Bigadv points change

[GreyWhiskers]

Deadlines adjusted for 6903 and 6904 'Big Bigadv' work units.

http://foldingforum.org/viewtopic.php?f=24&t=19180.

preferred deadlines are 5 days for 6903, 5.6 for 6904

I saw that the preferred deadline was shortened, not the final deadline. That cuts off the QRB if you don't make the preferred deadline. If you fall between the preferred and final deadline, you get base points. Since the QRB formula is based only on the final deadline, the points acceleration does not change for the über-folders, or any other folders who make the new preferred cutoff.

By my calculation, 5 days preferred deadline is a 72 minute TPF, and 5.6 days is 80:34 [mm:ss] TPF. If, per Grandpa_01, 2600K at 4.4Ghz takes 53 min. to do a frame of a 6903, then that 2600k (by my argument above) fully supports even the new timeline. Per the QRB calculator, it would get, still, 229,909.97 points (given a 12 minute non-overlapped upload time), or 68,480.23 ppd.

Now, I don't have a horse in this particular race, so its somewhat academic to me. But, PG does set their parameters to meet the science needs.

[GreyWhiskers]

Being a "Quant" as I have called it before, I like to look at the numbers. Again, not having a horse in the big big adv race, I threw the new thresholds into the excel sheet I put together a while ago to let me see what's happening to the points and bonuses for the different WUs vs TPF. I've organized this section by completion days (1.1 days to 5.9 days).

Observations:
I didn't think that the transition crossing the desired threshold would be so abrupt. I thought that you would sort of seque into it, and not come to a huge cliff.

That certainly is a motivator to not miss the desired threshold, but that value may still be set to long if these projects are designed to bring the big iron to the party. Where would one draw a new threshold, a new cliff. Is there any objective scientific project background for how quickly the project needs these WUs back to cause the threshold to be brought down even further?

I think I've got the formulas right in the spreadsheet, but I ask for constructive feedback if I don't.

EDIT: reformatted to show better in Code window. Preview wasn't right.

