I have 32-bit fp GPU, is it good for collatz?
log in

Advanced search

Message boards : Windows : I have 32-bit fp GPU, is it good for collatz?

Author Message
Saul
Send message
Joined: 5 Apr 12
Posts: 8
Credit: 450,399
RAC: 0
Message 13759 - Posted: 12 Apr 2012, 3:48:55 UTC

A yes or no would do. I hope I can help, I have 2 Crossfired ATI HD4670s. I'm running collatz in both GPUs and it gives lots or aborted by project WUs, I'm reading Supported ATI Cards and Lot of ATI errors threads. The AMD Supported Cards says my card is supported so I met collatz system requirements but it also states clearly it does not support 64-bit fpops. So I ask if collatz have 32-bit fp WUs? or am I just incorrectly trying to crunch 64-bit WUs?

Profile mikey
Avatar
Send message
Joined: 11 Aug 09
Posts: 3245
Credit: 1,699,257,919
RAC: 4,977,369
Message 13762 - Posted: 12 Apr 2012, 10:13:40 UTC - in response to Message 13759.

A yes or no would do. I hope I can help, I have 2 Crossfired ATI HD4670s. I'm running collatz in both GPUs and it gives lots or aborted by project WUs, I'm reading Supported ATI Cards and Lot of ATI errors threads. The AMD Supported Cards says my card is supported so I met collatz system requirements but it also states clearly it does not support 64-bit fpops. So I ask if collatz have 32-bit fp WUs? or am I just incorrectly trying to crunch 64-bit WUs?


I believe the project will just send you the units your pc hardware can handle. meaning yes you should be okay. One thing I would do though is go into the project settings, on the web page, and set them so you only get the mini Collatz units, not the longer ones. This will mean your cards won't work as long on each unit. Also I did not see ANY workunits with errors, just lots and lots of them waiting to be crunched. You might think about lowering your cache a little bit so you contact Collatz a bit more often and don't come so close to any deadlines. If you are gaming with those gpu's then cross-firing is a good thing, if no then it does not help in crunching. If you are gaming have you told Boinc to stop crunching while you are gaming? If not then you are dieing waaaay too often in the game.

Saul
Send message
Joined: 5 Apr 12
Posts: 8
Credit: 450,399
RAC: 0
Message 13769 - Posted: 12 Apr 2012, 18:37:56 UTC - in response to Message 13762.
Last modified: 12 Apr 2012, 18:43:20 UTC

I really hope the project can tell which cards are 32-bit and which are 64-bit and can have WUs for both, I can't afford any hardware right now, if that is the case I'm glad.

The mini collatz WUs are the ones that abort the most. No problem with dead lines, I'm crunching WUs doe in 6 days from now.

I tell BOINC to stop collatz project when using a 3D game/app, because even in no-3D apps. it causes lag. Well I'm using both GPUs so if they are used by collatz with CrossFire or not that's OK with me as long as both my GPUs are used, maybe collatz could be accelerated a bit by using CrossFire/SLI, I think this project should explore that possibility.

Thx for the reply.

Profile mikey
Avatar
Send message
Joined: 11 Aug 09
Posts: 3245
Credit: 1,699,257,919
RAC: 4,977,369
Message 13776 - Posted: 13 Apr 2012, 12:40:49 UTC - in response to Message 13769.

I really hope the project can tell which cards are 32-bit and which are 64-bit and can have WUs for both, I can't afford any hardware right now, if that is the case I'm glad.

The mini collatz WUs are the ones that abort the most. No problem with dead lines, I'm crunching WUs doe in 6 days from now.

I tell BOINC to stop collatz project when using a 3D game/app, because even in no-3D apps. it causes lag. Well I'm using both GPUs so if they are used by collatz with CrossFire or not that's OK with me as long as both my GPUs are used, maybe collatz could be accelerated a bit by using CrossFire/SLI, I think this project should explore that possibility.

Thx for the reply.


CrossFire makes both gpu's act like one gpu, that doesn't make for faster crunching but it does make for MUCH faster gaming. Transferring the data between gpu's takes time and that makes for slow crunching. And YES projects CAN tell what kind of card you have and send you the best workunit they have for your particular pc/gpu.

