run faster than the V7700?

Glad you got it working Black_Jac.

I don't think Philippe85 will have the same success as I believe his FirePro V5800 does not support double precision. With suitable drivers, FirePro V5800 should work for single precision projects such as Collatz Conjecture and DNETC@HOME.

I'm getting a new Nvidia GTX460 card tomorrow for my @home computer and was wondering if Milky Way will detect it and it's double-precision capability when it starts up or will I have to do something like detach/re-attach to the project.

The old cards although Cuda-capable did not support double-precision.

Thanks for any help that you can provide.
Steve

I'm getting a new Nvidia GTX460 card tomorrow for my @home computer and was wondering if Milky Way will detect it and it's double-precision capability when it starts up or will I have to do something like detach/re-attach to the project.

The old cards although Cuda-capable did not support double-precision.

Thanks for any help that you can provide.
Steve

Thanks for the information.

I should have mentioned that I'm running 64-bit Ubuntu 10.10 (the current version). Is there a Linux version?

Thanks,
Steve

I think you should be fine then as it was only on Windows that we had the issue.
____________

Ok. Great.

Thanks for the information.

Regards,
Steve

Ok, now I have a new problem.

When I first installed the @home apps a few weeks ago with my old Nvidia cards, all of the apps could see the GPUs.

Then, because I was having trouble getting Einstein to run, I did a complete detach, re-install, and re-attach. After that, none of the apps see the GPUs. The two apps that should use the GPUs are Cosmology and MW.

I'm running 64-bit Ubuntu (10.10) with the latest Nvidia driver. I've got a monitor attached to each graphics card and am using Xinerama to span the desktop.

Also, I've modified the config.xml file to "use all gpu's" option as suggested in the FAQ and I've run out of ideas.

At the moment, I'm thinking about setting the machine to not accept any new work,finishing the existing WU's and then detaching and re-attaching to each project.

Does anyone have any other suggestions for getting the GPUs recognized and engaged?

Thanks for any help that you can provide.

Steve

Your problem is BOINC does not see the video card so does not know to accept work for the GPU.

The user boinc is not included in the video group and needs to be added to it.

http://boinc.berkeley.edu/dev/forum_thread.php?id=3564

That should have hopefully relevant info for the problem.
____________

Thanks for the help.

The link had the information that I needed. BOINC now sees the GPUs.

Thanks again.

Steve

Could someone clarify things for me?

I have 2 Nvidia GPUs. GPU0 is a Fermi card, while GPU1 is a single-precision card (GTS8400) running under 64-bit Ubuntu on an AMD quad core.

Now that I've got everything running with the newly installed Fermi added, I see on the task screen that when MW downloads WU's, it will momentarily show the work unit and a notation that the it is using 1 cpu core and GPU0 or 1 core and GPU1. Then the lines disappear. I am downloading and processing WU's of some type, but I'm curious as to what kind.

A check of the Messages screen only shows a line reading "Requesting new task for GPU." The WU download lines say that I' using MW version 24, if that helps.

I am getting Arecibo Binary Pulsar WUs that stipulate they are using the Fermi.

Is Einstein monopolizing the Fermi such that MW decides to send me WUs that don't require the Fermi?

If so, is there anything that I can do to even things out?

Having said all that, it's great to have this crunching along.

Thanks for any help that you can provide.

Steve

Is there a wiki page or link that would have information on how to allocate @home tasks between multiple machines?

I've got my machine working from spares and it's a pretty decent machine. Now, I've got the bug to build a serious @home machine ("more power", aargh!). It will be a multi-core running under Ubuntu, with at least one Fermi card.

The projects that I'm interested in are:
Einstein (single and double precision wu's)
MW (ditto)
Cosmology (single)
Orbit (I think just single)
Seti (just single on Linux, has both on Windows)

My questions I'm sure have been addressed before if someone can just point me in the right direction. I haven't found it in the FAQs but I may have missed it.

Questions like, "should I concentrate the Fermi cards and double-precision projects on one machine and the single-precision tasks on the other?" Or does it matter? Those sorts of questions.

Thanks for any help that you can provide.

Regards,
Steve

You listed some projects you are interested in, great, however overall is the motivation to compete re credits (as such) or is there a driving motive on a particular genric type of project? Reason for asking is if this beast is being designed solely as a Cruncher, the over-arching motives need understanding in order to help you design it in the best possible way.

Questions like, "should I concentrate the Fermi cards and double-precision projects on one machine and the single-precision tasks on the other?" Or does it matter? Those sorts of questions

Zydor,

Thanks for the feedback and suggestions.

