ati 4770
log in

Advanced search

Message boards : Number crunching : ati 4770

Author Message
Profile verstapp
Avatar
Send message
Joined: 26 Jan 09
Posts: 585
Credit: 464,286,454
RAC: 898
Message 20889 - Posted: 30 Apr 2009 | 3:23:11 UTC

This ati 4770 that all the tech sites are raving about, eg X-Bit , has anyone tried it with MW?
____________
Cheers,

PeterV

.

Profile The Gas Giant
Avatar
Send message
Joined: 24 Dec 07
Posts: 1947
Credit: 240,865,573
RAC: 0
Message 20892 - Posted: 30 Apr 2009 | 5:54:21 UTC

Is it double-precision? I thought only the RV770 chips were.

Profile verstapp
Avatar
Send message
Joined: 26 Jan 09
Posts: 585
Credit: 464,286,454
RAC: 898
Message 20893 - Posted: 30 Apr 2009 | 6:43:43 UTC
Last modified: 30 Apr 2009 | 7:13:50 UTC

dunno, don't care.
But it turns out that there are claims that it does,
eg beyond3d
____________
Cheers,

PeterV

.

Profile arkayn
Avatar
Send message
Joined: 14 Feb 09
Posts: 914
Credit: 74,780,239
RAC: 225
Message 20896 - Posted: 30 Apr 2009 | 8:56:41 UTC - in response to Message 20893.

http://www.pcgameshardware.com/aid,682726/Review-Ati-RadeonHD-4770-vs-HD-4850-und-Geforce-9800-GT/Reviews/
____________

Profile Simplex0
Avatar
Send message
Joined: 11 Nov 07
Posts: 232
Credit: 178,221,048
RAC: 0
Message 20921 - Posted: 30 Apr 2009 | 13:35:34 UTC

128-bit floating point precision for all operations

Profile Simplex0
Avatar
Send message
Joined: 11 Nov 07
Posts: 232
Credit: 178,221,048
RAC: 0
Message 20925 - Posted: 30 Apr 2009 | 14:30:56 UTC

According to Wikipedia this 128-bit format is quadruple precision

Cluster Physik
Send message
Joined: 26 Jul 08
Posts: 627
Credit: 94,940,203
RAC: 0
Message 20947 - Posted: 30 Apr 2009 | 17:18:55 UTC - in response to Message 20921.
Last modified: 30 Apr 2009 | 17:21:55 UTC

128-bit floating point precision for all operations

That just means you have 128 bits (16 Bytes) per pixel, i.e. 4 Bytes (32bit single precision floats) per component (red, green, blue, alpha). It has nothing to do with the double precision capability.

Actually, I don't know if the HD4770 can work with doubles (but I hope so). PCGames are the only ones who even mentioned double precision in their review.

Profile arkayn
Avatar
Send message
Joined: 14 Feb 09
Posts: 914
Credit: 74,780,239
RAC: 225
Message 20959 - Posted: 30 Apr 2009 | 18:15:38 UTC


____________

Profile Simplex0
Avatar
Send message
Joined: 11 Nov 07
Posts: 232
Credit: 178,221,048
RAC: 0
Message 20964 - Posted: 30 Apr 2009 | 18:42:16 UTC

Is there a difference in the precision of the calculation if each number used in the calculation are stored in one 128 bit long memory location or in two 64 bit long memory locations?

Cluster Physik
Send message
Joined: 26 Jul 08
Posts: 627
Credit: 94,940,203
RAC: 0
Message 20967 - Posted: 30 Apr 2009 | 19:01:12 UTC - in response to Message 20964.

Is there a difference in the precision of the calculation if each number used in the calculation are stored in one 128 bit long memory location or in two 64 bit long memory locations?

Most of the stuff is stored in the registers of the GPU during the calculation ;)

I don't know what you are exactly referring to, but to clarify my first comment on the "128bit precision" issue I have to say that it is mainly PR FUD. As mentioned it just says the GPU has a completely float (single precision, i.e. 32 bits) capable shader pipeline. No graphics card on the market (not even commodity CPUs) supports 128bit precision calculations in the sense of quad precision and therefore also not the storage of values in that data format.