Saul
Send message
Joined: 5 Apr 12
Posts: 8
Credit: 450,399
RAC: 0
Message 13783 - Posted: 14 Apr 2012, 3:46:20 UTC - in response to Message 13776.


CrossFire makes both gpu's act like one gpu, that doesn't make for faster crunching but it does make for MUCH faster gaming. Transferring the data between gpu's takes time and that makes for slow crunching.

I know what CrossFire does, so if it works for games it should work for GPGPU, since it's 5 Gbps CrossFire bridge accelerating processing, of course it takes time, every synchronization in the PC takes a few cycles to adjust the transfers but this is not different: it takes the minimum amount of cycles to synchronize, it should work for anything, hence my exploring suggestion.

And YES projects CAN tell what kind of card you have and send you the best workunit they have for your particular pc/gpu.

Is not always the case I recently added a protein project and it failed and restarted after a few seconds, after a few fails it took another WU, this is when you get a 64-bit WU into a 32-bit GPU. I removed the project of course, that's why I made this question here, I was afraid I'd be successfully and wrongly crunching a 64-bit WU with my 32-bit GPUs

Profile mikey
Avatar
Send message
Joined: 11 Aug 09
Posts: 3245
Credit: 1,699,257,919
RAC: 4,977,369
Message 13785 - Posted: 14 Apr 2012, 12:22:40 UTC - in response to Message 13783.


CrossFire makes both gpu's act like one gpu, that doesn't make for faster crunching but it does make for MUCH faster gaming. Transferring the data between gpu's takes time and that makes for slow crunching.


I know what CrossFire does, so if it works for games it should work for GPGPU, since it's 5 Gbps CrossFire bridge accelerating processing, of course it takes time, every synchronization in the PC takes a few cycles to adjust the transfers but this is not different: it takes the minimum amount of cycles to synchronize, it should work for anything, hence my exploring suggestion.


Unfortunately it doesn't work the same way for crunching though. The best idea is to try crunching with the cross-fire on, then run some units with it off. If it helps leave it on, if it doesn't then either live with it or run two projects, one on each card.

Saul
Send message
Joined: 5 Apr 12
Posts: 8
Credit: 450,399
RAC: 0
Message 13791 - Posted: 15 Apr 2012, 20:18:26 UTC - in response to Message 13785.
Last modified: 15 Apr 2012, 20:19:45 UTC

CrossFire(CF) makes 2 cards work as 1, and I'm running 2 WUs, which means collatz is not using CrossFire. Your suggestion about testing with it on and off makes me think that you don't know if CF would have a improvement on collatz from the developer POV, so let them affirm or deny if this, what you write is your opinion and nothing else at the seems, if you know better explain me in detail, I love technical details.

Profile mikey
Avatar
Send message
Joined: 11 Aug 09
Posts: 3245
Credit: 1,699,257,919
RAC: 4,977,369
Message 13798 - Posted: 16 Apr 2012, 10:55:53 UTC - in response to Message 13791.

CrossFire(CF) makes 2 cards work as 1, and I'm running 2 WUs, which means collatz is not using CrossFire. Your suggestion about testing with it on and off makes me think that you don't know if CF would have a improvement on collatz from the developer POV, so let them affirm or deny if this, what you write is your opinion and nothing else at the seems, if you know better explain me in detail, I love technical details.


What I know is I have 15 pc's running here and a gpu in each one, several have more than one gpu in them, and NOT ONE is faster with CF on then off! But if that is how you feel have fun!
____________

terencewee*
Send message
Joined: 20 Dec 10
Posts: 12
Credit: 39,717,409
RAC: 0
Message 13800 - Posted: 16 Apr 2012, 16:30:14 UTC - in response to Message 13791.
Last modified: 16 Apr 2012, 16:30:57 UTC

Your suggestion about testing with it on and off makes me think that you don't know if CF would have a improvement on collatz from the developer POV, so let them affirm or deny if this, what you write is your opinion and nothing else at the seems, if you know better explain me in detail, I love technical details.


@Saul:
Simmer down - mikey is trying to help. It's a process of elimination.
Look at the date mikey joined the project & his credits.

If an experienced person is giving me advice, I will be *very-grateful*.

Saul
Send message
Joined: 5 Apr 12
Posts: 8
Credit: 450,399
RAC: 0
Message 13809 - Posted: 19 Apr 2012, 2:25:32 UTC - in response to Message 13800.

