【单卡双开－－跑FAH是不是真的提高了运算能力？】

金鹏

结论：双开能榨干你显卡最后一滴运算能力，避免运算平均PPD低于显卡最高PPD的情况出现，特别是在算分值较低的小包时双开效果明显（双开前提你的U不要太弱）


单卡能够双开！但是真的提高了全速运行下效率？（此试验使用3870毒药超频到878/2412进行）
单纯看PPD是绝对错误的，单卡单开的PPD最高显示2545，我刚双开FAH时FAHMON显示居然有4300＋，可是双开运行一短时间后总的PPD逐步下降到3814


[ 本帖最后由 金鹏 于 2008-10-4 10:46 编辑 ]

金鹏

为了进一步证实双开后真实的PPD，又下载了两个4742包进行双开，这个是GPU目录下的任务一

[ 本帖最后由 金鹏 于 2008-10-3 09:04 编辑 ]

金鹏

这个是GPU2目录下的任务二

金鹏

随着运行时间的增长总PPD逐渐下降，比较可喜的是同样显示99％负载，显卡温度由单开的68度升高到双开的71－73度，从另外一个角度是不是反映了显卡负载的增大？
双开运行了一个多小时后PPD降到2823，疑问来了：一是PPD会不会降到单卡单开的2545？二是运算结果能不能正常提交并下载新任务？

[ 本帖最后由 金鹏 于 2008-10-3 09:15 编辑 ]

金鹏

由于双开，显示完成任务时间是早上快7点了，所以上床睡觉，早上再来一看究竟
早上来看后，昨晚的两个任务完成并成功提交，新下载的任务也在正常运行中显示为2699 PPD，也是高于单开的2545  PPD
下面看看完成两个任务的时间，便于衡量双开是不是提高了运算能力？

# Windows GPU Console Edition #################################################
###############################################################################

                       Folding@Home Client Version 6.20r1

                          http://folding.stanford.edu

###############################################################################
###############################################################################

Launch directory: C:\Documents and Settings\HuHu\Application Data\Folding@home-gpu