Profile Simplex0
Avatar
Send message
Joined: 11 Nov 07
Posts: 232
Credit: 178,221,048
RAC: 0
Message 20968 - Posted: 30 Apr 2009 | 19:42:00 UTC
Last modified: 30 Apr 2009 | 19:46:58 UTC

Thank you Cluster.
I just try to understand the exact meaning of the expresion "double precision" aparently it is a standard according to wikipedia...
"Double precision floating point is an IEEE 754 standard for encoding binary or decimal floating point numbers in 8 bytes."

Cluster Physik
Send message
Joined: 26 Jul 08
Posts: 627
Credit: 94,940,203
RAC: 0
Message 20971 - Posted: 30 Apr 2009 | 20:33:13 UTC - in response to Message 20968.

Thank you Cluster.
I just try to understand the exact meaning of the expresion "double precision" aparently it is a standard according to wikipedia...
"Double precision floating point is an IEEE 754 standard for encoding binary or decimal floating point numbers in 8 bytes."

Exactly. Support for calculations with such double precision values (8 Bytes or 64 bits in size) is needed by MW.

uBronan
Avatar
Send message
Joined: 9 Feb 09
Posts: 166
Credit: 27,520,813
RAC: 0
Message 20978 - Posted: 30 Apr 2009 | 21:43:54 UTC

if i see the technical stats i guess it exactly places itself between the 4830 and the 4850, i just wonder myself if it would work with my mainbord gpu to create a x-crossfire solution with it.
The mainboard one reports itself to be a HD3200

Gecko
Avatar
Send message
Joined: 23 Feb 08
Posts: 20
Credit: 804,154
RAC: 0
Message 20983 - Posted: 30 Apr 2009 | 22:20:47 UTC

Asus 4770 TOP looks to be insanely fast w/ on board voltage tweak and smartdoctor performance control application for voltage mod to 1.2V.
Stock is 800 MHz (core), and 850 MHz (memory). Using the above, voltages can be upped from 0.95 V to 1.2 V w/ freqencies as high as 971 MHz, and 1150 MHz for the memory. That's 1.24 Tflps for @ $100 IF a person can find one.
Essentially same performance as reference 4870, but w/ a single 6-pin power connector.

Short enough to fit on the new mini-itx board and mini-itx case from Silverstone ; > ) I'm getting the itch to grab one when I can.

Cluster Physik
Send message
Joined: 26 Jul 08
Posts: 627
Credit: 94,940,203
RAC: 0
Message 21488 - Posted: 7 May 2009 | 11:35:29 UTC - in response to Message 20889.
Last modified: 7 May 2009 | 11:39:22 UTC

This ati 4770 that all the tech sites are raving about, eg X-Bit , has anyone tried it with MW?

The HD4770 has been tested to work. It is recognized by the GPU app as RV770 (looks like an overclocked HD4830, it does not know about the RV740, yet) and runs about 5% slower than a HD4850 at stock clocks. But remember there could be the same problem as with the HD4890 and the WindowsXP drivers. The current version does not run reliably under XP32 and XP64. Vista appears to work though.

CAL Runtime: 1.4.255
Found 1 CAL device

Device 0: ATI Radeon HD 4800 (RV770) 512 MB local RAM (remote 64 MB cached + 256 MB uncached)
GPU core clock: 750 MHz, memory clock: 800 MHz
640 shader units organized in 8 SIMDs with 16 VLIW units (5-issue), wavefront size 64 threads
supporting double precision

uBronan
Avatar
Send message
Joined: 9 Feb 09
Posts: 166
Credit: 27,520,813
RAC: 0
Message 22827 - Posted: 20 May 2009 | 21:40:30 UTC
Last modified: 20 May 2009 | 21:40:49 UTC

Can anyone tell me how i can see if my card is being found by mw or boinc
I have 2 x 4770 ati cards but they are only be able to install with cat 9.4 or 9.5
____________
Its new, its relative fast... my new bicycle

Profile caferace
Avatar
Send message
Joined: 4 Aug 08
Posts: 46
Credit: 8,255,900
RAC: 0
Message 22842 - Posted: 20 May 2009 | 22:57:58 UTC - in response to Message 22827.