Code: Select all

                                      6904                      6903
                         pts        31,541                    22,706
                           k         37.31                     38.05
                    deadline            17                        12
       TPF          threshold          5.6                         5
 dec min if>60     WU Compl
 else mm:ss    hrs     Days        Points      ppd           Points      ppd
      84.96    141.6     5.90     31,541.0     5,345.9      22,706.0     3,848.5
      83.52    139.2     5.80     31,541.0     5,438.1      22,706.0     3,914.8
      82.08    136.8     5.70     31,541.0     5,533.5      22,706.0     3,983.5
      80.64    134.4     5.60     31,541.0     5,632.3      22,706.0     4,054.6
      79.20    132.0     5.50    338,712.5    61,584.1      22,706.0     4,128.4
      77.76    129.6     5.40    341,834.3    63,302.7      22,706.0     4,204.8
      76.32    127.2     5.30    345,044.1    65,102.7      22,706.0     4,284.2
      74.88    124.8     5.20    348,346.0    66,989.6      22,706.0     4,366.5
      73.44    122.4     5.10    351,744.6    68,969.5      22,706.0     4,452.2
      72.00    120.0     5.00    355,244.6    71,048.9      22,706.0     4,541.2
      70.56    117.6     4.90    358,851.3    73,235.0     219,184.9    44,731.6
      69.12    115.2     4.80    362,570.0    75,535.4     221,456.3    46,136.7
      67.68    112.8     4.70    366,406.9    77,958.9     223,799.8    47,617.0
      66.24    110.4     4.60    370,368.1    80,514.8     226,219.3    49,178.1
      64.80    108.0     4.50    374,460.7    83,213.5     228,719.0    50,826.5
      63.36    105.6     4.40    378,692.1    86,066.4     231,303.5    52,569.0
      61.92    103.2     4.30    383,070.2    89,086.1     233,977.6    54,413.4
      60.48    100.8     4.20    387,603.7    92,286.6     236,746.7    56,368.3
      59:02     98.4     4.10    392,302.1    95,683.4     239,616.5    58,443.0
      57:36     96.0     4.00    397,175.6    99,293.9     242,593.2    60,648.3
      56:10     93.6     3.90    402,235.4   103,137.3     245,683.7    62,995.8
      54:43     91.2     3.80    407,493.6   107,235.1     248,895.4    65,498.8
      53:17     88.8     3.70    412,963.5   111,611.8     252,236.4    68,172.0
      51:50     86.4     3.60    418,659.8   116,294.4     255,715.7    71,032.1
      50:24     84.0     3.50    424,598.6   121,313.9     259,343.0    74,098.0
      48:58     81.6     3.40    430,797.4   126,705.1     263,129.3    77,391.0
      47:31     79.2     3.30    437,275.9   132,507.9     267,086.3    80,935.2
      46:05     76.8     3.20    444,055.8   138,767.4     271,227.4    84,758.6
      44:38     74.4     3.10    451,161.2   145,535.9     275,567.3    88,892.7
      43:12     72.0     3.00    458,618.9   152,873.0     280,122.5    93,374.2
      41:46     69.6     2.90    466,459.1   160,848.0     284,911.2    98,245.3
      40:19     67.2     2.80    474,715.6   169,541.3     289,954.3   103,555.1
      38:53     64.8     2.70    483,426.7   179,046.9     295,275.0   109,361.1
      37:26     62.4     2.60    492,635.7   189,475.3     300,899.8   115,730.7
      36:00     60.0     2.50    502,391.8   200,956.7     306,858.8   122,743.5
      34:34     57.6     2.40    512,751.5   213,646.5     313,186.5   130,494.4
      33:07     55.2     2.30    523,779.7   227,730.3     319,922.4   139,096.7
      31:41     52.8     2.20    535,551.5   243,432.5     327,112.6   148,687.5
      30:14     50.4     2.10    548,154.4   261,025.9     334,810.4   159,433.5
      28:48     48.0     2.00    561,691.1   280,845.6     343,078.6   171,539.3
      27:22     45.6     1.90    576,282.9   303,306.8     351,991.2   185,258.5
      25:55     43.2     1.80    592,074.4   328,930.2     361,636.6   200,909.2
      24:29     40.8     1.70    609,239.5   358,376.2     372,121.0   218,894.7
      23:02     38.4     1.60    627,989.7   392,493.6     383,573.5   239,733.4
      21:36     36.0     1.50    648,585.0   432,390.0     396,153.0   264,102.0
      20:10     33.6     1.40    671,349.3   479,535.2     410,057.3   292,898.1
      18:43     31.2     1.30    696,692.1   535,917.0     425,536.6   327,335.8
      17:17     28.8     1.20    725,140.1   604,283.4     442,912.5   369,093.8
      15:50     26.4     1.10    757,384.1   688,531.0     462,607.0   420,551.9
      14:24     24.0     1.00    794,351.2   794,351.2     485,186.4   485,186.4
      13:48     23.0     0.96    811,436.0   846,715.8     495,621.7   517,170.4
      13:12     22.0     0.92    829,672.7   905,097.5     506,760.6   552,829.8
      12:36     21.0     0.88    849,197.1   970,511.0     518,686.0   592,784.0
      12:00     20.0     0.83    870,168.1 1,044,201.7     531,495.0   637,794.0
      11:24     19.0     0.79    892,773.7 1,127,714.1     545,302.4   688,803.0
      10:48     18.0     0.75    917,237.7 1,222,983.6     560,245.0   746,993.3
      10:12     17.0     0.71    943,829.8 1,332,465.7     576,487.3   813,864.4
      09:36     16.0     0.67    972,877.5 1,459,316.3     594,229.5   891,344.3
      09:00     15.0     0.63   1,004,783. 1,607,653.7     613,717.6   981,948.2
      08:24     14.0     0.58   1,040,049. 1,782,942.6     635,258.1 1,089,013.8
      07:48     13.0     0.54   1,079,310. 1,992,573.6     659,238.5 1,217,055.6
      07:12     12.0     0.50   1,123,382. 2,246,764.4     686,157.1 1,372,314.3
      06:36     11.0     0.46   1,173,334. 2,560,002.4     716,667.8 1,563,638.7
      06:00     10.0     0.42   1,230,603. 2,953,448.5     751,647.5 1,803,954.0
      05:24      9.0     0.38   1,297,170. 3,459,120.1     792,306.0 2,112,816.0
      04:48      8.0     0.33   1,375,856. 4,127,569.8     840,367.4 2,521,102.3
      04:12      7.0     0.29   1,470,852. 5,042,923.1     898,390.6 3,080,196.3
      03:36      6.0     0.25   1,588,702. 6,354,809.4     970,372.7 3,881,490.9

[7im]

I didn't think that the transition crossing the desired threshold would be so abrupt.