[12:14:41] - Ask before connecting: No
[12:14:41] - User name: husq (Team 3213)
[12:14:41] - User ID: 18E13CFC5B14E2F6
[12:14:41] - Machine ID: 1
[12:14:41]
[12:14:41] Loaded queue successfully.
[12:14:41] Initialization complete
[12:14:41]
[12:14:41] + Processing work unit
[12:14:41] Core required: FahCore_11.exe
[12:14:41] Core found.
[12:14:41] Working on queue slot 08 [October 2 12:14:41 UTC]
[12:14:41] + Working ...
[12:14:42]
[12:14:42] *------------------------------*
[12:14:42] Folding@Home GPU Core - Beta
[12:14:42] Version 1.10 (Tue Aug 12 10:03:11 PDT 2008)
[12:14:42]
[12:14:42] Compiler  : Microsoft (R) 32-bit C/C++ Optimizing Compiler Version 14.00.50727.762 for 80x86
[12:14:42] Build host: amoeba
[12:14:42] Board Type: AMD
[12:14:42] Core      :
[12:14:42] Preparing to commence simulation
[12:14:42] - Looking at optimizations...
[12:14:42] - Files status OK
[12:14:42] - Expanded 87926 -> 447304 (decompressed 508.7 percent)
[12:14:42] Called DecompressByteArray: compressed_data_size=87926 data_size=447304, decompressed_data_size=447304 diff=0
[12:14:42] - Digital signature verified
[12:14:42]
[12:14:42] Project: 4742 (Run 0, Clone 423, Gen 2)
[12:14:42]
[12:14:42] Assembly optimizations on if available.
[12:14:42] Entering M.D.
[12:14:48] Will resume from checkpoint file
[12:14:48] Working on p4742_lam5w_300K
[12:14:48] Client config found, loading data.
[12:14:48] Starting GUI Server
[12:14:51] Resuming from checkpoint
[12:14:51] Verified work/wudata_08.log
[12:14:51] Verified work/wudata_08.edr
[12:14:51] Verified work/wudata_08.xtc
[12:18:10] Completed 1%
[12:22:20] Completed 2%
[12:28:33] Completed 3%
[12:34:47] Completed 4%
[12:40:59] Completed 5%
[12:47:13] Completed 6%
[12:53:27] Completed 7%
[12:59:39] Completed 8%
[13:05:53] Completed 9%
[13:12:06] Completed 10%
[13:17:04] Opening C:\Documents and Settings\HuHu\Application Data\Folding@home-gpu\MyFolding.html...
[13:18:20] Completed 11%
[13:24:37] Completed 12%
[13:30:55] Completed 13%
[13:37:13] Completed 14%
[13:43:27] Completed 15%
[13:49:45] Completed 16%
[13:56:03] Completed 17%
[14:02:20] Completed 18%
[14:08:37] Completed 19%
[14:14:57] Completed 20%
[14:21:21] Completed 21%
[14:27:38] Completed 22%
[14:33:59] Completed 23%
[14:40:21] Completed 24%
[14:46:50] Completed 25%
[14:53:06] Completed 26%
[14:59:24] Completed 27%
[15:05:42] Completed 28%
[15:11:58] Completed 29%
[15:18:15] Completed 30%
[15:24:32] Completed 31%
[15:30:49] Completed 32%
[15:37:15] Completed 33%
[15:43:33] Completed 34%
[15:49:49] Completed 35%
[15:56:04] Completed 36%
[16:02:18] Completed 37%
[16:08:30] Completed 38%
[16:14:44] Completed 39%
[16:20:57] Completed 40%
[16:27:08] Completed 41%
[16:33:17] Completed 42%
[16:39:29] Completed 43%
[16:45:42] Completed 44%
[16:51:54] Completed 45%
[16:58:08] Completed 46%
[17:04:20] Completed 47%
[17:10:35] Completed 48%
[17:16:47] Completed 49%
[17:22:58] Completed 50%
[17:29:11] Completed 51%
[17:35:24] Completed 52%
[17:41:36] Completed 53%
[17:47:48] Completed 54%
[17:53:59] Completed 55%
[18:00:10] Completed 56%
[18:06:24] Completed 57%
[18:12:35] Completed 58%
[18:14:41] + Working...
[18:18:47] Completed 59%
[18:25:00] Completed 60%
[18:31:11] Completed 61%
[18:37:24] Completed 62%
[18:43:38] Completed 63%
[18:49:46] Completed 64%
[18:55:56] Completed 65%
[19:02:10] Completed 66%
[19:08:22] Completed 67%
[19:14:33] Completed 68%
[19:20:46] Completed 69%
[19:27:01] Completed 70%
[19:33:10] Completed 71%
[19:39:25] Completed 72%
[19:45:37] Completed 73%
[19:51:48] Completed 74%
[19:57:59] Completed 75%
[20:04:09] Completed 76%
[20:10:22] Completed 77%
[20:16:35] Completed 78%
[20:22:48] Completed 79%
[20:28:59] Completed 80%
[20:35:12] Completed 81%
[20:41:25] Completed 82%
[20:47:35] Completed 83%
[20:53:46] Completed 84%
[21:00:00] Completed 85%
[21:06:12] Completed 86%
[21:12:26] Completed 87%
[21:18:37] Completed 88%
[21:24:50] Completed 89%
[21:31:03] Completed 90%
[21:37:16] Completed 91%
[21:43:29] Completed 92%
[21:49:40] Completed 93%
[21:55:53] Completed 94%
[22:02:06] Completed 95%
[22:08:21] Completed 96%
[22:14:33] Completed 97%
[22:20:45] Completed 98%
[22:26:56] Completed 99%
[22:33:07] Completed 100%
[22:34:07]
[22:34:07] Finished Work Unit:
[22:34:07] - Reading up to 30216 from "work/wudata_08.trr": Read 30216
[22:34:07] trr file hash check passed.
[22:34:07] - Reading up to 1959464 from "work/wudata_08.xtc": Read 1959464
[22:34:07] xtc file hash check passed.
[22:34:07] edr file hash check passed.
[22:34:07] logfile size: 76649
[22:34:07] Leaving Run
[22:34:08] - Writing 2067401 bytes of core data to disk...
[22:34:08] Done: 2066889 -> 1990395 (compressed to 96.2 percent)
[22:34:08]   ... Done.
[22:34:08] - Shutting down core

[ 本帖最后由 金鹏 于 2008-10-3 09:37 编辑 ]

金鹏

GPU下的任务1用时10小时15分，GPU2下的任务2用时也是差不多10小时15分，说明同样的4742任务双开用时应该差不多
计算一下两个任务平均值，即完成一个4742的任务双开需要5小时08分，比单开的5小时30分效率提高  7％ 左右，（按照PPD测算双开比单开提高6％－8％）
以上是计算误差？还是真的有所提高？

# Windows GPU Console Edition #################################################
###############################################################################

                       Folding@Home Client Version 6.20r1

                          http://folding.stanford.edu

###############################################################################
###############################################################################

