GPU FAQ

From FaHWiki

Jump to: navigation, search

Since the public release of the Folding@Home client for modern ATI GPUs several issues have come up. Some of them are known and included in the GPU FAQ. Others are being discussed in the [1] Folding-Community support forum.

If you have an issue which is not reported yet, please report it on the Folding-Community forum.


Contents

[edit]

Supported Graphics Cards

GPU1 client: ATI x1600, x1800, and x1900 class (discountinued June 6, 2008)

GPU2 client: ATI HD2xxx, HD3xxx class, HD4xxx and later (Anything with CAL support in the drivers.)
_a_n_d____: All nVidia 8xxx class GPUs and later* (Anything with CUDA support in the drivers.)

  • Note that nVidia cards with less than 256MB of video do not have support for CUDA
[edit]

Will earlier ATI cards such as the 9600PRO ever be supported?

No! 

The short technical answer is that those boards do not support looping constructs 
and are also 24-bit precision which is a problem for us. The X1K boards were the 
first to support 32-bit precision and branching/looping on the ATI/AMD side of the world.
---mhouston [2]
[edit]

Known Issues

  • If Windows enters the "fast user switching" screen the GPU core will quit. Solution for now is to turn off the screensaver going to login screen.
  • Doing a 3-finger salute [ctrl][alt][del] to bring up the task manager will also cause the GPU to exit. Solution: right-click on the task bar then click "Task Manager"

Note: The GPU2 clients based on CAL and CUDA have resolved many of these DX9 context based errors.

[edit]

Version Info

Please note: the version info below is a future roadmap for cores and not all cores are publicly available.

Recent ATI versions:

Ati v1.13 09/16/08 Fixed a bug in corewrapper code; added more diagnostics and include restriction: T_calc - T_tpr > 200 for triggering EUE; previously the constraint had been T_calc/T_tpr > 2.

  • 951446 Sep 16 11:17 Core_12.fah md5sum=322281cfc0176592b42e36d006864c9d
  • 3670016 Sep 16 11:17 FahCore_12 md5sum=7a4e390764d86455a87f5c9fc92ed3a4

Ati v1.14 09/29/08 Core_11 version of v1.13

  • 955707 Sep 29 15:41 Core_11.fah md5sum=84b00dc0fef800ea6060b4cd28484bde
  • 3710976 Sep 29 15:41 FahCore_11 md5sum=447da667df4d34fdeb229c6d1649e036

Ati v1.15 09/30/08 Core_12 -- includes mods to integrator to enhance numerical stability

  • 953855 Sep 30 12:12 Core_12.fah md5sum=3447fcac6cde5738d500dbfc55e1d55c
  • 3719168 Sep 30 12:12 FahCore_12 md5sum=9f7aa49e47f7f02702fab3d1dfffb8ed

Ati v1.16 10/08/08 Core_12 -- returns UNSTABLE_MACHINE instead of INTERRUPTED when NAN, Self-test failure, ... errors are detected.

  • 954004 Oct 8 16:09 Core_12.fah md5sum=fd534c57b3ad01c2fe81694c576bf097
  • 3719168 Oct 8 16:09 FahCore_12 md5sum=abad0a9b89cdb3706f6f6e7dea1a12dc

Ati v1.17 10/13/08 Core_11 -- same changes as w/ Core_12 v1.16: returns UNSTABLE_MACHINE instead of INTERRUPTED when NAN, Self-test failure, ... errors are detected.

  • 955814 Oct 13 10:55 Core_11.fah md5sum=b8cd622ef057678d38a5b69d4227a2eb
  • 3710976 Oct 13 10:55 FahCore_11 md5sum=3be79f78d8c0cc90b4d7b5514c044e23

Ati v1.18 10/14/08 Core_11 includes mods to integrator to enhance numerical stability

  • 954612 Oct 14 13:06 Core_11.fah md5sum=3dcdb35a83b3eed6661dff1c2cb69c0c
  • 3719168 Oct 14 13:06 FahCore_11 md5sum=5e66bac7d27efc0f8fe9f71ee5ba1b2f

Ati v1.19 11/18/08 Core_12 core for testing performance on 7XX boards

  • 963672 Nov 18 16:21 Core_12.fah md5sum=9a76a9fd8a4e3a81f1f40135625e10eb
  • 3805184 Nov 18 16:21 FahCore_12 md5sum=13fb2571295e1810b555822c794f37f8

Ati v1.20 11/19/08 Manual download

Ati v1.21 11/25/08 Core_12 AMD bug fix

  • 1056128 Nov 25 16:40 Core_12.fah md5sum=caa8c5a48a1f9ed3d8340454165e73ef
  • 3035136 Nov 25 16:40 FahCore_12 md5sum=37454bcf4030be3f982c3b2aebc1843c

