Join PrimeGrid
Returning Participants
Community
Leader Boards
Results
Other
drummers-lowrise
|
Message boards :
Sieving :
MSR Sieving - P / watt for GPUs
| Author |
Message |
|---|
|
|
|
On Discord we were calculating some our of GPU's P / watt ratios. How efficiently they sieve.
Here are some results:
Turing cards:
Two different 1650s, and MSI Ventus, and Gigabyte
~240 P/day at ~67 watts - 3.582 P/watt
A Gigabyte 2080
~820 P/day at ~260 watts - 3.150 P/watt
An EVGA 1660 Ti
~420 P/day at ~120 watts - 3.500 P/ watt | |
|
|
|
|
My results:
Sapphire RX 550 (Lexa)
20 P/day at ~25 watts - 0.800 P/watt | |
|
 Dave 
Send message
Joined: 13 Feb 12
Posts: 3171
ID: 130544
Credit: 2,233,021,669
RAC: 614,280
                           
|
|
Ye Olde Doorstope GTX580: 240W (including 40W idle!) 91P therefore 0.38P/W.
GT1050: 75W 51P 0.68P/W.
:(. Still to break that 1:1 barrier. | |
|
compositeVolunteer tester
Send message
Joined: 16 Feb 10
Posts: 1140
ID: 55391
Credit: 1,020,885,942
RAC: 1,702,094
                      
|
|
That's not right - calculating P per watt. Watt is a unit of measure of instantaneous power, so you need to integrate power with respect to time to make an apples-to-apples comparison for P/day and watt-hours per day, usually scaled to kilowatt hours per day. Then it would make sense to talk about energy efficiency in terms of watt-hour per P, or it's reciprocal P per watt-hour. To get this value, divide the numbers quoted in this forum by 24. | |
|
dukebgVolunteer tester
Send message
Joined: 21 Nov 17
Posts: 242
ID: 950482
Credit: 23,670,125
RAC: 0
                 
|
|
Why is noone telling which GFN n this P/ ratios are for? It runs quite differently on different n (lower n limited by CPU more) and it's best to compare values only on GFN-22, for example. | |
|
Dave
Send message
Joined: 13 Feb 12
Posts: 3171
ID: 130544
Credit: 2,233,021,669
RAC: 614,280
|
Ye Olde Doorstope GTX580: 240W (including 40W idle!) 91P therefore 0.38P/W.
GT1050: 75W 51P 0.68P/W.
:(. Still to break that 1:1 barrier.
Mine are GFN22. | |
|
|
|
That's not right - calculating P per watt.
You are right, but they are really dividing "points/day" by watts, so this does make sense, but comes out in "funny" units.
If you like "days" as a unit, then
1 watt = 1 joule/second = 86400 joules/day = 0.0864 megajoules/day
and when you divide "points/day" by that, the "per day" part cancels, and you end up with the unit "points/megajoule". That is the number of points you get for each microjoule of power.
As an example:
3.5 (points/day)/watt = 3.5 (points/day)/(0.0864 megajoules/day) = 40.5 points/megajoule
For every megajoule of energy "used", you get 40.5 points.
If you have a perverse preference for the mixed unit "kilowatthours", it is going to be:
3.5 (points/day)/watt = 3.5 (points/day)/(1 watthour/hour) = 3.5 (points/day)/(0.024 kilowatthour/day) = 146 points/kilowatthour
as you said!
/JeppeSN | |
|
dthononVolunteer tester
Send message
Joined: 6 Dec 17
Posts: 435
ID: 957147
Credit: 1,737,342,032
RAC: 18,611
                             
|
|
Agreed, except for the tiny detail that my electricity bill is in kWh and not MJ. So I prefer the perverse presentation of points/kWh ;-) | |
|
AzmodesVolunteer tester
Send message
Joined: 30 Dec 16
Posts: 184
ID: 479275
Credit: 2,197,504,179
RAC: 13,692
          
|
|
The "P" doesn't stand for points, though. It's peta, as in x quadrillion of the factor range covered per day. Which makes it important to specify the n value sieved.
____________
Long live the sievers.
+ Encyclopaedia Metallum: The Metal Archives + | |
|
compositeVolunteer tester
Send message
Joined: 16 Feb 10
Posts: 1140
ID: 55391
Credit: 1,020,885,942
RAC: 1,702,094
|
|
Precisely, "power" is not a measure of energy, it is a measure of energy per unit time. We pay for electicity by energy usage, hence the billing in kilowatthours. That applies in France as well as North America. | |
|
|
|
The "P" doesn't stand for points, though. It's peta, as in x quadrillion of the factor range covered per day. Which makes it important to specify the n value sieved.
Thank you for explaining that. /JeppeSN | |
|
darkclownVolunteer tester
Send message
Joined: 3 Oct 06
Posts: 328
ID: 3605
Credit: 1,422,865,129
RAC: 337,605
      