For the non-quants, what do you mean by this?

This transition here...?

5.00 22,706.0 4,541.2
4.90 219,184.9 44,731.6

[ChasR]

I think he's referring to the preferred deadline, going from 31,541 points to 338,712 points for the difference in 2.4 hours. That is indeed a cliff.

[Leonardo]

Thanks for the numbers crunching - clear explanation beyond a thousand words.

[7im]

I think he's referring to the preferred deadline, going from 31,541 points to 338,712 points for the difference in 2.4 hours. That is indeed a cliff.

Yep, same thing. I used p6903 numbers, you did 6904.

Isn't this just the same cliff in a different place on the time line? Sure, the drop might be a little higher a little sooner, but it's always been a big drop off for -bigadv. What's the big deal?

[ChasR]

The other way to look at it is it's a huge bonus for beating the preferred deadline by a minute.

[Grandpa_01]

There is also another way to look at it a 2500K @ 5Ghz with a frame time of around 60:00 min. will get a 6903 done long before the deadline I do not think they will be able to get a 6904 done so they will have to dump those. Me thinks they either need to raise it back up or lower it more because I do not think that all of those that are using the hack are not going to care about dumping a 6904 every once in a while. Some may care but I would say most will not. Which will most likely slow the project down even more.

[7im]

The other way to look at it is it's a huge bonus for beating the preferred deadline by a minute.

The other way? Seems like that's the first way a lot of people look at it lately. Completing (and uploading) the work unit with one minute to spare seems like the only permission some people need to run all kinds of unusual configurations...

So as a result, they had to shorten the deadlines to promote best behavior.

[mdk777]

I didn't think that the transition crossing the desired threshold would be so abrupt.

I think the point revolves around the continuity(or lack) of the reward curve.

Again, only looking at the points: ppd double counts the affect.

Is there a curve on the standard -smp that would transition for 31,541points to 338,712 points using similar equipment in a similar time frame ?

If these reward curves overlapped and had smooth transitions, there would be no motivation to over-extend your hardware.

[GreyWhiskers]

I think the point revolves around the continuity(or lack) of the reward curve.

Is there a curve on the standard -smp that would transition for 31,541points to 338,712 points using similar equipment in a similar time frame ?

If these reward curves overlapped and had smooth transitions, there would be no motivation to over-extend your hardware.

+1 to mdk777. That has stated the point I've been thinking about for a while. What's below is to garner discussion on the questions that have been the subject of the hundreds of posts above. For the bigbeta p6903/4, I don't have a personal stake in the issue, so I'm just playing with the numbers.

I've thrown together another wrinkle here. Sandwiched between the col for the current "flagship" bigadv p6900 and the bigbeta 6903 is a what-if shortening of both the deadline and threshold times. Shortening the threshold changes the cliff where you start earning the QRB points. Shorting the deadline changes the computation of the QRB, and establishes when you lose points totally.

Nothing magic about these numbers - no one should get hung up in thinking this is a formal proposal or anything - but I wanted to put forth a stalking horse to objectively guide the discussions about the deadlines. This doesn't put more continuity into the project, and a smoother transition between regular SMP, bigadv, and "big bigadv" - that's another step way beyond my little exercise.

As I've stated previously, I think that the thresholds should follow the scientific project need and motivate bringing the high iron forward, while giving guidance on the donors' purchasing decisions.

Thanks to PG for keeping things interesting!

