GeForce GPU Runs FAH Protein Simulations 140x Faster Than Traditional Processors

Edison · 发表于 2008-7-29 13:54

http://biz.yahoo.com/prnews/080724/aqth016.html?.v=61

NVIDIA Dramatically Accelerates the Search For a Cure
Thursday July 24, 9:00 am ET
GeForce GPU Runs Folding@home Protein Simulations 140 Times Faster Than Traditional Processors

SANTA CLARA, Calif., July 24 /PRNewswire-FirstCall/ -- Stanford University's distributed computing program Folding@home has become a major force in researching cures to life-threatening diseases such as cancer, cystic fibrosis, and Parkinson's disease by combining the computing horsepower of millions of processors to simulate protein folding. The Folding@home project is the latest example in the expanding list of non-gaming applications for graphics processing units (GPU). By running the Folding@home client on an NVIDIA® GeForce® GPU, protein-folding simulations can be done 140 times faster than on some of today's traditional CPUs.

ADVERTISEMENT
(Logo: http://www.newscom.com/cgi-bin/prnh/20020613/NVDALOGO)

"The impact of GeForce GPUs on protein folding simulations was immediate and dramatic," said Vijay Pande, associate professor of chemistry, Stanford University and director of the Folding@home project. "Teams that are folding with GeForce GPUs are seeing their production skyrocket. Applying that kind of processing power to Folding@home changes the whole dynamic of the project and could significantly reduce the time it takes to carry out our biomedical research."

The Folding@home project has amassed a large following of computer enthusiasts who compete in teams to churn through as many data units as possible. Their unofficial stats are organized by and displayed at ExtremeOverclocking.com. It took the NVIDIA internal folding team only two weeks to move ahead of 90% of all teams, using only 10 machines. After expanding the team to include more GPUs, the NVIDIA team has moved inside the top 0.1% of teams in all-time total production in less than a month.

Other folding teams are also seeing their status rise as a result of the NVIDIA Folding@home client.

"We saw the completed work double for our PC Games Hardware Folding team as a result of many team members installing the NVIDIA Folding client," said Carsten Spille, editor at PC Games Hardware. "We are passing many teams every day and we have finally reached our goal of being one of the top 100 folding teams in the world

boris_lee · 发表于 2008-7-29 13:56

CUDA又一次后来居上:lol:
期待larabee的表现

80x87 兮归来~

tomsmith123 · 发表于 2008-7-29 14:10

140X，就算按照两个GTX280 算，这传统CPU 也太渣了吧？

boris_lee · 发表于 2008-7-29 14:12

原帖由 tomsmith123 于 2008-7-29 14:10 发表
140X，就算按照两个GTX280 算，这传统CPU 也太渣了吧？

F@H这种应用也很正常啊。
FFT传统cpu还被三四百M的Ti虐呢{titter:]

tomsmith123 · 发表于 2008-7-29 14:20

按照峰值计算，理论峰值，GTX280 就算1.3TGFlops，这种事情很难发生，即使是FAH，两个2.6TFlops，除以140X，那么所谓传统CPU 的性能不过是不足20G。
20G 什么概念？一个Q6600不超频可以达到32G。

gz_easy · 发表于 2008-7-29 14:28

古老的E6700的峰值都达到21.28GFLOPS。
http://www.pcinlife.com/article/ ... 4158113d191_24.html

Edison · 发表于 2008-7-29 14:35

NVIDIA在6月20日开始提供CUDA FAH，下面是这半年来GPU FAH的变化状况：

Marth 25th 2008 - 27 TFLOPS - 465 Active CPUs - 5261 Total CPUs

June 16th 2008 - 143 TFLOPS - 2068 Actice CPUs - 9059 Total CPUs (Right before GPU2 was launched)