Launch directory: C:\Program Files\Folding@home\Folding@home-gpu2


[12:20:01] - Ask before connecting: No
[12:20:01] - User name: husq (Team 3213)
[12:20:01] - User ID: 18E13CFC5B14E2F6
[12:20:01] - Machine ID: 2
[12:20:01]
[12:20:01] Loaded queue successfully.
[12:20:01] Initialization complete
[12:20:01]
[12:20:01] + Processing work unit
[12:20:01] Core required: FahCore_11.exe
[12:20:01] Core found.
[12:20:01] Working on queue slot 08 [October 2 12:20:01 UTC]
[12:20:01] + Working ...
[12:20:02]
[12:20:02] *------------------------------*
[12:20:02] Folding@Home GPU Core - Beta
[12:20:02] Version 1.10 (Tue Aug 12 10:03:11 PDT 2008)
[12:20:02]
[12:20:02] Compiler  : Microsoft (R) 32-bit C/C++ Optimizing Compiler Version 14.00.50727.762 for 80x86
[12:20:02] Build host: amoeba
[12:20:02] Board Type: AMD
[12:20:02] Core      :
[12:20:02] Preparing to commence simulation
[12:20:02] - Looking at optimizations...
[12:20:02] - Files status OK
[12:20:02] - Expanded 87890 -> 447304 (decompressed 508.9 percent)
[12:20:02] Called DecompressByteArray: compressed_data_size=87890 data_size=447304, decompressed_data_size=447304 diff=0
[12:20:02] - Digital signature verified
[12:20:03]
[12:20:03] Project: 4742 (Run 7, Clone 378, Gen 1)
[12:20:03]
[12:20:03] Assembly optimizations on if available.
[12:20:03] Entering M.D.
[12:20:09] Will resume from checkpoint file
[12:20:09] Working on p4742_lam5w_300K
[12:20:09] Client config found, loading data.
[12:20:09] Starting GUI Server
[12:20:11] Resuming from checkpoint
[12:20:11] Verified work/wudata_08.log
[12:20:11] Verified work/wudata_08.edr
[12:20:11] Verified work/wudata_08.xtc
[12:26:26] Completed 1%
[12:32:37] Completed 2%
[12:38:50] Completed 3%
[12:45:02] Completed 4%
[12:51:12] Completed 5%
[12:57:25] Completed 6%
[13:03:38] Completed 7%
[13:09:51] Completed 8%
[13:16:04] Completed 9%
[13:17:19] Opening C:\Program Files\Folding@home\Folding@home-gpu2\MyFolding.html...
[13:22:20] Completed 10%
[13:28:38] Completed 11%
[13:34:58] Completed 12%
[13:41:13] Completed 13%
[13:47:30] Completed 14%
[13:53:47] Completed 15%
[14:00:05] Completed 16%
[14:06:21] Completed 17%
[14:12:39] Completed 18%
[14:19:05] Completed 19%
[14:25:26] Completed 20%
[14:31:44] Completed 21%
[14:38:06] Completed 22%
[14:44:34] Completed 23%
[14:50:55] Completed 24%
[14:57:11] Completed 25%
[15:03:30] Completed 26%
[15:09:46] Completed 27%
[15:16:03] Completed 28%
[15:22:22] Completed 29%
[15:28:36] Completed 30%
[15:35:01] Completed 31%
[15:41:22] Completed 32%
[15:47:39] Completed 33%
[15:53:55] Completed 34%
[16:00:06] Completed 35%
[16:06:20] Completed 36%
[16:12:31] Completed 37%
[16:18:42] Completed 38%
[16:24:53] Completed 39%
[16:31:09] Completed 40%
[16:37:22] Completed 41%
[16:43:34] Completed 42%
[16:49:45] Completed 43%
[16:55:57] Completed 44%
[17:02:08] Completed 45%
[17:08:19] Completed 46%
[17:14:33] Completed 47%
[17:20:44] Completed 48%
[17:26:56] Completed 49%
[17:33:09] Completed 50%
[17:39:22] Completed 51%
[17:45:33] Completed 52%
[17:51:46] Completed 53%
[17:57:59] Completed 54%
[18:04:10] Completed 55%
[18:10:22] Completed 56%
[18:16:34] Completed 57%
[18:20:01] + Working...
[18:22:49] Completed 58%
[18:29:01] Completed 59%
[18:35:12] Completed 60%
[18:41:24] Completed 61%
[18:47:37] Completed 62%
[18:53:51] Completed 63%
[19:00:04] Completed 64%
[19:06:17] Completed 65%
[19:12:30] Completed 66%
[19:18:41] Completed 67%
[19:24:52] Completed 68%
[19:31:05] Completed 69%
[19:37:15] Completed 70%
[19:43:27] Completed 71%
[19:49:39] Completed 72%
[19:55:54] Completed 73%
[20:02:09] Completed 74%
[20:08:21] Completed 75%
[20:14:32] Completed 76%
[20:20:43] Completed 77%
[20:26:56] Completed 78%
[20:33:08] Completed 79%
[20:39:19] Completed 80%
[20:45:33] Completed 81%
[20:51:46] Completed 82%
[20:57:58] Completed 83%
[21:04:11] Completed 84%
[21:10:22] Completed 85%
[21:16:33] Completed 86%
[21:22:45] Completed 87%
[21:28:59] Completed 88%
[21:35:08] Completed 89%
[21:41:21] Completed 90%
[21:47:32] Completed 91%
[21:53:44] Completed 92%
[21:59:56] Completed 93%
[22:06:07] Completed 94%
[22:12:17] Completed 95%
[22:18:28] Completed 96%
[22:24:43] Completed 97%
[22:30:56] Completed 98%
[22:35:40] Completed 99%
[22:41:52] Completed 100%
[22:42:52]
[22:42:52] Finished Work Unit:
[22:42:52] - Reading up to 30216 from "work/wudata_08.trr": Read 30216
[22:42:52] trr file hash check passed.
[22:42:52] - Reading up to 1960964 from "work/wudata_08.xtc": Read 1960964
[22:42:52] xtc file hash check passed.
[22:42:52] edr file hash check passed.
[22:42:52] logfile size: 76485
[22:42:52] Leaving Run
[22:42:53] - Writing 2068737 bytes of core data to disk...
[22:42:53] Done: 2068225 -> 1991518 (compressed to 96.2 percent)
[22:42:53]   ... Done.
[22:42:53] - Shutting down core