Can anyone tell me how i can see if my card is being found by mw or boinc
I have 2 x 4770 ati cards but they are only be able to install with cat 9.4 or 9.5

Look at the Task Details for a completed WU done by your box with the 4770's. Here is one for my Q6600 with a HD3850:

<core_client_version>6.4.6</core_client_version>
<![CDATA[
<stderr_txt>
Running Milkyway@home ATI GPU application version 0.19e by Gipsel
CPU: Intel(R) Core(TM)2 Quad CPU Q6600 @ 2.40GHz (4 cores/threads) 2.39898 GHz (549ms)

CAL Runtime: 1.3.145
Found 1 CAL device

Device 0: ATI Radeon HD 3800 (RV670) 512 MB local RAM (remote 28 MB cached + 512 MB uncached)
GPU core clock: 669 MHz, memory clock: 829 MHz
320 shader units organized in 4 SIMDs with 16 VLIW units (5-issue), wavefront size 64 threads
supporting double precision

1 WUs already running on GPU 0
Starting WU on GPU 0

main integral, 160 iterations
predicted runtime per iteration is 408 ms (33.3333 ms are allowed), dividing each iteration in 13 parts
borders of the domains at 0 123 246 369 492 615 738 861 984 1107 1230 1353 1476 1600
Calculated about 3.70012e+012 floatingpoint ops on GPU, 6.34181e+007 on FPU. Approximate GPU time 85.2656 seconds.

cut number 0, 40 iterations
predicted runtime per iteration is 51 ms (33.3333 ms are allowed), dividing each iteration in 2 parts
borders of the domains at 0 200 400
Calculated about 1.15629e+011 floatingpoint ops on GPU, 5.38792e+006 on FPU. Approximate GPU time 2.64063 seconds.
Calculated about 2.48101e+009 floatingpoint ops on FPU (stars).

WU completed.
CPU time: 15.4375 seconds, GPU time: 87.9063 seconds, wall clock time: 286.817 seconds, CPU frequency: 2.40012 GHz

</stderr_txt>
]]>


-jim

uBronan
Avatar
Send message
Joined: 9 Feb 09
Posts: 166
Credit: 27,520,813
RAC: 0
Message 22933 - Posted: 21 May 2009 | 20:25:02 UTC
Last modified: 21 May 2009 | 21:15:36 UTC

Ok i found that i looks like the famous agp can't write error which should be fixed with cat 8.12 and the fix.But the cards are only seen with the cat 9.5 drivers. I wonder if is needed on pci-e cards.
I have been messing with other versions of catalyst but they simply won't work.
The cat 9.4 was looking promising but all of a sudden i kept getting blue screens. I have not checked the cd from manufactor is there is a older driver on it but i guess not.
I also took out 4 Gb memory since i had some weird sluggish startups if i ran with all 4 memory banks in it.

I have a multiboot system with win xp 32 ,vista 32, and win7 x64 on seperate drives. I have updated directx 9.0c and have all .net versions running and updated to the latest versions

I have put only 1 card in the system and the other is waiting in the box untill i get it running stable again.
I had been messing with running my old gt9600 nv card on gpugrid and the ati together doing seperate tasks ... but it simply no longer was seen as a cuda device and all of a sudden was bleu screening till death ;)

So i cleaned out all and reinstalled the ATI cleanly.

This is the error i see in boinc :

21-May-2009 01:55:04 [Milkyway@home] Not reporting or requesting tasks
21-May-2009 01:55:09 [Milkyway@home] Scheduler request completed: got 0 new tasks
21-May-2009 01:55:19 [climateprediction.net] Sending scheduler request: To send trickle-up message.
21-May-2009 01:55:19 [climateprediction.net] Not reporting or requesting tasks
21-May-2009 01:55:55 [Milkyway@home] task ps_s26_17_15857_1242859528_0 resumed by user
21-May-2009 01:55:55 [Milkyway@home] Starting ps_s26_17_15857_1242859528_0
21-May-2009 01:55:55 [Milkyway@home] [error] Process creation failed: Het systeem kan het opgegeven bestand niet vinden. (0x2)
21-May-2009 01:55:56 [Milkyway@home] [error] Process creation failed: Het systeem kan het opgegeven bestand niet vinden. (0x2)
21-May-2009 01:55:57 [Milkyway@home] [error] Process creation failed: Het systeem kan het opgegeven bestand niet vinden. (0x2)
21-May-2009 01:55:57 [Milkyway@home] [error] Process creation failed: Het systeem kan het opgegeven bestand niet vinden. (0x2)
21-May-2009 01:55:58 [Milkyway@home] [error] Process creation failed: Het systeem kan het opgegeven bestand niet vinden. (0x2)
21-May-2009 01:55:59 [Milkyway@home] Computation for task ps_s26_17_15857_1242859528_0 finished
21-May-2009 01:55:59 [Milkyway@home] Output file ps_s26_17_15857_1242859528_0_0 for task ps_s26_17_15857_1242859528_0 absent