@Mikey: you misunderstood, I'm not saying CF helps in the current state of collatz, what I'm saying, again, is that if developers choose to use CF to use 2+ GPUs as a way to explore WU processing acceleration that something you don't know, so you don't have the technical authority to say if it'd be useful or not.

@terencewee*: No matter how experience he has crunching he doesn't have a say what the devs can or can not do about CF and if you were misunderstanding like mikey did and you were referring to CF crunching possibility then the above paragraph explains it.

I'm grateful with the info that was relevant to my question and I did said so.

Profile mikey
Avatar
Send message
Joined: 11 Aug 09
Posts: 3245
Credit: 1,699,257,919
RAC: 4,977,369
Message 13813 - Posted: 19 Apr 2012, 10:48:29 UTC - in response to Message 13809.

@Mikey: you misunderstood, I'm not saying CF helps in the current state of collatz, what I'm saying, again, is that if developers choose to use CF to use 2+ GPUs as a way to explore WU processing acceleration that something you don't know, so you don't have the technical authority to say if it'd be useful or not.

@terencewee*: No matter how experience he has crunching he doesn't have a say what the devs can or can not do about CF and if you were misunderstanding like mikey did and you were referring to CF crunching possibility then the above paragraph explains it.

I'm grateful with the info that was relevant to my question and I did said so.


Have you tried crunch for Moo yet? They do exactly what you are talking about and NO CF does NOT help!! On Moo the project BY DEFAULT uses ALL of your gpu's on one workunit, but using CF actually makes things SLOWER!!! I have waaaay too many t-shirts of the 'been there, done that' type to be surprised but you are of course welcome to try it on your own and see what happens. I am learning new things everyday, if you don't do that you are DEAD!!!

Saul
Send message
Joined: 5 Apr 12
Posts: 8
Credit: 450,399
RAC: 0
Message 13816 - Posted: 19 Apr 2012, 22:43:29 UTC

SO now you think you'll have an authority by yelling at me? grow up...
Well Moo is a stupid project IMO, not worth investing CPU/GPU time. If they do it slower well maybe that's their fault, maybe they're not doing it right, that's only the first reference to CF GPU crunching I have so I'm not convinced by your experienced, you shouldn't either if it's also your first reference.

Profile mikey
Avatar
Send message
Joined: 11 Aug 09
Posts: 3245
Credit: 1,699,257,919
RAC: 4,977,369
Message 13819 - Posted: 20 Apr 2012, 10:01:15 UTC - in response to Message 13816.

SO now you think you'll have an authority by yelling at me? grow up...
Well Moo is a stupid project IMO, not worth investing CPU/GPU time. If they do it slower well maybe that's their fault, maybe they're not doing it right, that's only the first reference to CF GPU crunching I have so I'm not convinced by your experienced, you shouldn't either if it's also your first reference.


Saul I was NOT yelling at you...let's chalk it up to you being from Bolivia and me being from the US and it being a language translation thing, okay!

As for whether you chose to take my advice or not, we are human beings you can take it or leave it. There are no strings attached, if you think I am full of it, just don't listen to anything I have to say and ask someone else, I have no problems with that! Message Boards are full of people freely trying ot help someone else, if you feel that I am not helping you then just stop replying and move on, I won't feel bad! As I told the Admin over at another project trying to fix a problem, he is barking up the wrong tree, but if he wants to continue go right ahead, he is the one doing the barking!

Saul have you ever tried clicking on someones name over in the left hand column? If not you should, it might show you things you never thought of before.

Profile Crunch3r
Volunteer moderator
Project developer
Project tester
Avatar
Send message
Joined: 30 Jun 09
Posts: 219
Credit: 7,515,663,568
RAC: 10,847
Message 13826 - Posted: 21 Apr 2012, 19:07:10 UTC - in response to Message 13759.
Last modified: 21 Apr 2012, 19:08:07 UTC

A yes or no would do.


It doesn't matter since collatz uses integers...

Besides all that there are no such thing as 32 bit or 64 bit WU(there are 64 and 32 bit APPs though... that's a difference. Although that only matters if your crunching on CPU). All WUs are the same regardless of GPU or CPU type. The only difference is the length (mini collatz or regular collatz WU)

Hope that helps.
____________

Team BOINC United.Join Science that matters.