|
|
Yeah, this is more a tracking of "P" sieved/watt out of curiosity on card efficiency
____________
My lucky #: 60133106^131072+1 (GFN 17-mega) | |
|
|
|
That's not right - calculating P per watt. Watt is a unit of measure of instantaneous power, so you need to integrate power with respect to time to make an apples-to-apples comparison for P/day and watt-hours per day, usually scaled to kilowatt hours per day. Then it would make sense to talk about energy efficiency in terms of watt-hour per P, or it's reciprocal P per watt-hour. To get this value, divide the numbers quoted in this forum by 24.
for 1 hr, it uses that many watts. Fairly simple to understand.
| |
|
|
|
Why is noone telling which GFN n this P/ ratios are for? It runs quite differently on different n (lower n limited by CPU more) and it's best to compare values only on GFN-22, for example.
For this it is the fastest your GPU can sieve at. For most people this is GFN22, but you can do it for GFN23 or higher and it will see be as fast as you CPU can feed it, for GFN22 and above this is usually the MAX P/day of your GPU as it does not take close to 1 CPU core even on a slow 8350 to max out any given card. Over 1000P/day for most CPUs for GFN22 for any given card.
| |
|
|
|
Yeah, this is more a tracking of "P" sieved/watt out of curiosity on card efficiency
Exactly what it is. How efficient your card is at doing X number of P per watts.
For those that don't know what manual sieving is and don't know what a P range is than this is not for you.
to do X number or P per day at a constant rate of watts can be calculated also to figure you electrical cost to figure our how much it will cost you to do that many P per day on a given card.
I'm really not sure why there is confusion.
It shows how many P/day it can do using X number of watts. Simple efficiency. Don't read into.
| |
|
AzmodesVolunteer tester
Send message
Joined: 30 Dec 16
Posts: 184
ID: 479275
Credit: 2,197,504,179
RAC: 13,692
|
|
Posting some stuff from my evergrowing BOINC spreadsheet. I didn't measure power draw, so this is just P/day (and largely approximate at that). All at GFN21 and with b13 and w1 unless otherwise noted. The cards' memory clocks are also as far down as they go.
RTX 2080: 850 (Windows 10, +150 core OC, two instances in tandem)
RTX 2070 Founder's edition: 605 (Ubuntu, +130 core OC)
GTX 1660 Ti: 445 (Ubuntu, +130 core OC)
GTX 1080 Ti: 270 (Ubuntu)
GTX 980: 120 (Ubuntu)
GTX 1070 Ti: 180 (Ubuntu)
GTX 1060 6GB 100
GTX 1060 3GB: 95 (Windows 10, +200 core OC)
GTX 1050 Ti: 63 (Windows 10, +150 core OC)
GT 1030: 28 (Ubuntu)
GTX 970: 78 (Ubuntu)
HD 7970: 78 (Windows 7, +100 core OC)
Intel HD 4000: 2.5 (b7)
R9 280X: 70 (Windows 7, +100 core OC, b11)
GT 540M: 12 (Windows 7, +180 core OC, b9)
____________
Long live the sievers.
+ Encyclopaedia Metallum: The Metal Archives + | |
|
|
|
Posting some stuff from my evergrowing BOINC spreadsheet. I didn't measure power draw, so this is just P/day (and largely approximate at that). All at GFN21 and with b13 and w1 unless otherwise noted. The cards' memory clocks are also as far down as they go.
RTX 2080: 850 (Windows 10, +150 core OC, two instances in tandem)
RTX 2070 Founder's edition: 605 (Ubuntu, +130 core OC)
GTX 1660 Ti: 445 (Ubuntu, +130 core OC)
GTX 1080 Ti: 270 (Ubuntu)
GTX 980: 120 (Ubuntu)
GTX 1070 Ti: 180 (Ubuntu)
GTX 1060 6GB 100
GTX 1060 3GB: 95 (Windows 10, +200 core OC)
GTX 1050 Ti: 63 (Windows 10, +150 core OC)
GT 1030: 28 (Ubuntu)
GTX 970: 78 (Ubuntu)
HD 7970: 78 (Windows 7, +100 core OC)
Intel HD 4000: 2.5 (b7)
R9 280X: 70 (Windows 7, +100 core OC, b11)
GT 540M: 12 (Windows 7, +180 core OC, b9)
Very nice, but do you have the power draw for each card at those P rates?
Your post might be better referenced in this thread ;) -
http://www.primegrid.com/forum_thread.php?id=8640 | |
|
AzmodesVolunteer tester
Send message
Joined: 30 Dec 16
Posts: 184
ID: 479275
Credit: 2,197,504,179
RAC: 13,692
|
|
Nah, as I wrote, I didn't measure power draw and I no longer have some of those cards. In my experience it's usually more or less on target to just go with TDP. Msieving maxes out GPUs if done correctly/efficiently, so they're pushed to their power limits which aligns with TDP.
Do we really need two threads? I'd say just make it one for msieving swiftness by GPU in general and include wattage if available.
____________
Long live the sievers.
+ Encyclopaedia Metallum: The Metal Archives + | |
|
|
|
Nah, as I wrote, I didn't measure power draw and I no longer have some of those cards. In my experience it's usually more or less on target to just go with TDP. Msieving maxes out GPUs if done correctly/efficiently, so they're pushed to their power limits which aligns with TDP.
Do we really need two threads? I'd say just make it one for msieving swiftness by GPU in general and include wattage if available.
My goal with this thread was an efficiency thread. Performance was in the other thread. I see how we could keep them together. | |
|
Message boards :
Sieving :
MSR Sieving - P / watt for GPUs |