June 20th 2008 - 541 TFLOPS

June 21st 2008 - 604 TFLOPS

June 22nd 2008 - 670 TFLOPs - 6093 Active CPUs - 13282 Total CPUs

June 23rd 2008 - 718 TFLOPS - 6527 Active CPUs - 13735 Total CPUs

June 24th 2008 - 771 TFLOPS - 7006 Active CPUs - 14226 Total CPUs

June 25th 2008 - 804 TFLOPS - 7311 Active CPUs - 14531 Total CPUs

June 26th 2008 - 834 TFLOPS - 7585 Active CPUs - 14828 Total CPUs

June 27th 2008 - 871 TFLOPS - 7917 Active CPUs - 15179 Total CPUs

June 28th 2008 - 894 TFLOPS - 8124 Active CPUs - 15480 Total CPUs

June 29th 2008 - 902 TFLOPS - 8197 Active CPUs - 15763 Total CPUs

June 30th 2008 - 895 TFLOPS - 8139 Active CPUs - 16013 Total CPUs (Decrease! no!)

July 1st 2008 - 890 TFLOPS - 8089 Active CPUs - 16239 Total CPUs

July 2nd 2008 - 886 TFLOPS - 8058 Active CPUs - 16498 Total CPUs (1.07 Core and Multiple GPU2's possible)

July 3rd 2008 - 894 TFLOPS - 8127 Active CPUs - 16768 Total CPUs

July 4th 2008 - 904 TFLOPS - 8217 Active CPUs - 17191 Total CPUs (2.5 Petaflops for the total F@H effort was passed)

July 5th 2008 - 902 TFLOPS - 8199 Active CPUs - 17284 Total CPUs (GPU2 server went down for a bit)

July 6th 2008 - 914 TFLOPS - 8311 Active CPUs - 17586 Total CPUs

July 7th 2008 - 924 TFLOPS - 8404 Active CPUs - 17890 Total CPUs

July 8th 2008 - 933 TFLOPS - 8480 Active CPUs - 18115 Total CPUs

July 9th 2008 - 942 TFLOPS - 8566 Active CPUs - 18374 Total CPUs

July 10th 2008 - 953 TFLOPS - 8666 Active CPUs - 18650 Total CPUs (Linux and Windows clients seeing slight drops)

July 11th 2008 - 956 TFLOPS - 8695 Active CPUs - 18889 Total CPUs

July 12th 2008 - 972 TFLOPS - 8837 Active CPUs - 19173 Total CPUs

July 13th 2008 - 977 TFLOPS - 8878 Active CPUs - 19383 Total CPUs

July 14th 2008 - 984 TFLOPS - 8943 Active CPUs - 19623 Total CPUs

July 15th 2008 - 987 TFLOPS - 8975 Active CPUs - 19836 Total CPUs

July 16th 2008 - 997 TFLOPS - 9068 Active CPUs - 20000 Total CPUs

July 17th 2008 - 1008 TFLOPS - 9168 Active CPUs - 20261 Total CPUs (GPU2 passes 1 PetaFlop)

July 18th 2008 - 1020 TFLOPS - 9275 Active CPUs - 20521 Total CPUs

July 19th 2008 - 1023 TFLOPS - 9298 Active CPUs - 20735 Total CPUs

July 20th 2008 - 1032 TFLOPS - 9378 Active CPUs - 20987 Total CPUs

July 21st 2008 - 1039 TFLOPS - 9442 Active CPUs - 21321 Total CPUs

July 22nd 2008 - 1045 TFLOPS - 9496 Active CPUs - 21425 Total CPUs

boris_lee · 发表于 2008-7-29 14:41

原帖由 tomsmith123 于 2008-7-29 14:20 发表
按照峰值计算，理论峰值，GTX280 就算1.3TGFlops，这种事情很难发生，即使是FAH，两个2.6TFlops，除以140X，那么所谓传统CPU 的性能不过是不足20G。
20G 什么概念？一个Q6600不超频可以达到32G。

这个我不是很确定了，可能FAH的算法对CPU优化的不够好，中间等待I/O的时间过多？
反正标称几Gflops的p43.0 算FFT比标称1.8G FLOPS的虎鲨101差得远

