PPL - Performance Per Liter, do you SFF?

petricor

Airflow Optimizer
Bronze Supporter
May 12, 2018
336
1,820
Decided to push even higher and managed to hit 5017.84MHz on my NUC, and benching again raised my multi-thread score to a whopping 2930.

New calculations puts my score at:

(2930 * 1426) / (1.14 + 0.48) = 2579123 Performance Per Liter Score

Which is a solid 3rd place now.
Thats pretty impressive and close enough to make for a challenge ;) I’ll be back...
 

Duality92

Airflow Optimizer
Original poster
Apr 12, 2018
307
330
Updated! We're at 20 entries so far! When I get my HDPlex 400W and can finish my 3D printed case, I might be able to get a second score in appart from my ECS Liva Z.
 
  • Like
Reactions: Obelisk

Rysen

Trash Compacter
Jan 13, 2017
40
61
My personal PC as is used everyday. No overclocking on anything.

CPU: i7 7700k (at Stock)
Graphic Card: GTX 1080 Ti
Validation: https://valid.x86.fr/qsn5uc
Total score: (2620.7*4589)/7.2 = 1670332
Liters: 7.2 with the added fan (Can also run at the same speed without the fan but didn't want to take it off. Obviously at higher temperatures. I've tested it before.)


 
Last edited:
  • Like
Reactions: Arie and Damascus

petricor

Airflow Optimizer
Bronze Supporter
May 12, 2018
336
1,820
Just came across this and quickly benchmarked my current build - as-is right now, with no optimisation or overclocking whatsoever (running rock stable), incl overhead for remote access as running headless...

score:

2489451 - petricor - i7 8700k - GTX 1080 - (3806.5*3270)/5.0
validation link: https://valid.x86.fr/vtuftp

benchmarks:

https://valid.x86.fr/7tx1l5

build (slightly ahead of my log.. am a few posts behind!):

OK just had my first take at overclocking my build - using the cookie cutter "33%" ASUS ROG CPU template pushing it near 5 ghz (just under - suspect there is more headroom!) , +197Mhz on the GPU core and +439Mhz on the GPU mem clock, no voltage change whatsoever to CPU/GPU - running stable so far with the GPU stabilising at 80 deg at 25 ambient running Heaven, in horizontal case position with all panels on. This time with a screen connected directly rather than remoting the image, taking overhead off the system...

Long story short: it gets me to a rather unexpected 3,380,913 points ;)

my new line:

3380913 - petricor - i7 8700k - GTX 1080 - (4349x3887)/5.0

Validation link: https://valid.x86.fr/7tx1l5

benchmarks:


 

Rysen

Trash Compacter
Jan 13, 2017
40
61
OK just had my first take at overclocking my build - using the cookie cutter "33%" ASUS ROG CPU template pushing it near 5 ghz (just under - suspect there is more headroom!) , +197Mhz on the GPU core and +439Mhz on the GPU mem clock, no voltage change whatsoever to CPU/GPU - running stable so far with the GPU stabilising at 80 deg at 25 ambient running Heaven, in horizontal case position with all panels on. This time with a screen connected directly rather than remoting the image, taking overhead off the system...

Long story short: it gets me to a rather unexpected 3,380,913 points ;)

my new line:

3380913 - petricor - i7 8700k - GTX 1080 - (4349x3887)/5.0

Validation link: https://valid.x86.fr/7tx1l5

benchmarks:


Damn that would put you in second place. The score gets very unfair with an i9 because even though the multi-threaded score is higher the performance doesn't really scale that well in any non-benchmark practical application. Plus the fact that the CPU score can be so much higher than the GPU doesn't do the whole number justice because GPU power of a 1080ti (750$) for example, say a score of 4300 is equivalent to a 8700k (400$) at 5GHZ (In terms of score) when it should really be equivalent to the score of an i9 (900$) which is closer in price to a 1080ti or maybe a titan x (1200$).
 
Last edited:

rfarmer

Spatial Philosopher
Jul 7, 2017
2,605
2,720
OK just had my first take at overclocking my build - using the cookie cutter "33%" ASUS ROG CPU template pushing it near 5 ghz (just under - suspect there is more headroom!) , +197Mhz on the GPU core and +439Mhz on the GPU mem clock, no voltage change whatsoever to CPU/GPU - running stable so far with the GPU stabilising at 80 deg at 25 ambient running Heaven, in horizontal case position with all panels on. This time with a screen connected directly rather than remoting the image, taking overhead off the system...

Long story short: it gets me to a rather unexpected 3,380,913 points ;)

my new line:

3380913 - petricor - i7 8700k - GTX 1080 - (4349x3887)/5.0

Validation link: https://valid.x86.fr/7tx1l5

benchmarks:



Great score and your cable sleeving/management skills are legendary. :thumb:
 

Arie

Trash Compacter
Jul 4, 2018
37
70
Damn that would put you in second place. The score gets very unfair with an i9 because even though the multi-threaded score is higher the performance doesn't really scale that well in any non-benchmark practical application. Plus the fact that the CPU score can be so much higher than the GPU doesn't do the whole number justice because GPU power of a 1080ti (750$) for example, say a score of 4300 is equivalent to a 8700k (400$) at 5GHZ (In terms of score) when it should really be equivalent to the score of an i9 (900$) which is closer in price to a 1080ti or maybe a titan x (1200$).

