GTX 1080 vs. 1070 in Folding@Home and PrimeGrid

Discussion in 'Distributed Computing' started by StefanR5R, Jan 2, 2017.

  1. StefanR5R

    StefanR5R Member

    Joined:
    Dec 10, 2016
    Messages:
    181
    Likes Received:
    79
    I already posted a few observations regarding efficiency of NVIDIA GTX 1080 and 1070 at different core frequencies in the PrimeGrid LLR Races Thread 2016. Here are some more numbers.

    The GTX 1080 used here is a reference board but water-cooled (for fun, not for extreme OC'ing).
    The GTX 1070 is a factory-overclocked partner board (Gainward Phoenix GS), air-cooled.
    Both cards were sitting in the same PC with i7-6950X @ 4.0 GHz, each card in a PCIe v3.0 16-lane slot, Windows 7, NVIDIA driver 373.06.

    Folding@Home

    These results come from a few sessions with different card settings, each session spanning several days, with a random mixture of WU types. As you know, PPD varies a lot between different WUs at F@H, making it cumbersome to benchmark for F@H.
    Code:
    card    core        points/day  average   energy
            frequency   (PPD)       power     per point
    ----------------------------------------------------
    1080    1.61 GHz    680'000     109 W     14 J
    1080    1.82 GHz    770'000     152 W     17 J
    1080    1.82 GHz    760'000     152 W     17 J
    ----------------------------------------------------
    1070    1.90 GHz    630'000     147 W     20 J
    1070    2.01 GHz    710'000     150 W     18 J
    
    Memory frequencies were left unmodified at effective 4.5 GHz (1080) and 3.8 GHz (1070) in all sessions.

    Power was not measured externally but simply taken from the board sensors. Thus these numbers do not include power usage of the host PC (e.g. consumption of the supporting CPU process), and they are only as precise as the board sensors.

    Power target of the cards was left at 100 % in each of these sessions. I achieved the different frequencies in case of the 1080 simply by not running Firefox in the 1.61 GHz session, and by having a Firefox window sitting minimized in the two 1.82 GHz session. In case of the 1070, 2.01 GHz were achieved simply by letting the card run with factory OC parameters, and 1.90 GHz by setting a negative frequency offset in MSI Afterburner.

    Conclusion so far:
    The 1080 is not only a bit faster than the 1070 (surprise), but it is also more energy-efficient while being faster. Plus, absolute power consumption is the same, hence same cooling solutions would produce same noise. But are these advantages of the 1080 over the 1070 worth its dramatically higher cost of purchase?

    Remark:
    Unfortunately, this particular factory-OCed 1070 turned out to be unstable at 2.0 GHz with F@H load. It has been stable at 1.9 GHz so far. However, my primitive way of downclocking it by a frequency offset without lowering the power target looks to be detrimental to efficiency. See below.

    Edit, February 13:
    This GTX 1070 from Gainward is in fact stable at the factory overclock, and at a trivial manual overclock to 2.1 GHz as well. I had underestimated the power draw of OC'd CPU + 2 GPUs; fixed this by a beefier PSU.


    PrimeGrid

    These are the same numbers as previously posted, only the 1080@1.8GHz row is new. All of these results are from "PPS (Sieve) 1.39 (cudaPPSsieve)" tasks.
    Code:
    card     power    core        time       power     energy    comment
             target   frequency   per task             per task
    ----------------------------------------------------------------------------------------------
    1080     none     1.6 GHz     292 s      125 W     37 kJ     baseline
    1080     none     1.8 GHz     264 s      173 W     46 kJ   
    ----------------------------------------------------------------------------------------------
    1070     46 %     1.6 GHz     386 s       80 W     31 kJ     same core frequency as baseline
    1070     65 %     1.8 GHz     348 s      105 W     37 kJ     same energy per task as baseline
    1070     77 %     1.9 GHz     334 s      125 W     42 kJ     same power as baseline
    1070     none     2.0 GHz     319 s      165 W     53 kJ     same power target as baseline
    
    Again, the power column shows numbers from board sensors, not from external measurements. The cards were neither temperature-limited nor CPU-limited in these tests. Only one GPU task per card was running at the same time. Core frequency, time per task, and power were averaged over several tasks. Variance between tasks of the same session was low, very much unlike the variance of F@H WUs. Another important difference between PrimeGrid and Folding@Home is that the former gives credit linearly to the amount of work done, while the latter gives extra credit for quicker task completion.