Ati v1.22 12/08/08 Core_11 version of Core_12 v1.21 w/ a fix to SHAKE code

  • 1056837 Dec 8 14:09 Core_11.fah md5sum=7806799a675a0c7cf9df75e9361dc58a
  • 3035136 Dec 8 14:09 FahCore_11 md5sum=6d66f383666c58e3b1abeda5b365aa8e

Ati v1.23 01/20/09 Core_12 Shake fix and improved checkpointing

  • 1091279 Jan 20 19:23 Core_12.fah md5sum=6a0b1b53780b127780aaac4abb44e564
  • 3198976 Jan 20 19:23 FahCore_12 md5sum=b12e178009ef6b5e641093ee9f684560

Ati v1.24 02/06/09 Core_12 New libs from AMD; allows CAL_NO_FLUSH to be turned on, fixes CAL resources issue, ...

  • 1091241 Feb 6 12:47 Core_12.fah md5sum=119d153258c331039ecee3c68b6f739b
  • 3203072 Feb 6 12:44 FahCore_12 md5sum=0b34e94e54233b329cac70c96407d10a

Ati v1.24 02/09/09 Core_11 Same as Core_12 v1.24

  • 1091180 Feb 9 12:05 Core_11.fah md5sum=9369d87f2dc39e59965698ec04929267
  • 3203072 Feb 9 12:05 FahCore_11 md5sum=b7c4ce9c7f921e609d979b03407cbb00


Recent Nvidia versions:

Nvidia v1.11 09/16/08 Fixed a bug in corewrapper code; added more diagnostics and include restriction T_calc - T_tpr > 200 for triggering a EUE; previously the constraint had been T_calc/T_tpr > 2.

  • 640817 Sep 16 11:28 Core_13.fah md5sum=9e47e670de9bd7b4a9ec9f941cab56ab
  • 1822720 Sep 16 11:28 FahCore_13 md5sum=cfef83d588bd8f3ac8a7ada585300aba

Nvidia v1.12 09/25/08 Core_13: Fixed memory corruption problem; more diagnostics in corewrapper

  • 640590 Sep 25 11:10 Core_13.fah md5sum=96932213269cb91f3d6c14fc2006fd29
  • 1822720 Sep 25 11:10 FahCore_13 md5sum=56b8fdaeb13c52fa1ddb6bcc2faa368d

Nvidia v1.13 09/29/08 Core_11 version of v1.12

  • 640739 Sep 29 15:25 Core_11.fah md5sum=626a6c05b6cde91ef654a1a6c67c58be
  • 1822720 Sep 29 15:25 FahCore_11 md5sum=4bd1c1de6b572b71ddc093387fa47da5

Nvidia v1.14 10/02/08 Core_13 contains fix to corewrapper code that was causing the following type error:

[03:13:12] CoreStatus = C0000005 (-1073741819)
[03:13:12] Client-core communications error: ERROR 0xc0000005
[03:13:12] This is a sign of more serious problems, shutting down.

  • 640771 Oct 2 16:27 Core_13.fah md5sum=3f37f45d9fdf594df9d12da3fe944528
  • 1822720 Oct 2 16:27 FahCore_13 md5sum=23441a76782c4d4dc06406a4d2ee6bbc

Nvidia v1.15 10/13/08 Core_11: Now returns 'UNSTABLE_MACHINE' instead of 'INTERRUPTED' if there is a problem w/ the simulation.

  • 640343 Oct 13 12:14 Core_11.fah md5sum=ff9955211f647fd9f0ca55d49f2a9f1d
  • 1822720 Oct 13 12:14 FahCore_11 md5sum=3b34801b5acc5af92a4132c35ce614e2

Nvidia v1.16 10/16/08 Core_13: includes mods to integrator to enhance numerical stability

  • 643319 Oct 16 14:26 Core_13.fah md5sum=af593f3c7283f055df86529549136ed6
  • 1843200 Oct 16 14:26 FahCore_13 md5sum=a7805e3d3d30b00d54d34660f8f48efc

Nvidia v1.17 10/20/08 Core_13: essentially the same as v1.16. Diagnostic output was causing FAH to EUE; the output has been disabled.

  • 642843 Oct 20 15:15 Core_13.fah md5sum=a948b12c8c49bf3d48aff343f31c7a1a
  • 1843200 Oct 20 15:15 FahCore_13 md5sum=77ca6d19efe87a091680aec06d17bdef

Nvidia v1.18 10/23/08 Core_11: Core_11 version of v1.17; may have problems on 8800 boards

  • 642615 Oct 23 12:42 Core_11.fah md5sum=1283e8b7cdf3546d3316bb0066b8bfdf
  • 1843200 Oct 23 12:42 FahCore_11 md5sum=9e7a0356ffefa0a205309e017777a3f8

Nvidia v1.19 10/31/08 Core_13: Code based on union of v1.09 and v1.18. Includes update to integrator, memory corruption problem fix, ... The main difference between this version and v1.18 is that the performance optimizations between v1.09 & v1.13 were removed.

