http://folding.extremeoverclocking....php?s=&u=471448

隊友 rickytzai的日產圖表，

看起像是 做一休一...有時還有周休2日:flash:

這看起來是CPU的WU 非常大
要2天才能算完

周期剛好都一樣?

這是傳說中的extra-large WU, 期限內完成基本分有25000, 提前完成有bonus。
rickytzai隊友50多小時完成一個, bonus 33000左右, 一個WU就可得58000多分,
換算約25000 ppd。

這種WU門檻不低, 要有至少4C/8T及6G RAM。有此種配備的隊友可考慮, 若以
GPU達成25000 ppd至少要有三張GTX275, 電費會比extra-large WU來得高。

kazama前輩，
如果想要跑這種 WU，需要什麼具體設定嗎?如OS與cfg等，
過一陣子WCG各專案都衝到4年後，
小弟也想來跑這種WU. :)

25000 PPD也大約是2.5張9800GX2...
算起來，也是extra-large WU比較省電..

對extra-large WU有興趣的網兄請參考以下overclockers網站文章:
http://www.overclockers.com/forums/...ad.php?t=627195

The Computer:

1) Must have 8 cores to run them, either real cores, or hyperthreaded.

2) Must be an i7 920 or better for a single cpu system. Stanford advises other single cpu quad core systems, can NOT meet the deadline, even folding 24/7. Multiple quad core systems may meet the deadline, however.

3) Must have 4GB RAM minimum, 8GB is requested by Stanford, because the memory requirements will grow as the project continues.

4) This client requires Linux, (either virtualized or native), or OSX, as it's OS.

This is the link to the drop in binary file for PC's:
http://www.stanford.edu/~kasson/folding/linux/fah6

And for Apple OSX:
http://www.stanford.edu/~kasson/folding/osx/fah6

You can just drop this into your folding folder and replace your current fah6 file.


To Participate:

1) You must have the new client, and a passkey. Passkey info and d/l link, is here: http://folding.stanford.edu/English/FAQ-passkey

You may receive an extra large WU to fold without a passkey, but your bonus factor will be only one.

2) You must have folded 10 SMP WU's at least, using that passkey. 80% of these SMP WU's, must have been completed and returned by the deadline.

3) The FAH client configuration requires big WU's to be answered y or yes, and you must NOT have the advanced WU flag present. You need these flags: -bigadv -smp 8

The 8 after the smp flag can be lowered, and some have found leaving one core off their total number of cores, helpful.

Link to the forum for Extra Large WU's:
http://foldingforum.org/viewforum.ph...fdcfbb4d63d197

Points:

1) Points for the Extra Large WU's = base points + bonus points. (or base points * bonus factor).

The base points are approximately 25,000. These points match what the test systems yield folding regular sized SMP WU's, in the same amount of time.

2) Bonus points are calculated on a time-sensitive, sliding scale, using these formulas:

If the WU is returned after it is marked to be reissued, then the bonus factor is one.

If the WU is returned before it is marked to be reissued, and the folder used their passkey, then the bonus factor is calculated by:


bonus factor = square root(deadline_time * k / WU_time)

WU_time: time WU results are received back minus the time WU was issued

deadline_time: when the WU would expire. Also, the time between when the WU was issued, and the time when it would expire.

k: a constant factor set by Stanford. Since k is a multiplier, not a divisor, think of it as "encouragement".

3) A few examples of folding extra large WU's:

CPU Speed Points per WU Points per Day Time to Complete
===========================================================================
1x i7 940 3.67GHz 55,000 avg. 24,000 avg. 2 days 6 hours

2x Xeon 5420 2.5GHz 53,000 avg. 20,380 avg. 2 days 12 hours
============================================================================

感謝kazama前輩的extra-large WU指南，

看起來真的是不低的需求，
比i7等級低的4核心CPU即使全力跑還是可能會逾期，
還有大於4G記憶體需求，如果要用VM跑，要先升級成64bit的windows...