    Conclusion so far:
    Out of the box, the 1080 is both faster and more energy-efficient as the factory-OCed 1070. By lowering the power target substantially, you can make a 1070 as efficient, or even more efficient, than a stock 1080 in PrimeGrid (but not in F@H). However, it looks borderline impossible to clock a 1070 so high that it approaches the absolute performance of a stock 1080 in PrimeGrid. (But you can accomplish this in F@H with a good 1070 specimen - perhaps because then these cards begin to be CPU-limited with their single-threaded supporting CPU process.)

    Still, the performance and efficiency differences that I am seeing here do not quite justify the difference of purchase price between these cards, do they?
     
    #1 StefanR5R, Jan 2, 2017
    Last edited: Feb 13, 2017
  2. Loading...

    Similar Threads - Folding@Home PrimeGrid Forum Date
    Getting back into folding... Distributed Computing Feb 28, 2017
    Seti@Home Stats - 28 Jan 2017 Distributed Computing Jan 28, 2017
    PrimeGrid: CPU benchmarks Distributed Computing Jan 5, 2017
    The BEST fan for your home 'server rack/room' Distributed Computing Dec 29, 2016
    Folding@Home stats issue Distributed Computing Dec 23, 2016

  3. TennesseeTony

    TennesseeTony Elite Member

    Joined:
    Aug 2, 2003
    Messages:
    1,609
    Likes Received:
    235
    Not yet. :) I'm really hoping the 1080Ti gets announced on the 4th, and thus the 1080's start hitting ebay/getting rebates/markdowns. :D
     
    Ken g6 likes this.
  4. StefanR5R

    StefanR5R Member

    Joined:
    Dec 10, 2016
    Messages:
    181
    Likes Received:
    79
    I wonder what the ratio is of people waiting to displace their 1080 quickly, to people waiting for 1080 prices to become palatable.
     
  5. TennesseeTony

    TennesseeTony Elite Member

    Joined:
    Aug 2, 2003
    Messages:
    1,609
    Likes Received:
    235
    I'm genuinely shocked, there was no 1080Ti announcement at CES. Maybe I can hang on to my 980Ti's a bit longer while it's cold and not lose much capital. :)
     
  6. StefanR5R

    StefanR5R Member

    Joined:
    Dec 10, 2016
    Messages:
    181
    Likes Received:
    79
    Perhaps NVIDIA knew that AMD wouldn't announce precise products and release dates today either, and figured they can continue selling GP104 and GP102 in the way they are doing currently.
     
  7. StefanR5R

    StefanR5R Member

    Joined:
    Dec 10, 2016
    Messages:
    181
    Likes Received:
    79
    I have been running the 1070 with power limit = 75 % and core clock offset = -50 MHz for a few days now, and it is fully stable too. These parameters give 1.9 GHz at 0.96 V average / 122 W average / 157 W peak, instead of 1.9 GHz at 1.04 V average / 147 W average / 192 W peak.

    However, I have been loading the CPU with primegrid a lot, hence F@H might be a little CPU-starved at the moment. Therefore I haven't bothered to calculate the PPD since I applied the lower power target. It stands to reason that some of the power reduction came with a penalty to PPD.
     
  8. StefanR5R

    StefanR5R Member

    Joined:
    Dec 10, 2016
    Messages:
    181
    Likes Received:
    79
    Maybe something will surface in March:

     
    TennesseeTony likes this.
  9. StefanR5R

    StefanR5R Member

    Joined:
    Dec 10, 2016
    Messages:
    181
    Likes Received:
    79
    NVIDIA put up a countdown towards a 1080TI announcement.
     