  • 642652 Oct 31 11:27 Core_13.fah md5sum=82710b2d639a008420d54f83b792be26
  • 1843200 Oct 31 11:27 FahCore_13 md5sum=244143b9fb7366156280fd89a79f66cd

Nvidia v1.19 11/03/08 Core_11: Same as Core_13 v1.19

  • 642475 Nov 3 10:40 Core_11.fah md5sum=bcfad16b82343cbda83391e27dbcdf5a
  • 1843200 Nov 3 10:40 FahCore_11 md5sum=c6740b5decdbad5a3de8d8731d208025

Nvidia v1.21 11/24/08 Core_13: Born radii updated prior to calculation of forces; mdp/tpr Shake tolerance setting is now honored; reposted on 11/24 (original 11/19/08) w/ correct version id in corewrapper code; md5sum values updated here

  • 642693 Nov 19 15:50 Core_13.fah md5sum=f7e73aae83a824780fc3fa7351ba4d8a
  • 1843200 Nov 19 15:47 FahCore_13.exe md5sum=a48c89cd207a37fe10ef1606e6686972

Nvidia v1.22 12/01/08 Core_11: Same as Core_13 v1.21, but Core_11

  • 642687 Dec 1 11:46 Core_11.fah md5sum=ffe19a839c225df3af746df0323dd1dc
  • 1843200 Dec 1 11:46 FahCore_11.exe md5sum=5f894303d6e90a6066cfb0a7d907bb56

Nvidia v1.23 01/21/09 Core_13: improved checkpointing

  • 648968 Jan 21 16:34 Core_13.fah md5sum=262e45e9c46227f16c4accfbdd4e3bcd
  • 1859584 Jan 21 16:34 FahCore_13.exe md5sum=a4ea39254d1c2747f5d8f8d17ed4def5

Nvidia v1.24 03/23/09 Core_13: checkpoint file and other frame files removed, if error detected in reading checkpoint file

  • 649454 Mar 23 15:40 Core_13.fah md5sum=a2ec20e13145ae843bb5f3fde6e84e54
  • 1859584 Mar 23 15:40 FahCore_13.exe md5sum=3f5a5b6ec19e3b92fb92fcb81a8a59f1

Nvidia v1.25 05/06/09 Core_11: includes checkpoint fixes, but Born radii changes from 12/08 rolled back

  • 682864 May 6 11:52 Core_11.fah md5sum=ff3de5722abcdfd2ad0f1235a61e9ed1
  • 2166784 May 6 11:52 FahCore_11.exe md5sum=edbbf1dc78c0b7474437cad6440b490f

Nvidia v1.26 06/09/09 Core_11: experimental core that allows donors to set the environment variable, FAH_GPU_IDLE, to idle the board for a percentage of the total time. The goal is to reduce heating problems for larger work units

  • 1904640 June 9 14:10 FahCore_11.exe md5sum=c48beeb242a3fd52b9bfe6b29474ab22

Nvidia v1.27 06/18/09 Core_11: fixes logic error w/ v1.26 -- the actual idle time was 5 times smaller than requested. Also fixed 'usage' output to FAHlog.txt -- it was not being updated after the change to using the environment variable to set idle time.

  • 1904640 Jun 18 15:02 FahCore_11.exe md5sum=1ced770b67f1111bfbe8bcefacef9d1f

Nvidia v1.28 09/01/09 Core_11: Imran's fix for machines w/ multiple gpus (limited test)

Nvidia v1.29 09/10/09 Core_13: includes Peter Kasson's mods for hash files; also fix for machines w/ multiple gpus (Imran's fix)

  • 664626 Sep 10 12:24 Core_13.fah md5sum=280240fc33b445c9cfd058c453380d91
  • 1908736 Sep 10 12:24 FahCore_13.exe md5sum=6bb1315870659d16cc8e415c98ba9878

Nvidia v1.30 09/14/09 Core_11: same as v1.29

  • 664933 Sep 14 10:43 Core_11.fah md5sum=12d07e55caf100b7e2f0ca08e6bfb124
  • 1908736 Sep 14 10:40 FahCore_11.exe md5sum=4953e5c2de4072c70d6b88ad52b57664

Nvidia v1.31 09/15/09 Core_11: fixed bug in generation of hash; updated corecomp

  • 665067 Sep 15 13:16 Core_11.fah md5sum=470f020f271b2667a5c2c2f8b9938ecb
  • 1908736 Sep 15 13:16 FahCore_11.exe md5sum=832e16e37fe376a06f46bbf68cd98850
[edit]

Links

Retrieved from "http://fahwiki.net/index.php/GPU_FAQ"
Personal tools