PPL - Performance Per Liter, do you SFF?

GuilleAcoustic

Chief Procrastination Officer
SFFn Staff
LOSIAS
Jun 29, 2015
2,984
4,421
guilleacoustic.wordpress.com
Informative Post: Not mine, gleaned from the 'net and approximated...

Intel Hades Canyon NUC: i7-8809g 3.8GHz with Vega M GH - (2200 * 2000)/1.2 = 3,666,666

TL;DR : Intel cheats.

Does that volume includes the AC-DC brick, which is almost as big as the NUC itself (0.5L) :p ?



Edit: At 2.58M, with AC-DC included, this is still an impressive score.
 
Last edited:
  • Like
Reactions: thewizzard1

Duality92

Airflow Optimizer
Original poster
Apr 12, 2018
307
330
I'll update later today if I get the chance, this is quite interested as far as PPL is going with the Silverstone case raising the bar by a considerable amount!
 
  • Like
Reactions: VegetableStu

rokabeka

network packet manipulator
Jul 9, 2016
248
268
this is a bit offtopic, sorry about it.

GermanDrifter97's result made me think about my config. I have the same setup but in an S4M-C. so same results, lower divider (maybe below 6L with the 330W brick?)
but after that I got an apetite to beat confusis's result :D I am not sure I can do that but I swapped my other S4M-C's CPU from a 12 core low power Broadwell to a 14C 2.1GHz Haswell (105W!) tada.wav. from this point on theoritically I could add the 1080 and voila!
but this won't fly. Scramming this CPU with the low profile dynatron t318 and a 12cm slim fan above, makes this thing too loud even when idling.

long story short: so I am not beating confusis's result today :D
 
  • Like
Reactions: confusis

Arie

Trash Compacter
Jul 4, 2018
37
70
Fantastic. Do you have more pics and details of your setup? Which cooling, PSU, etc?

Sure, I'll add some more pictures later when I'm at a LAN party with my two systems. Switched some of the guts around though.

This was my setup for the i9 7980XE and the 1080Ti in the SG13:
Front intake is done by a 120mm Noctua NF-F12 fan on a Corsair H80iv2 radiator. I was hoping to use the 120mm fan of an SFX-L PSU to help with outtake, but I haven't found one that doesn't turn into a jet engine (BeQuiet 600W, Silverstone SX800) or goes into OTP (Sharkoon 500W) with this specific setup. So for now the PSU is a Corsair SF600 which stays suspiciously quiet despite the high temps in the case.
The GPU takes in its own air directly from the side and dumps a lot of it in the case, but luckily the SG13 has a lot of ventilation holes.

I got a used 7980XE and it was already delidded and the TIM replaced with LM. On stock clocks it's not too hard to cool, except in AVX based workloads where it will pull 200+W (just the CPU). That kind of power usage sends the liquid temps of the H80i to 50 celsius with the NF-F12 fan at 2000RPM.

For benchmarking purposes I applied a 4Ghz all core overclock with a slight undervolt. AVX workloads now pull 330W and CPU temps will quickly go into the 80s and 90s. The VRM on the Asrock X299 ITX board doesn't have a huge heatsink and airflow over it isn't amazing in the SG13, so I'm not comfortable with running this 24/7. It was fine for gaming workloads though :)
 
Last edited:
  • Like
Reactions: Gautam and ZMan

Gautam

Cable-Tie Ninja
Sep 5, 2016
148
123
Great post and I'm glad that you're noting AVX workloads. It's common to just pretend that they don't exist. The power draw is a little higher than I expected, but it's pretty impressive for a single 120mm rad to keep over 300W at below 100C.
 

chx

Master of Cramming
May 18, 2016
553
282
All of this is nice but -- as with the entire SFF movement -- I feel a bit neglected :) Where is thin? While I consider laptops below one inch chasing thinner and thinner an absolute craziness, there is no thin desktop which is truly odd given how the laptop world has been taken over.