    TennesseeTony likes this.
  10. Markfw

    Markfw CPU Moderator VC&G Moderator Elite Member
    Super Moderator

    Joined:
    May 16, 2002
    Messages:
    14,184
    Likes Received:
    391
    I would have to put the killalwatt back on both of my systems, and kill the CPU on the 2683, but the 1080 has 2 power ports, and the 1070 has one. I think from what I remember, the 1070 was about 150 watts, and the 1080 about 180-190. My cards are all EVGA, and all "factory" overclocked using Xprecision. My 1080's usually beat the 1070's by 100k pd, but sometimes by 200-300k. Th record is one mill ppd on my 1080, and the record for my 1070's is about 750k. Tony and I did a week long study, and the 1080's averaged 730k ppd, and the 1070's averaged 630k ppd.

    I have 3 1070's and 2 1080's
     
  11. crashtech

    crashtech Diamond Member

    Joined:
    Jan 4, 2013
    Messages:
    6,241
    Likes Received:
    152
    I may have to put this to more scrutiny, because it was my understanding that the performance per dollar was a lot worse with the 1080. But depending on the cost of power and the expected duration it will be utilized, the 1080 may be the perf/$ king after all.
     
  12. Markfw

    Markfw CPU Moderator VC&G Moderator Elite Member
    Super Moderator

    Joined:
    May 16, 2002
    Messages:
    14,184
    Likes Received:
    391
    Well, after my test, I sure plan to get 1070's until further notice, they are the king IMO, and Tony agrees (unless he has changed his mind). And your SR-2 board has had 3 1070's in it, and it only draws 700 watt with all 3 cards, and all 24 threads of the CPU's@100% (unless you raised the vcore a lot)

    Edit: $1 per thousand ppd for 1080 and $0.62 per thousand for 1070 per our results
     
    #11 Markfw, Feb 23, 2017
    Last edited: Feb 23, 2017
    crashtech likes this.
  13. TennesseeTony

    TennesseeTony Elite Member

    Joined:
    Aug 2, 2003
    Messages:
    1,609
    Likes Received:
    235
    If I remember correctly, no time to hunt the threads right now, taking into consideration the cost of the card, the performance, and the energy use, I did once proclaim the 1070 to be the king.

    A day or so later though, I realized I had calculated all that using a 250w TDP for the 1080, when it should have been 180w, only 30w more than the 1070. Doing the numbers again with the correct TDP, the 1080 came out on top for price/performance/energy use. I even tried to calculate how long it would take for the energy $ saved, to pay for the extra cost of the 1080. It was something like 2+ years.

    I think my final conclusion was that it was nearly a tie, taking so long to pay for the 1080 with energy savings, that it'd be best, for me at least, to wait until the 1080Ti came out, and the 1080 prices dropped. A $400-$450 1080 is then a no-brainer, new or used. Well, depending on how far the prices drop on the 1070s, lol.

    I think my biggest personal concern with a 1080 for Folding, is that Mark's points seemed to vary a good deal more than the 1070s that I had. Certain tasks killed the performance of both cards, but only a few tasks seemed to really shine for the 1080, whereas my 1070s were far more stable regarding ppd. Lower, but more consistent.

    [​IMG]
     
    #12 TennesseeTony, Feb 23, 2017
    Last edited: Feb 23, 2017
    crashtech likes this.
  14. Markfw

    Markfw CPU Moderator VC&G Moderator Elite Member
    Super Moderator

    Joined:
    May 16, 2002
    Messages:
    14,184
    Likes Received:
    391
    I may have accounted for those variables. You had NO CPU contention, and I found that, since I was using it. Also, the OS. I learned from this to use win10, and leave plenty if not all CPUs available for the cards. But after observing mine for quite some time, MOST of the time they are now consistent. Not at the moment (I wish I could easily post screen shots)
     
  15. StefanR5R

    StefanR5R Member

    Joined:
    Dec 10, 2016
    Messages:
    181
    Likes Received:
    79
    @Markfw, about wattmeter versus TDP: With cards so close in performance and efficiency like 1080 and 1070, actual power measurements are indeed necessary for conclusive results. TDP as a ballpark number seems too imprecise. From what I have seen on my systems, most F@H work units cause the cards to work at quite a bit less than their TDP, on average, when not overclocked. (There may be very brief peaks of power usage even in excess of TDP, if sensors are to be believed. But these are negligible for average power consumption.)