I agree that the PPL formula heavily favors extreme core counts CPUs, but I wouldn't call it unfair. I actually make use of all those cores in my system, either while using it as my programming workstation to compile and run tests (scaling extremely well over the cores), or while gaming and streaming at the same time with high bitrate and higher quality x264 presets than would be possible with a "measly" 6 core CPU ;)
 
  • Like
Reactions: Gautam

Gautam

Cable-Tie Ninja
Sep 5, 2016
148
123
Damn that would put you in second place. The score gets very unfair with an i9 because even though the multi-threaded score is higher the performance doesn't really scale that well in any non-benchmark practical application. Plus the fact that the CPU score can be so much higher than the GPU doesn't do the whole number justice because GPU power of a 1080ti (750$) for example, say a score of 4300 is equivalent to a 8700k (400$) at 5GHZ (In terms of score) when it should really be equivalent to the score of an i9 (900$) which is closer in price to a 1080ti or maybe a titan x (1200$).
It's actually quite fair from the standpoint of case design. CPU-Z scales almost linearly with number of cores, as does heat output and power draw. If you have a system that can handle double the heat, it should be capable of scoring twice as high. Price and popularity aren't being measured here. There are ways to take advantage of the relatively low power draw and heat and translate that into low volume to compensate, like @confusis setup.
 
  • Like
Reactions: petricor and ZMan

K888D

SFF Guru
Lazer3D
Feb 23, 2016
1,483
2,970
www.lazer3d.com
It's actually quite fair from the standpoint of case design. CPU-Z scales almost linearly with number of cores, as does heat output and power draw. If you have a system that can handle double the heat, it should be capable of scoring twice as high. Price and popularity aren't being measured here. There are ways to take advantage of the relatively low power draw and heat and translate that into low volume to compensate, like @confusis setup.

I agree, its not the CPU's fault that every day software doesn't take full advantage of it. Its like driving a Ferrari on around the streets, it has the performance potential but you can't take advantage of it with a 30 mph speed limit.
 
  • Like
Reactions: petricor

Gautam

Cable-Tie Ninja
Sep 5, 2016
148
123
I agree, its not the CPU's fault that every day software doesn't take full advantage of it. Its like driving a Ferrari on around the streets, it has the performance potential but you can't take advantage of it with a 30 mph speed limit.
Yeah. And as @Arie said, there are applications that do. For video encoding, a high core count is almost a must, and Adobe Media Encoder and Handbrake both make use of AVX instructions as well. Professional use cases can demand a lot more CPU horsepower than gaming.
Decided to push even higher and managed to hit 5017.84MHz on my NUC, and benching again raised my multi-thread score to a whopping 2930.

New calculations puts my score at:

(2930 * 1426) / (1.14 + 0.48) = 2579123 Performance Per Liter Score

Which is a solid 3rd place now.
This is just nuts. Even in a full-size desktop, my 2 year-old 6700K would have no hope of clocking that high, and yet here we are with a PC the size of a small book, and half the CPU TDP. Quite amazing how quickly things progress.
 
  • Like
Reactions: Biowarejak

petricor

Airflow Optimizer
Bronze Supporter
May 12, 2018
336
1,820
I agree, its not the CPU's fault that every day software doesn't take full advantage of it. Its like driving a Ferrari on around the streets, it has the performance potential but you can't take advantage of it with a 30 mph speed limit.
Also adds to the fun - technically it should be possible to beat the “extreme” series cpus using the PPL formula as there is a natual limit to how compact one can go looking at the power and thermal footprint of those beasts- they come with a whole host of challenges of their own... with a 1080ti in a brickless 5l build the figures should pan out well over 4m, so the crown is up for grabs - all a question of doing it ;)
 

petricor

Airflow Optimizer
Bronze Supporter
May 12, 2018
336
1,820
@petricor mind letting me in on how you got a brickless s4m with those specs? I'd like to do the same with my 6950x and an 1180/70
A mixture of research into previous builds, planning, and a lot of trial-and-error: I didn’t really know whether my power solution would fire up at all before building it (it was an untested idea and turned out to be pretty robust) - have a look at my log (link in my benchmark post), I just posted an update on airflow which is particularly relevant to the GPU
 
  • Like
Reactions: Damascus

Damascus

Master of Cramming
Feb 27, 2018
551
388
A mixture of research into previous builds, planning, and a lot of trial-and-error: I didn’t really know whether my power solution would fire up at all before building it (it was an untested idea and turned out to be pretty robust) - have a look at my log (link in my benchmark post), I just posted an update on airflow which is particularly relevant to the GPU
Thanks, took a look and it's very informative. How much did you pay for your meanwell unit? I see them on mouser for 70 CAD/ 50 USD
 
  • Like
Reactions: Biowarejak