At the moment, I'm going to have to put the idea of another machine aside since I just found out that I am not returning good double-precision WUs. All of the units are getting rejected for errors. When I checked the website I found the following:

<core_client_version>6.10.58</core_client_version>
<![CDATA[
<message>
process exited with code 1 (0x1, -255)
</message>
<stderr_txt>
Device index specified on the command line was 0
Looking for a Double Precision capable NVIDIA GPU
The device GeForce GTX 460 from the command line cannot be used because a device supporting compute capability 1.3 (Double Precision) is required
Found 1 CUDA cards
Found a GeForce GTX 460
Device cannot be used, it does not have compute capability 1.3 support
No compute capability 1.3 cards have been found, exiting...

</stderr_txt>

The GTX 460 is 2.1 compute capability. It's got to be backward-compatible! I wonder if the app has "1.3" hardwired into the code and that's why it's not working?

I previously had an additional 8400GS card in which I physically pulled so it's just the Fermi now.


I'm no programmer but I'm beginning to wonder if it's a driver issue. I thought the one (the "restricted" driver that Ubuntu installed) was the latest but now I'm wondering. I see where Nvidia has a newer driver so I may try that one. Also, I noticed that when Boinc starts up and correctly sees and describes the Fermi card, it also says "driver unknown."

I've just about hit the wall on this one.

Regards,
Steve

Matt,

Thanks, I definitely will.

As an aside, a little while ago, I found a message thread over at Seti with someone having the exact same problem with the exact same error message about his Fermi card not being 1.3-capable.

Regards,
Steve

"As long as we've got computers, who needs enemies!"

I got this when I tried to run the NVIDIA GPU Wu's on my M8700 GT ???

stderr out

<core_client_version>6.10.58</core_client_version>
<![CDATA[
<message>
Incorrect function. (0x1) - exit code 1 (0x1)
</message>
<stderr_txt>
<search_application> milkywayathome separation 0.48 Windows x86 double OpenCL </search_application>
Found 1 platforms
Platform 0 information:
Platform name: NVIDIA CUDA
Platform version: OpenCL 1.0 CUDA 3.2.1
Platform vendor:
Platform profile:
Platform extensions: cl_khr_byte_addressable_store cl_khr_icd cl_khr_gl_sharing cl_nv_d3d9_sharing cl_nv_d3d10_sharing cl_khr_d3d10_sharing cl_nv_d3d11_sharing cl_nv_compiler_options cl_nv_device_attribute_query cl_nv_pragma_unroll
Using device 0 on platform 0
Found 1 CL devices
Device GeForce 8700M GT (NVIDIA Corporation:0x10de)
Type: CL_DEVICE_TYPE_GPU
Driver version: 260.99
Version: OpenCL 1.0 CUDA
Compute capability: 1.1
Little endian: CL_TRUE
Error correction: CL_FALSE
Image support: CL_TRUE
Address bits: 32
Max compute units: 4
Clock frequency: 1250 Mhz
Global mem size: 519634944
Max mem alloc: 134217728
Global mem cache: 0
Cacheline size: 0
Local mem type: CL_LOCAL
Local mem size: 16384
Max const args: 9
Max const buf size: 65536
Max parameter size: 4352
Max work group size: 512
Max work item dim: 3
Max work item sizes: { 512, 512, 64 }
Mem base addr align: 2048
Min type align size: 128
Timer resolution: 1000 ns
Double extension: MW_NONE_DOUBLE
Extensions: cl_khr_byte_addressable_store cl_khr_icd cl_khr_gl_sharing cl_nv_d3d9_sharing cl_nv_d3d10_sharing cl_khr_d3d10_sharing cl_nv_d3d11_sharing cl_nv_compiler_options cl_nv_device_attribute_query cl_nv_pragma_unroll cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics

Compiler flags:
-cl-mad-enable -cl-no-signed-zeros -cl-strict-aliasing -cl-finite-math-only -DUSE_CL_MATH_TYPES=0 -DUSE_MAD=0 -DUSE_FMA=0 -cl-nv-verbose -DDOUBLEPREC=1 -DMILKYWAY_MATH_COMPILATION -DNSTREAM=3 -DFAST_H_PROB=1 -DAUX_BG_PROFILE=0 -DUSE_IMAGES=1 -DI_DONT_KNOW_WHY_THIS_DOESNT_WORK_HERE=0