Code: Select all

|                               6900                     6903                       6903
                   pts         7,164                   22,706                     22,706
                     k          26.4                    38.05                      38.05
      TPF       deadlin            6                        5                         12
  dec min       thresho            4                        3                          5
  if >60  WU Compl                                Possible shortening        Current (7-20-2011)
    mm:ss  hrs   Days     Points        ppd         Points        ppd          Points        ppd
     84.96 141.6   5.90      7,164.0     1,214.2          0.0           0.0     22,706.0       3,848.5
     83.52 139.2   5.80      7,164.0     1,235.2          0.0           0.0     22,706.0       3,914.8
     82.08 136.8   5.70      7,164.0     1,256.8          0.0           0.0     22,706.0       3,983.5
     80.64 134.4   5.60      7,164.0     1,279.3          0.0           0.0     22,706.0       4,054.6
     79.20 132.0   5.50      7,164.0     1,302.5          0.0           0.0     22,706.0       4,128.4
     77.76 129.6   5.40      7,164.0     1,326.7          0.0           0.0     22,706.0       4,204.8
     76.32 127.2   5.30      7,164.0     1,351.7          0.0           0.0     22,706.0       4,284.2
     74.88 124.8   5.20      7,164.0     1,377.7          0.0           0.0     22,706.0       4,366.5
     73.44 122.4   5.10      7,164.0     1,404.7          0.0           0.0     22,706.0       4,452.2
     72.00 120.0   5.00      7,164.0     1,432.8          0.0           0.0     22,706.0       4,541.2
     70.56 117.6   4.90      7,164.0     1,462.0     22,706.0       4,633.9    219,184.9      44,731.6
     69.12 115.2   4.80      7,164.0     1,492.5     22,706.0       4,730.4    221,456.3      46,136.7
     67.68 112.8   4.70      7,164.0     1,524.3     22,706.0       4,831.1    223,799.8      47,617.0
     66.24 110.4   4.60      7,164.0     1,557.4     22,706.0       4,936.1    226,219.3      49,178.1
     64.80 108.0   4.50      7,164.0     1,592.0     22,706.0       5,045.8    228,719.0      50,826.5
     63.36 105.6   4.40      7,164.0     1,628.2     22,706.0       5,160.5    231,303.5      52,569.0
     61.92 103.2   4.30      7,164.0     1,666.0     22,706.0       5,280.5    233,977.6      54,413.4
     60.48 100.8   4.20      7,164.0     1,705.7     22,706.0       5,406.2    236,746.7      56,368.3
     59:02  98.4   4.10      7,164.0     1,747.3     22,706.0       5,538.0    239,616.5      58,443.0
     57:36  96.0   4.00      7,164.0     1,791.0     22,706.0       5,676.5    242,593.2      60,648.3
     56:10  93.6   3.90     45,656.3    11,706.7     22,706.0       5,822.1    245,683.7      62,995.8
     54:43  91.2   3.80     46,253.2    12,171.9     22,706.0       5,975.3    248,895.4      65,498.8
     53:17  88.8   3.70     46,874.0    12,668.7     22,706.0       6,136.8    252,236.4      68,172.0
     51:50  86.4   3.60     47,520.6    13,200.2     22,706.0       6,307.2    255,715.7      71,032.1
     50:24  84.0   3.50     48,194.7    13,769.9     22,706.0       6,487.4    259,343.0      74,098.0
     48:58  81.6   3.40     48,898.3    14,381.9     22,706.0       6,678.2    263,129.3      77,391.0
     47:31  79.2   3.30     49,633.6    15,040.5     22,706.0       6,880.6    267,086.3      80,935.2
     46:05  76.8   3.20     50,403.2    15,751.0     22,706.0       7,095.6    271,227.4      84,758.6
     44:38  74.4   3.10     51,209.7    16,519.3     22,706.0       7,324.5    275,567.3      88,892.7
     43:12  72.0   3.00     52,056.2    17,352.1     22,706.0       7,568.7    280,122.5      93,374.2
     41:46  69.6   2.90     52,946.1    18,257.3    183,909.4      63,417.0    284,911.2      98,245.3