I have my eyes on the MSI Vortex G25 8RE. It's by far the most carry on friendly machine I have seen in my life. And, I am guessing the score here would be above 1.9M -- it's a guess because I do not know how much the mobile 1070 is slower than a desktop I guessed 85% and I do not know the physical dimensions of the brick I guessed a full litre. I asked the latter from MSI, I will post when I know more.
 

CC Ricers

Shrink Ray Wielder
Bronze Supporter
Nov 1, 2015
2,234
2,557
As for thin desktop systems I also found this BBen GB01 a while ago. Looks like one of those compact towers that would be used in LAN cafes.

It is being sold with a GTX 1060 so probably has a MXM card.
 

chx

Master of Cramming
May 18, 2016
553
282
VERY NICE! That's a lot cheaper (Aliexpress 32822320629, 32835102739) and the 180W adapter surely is much smaller than a 330W. It's a bit thicker but still tolerable. And an 1070 is an overkill for an 1080p monitor I think.
 

ZMan

Average Stuffer
Oct 12, 2017
69
87
All of this is nice but -- as with the entire SFF movement -- I feel a bit neglected :) Where is thin? While I consider laptops below one inch chasing thinner and thinner an absolute craziness, there is no thin desktop which is truly odd given how the laptop world has been taken over.

I have my eyes on the MSI Vortex G25 8RE. It's by far the most carry on friendly machine I have seen in my life. And, I am guessing the score here would be above 1.9M -- it's a guess because I do not know how much the mobile 1070 is slower than a desktop I guessed 85% and I do not know the physical dimensions of the brick I guessed a full litre. I asked the latter from MSI, I will post when I know more.

Thin is just one way to get to Small. The heat produced by the powerful desktop components needs radiators. The bigger - the better. Hence Thin has its limits.
All thin builds end up looking like Steam Machine console. And their volume is not optimal.
In this thread the volume makes a big difference.

besides.. where is the fun in buying a ready made PC...
 

el01

King of Cable Management
Jun 4, 2018
770
588
F
Rules to enter
  • Picture of actual PC
  • Volume calculated with exterior dimensions of case WITH power brick(s)
  • CPU-Z / GPU-Z screenshots obligatory, see post #6
  • CPU-Z must be run on "Benchmark" and validation link saved with your username
  • GPU-Z must be on the first tab when doing screenshot with a dedicated GPU
  • Heaven Benchmark on "Extreme" preset
  • PPL is calculated by (CPU-Z Multi Thread score * Heaven Benchmark score)/Build Volume rounded to nearest thousand.
  • This is a friendly competition, keep it clean and civil

Leaderboard (PPL - Username - CPU - GPU - (CPUZMT*Heaven)/Volume - Post #
  1. 2297890 - confusis - i7 8700 - GTX 1060 6GB - (3879.7*2073)/3.5 - Post #121
  2. 2280611 - GermanDrifter97 - R7 1700 - GTX 1080 - (4606.6*3317)/6.7 - Post #102
  3. 2139513 - rfarmer - i7 8700K - GTX 1070 - (4520*3092)/6.7 - Post #74
  4. 1968770 - ZMan - i7 7700K - GTX 1070 - (2533*2938)/3.78 - Post #114
  5. 1481282 - Testifier - i7 7800X - GTX 1080 Ti SLI - (3297.8*4887)/10.88 - #33
  6. 1206206 - Freeks - R5 1600 - GTX 1080 - (3151.7*3100)/8,1 - Post #48
  7. 1185221 - TheHIg - R5 2600- GTX 1060 6G - (3388*2050)/5.86 - Post #56
  8. 1153142 - firewolfy - i7 6700K - GTX 1070 - (2655*2910)/6.7 - Post #57
  9. 983850 - axm - i7 6700K - GTX 1080 Ti - (2585.3*4795)/12.6
  10. 886555 - theGryphon - R7 1800X - GTX 1050 Ti - (4749.4*1232)/6.6 - Post #26
  11. 561911 - Curiosity - i5 3570k - GTX 1060 6GB - (1441.7*1598)/4.11 - Post #7
  12. 410884 - cstein08 - i5 2320 - GTX 970 - (1211*1537)/4.53 - Post #51
  13. 402835 - Frazhna - i5 4690 - GTX 1050 2G - (1471.9*1092)/3.99 - Post #81
  14. 400610 - Duality92 - i5 7300U ES - Intel HD Graphics - (1051.6*240)/0.63 - Post #6
  15. 211909 - jmarin - R5 2400G - GTX 1050 Ti - (2399*1181)/13.37 - Post #37