The error given is the system can't find given file
I have the 3 renamed files in the system32 map

One of the errored units Here
____________
Its new, its relative fast... my new bicycle

Cluster Physik
Send message
Joined: 26 Jul 08
Posts: 627
Credit: 94,940,203
RAC: 0
Message 22949 - Posted: 21 May 2009 | 21:47:59 UTC - in response to Message 22933.
Last modified: 21 May 2009 | 21:50:19 UTC

Ok i found that i looks like the famous agp can't write error which should be fixed with cat 8.12 and the fix.
[..]

stderr out
<core_client_version>6.6.28</core_client_version>
<![CDATA[
<message>
- exit code -1073741515 (0xc0000135)
</message>
]]>

The error you are seeing means that some dll is missing. It has nothing to do with AGP whatsoever.

You should recheck if you have the correct version (32bit vs 64bit), all dll files (brook.dll and the three CAL dlls) and .NET framework (3.5 with all updates) installed. Maybe a restart of BOINC (or the whole computer to be sure) will help (usually does if boinc was not shut down during some changes).

uBronan
Avatar
Send message
Joined: 9 Feb 09
Posts: 166
Credit: 27,520,813
RAC: 0
Message 22952 - Posted: 21 May 2009 | 22:07:31 UTC

Rechecking all parameters again

____________
Its new, its relative fast... my new bicycle

uBronan
Avatar
Send message
Joined: 9 Feb 09
Posts: 166
Credit: 27,520,813
RAC: 0
Message 22969 - Posted: 22 May 2009 | 0:27:48 UTC

Yes i found the culprits first one is when installing the new 9.5 deletes the amdcal* files, and second i found that in general some info is added to the path in windows.
This should not be a problem however most of the time the pathnames are too long which make it fail.
So in my case the root paths where missing which simply means that it does not find the windows and windows\system32 files anymore

its easily solved by goto configuration panel > select system > Advanced >
Variables > System Variables then select the line path and press edit
Then remove lenghty sentences like c:\Program Files\Quicktime\QT...jadajada
And i found many are not even needed ever now the ones which are needed are :
%SystemRoot%;%SystemRoot%\system32;%SystemRoot%\System32\WBEM
____________
Its new, its relative fast... my new bicycle

uBronan
Avatar
Send message
Joined: 9 Feb 09
Posts: 166
Credit: 27,520,813
RAC: 0
Message 23095 - Posted: 23 May 2009 | 8:27:59 UTC

Well after i solved the removed files it was starting and seemed to work untill it actually got some work.

The gipsel application crashed the gpu so i start to think something is wrong with these cards.

Both end up with either a white/black screen and machine being completely frozen, i have send a message to ati support.

____________
Its new, its relative fast... my new bicycle

Profile JockMacMad TSBT
Avatar
Send message
Joined: 28 Jan 09
Posts: 31
Credit: 84,754,088
RAC: 0
Message 23814 - Posted: 31 May 2009 | 23:20:04 UTC - in response to Message 23095.

I have just installed Catalyst 9.5 and it's not working for me at all. Computation Error on 1 unit at a time.

Going back to try 9.4 and see what happens as that was all working just fine for me.

uBronan
Avatar
Send message
Joined: 9 Feb 09
Posts: 166
Credit: 27,520,813
RAC: 0
Message 23963 - Posted: 2 Jun 2009 | 13:52:32 UTC

somehow i am wondering why i could not install 9.4
it simply could not find a ati device to install on
i'll do a final test and see what happens
____________
Its new, its relative fast... my new bicycle

uBronan
Avatar
Send message
Joined: 9 Feb 09
Posts: 166
Credit: 27,520,813
RAC: 0
Message 23966 - Posted: 2 Jun 2009 | 15:14:32 UTC

I have tested on every os and whatever i do it simply crashes either it gives a core error or it reports 128 (0x80).
While all needed files are in the system32 map it still not work.
xp, vista, and win 7 all 32 or 64 it doesn't matter they only see my card with 9.5 driver i guess my gpu has a newer model number or something.

____________
Its new, its relative fast... my new bicycle

Post to thread

Message boards : Number crunching : ati 4770


Main page · Your account · Message boards


Copyright © 2013 AstroInformatics Group