Edison · 发表于 2008-7-29 14:46

大致上这半年的GPU FAH状况就是从AMD only时代的27TFLOPS，变成NVIDIA内测开始时代的143TFLOPS，然后是媒体发放（带动了诸如HardOCP等团队）的6月20日的541 TFLOPS，现在增长到1PFLOPS。

harleylg · 发表于 2008-7-29 14:54

原帖由 tomsmith123 于 2008/7/29 14:20 发表
按照峰值计算，理论峰值，GTX280 就算1.3TGFlops，这种事情很难发生，即使是FAH，两个2.6TFlops，除以140X，那么所谓传统CPU 的性能不过是不足20G。
20G 什么概念？一个Q6600不超频可以达到32G。

CPU按照单核计算就差不多了……1.3T/140=9G，32G/4=8G，大概差不多。

Edison · 发表于 2008-7-29 14:57

6月20日后发生的事情，留意GPU的咖啡线，PS3出现了巨大的抖动，据说是因为MGS4发售 {shocked:] ：

http://foldingforum.org/viewtopic.php?f=16&t=3977

boris_lee · 发表于 2008-7-29 15:04

确定了
F@H的traditional是non-SMP的。{shocked:]
另外有个smp版本的，估计不作为tradtional计算

Edison · 发表于 2008-7-29 15:07

原帖由 boris_lee 于 2008-7-29 15:04 发表
确定了
F@H的traditional是non-SMP的。{shocked:]
另外有个smp版本的，估计不作为tradtional计算

HPC client中的SMP是包含到OS statistics统计的。

gz_easy · 发表于 2008-7-29 15:09

FAH FAQ中有提到目前GPU2使用到CPU资源，完全off load to GPU希望在将来实现。

只看该作者 · 发表于 2008-7-29 15:49

提示: 作者被禁止或删除内容自动屏蔽

天下18 · 发表于 2008-8-1 14:27

提示: 作者被禁止或删除内容自动屏蔽

内个 · 发表于 2008-8-1 14:31

{lol:] FAH成测试主流后 INTEL在CPU里面硬塞硬件解释电路既SUPERPI之后再次无敌于世{victory:]

ATI专业卡 · 发表于 2008-8-1 16:56

哈哈脱离原来的意义了.

只看该作者 · 发表于 2008-8-1 17:25

提示: 作者被禁止或删除内容自动屏蔽

天下18 · 发表于 2008-8-15 11:50

提示: 作者被禁止或删除内容自动屏蔽

帐号		自动登录	找回密码
密码			注册

kamuiyay 该用户已被删除	15^# 发表于 2008-7-29 15:49 \| 只看该作者提示: 作者被禁止或删除内容自动屏蔽
kamuiyay 该用户已被删除
	回复支持反对使用道具举报显身卡

天下18 天下18 当前离线积分 24 IP卡狗仔卡头像被屏蔽	16^# 发表于 2008-8-1 14:27 \| 只看该作者提示: 作者被禁止或删除内容自动屏蔽
天下18 天下18 当前离线积分 24 IP卡狗仔卡头像被屏蔽
	回复支持反对使用道具举报显身卡

kamuiyay 该用户已被删除	19^# 发表于 2008-8-1 17:25 \| 只看该作者提示: 作者被禁止或删除内容自动屏蔽
kamuiyay 该用户已被删除
	回复支持反对使用道具举报显身卡

天下18 天下18 当前离线积分 24 IP卡狗仔卡头像被屏蔽	20^# 发表于 2008-8-15 11:50 \| 只看该作者提示: 作者被禁止或删除内容自动屏蔽
天下18 天下18 当前离线积分 24 IP卡狗仔卡头像被屏蔽
	回复支持反对使用道具举报显身卡

GeForce GPU Runs FAH Protein Simulations 140x Faster Than Traditional Processors

回复 4# boris_lee 的帖子