     40:19  67.2   2.80     53,883.3    19,244.0    187,164.7      66,844.5    289,954.3     103,555.1
     38:53  64.8   2.70     54,872.1    20,323.0    190,599.2      70,592.3    295,275.0     109,361.1
     37:26  62.4   2.60     55,917.3    21,506.7    194,230.0      74,703.8    300,899.8     115,730.7
     36:00  60.0   2.50     57,024.7    22,809.9    198,076.5      79,230.6    306,858.8     122,743.5
     34:34  57.6   2.40     58,200.6    24,250.3    202,161.0      84,233.7    313,186.5     130,494.4
     33:07  55.2   2.30     59,452.4    25,848.9    206,509.0      89,786.5    319,922.4     139,096.7
     31:41  52.8   2.20     60,788.6    27,631.2    211,150.3      95,977.4    327,112.6     148,687.5
     30:14  50.4   2.10     62,219.1    29,628.1    216,119.2     102,913.9    334,810.4     159,433.5
     28:48  48.0   2.00     63,755.6    31,877.8    221,456.3     110,728.1    343,078.6     171,539.3
     27:22  45.6   1.90     65,411.8    34,427.3    227,209.3     119,583.9    351,991.2     185,258.5
     25:55  43.2   1.80     67,204.3    37,335.7    233,435.4     129,686.3    361,636.6     200,909.2
     24:29  40.8   1.70     69,152.6    40,678.0    240,203.0     141,295.9    372,121.0     218,894.7
     23:02  38.4   1.60     71,280.9    44,550.6    247,595.6     154,747.3    383,573.5     239,733.4
     21:36  36.0   1.50     73,618.6    49,079.1    255,715.7     170,477.1    396,153.0     264,102.0
     20:10  33.6   1.40     76,202.5    54,430.3    264,690.9     189,064.9    410,057.3     292,898.1
     18:43  31.2   1.30     79,079.1    60,830.0    274,682.7     211,294.4    425,536.6     327,335.8
     17:17  28.8   1.20     82,308.1    68,590.1    285,898.8     238,249.0    442,912.5     369,093.8
     15:50  26.4   1.10     85,968.0    78,152.7    298,611.6     271,465.1    462,607.0     420,551.9
     14:24  24.0   1.00     90,164.0    90,164.0    313,186.5     313,186.5    485,186.4     485,186.4
     13:48  23.0   0.96     92,103.2    96,107.7    319,922.4     333,832.1    495,621.7     517,170.4
     13:12  22.0   0.92     94,173.2   102,734.4    327,112.6     356,850.1    506,760.6     552,829.8
     12:36  21.0   0.88     96,389.4   110,159.3    334,810.4     382,640.5    518,686.0     592,784.0
     12:00  20.0   0.83     98,769.7   118,523.7    343,078.6     411,694.3    531,495.0     637,794.0
     11:24  19.0   0.79    101,335.6   128,002.8    351,991.2     444,620.4    545,302.4     688,803.0
     10:48  18.0   0.75    104,112.4   138,816.6    361,636.6     482,182.1    560,245.0     746,993.3
     10:12  17.0   0.71    107,130.8   151,243.5    372,121.0     525,347.2    576,487.3     813,864.4
     09:36  16.0   0.67    110,427.9   165,641.8    383,573.5     575,360.3    594,229.5     891,344.3
     09:00  15.0   0.63    114,049.4   182,479.1    396,153.0     633,844.8    613,717.6     981,948.2
     08:24  14.0   0.58    118,052.4   202,375.5    410,057.3     702,955.4    635,258.1   1,089,013.8
     07:48  13.0   0.54    122,508.8   226,170.0    425,536.6     785,606.0    659,238.5   1,217,055.6
     07:12  12.0   0.50    127,511.2   255,022.3    442,912.5     885,825.1    686,157.1   1,372,314.3
     06:36  11.0   0.46    133,181.1   290,576.8    462,607.0   1,009,324.5    716,667.8   1,563,638.7
     06:00  10.0   0.42    139,681.5   335,235.5    485,186.4   1,164,447.3    751,647.5   1,803,954.0
     05:24   9.0   0.38    147,237.2   392,632.5    511,431.3   1,363,816.9    792,306.0   2,112,816.0
     04:48   8.0   0.33    156,168.6   468,505.9    542,454.8   1,627,364.5    840,367.4   2,521,102.3
     04:12   7.0   0.29    166,951.3   572,404.4    579,908.6   1,988,258.1    898,390.6   3,080,196.3
     03:36   6.0   0.25    180,328.0   721,312.0    626,372.9   2,505,491.6    970,372.7   3,881,490.9