    @crashtech, the 1080 looks quite a bit better at perf/Watt than the 1070 in the top post. But in retrospect I need to point out:
    • Owing to my inexperience with GPU computing, I may have been a bit more lucky with my Afterburner settings with the 1080 than with the 1070 during those tests. (Power target versus GPU clock offset...)
    • Card BIOS presets are not entirely comparable. I tested a stock 1080 against a factory-overclocked 1070. Even a mild OC may push a card quickly away from the energy-efficiency sweet spot.
    • Another factor which is supposed to influence energy efficiency is GPU temperature. I tested a watercooled 1080 vs. aircooled 1070. But temperatures of both cards were very close nevertheless, since CPU and 1080 share only a 360 mm radiator with slowed-down fans; and the 1070 has got a very effective triple-slot cooler.
    • As Markfw wrote, a different OS as well as CPU availability is influencing GPU performance more or less. This is bound to affect both PPD and power consumption in nonlinear ways, shifting the odds between 1080 and 1070 once more.
    On the other hand, power consumption of the host (CPU, chipset, PSU...) was not included in my post, as mentioned. This baseline power consumption speaks of course for fewer bigger cards/ against more smaller cards. But the difference between 1070 and 1080 is not so large in this respect, I'd say.

    @TennesseeTony, about work unit variability: I took the sum of estimated PPD as logged by the F@H client from several days long sessions (almost a week each), with whatever WUs came in at the time. Hard to say whether those sessions were long enough to average out the WU variability. On the other hand, there are those PrimeGrid numbers which are highly consistent on 1080 as well as on 1070.
     
  16. Markfw

    Markfw CPU Moderator VC&G Moderator Elite Member
    Super Moderator

    Joined:
    May 16, 2002
    Messages:
    14,184
    Likes Received:
    391
    As far as power usage, I care about what it takes from the wall. I have a $300 a month electric bill, so the kill-a-watt vs the reported PPD from Stanford is the $/perf that I care about. I want the most points for my $$$
     
  17. PCTC2

    PCTC2 Diamond Member

    Joined:
    Feb 18, 2007
    Messages:
    3,878
    Likes Received:
    7
    For how close the GTX 1070 is to the GTX 1080 in performance (10%?), the extra cost of power is negligible for being $250+ less. $250 to cover the difference in power would be a long time in the making.
     
  18. Markfw

    Markfw CPU Moderator VC&G Moderator Elite Member
    Super Moderator

    Joined:
    May 16, 2002
    Messages:
    14,184
    Likes Received:
    391
    Exactly, hance why I currently recommend the 1070 for F@H
     
  19. StefanR5R

    StefanR5R Member

    Joined:
    Dec 10, 2016
    Messages:
    181
    Likes Received:
    79
    That's 730 / 630 - 1 = 16 % performance delta.

    According to my data, a 1080 delivering 730 kPPD would draw slightly less power than a 1070 with 630 kPPD (not more, unlike TDP specs suggest).

    Would be nice if the 1080 would retail at no more than 16 % above the 1070. ;-)
     
  20. Markfw

    Markfw CPU Moderator VC&G Moderator Elite Member
    Super Moderator

    Joined:
    May 16, 2002
    Messages:
    14,184
    Likes Received:
    391
    Thats the rub ! $700 for the 1080, but right now, newegg has the OC'ed 1070 black edition for $360 ! Almost half the price, and less power draw for ~20% less performance.


    I did also find the 1080 superclocked 6286 that I have 2 of for $560 each, but still, $200 more each for again, 16-20% performance.
     
  21. StefanR5R

    StefanR5R Member

    Joined:
    Dec 10, 2016
    Messages:
    181
    Likes Received:
    79
    Let's consider another source for F@H PPD.