Build status: CL_BUILD_ERROR
Build log:
ptxas application ptx input, line 58; error : Instruction 'ld' requires SM 1.3 or higher, or map_f64_to_f32 directive
ptxas application ptx input, line 74; error : Instruction 'ld' requires SM 1.3 or higher, or map_f64_to_f32 directive
ptxas application ptx input, line 75; error : Instruction 'ld' requires SM 1.3 or higher, or map_f64_to_f32 directive
ptxas application ptx input, line 177; error : Instruction 'ld' requires SM 1.3 or higher, or map_f64_to_f32 directive
ptxas application ptx input, line 178; error : Instruction 'ld' requires SM 1.3 or higher, or map_f64_to_f32 directive
ptxas application ptx input, line 179; error : Instruction 'ld' requires SM 1.3 or higher, or map_f64_to_f32 directive
ptxas application ptx input, line 180; error : Instruction 'ld' requires SM 1.3 or higher, or map_f64_to_f32 directive
ptxas application ptx input, line 181; error : Instruction 'ld' requires SM 1.3 or higher, or map_f64_to_f32 directive
ptxas application ptx input, line 182; error : Instruction 'ld' requires SM 1.3 or higher, or map_f64_to_f32 directive
ptxas application ptx input, line 183; error : Instruction 'ld' requires SM 1.3 or higher, or map_f64_to_f32 directive
ptxas application ptx input, line 184; error : Instruction 'ld' requires SM 1.3 or higher, or map_f64_to_f32 directive
ptxas application ptx input, line 185; error : Instruction 'ld' requires SM 1.3 or higher, or map_f64_to_f32 directive
ptxas application ptx input, line 198; error : Instruction 'ld' requires SM 1.3 or higher, or map_f64_to_f32 directive
ptxas application ptx input, line 199; error : Instruction 'ld' requires SM 1.3 or higher, or map_f64_to_f32 directive
ptxas application ptx input, line 200; error : Instruction 'ld' requires SM 1.3 or higher, or map_f64_to_f32 directive
ptxas application ptx input, line 201; error : Instruction 'mov' requires SM 1.3 or higher, or map_f64_to_f32 directive
ptxas application ptx input, line 211; error : Instruction 'ld' requires SM 1.3 or higher, or map_f64_to_f32 directive
ptxas application ptx input, line 212; error : Instruction 'ld' requires SM 1.3 or higher, or map_f64_to_f32 directive
ptxas application ptx input, line 213; error : Instruction 'ld' requires SM 1.3 or higher, or map_f64_to_f32 directive
ptxas application ptx input, line 214; error : Instruction 'ld' requires SM 1.3 or higher, or map_f64_to_f32 directive
ptxas application ptx input, line 215; error : Instruction 'ld' requires SM 1.3 or higher, or map_f64_to_f32 directive
ptxas application ptx input, line 216; error : Instruction 'ld' requires SM 1.3 or higher, or map_f64_to_f32 directive
ptxas application ptx input, line 219; error : Instruction 'mov' requires SM 1.3 or higher, or map_f64_to_f32 directive
ptxas application ptx input, line 220; error : Instruction 'mov' requires SM 1.3 or higher, or map_f64_to_f32 directive
ptxas application ptx input, line 221; error : Instruction 'mov' requires SM 1.3 or higher, or map_f64_to_f32 directive
ptxas application ptx input, line 230; error : Instruction 'mul' requires SM 1.3 or higher, or map_f64_to_f32 directive
ptxas application ptx input, line 231; error : Instruction 'mul' requires SM 1.3 or higher, or map_f64_to_f32 directive
ptxas application ptx input, line 232; error : Instruction 'mul' requires SM 1.3 or higher, or map_f64_to_f32 directive
ptxas application ptx input, line 233; error : Instruction 'mul' requires SM 1.3 or higher, or map_f64_to_f32 directive
ptxas application ptx input, line 234; error : Instruction 'add' requires SM 1.3 or higher, or map_f64_to_f32 directive
ptxas application ptx input, line 235; error : Instruction 'mul' requires SM 1.3 or higher, or map_f64_to_f32 directive
ptxas application ptx input, line 236; error : Instruction 'mul' requires SM 1.3 or higher, or map_f64_to_f32 directive
ptxas application ptx input, line 237; error : Instruction 'mul' requires SM 1.3 or higher, or map_f64_to_f32 directive
ptxas application ptx input, line 238; error : Instruction 'add' requires SM 1.3 or higher,
clBuildProgram: Build failure: CL_INVALID_BINARY
Error creating kernel 'mu_sum_kernel': CL_INVALID_PROGRAM_EXECUTABLE
Error building program from source: CL_INVALID_PROGRAM_EXECUTABLE
Error creating program from source: CL_INVALID_PROGRAM_EXECUTABLE
Failed to setup up CL
15:00:59 (3704): called boinc_finish

</stderr_txt>

____________
STE\/E

Matt,

As far as I can tell, I properly installed things to Ubuntu and I re-set the project, however I'm still downloading tasks that say cuda23.

The one new task that I completed was still bad with the same message about needing a double-precision card, even though it's a Fermi.

Should the work units still be marked as cuda 23or should I have un-installed the boinc client 6 before extracting the tar?

Thanks for any help that you can provide.

Steve

Okay Thanks, I didn't really think it would work anyway but tried them for the Heck of it, guess I'll have to get some cards that work ... lol
____________
STE\/E

One question, maybe someone got it before;
Is it possible to run a 5850 combined with a 5870 in one Box crunching?
I got a 5850 and can borrow a 5870 for some weeks now. But only have one Box for multible Cards, MSI 890FXA-GD70.
____________