[7im]

I think the point revolves around the continuity(or lack) of the reward curve.

Is there a curve on the standard -smp that would transition for 31,541points to 338,712 points using similar equipment in a similar time frame ?

If these reward curves overlapped and had smooth transitions, there would be no motivation to over-extend your hardware.

+1 to mdk777. That has stated the point I've been thinking about for a while.

A lot of us have been thinking that same thing for a while now, and have said so. For example, more than a six weeks ago, back on page 4 of this same thread, it was one of the items in my laundry list of changes that I suggested...

The points gap between smp and -bigadv needs to be reduced.

http://foldingforum.org/viewtopic.php?p=188205#p188205

I don't think the two will ever have a "smooth" transition, because either the SMP points would have to be blown up, or the -bigadv points blown down to bring them together.

Or if you make it smooth for 8 cores, there will be a gap for 12 cores and above.

But getting the 2 curves in to the same "neighborhood" would be a good start.

[bruce]

Is there a curve on the standard -smp that would transition for 31,541points to 338,712 points using similar equipment in a similar time frame ?

If these reward curves overlapped and had smooth transitions, there would be no motivation to over-extend your hardware.

I think this is a key concept, but I don't understand how you can conclude there would be no motivation. I'm not sure we should call it a "motivational factor" but at least in this post, I'm going to use your word.

Let's start with the assumption that the type of equipment is constant. Further, let's presume the hardware is high-end, capable of meeting the deadlines and you're requesting -bigadv. For projects which make similar demands on your computer, the PPD should be (more or less) constant, thought that's never an accurate statement unless your hardware exactly matches the benchmark machine.

Now assume the server runs out of bigadv-12 and you're assigned a bigadv-8 class of WU. The PPD will decrease by the motivational factor that you're talking about. For a second scenario, now assume that there are no bigadv at all and you're assigned a standard SMP assignment. Again, the PPD will decrease by a second motivational factor. With each of those steps, less of the machine will be taxed, but that is out of your control since it's due to a server running out of work or crashing.

Now look at it in reverse. Start with -smp (but not -bigadv) and ask yourself if the motivational factor is large enough to encourage folks to switch from -smp to -bigadv. Similarly, ask yourself if you can decide whether your client reports that it is eligible or ineligible for the BigBigadv WU, is the motivational factor is big enough to encourage you to over-extend your hardware.

Don't get me wrong. I'm not implying that the only thing that influences participation is a "motivational factor." I'm not implying that such a factor is a major consideration when designing the points system. There are many factors that have gone into establishing the points system. Certainly the basic benefit to science is a lot more significant that the concept of a "motivational factor" but since you asked about it from that perspective, let's look at it a little deeper from that perspective.

When Kasson adjusted the baseline points, he said that "We are therefore dropping the bigadv base points bonus from 50% to 20%, effective for all work units issued this time onwards." That's figured without the QRB and it's assuming that the WU is returned AFTER the preferred deadline. There are essentially zero folks talking about what happens after the preferred deadline; everybody is analyzing what happens to the QRB when the WU is returned BEFORE the preferred deadline. So now the real question is what happens to the PPDs after the QRB is added. Is the final motivational factor with QRB still 20% or some other number that's radically different? Just how big does it need to be to motivate people?

[mdk777]

Bruce,

I'm pointing out the all or nothing nature of the stepping.

If there were a more uniform QRB system;

3 x -smp4 on the same 12 core machine should generate something in the neighborhood(not equal but closer) of the one 12 core bigadv.

When I say no motivation, I really meant significantly diminished motivation. (some people will do anything for a few percent, but rational people recognize the diminishing return.)

I agree with 7im, you will never achieve zen balance.

However, the point being made by GreyWiskers is that he didn't realize the "step" in the step function was such a large one; really a cliff rather than a step.

I don't agree with 7im that there is little possibility to ramp the -smp up.

I agree that in order to differentiate different classes, there will always be these discontinuities.

GreyWiskers was just expressing surprise that the curves begin and end so far apart.

[Jester]

That's a good question Bruce,
As you said, even if some were to build a high end system to easily handle the new Big bigadv Wu's there's no guarantee of getting those Wu's exclusively,
so from the guidelines, such a high end machine would be a dual Cpu board and the two cheapest Xeon Hex core Cpu's even though they are capable of hyperthreading (assuming an Intel build),
with that would go a larger case for the bigger board size and a big Psu, double the memory for the 2 banks (1 per Cpu) etc, etc,
that adds up quickly compared to a dual core or better Cpu based system to run standard smp Wu's, and still no guarantee that it will get even a steady supply
of bigadv Wu's let alone the new "bigger" ones, and as we've seen lately, and as stated about all "beta" projects, no guarantee that the points reward will stay
at a certain level,
So for those who think that the new Wu's are out of proportion to the rest of the project, taken in isolation and presuming a constant supply I'd tend to agree,
but price up a system similar to the one mentioned and add components to allow running 24/7 in the hotter months, and then see if the cost is justified (and affordable)
enough to warrant the gamble for the estimated average ppd over the next week, month, year, much the same as any investment "the greater the risk the greater the reward".