[ 本帖最后由 金鹏 于 2008-10-3 10:01 编辑 ]

金鹏

这两张是今天早上双开新的任务，一个4743的，另一个4742的，总的PPD为2699
稍后再看看完成任务的时间，另外其他使用A卡的兄弟也可以试试双开是不是提高了A卡的总运算能力

[ 本帖最后由 金鹏 于 2008-10-3 09:46 编辑 ]

金鹏

这个是从统计网页上查来的，说明双开提交的两个548分值的结果有效

[ 本帖最后由 金鹏 于 2008-10-3 10:14 编辑 ]

shouldbe

不错。和双核CPU用affinity软件的效果很像。但不如4核CPU用双开的效果。

cicikml

很有益的参考，谢谢楼主
可惜的是提升非常小，只有两个548PPD。

zxy0714

原帖由 金鹏 于 2008-10-3 10:12 发表
这个是从统计网页上查来的，说明双开提交的两个548分值的结果有效

3870x2不算什么单卡吧,其实就是里面两个3870显示核心,你想想啊,3870x2功耗230W,就算给你双开跑出4800PPD,还是比不上88GT超到5300PPD的性能.而且,88GT功耗仅有100W...

金鹏

原帖由 cicikml 于 2008-10-3 11:47 发表
很有益的参考，谢谢楼主
可惜的是提升非常小，只有两个548PPD。

呵呵，兄弟是两个548分值的WU，单卡双开最后稳定在2699 PPD，比单卡单开多了150 PPD
提升有限

金鹏

原帖由 zxy0714 于 2008-10-3 11:50 发表


3870x2不算什么单卡吧,其实就是里面两个3870显示核心,你想想啊,3870x2功耗230W,就算给你双开跑出4800PPD,还是比不上88GT超到5300PPD的性能.而且,88GT功耗仅有100W... ...

兄弟说的是，可是很多买了A卡的兄弟们不会都像我一样马上再换更高端的卡滴

这个想法还是受你的800SP没完全利用的说法提醒，主要验证一下是不是全部利用上了，现在看来双开后不存在800SP没充分利用的效率问题，而是MOON.....兄弟说的是架构和运算密度的问题，看来CUDA不是炒作的，是真货

[ 本帖最后由 金鹏 于 2008-10-3 15:06 编辑 ]

金鹏

原帖由 fdb2002 于 2008-10-3 18:35 发表
我的双卡（HD4850+9600gt）怎么双开呢？

双卡双开我也正在尝试呢，你的问题也是我的问题，最希望是4870主卡＋9600GT做物理加速卡实现FAH的CUDA计算这样的结果
兄弟让我们一起尝试吧