Profile Crunch3r
Volunteer moderator
Project developer
Project tester
Avatar
Send message
Joined: 30 Jun 09
Posts: 219
Credit: 7,515,663,568
RAC: 10,847
Message 13827 - Posted: 21 Apr 2012, 19:10:27 UTC - in response to Message 13819.
Last modified: 21 Apr 2012, 19:13:09 UTC



Saul I was NOT yelling at you...


Writing in Caps is considered like Shouting/Yelling almost anywhere ;)

http://email.about.com/od/netiquettetips/qt/Writing-In-All-Caps-Is-Like-Shouting.htm

So i do understand Saul's point.
____________

Team BOINC United.Join Science that matters.

Saul
Send message
Joined: 5 Apr 12
Posts: 8
Credit: 450,399
RAC: 0
Message 13829 - Posted: 21 Apr 2012, 20:53:12 UTC - in response to Message 13827.

@mikey:I don't doubt you have experienced, I already saw your 1 time signature you posted, no need to indicate me to click on your profile to show off me the same thing and yes I have clicked on some profiles not only in this forums but in many and you're not the first one making huge amount of points/projects, but my point still stands about the developer part.

My 2 ATI 4670s are don't have 64-bit floating point (fp) only 32-bit so is not suitable for many GPU projects that might use CF so I can't experience them and searching forums to find out about CF performance on GPU crunching will be very time consuming; I'll let that to faith. However I know about computer architecture enough to tell you that CF not only let's you work with 2 + GPUs but it adds more bandwidth (5 Gbps) to communication between video cards besides the PCI-E 2.0 bus (also 5 Gbps or more), so if the GPU crunching is slower is not in the bandwidth, maybe the is just not optimal code.

You might have not yelling at me but your attitude tells me otherwise and yes as Crunch3r wrote, using CAPS and '!!!!' is impolite on the Internet.


@Crunch3r: Sorry, but there is such thin as 32-/64-bit GPU WUs it doesn't refer to the data bit length, as is in CPUs, but to the fp precision; if this projects uses integers well I think most GPUs can handle that.

Luckily as you write if this is a integer-only math project and I can participate, most GPU crunching are protein folding which uses 64-bit fp, most if not all other GPU projects also use 64-bit fp.

The ones that used to use 32-bit are changing to 64-bit because of OpenCL 1.0 specification, my GPU supports only partially OpenCL 1.0, it supports OpenCL 1.0b (beta), which is a little behind the official first official 1.0 spec. (1.1 is out and 1.2 in in beta-testing). I don't know if 64-bit flops GPUs have bigger integer capability than 32-bit flops GPU though, may be since their data bus is twice as wide.

Saul
Send message
Joined: 5 Apr 12
Posts: 8
Credit: 450,399
RAC: 0
Message 13837 - Posted: 23 Apr 2012, 10:25:00 UTC - in response to Message 13829.
Last modified: 23 Apr 2012, 10:30:28 UTC

Edit: The data bus I was referring to in the previous post was: 64-bt floating pointp (fp)capable GPUs have a 128-bit bus and the 32-bit fp capable ones a 64-bit bus.

I just wanted to add that most GPU project are changing to OpenCL as a universal platform for both ATI & nVidia both had their own: CAL & CUDA, respectively. So GPU's talking should refer to maximum data bit width like when refering to CPUs, but that is not the issue in most, if not all, GPU projects that use GPU's fp for calculations refer to the fp capability requirements of the volunteering GPUs unfortunately, hence the switch to focus on fp numbers in GPU talk in most forums.

In CPU talk, 32-bit CPUs have a 64-bit fp Unit (FPU); in 64-bit CPUs the FPU is 128-bit, but in GPUs it seems that 32-bit GPUs have 32-bit fp, I could be mistaking in this part. In the ATI Radeon™ HD 4600 Series GPU Specifications You read 128-bit fp for all operations, I think this means 4 32-bit separate numbers that can be send into the Video Card, I was told this by a AMD person a while ago by e-mail but it doesn't seem to be the meaning in the spec webpage.

Anyway, I was trying to clear my point.
Cheers.


Post to thread

Message boards : Windows : I have 32-bit fp GPU, is it good for collatz?


Main page · Your account · Message boards


Copyright © 2018 Jon Sonntag; All rights reserved.