硬體方面都不是問題，
但是要換OS就覺得工程浩大.. :stupefy:

SMP2 preparing for -advmethods, new benchmark
 

http://en.fah-addict.net/news/news-...-benchmarks.php

Peter Kasson has announced that following the recent release of the v6.29 client to the public, the first SMP2 units will be arriving shortly on all 3 platforms for users running SMP with advmethods. The new a3 core is not yet finished but the Pande Group have decided it is ready for open beta as they continue with their refinements.

Unfortunately (especially for Windows users) the SMP2 client will still require MPI to be installed at first as a1/a2 units will continue to be available as a fallback if a3 units run out, but the Pande Group anticipate this to be a short term measure and the MPI requirement will be removed a few weeks into the SMP2 open beta phase.

New clients aside, the SMP2 clients represent an expansion of the early completion bonus used in -bigadv. In line with the client's classification as a "high-performance" client, and to align points awarded to the Pande Group's wish of getting the speediest possible unit returns, all SMP2 units will be subject to the bonus. This means that the points for SMP2 units will appear abnormally low if the bonus is not calculated. Currently FahMon does not calculate the bonus (though it should be able to in the next release), but HFM.NET can.

The requirements to receive the bonus remain the same:
1. Must use a passkey
2. Must successfully return >=10 A2 or SMP2 work units with their passkey
3. Must successfully return >80% of A2 or SMP2 work units, within the deadline

A third change is that the benchmarks machines have been completely updated: SMP and regular CPU clients are now benchmarked by one machine. This machine has a Core i5-750 quad core processor with turbo mode disabled. Points are awarded based on comparisons with the old Pentium 4 benchmark machine (for single CPU clients), and three SMP systems: a 2.4GHz E6600, a 2.4GHz Q6600, and an overclocked 3.2GHz Q6600.

A full explanation of the new benchmarks can be found at the link below, but the new benchmarks are set up so that with the bonus, a 2.4GHz Q6600 is the crossover point for SMP2. Machines slower than this will earn less on SMP2 than with the a2 core, machines more powerful will earn more than with the a2 core.

Happy Folding!

SMP2 Source: Upcoming release of SMP2 cores
New benchmarks Source: Points system for SMP2 work units

jmn On: 01/24/10

就看CPU的extra-large WU長期可得性, 穩定性,運算力與得分間是否有合適的對價關係.

(出現很久了嗎? 需求要八緒應該不會太久之前~ credit不知是否會再調整)

給高額bonus的做法似乎隱含push大家升高階平台跑CPU的意味. (或補償部分運算失敗?)

顯卡還能依照遊戲需求而隨之更新, 看片轉檔之餘順便有空跑跑運算, 換了不會用不到.

若組了個CPU專跑的昂貴平台, 平時高速雙核已夠使用, 沒專案可跑的話就閑置可惜了.

個人以為Stanford研究團隊動機很單純，就是希望能及早取得WU結果以加速研究。
舊有的計分制度只要在deadline前達交WU都得相同分數，並未考慮達交時間差異。
而同一個WU能較早完成，代表的是較少的停機時間或較強的運算能力，而這兩樣
都牽涉到貢獻者投入的resource多寡，bonus制顯然是要反映這項差異並獎勵較早
回傳WU結果對研究的助益。

SMP2 client也實施了bonus制，GPU client將來若比照也不令人意外。
extra-large WU先實施bonus制或許會提供部分貢獻者更換4C/8T平台的動力，
但整體而言個人仍樂見能反映投入resource與達交時效差異的計分制度的實施。

因達交時間的差異而實施差異給分.....這倒是之前沒想過的, 因為資源強,能先運算完的,

它的回報就是能更快開始再算新的任務, 運算資源強的已經在總績分/日產都會反映出來.

而同一個WU, 算快算慢, 貢獻的總運算力仍一致, 故相同計分的原制度也有其公平的出發點.

不過, 既然有新的計分方式, 就再多觀察看看, 相信Pande Lab的人有智慧能處理好遊戲規則. :)