    Overclock.net's GPU Projects PPD Database currently lists GTX 1080 with 760 kPPD (sample size: 17) and GTX 1070 with 700 kPPD (sample size: 125, more by a magnitude). So their performance delta is merely 9 %.

    According to my board sensor based data, 1080@760kPPD and 1070@700kPPD will draw virtually the same power.

    I do wonder though whether the GPU Projects PPD Database's contents are somewhat biased towards higher credited WUs, as I suspect that some contributors might have been tempted to report those more often than low credited WUs.

    So here is a closer look at overclock.net's database:

    There is a Titan XP entry there with a sample size of 8, all samples from only a single contributor. This entry shows 1.9 times the performance of the 1080 entries, and 2.0 times the performance of the 1070 entries. Shader count of Titan XP, GTX 1080, and GTX 1070:
    3584 : 2560 : 1920 = 1.00 : (1/1.40) : (1/1.87) = 1.87 : 1.33 : 1.00​
    If we neglect nonlinear effects of Folding@Home's credit scheme due to early return bonus, and assume proportionality between PPD and shader count times GPU clock, and make the very much unjustified assumption that distribution of low-credit WUs vs. high-credit WUs is the same in the Titan XP, GTX 1080, and GTX 1070 entries, then a relative performance of
    1420 kPPD : 760 kPPD : 700 kPPD = 1.00 : (1/1.87) : (1/2.03) = 2.03 : 1.09 : 1.00​
    would mean that average GPU clocks of these samples should have been
    1.00 : (1/1.34) : (1.09) = 1.09 : 0.82 : 1.00, for example 2.0 GHz : 1.5 GHz : 1.8 GHz.​
    But according to my data, a 1080@760kPPD should be running at 1.8 GHz, and a 1070@700kPPD at 1.9...2.0 GHz. (The clocks, averaged over samples which the users reported to overclock.net are 1.99 GHz : 1.99 GHz : 2.06 GHz. So, users appear to have mostly reported meaningless peak clocks, not clocks that were averaged over the runtime of the WUs.)

    This suggests that WU type distribution in overclock.net's database is not representative in case of their GTX 1080 and Titan XP entries.

    Actually, given that Mark, Tony and I took observations over periods at the order of magnitude of a week per card, and seeing in my EOC stats from late January...early February an output of about 100 WUs per week with two cards, our observations should be at the order of roughly 50 WUs per card per observation. Compare and contrast to the sample size at overclock.net (8 : 17 : 127 WUs).

    Maybe all entries in overclock.net's database which are below a sample size of, say, 20, should be disregarded entirely.

    Disclaimer: I have read only the top post of overclock.net's "GPU Projects PPD Database" thread, not any of the >1800 replies. I.e. I am ignorant of any discussion of overclock.net's users' methodology and potential biases. Also, I have not attempted to compare which particular WU types are present at overclock.net, in Mark's and Tony's tables, or my own logs.
     
  22. Orange Kid

    Orange Kid Elite Member

    Joined:
    Oct 9, 1999
    Messages:
    3,035
    Likes Received:
    103
    ...and now the 1080ti comes into play... and a $100 drop on the 1080. Will the 1070 get a reduction in price too? Can only hope:)
     
  23. Markfw

    Markfw CPU Moderator VC&G Moderator Elite Member
    Super Moderator

    Joined:
    May 16, 2002
    Messages:
    14,184
    Likes Received:
    391
    wow, I have 9 available slots now......
     
  24. bds71

    bds71 Member

    Joined:
    Nov 29, 2016
    Messages:
    58
    Likes Received:
    30
    just remember that the new cards are 250W TDP (vs 180W TDP 1080 and 150W 1070 rated). all that extra power costs extra power :)
     
  25. Markfw

    Markfw CPU Moderator VC&G Moderator Elite Member
    Super Moderator

    Joined:
    May 16, 2002
    Messages:
    14,184
    Likes Received:
    391
    I have 3 AX1200 Corsairs to feed them, no problem, and just ordered a AX860 as a backup.