折腾了一下双开蛋疼的764X包--还是吃不饱

金鹏

VB下smp 7带着480跑764X显卡始终吃不饱，PPD也才13K 无聊试了试双开跑，提升PPD大概1K，GPU吃到80%-90%，显存占用翻倍达到1279M




关闭CPU客户端，空闲CPU环境下双开PPD 16.1K, 单开PPD  15.6K  ,结论：蛋疼，怎么也吃不饱

rawarrior

fah严重地打击了本次竞赛队员的心情..
不过鉴于大家的GPU接的都是764x的包,对于比赛结果影响较小..

金版这样的cpu大户要占优势咯...


回复 1# 金鹏
GTS450从803x的1w ppd,变成了764x的6K...苦不堪言..

金鹏

fah严重地打击了本次竞赛队员的心情..
不过鉴于大家的GPU接的都是764x的包,对于比赛结果影响较小..

金版这 ...
rawarrior 发表于 2012-4-6 23:37

我和其他几位兄弟不在6K-50K范围内，所以我们只管跑分，兄弟你们尽量冲击560SE

bugfix

01:08:40:    <client-type v='beta'/>
01:08:40:  </slot>
01:08:40:</config>
01:08:40:Trying to access database...
01:08:40:Successfully acquired database lock
01:08:40:Enabled folding slot 00: READY gpu:0:"GF100 [GeForce GTX 470]"
01:08:40:WU00:FS00:Starting
01:08:40:WU00:FS00:Running FahCore: D:\FAHClient/FAHCoreWrapper.exe "d:/Application Data/FAHClient/cores/www.stanford.edu/~pande/Win32/x86/NVIDIA/Fermi/beta/Core_16.fah/FahCore_16.exe" -dir 00 -suffix 01 -version 701 -checkpoint 3 -gpu 0 oneunit forceasm
01:08:41:WU00:FS00:Started FahCore on PID 3156
01:08:42:WU00:FS00:Core PID:5892
01:08:42:WU00:FS00:FahCore 0x16 started
01:08:43:Server connection id=1 on 0.0.0.0:36330 from 127.0.0.1
01:08:43:WU00:FS00:0x16:
01:08:43:WU00:FS00:0x16:*------------------------------*
01:08:43:WU00:FS00:0x16:Folding@Home GPU Core
01:08:43:WU00:FS00:0x16:Version                2.24 (Thu Feb 23 13:21:02 PST 2012)
01:08:43:WU00:FS00:0x16:Build host             ASUS_Laptop
01:08:43:WU00:FS00:0x16:Board Type             NVIDIA/CUDA
01:08:43:WU00:FS00:0x16:Core                   16
01:08:43:WU00:FS00:0x16:
01:08:43:WU00:FS00:0x16:Window's signal control handler registered.
01:08:43:WU00:FS00:0x16:Preparing to commence simulation
01:08:43:WU00:FS00:0x16:- Looking at optimizations...
01:08:43:WU00:FS00:0x16:- Files status OK
01:08:43:WU00:FS00:0x16:sizeof(CORE_PACKET_HDR) = 512 file=<>
01:08:43:WU00:FS00:0x16:- Expanded 59587 -> 269406 (decompressed 452.1 percent)
01:08:43:WU00:FS00:0x16:Called DecompressByteArray: compressed_data_size=59587 data_size=269406, decompressed_data_size=269406 diff=0
01:08:43:WU00:FS00:0x16:- Digital signature verified
01:08:43:WU00:FS00:0x16:
01:08:43:WU00:FS00:0x16:Project: 8023 (Run 18, Clone 0, Gen 30)
01:08:43:WU00:FS00:0x16:
01:08:43:WU00:FS00:0x16:Assembly optimizations on if available.
01:08:43:WU00:FS00:0x16:Entering M.D.
01:08:45:WU00:FS00:0x16:Will resume from checkpoint file 00/wudata_01.ckp
01:08:45:WU00:FS00:0x16:Tpr hash 00/wudata_01.tpr:  4278448954 3866793941 2448799410 2420824293 3542095267
01:08:45:WU00:FS00:0x16:GPU device id=0
01:08:45:WU00:FS00:0x16:Working on Protein
01:08:45:WU00:FS00:0x16:Client config unavailable.

明鏡止水

现在又都是80XX了。。大家继续飙吧

rawarrior

回复 5# 明鏡止水


    7641跑了46%..还要15个小时才完事...扔是不扔...

rawarrior

回复  明鏡止水


    7641跑了46%..还要15个小时才完事...扔是不扔...
rawarrior 发表于 2012-4-7 10:27

    刚才弄了一下，丢了七八个764x，还是没有803x，只好继续跑7641了。。

玩家共和国

擦，480的PPD居然是5234，GPU使用率40%，跑了16小时还剩5小时。昨晚在枕头边那一宿噪音真不值得啊

bugfix

This project (p7641) is set for 7718.00 points, a preferred deadline of 13.00 days, and a final deadline 18.00 days. This project uses the OPENMMGPU FAH core software and is hosted by Folding@home server 171.64.65.93.

金鹏

目前摆脱764X包的方法就是加上参数

1. -verbosity 9 -betateam

复制代码

跑7612 7613 内测包,虽比不上803X的PPD高，但和以前的持平

rawarrior

回复 10# 金鹏


    请问,v7客户端怎么弄...

玩家共和国

回复 10# 金鹏


  这包好久没见了啊，谢金版良方~

cuda

回复 1# 金鹏

奇怪，我刚才进行测试，p7641单开GPU占用96%-97%，没发现吃不饱的现象。
TPF 6m45s，PPD算起来和一般GPU包差别不大(显卡为GTX580)：

1. [09:47:00] Project: 7641 (Run 348, Clone 0, Gen 4)
   
2. [09:47:00]
   
3. [09:47:00] Assembly optimizations on if available.
   
4. [09:47:00] Entering M.D.
   
5. [09:47:02] Tpr hash work/wudata_02.tpr:  4281520380 1884500861 1965718380 3134043502 573556625
   
6. [09:47:02] GPU device info: vendor=0 device=0 name=<NA> match=0
   
7. [09:47:02] Working on Protein in water
   
8. [09:47:02] Client config found, loading data.
   
9. [09:47:02] Starting GUI Server
   
10. [09:48:52] Setting checkpoint frequency: 25000
    
11. [09:48:52] Completed         3 out of 2500000 steps (0%).
    
12. [09:55:37] Completed     25000 out of 2500000 steps (1%).
    
13. [10:02:22] Completed     50000 out of 2500000 steps (2%).
    
14. [10:09:08] Completed     75000 out of 2500000 steps (3%).
    
15. [10:15:53] Completed    100000 out of 2500000 steps (4%).

复制代码

金鹏

回复 11# rawarrior

V7没弄过

貌似在GPU客户端设置那里加入这个参数

金鹏

回复 13# cuda

99%才算吃饱。兄弟的580才跑出16.5K PPD，比起803X的25K+ PPD差了接近10K

这个包最大缺陷就是不能和其他分布式程序一块跑，否则GPU占用陡降，

3213兄弟们大多是CPU+GPU客户端一块跑，所以很蛋疼