AMD ATI Radeon HD 4700/4800 (RV740/RV770)

pragmatic prancing periodic problem child, left
pragmatic pranc...
Joined: 26 Jan 05
Posts: 153
Credit: 70000
RAC: 0
Topic 84796

OK, everyone with an AMD ATI Radeon HD 4700/4800 (RV740/RV770) or similar low level OpenCL 1.0 only GPU, who managed to get any work validated, hold up your hand. Tell what OS you use, which drivers.

I don't believe any work around here for the HD4xx0 cards will validate, when put against anything else. I looked around a bit and see that all ATIOpenCL will not validate against other ATIOpenCL (not even when the same model GPUs are used), and not against ATIOpenCLLion, CUDA32, CUDA32OSX, CUDA32NV270, CUDA32FULLCPU, or CPUSSE.

Since the developers here have fallen very silent on this subject, or are spending their time on fixing the upload and download servers, I thought I'd ask.

pragmatic prancing periodic problem child, left
pragmatic pranc...
Joined: 26 Jan 05
Posts: 153
Credit: 70000
RAC: 0

AMD ATI Radeon HD 4700/4800 (RV740/RV770)

Windows 7 - 64bit
HD4850 - 1GB
Catalysts 11.11

17933 was done yesterday, it will not validate as it's paired against a CUDA32.
17971 did not validate as it was paired against a CUDA32.
17950 is what I am running now, I've never seen what it does when paired against a CUDA32OSX, but somehow I doubt it makes any difference.

TRuEQ & TuVaLu
TRuEQ & TuVaLu
Joined: 11 Sep 06
Posts: 75
Credit: 615315
RAC: 0

I am normaly running on ATI

I am normaly running on ATI 5850

Now attached a second ATI card 4850

Now running on 2 cards.

The 5850 performs almost normal.... Now it uses 100% CPU to feed the GPU.
It hasn't used that before when I ran it alone.

The 4850 tasks all ends up "computation error" and the messages box says "output file abscent".

I will now remove the 4850 from the computor.

I hope to see that the CPU% gets lower when feeding the 5850 alone. 1 card in the computor.

TRuEQ & TuVaLu
TRuEQ & TuVaLu
Joined: 11 Sep 06
Posts: 75
Credit: 615315
RAC: 0

Yep. As I thought. Running

Yep.

As I thought. Running only 1 card then the CPU goes back to normal 15-25% to feed the GPU...

Sorry about adding a note about this when the thread as about 48xx cards.

pragmatic prancing periodic problem child, left
pragmatic pranc...
Joined: 26 Jan 05
Posts: 153
Credit: 70000
RAC: 0

RE: 17933 was done

Message 78863 in response to message 78860

Quote:
17933 was done yesterday, it will not validate as it's paired against a CUDA32.


As said, it did not pair against that CUDA32 and is sent to another CUDA32.

Quote:
17971 did not validate as it was paired against a CUDA32.


It was sent to another CUDA32 and did not pair against that one either, so it's sent to yet another CUDA32 to see what that one makes of it. The suspense.... Oh.

Alexone
Alexone
Joined: 21 Nov 11
Posts: 6
Credit: 1000
RAC: 0

ATI HD 4850 1GB Drivers

ATI HD 4850 1GB
Drivers 11.8
BOINC 7.0.2
Windows 7 Ult x64

WU's are inconclusive. And I think that they will be invalid =(
Computer is 1393
http://albertathome.org/workunit/18378
and
http://albertathome.org/workunit/18184

But thanks anyway for your work. =)

pragmatic prancing periodic problem child, left
pragmatic pranc...
Joined: 26 Jan 05
Posts: 153
Credit: 70000
RAC: 0

RE: RE: 17933 was done

Message 78865 in response to message 78863

Quote:
Quote:
17933 was done yesterday, it will not validate as it's paired against a CUDA32.

As said, it did not pair against that CUDA32 and is sent to another CUDA32.


It's now Completed, marked as invalid for me and another CUDA32, the other two CUDA32 have walked away with the money.

Quote:
17950 is what I am running now, I've never seen what it does when paired against a CUDA32OSX, but somehow I doubt it makes any difference.


It finished, Completed, validation inconclusive.

Seeing how the other 3 tasks I had were all paired against CUDA32 tasks, I aborted them already. There's in my opinion no use in burning electrons when they will never validate anyway. I've gotten 6 new tasks, in 3 of which I am paired against CUDA32 (which I'll abort again), 2 against other ATIOpenCL but higher denominations GPUs (a 58xx and 6xxx) and one undecided as at this time it's unsent.

pragmatic prancing periodic problem child, left
pragmatic pranc...
Joined: 26 Jan 05
Posts: 153
Credit: 70000
RAC: 0

RE: Seeing how the other 3

Message 78866 in response to message 78865

Quote:
Seeing how the other 3 tasks I had were all paired against CUDA32 tasks, I aborted them already. There's in my opinion no use in burning electrons when they will never validate anyway. I've gotten 6 new tasks, in 3 of which I am paired against CUDA32 (which I'll abort again), 2 against other ATIOpenCL but higher denominations GPUs (a 58xx and 6xxx) and one undecided as at this time it's unsent.


The Unsent was resent eventually and of course to a CUDA32. So I aborted that one as well. Will only run the two paired against other ATIOpenCL ones now, but don't have much hope they'll do much of anything. It'll be more like we both will end up with nothing, since all that work will be resent twice to two CUDA32 who in the end will get the work validated.

We know already that the lower end ATI GPUs don't validate against the rest of the world or each other. Is this information useful for the project, or not?

pragmatic prancing periodic problem child, left
pragmatic pranc...
Joined: 26 Jan 05
Posts: 153
Credit: 70000
RAC: 0

14141 had me paired against a

14141 had me paired against a HD69x0 GPU with Catalysts 11.11; Completed, validation inconclusive.

Is this of any use here, or are we just burning electrons because we can?

I've aborted the last one as well and set BOINC to ignore my GPU at start-up. There's at this time no use to run work here, other than to add to an electricity bill because we can. I'll be back when very strict changes have been done to the validator. It's been over a month now that I first reported that ATIOpenCL 1.0 does not validate against itself or 1.1, I've just done it again. Maybe that this time someone gives a damn.

Bikeman (Heinz-Bernd Eggenstein)
Bikeman (Heinz-...
Joined: 28 Aug 06
Posts: 164
Credit: 1864017
RAC: 0

Hi! Nevertheless, thanks for

Message 78868 in response to message 78867

Hi!

Nevertheless, thanks for your input and tenacity. I think it does make sense to retest this whenever a) a new driver version comes out or b) a new app version (I haven't read anything that excludes a driver problem as a factor here, since it's the same app, and it's doing reasonably well for openCL 1.1 cards).

I think unless either a new app version or driver version comes out, it really makes little sense to try this particular card any further here.

HBE

pragmatic prancing periodic problem child, left
pragmatic pranc...
Joined: 26 Jan 05
Posts: 153
Credit: 70000
RAC: 0

A retest only makes sense

Message 78869 in response to message 78868

A retest only makes sense when the validator has been adjusted, or a new app comes out.

Perhaps it's wiser though that the project should just throw their hands in the air and give up on trying to force a support for OpenCL 1.0 cards, make the requirement OpenCL 1.1 for ATI cards and be done with it. Exclude the 43xx, 45xx, 46xx, 47xx and 48xx GPUs, perhaps all the others with Beta support as well.

Want to do work with those GPUs? Go to the competition projects or upgrade. ATI cards are cheap, very cheap when compared to hard drive prices.

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.