Top CPU score : theGryphon - 4749.4 - Ryzen 7 1800X
Top GPU score : Testifier - 4887 - GTX 1080 Ti
Lowest volume : Duality92 - 0.63L - ECS Liva Z Plus
For power bricks, this seems kinda silly, but do we need to include the volume of the cable and the DC-in jack? That might be kinda a pain...
 

confusis

John Morrison. Founder and Team Leader of SFF.N
SFF Network
SFF Workshop
SFFn Staff
Jun 19, 2015
4,324
7,425
sff.network
F

For power bricks, this seems kinda silly, but do we need to include the volume of the cable and the DC-in jack? That might be kinda a pain...

Power brick general volume. Don't be pedantic about the cable stuff, this is an ollllllllllllllllllllllllllllllllllllllld argument on SFFn. The rules are clear :)
 
  • Like
Reactions: el01

InfernoZeus

SFF Lingo Aficionado
May 23, 2017
98
53
Thin is just one way to get to Small. The heat produced by the powerful desktop components needs radiators. The bigger - the better. Hence Thin has its limits.
All thin builds end up looking like Steam Machine console. And their volume is not optimal.
In this thread the volume makes a big difference.

besides.. where is the fun in buying a ready made PC...

Being thinner gives you more surface area though (for the same volume). Isn't this technically better for heat radiation?
 

ZMan

Average Stuffer
Oct 12, 2017
69
87
Being thinner gives you more surface area though (for the same volume). Isn't this technically better for heat radiation?
It is not that simple unfortunately.
Yes, more radiator surface area = better heat dissipation.
But there are socket specific limitations regarding radiator footprint on the motherboard. So in order to get more surface area the radiator design should go vertical using "fins". More and bigger fins = more surface area. But that obstructs the airflow. So you have to force more air through the fins using a fan. And that adds even more height to the cooling system.
So in reality Thin = poor heat dissipation.
 
  • Like
Reactions: Elerek

Elerek

Cable-Tie Ninja
Jul 17, 2017
228
165
Being thinner gives you more surface area though (for the same volume). Isn't this technically better for heat radiation?

edit: rewritten to be more clear.

A heatsink isn't a simple box, it's a stack of fins. Adding up the surface area of a stack of fins is almost exactly how you learn to calculate volume at the start of calculus. Heatsink surface area scales linearly with volume, not external area.

Secondly, case thickness is not 1:1 with heatsink size. The heatsink has to share that space with other things: Case panels, motherboard, cpu (or gpu die), mounting hardware, required clearances under the pcb, etc.

But all of those numbers are fixed, they don't change when you make the case thicker or thinner. So the smaller you make the case, the larger percentage of room they take up.

So if all the extra stuff takes up 25mm, and your case is 100m thick, your heatsink can use 75mm (75%) of the thickness, but if your case is 50mm thick, you only have 25mm (50%) of the room for it. So reducing your case thickness by 2x reduced your max heatsink size by 3x!
(Note: these are made up numbers and probably exaggerate the issue.)

So it's actually the reverse of your claim, thermals get worse at a greater than 1:1 ratio as you shrink the case thickness.
 
Last edited:
  • Like
Reactions: InfernoZeus

Elerek

Cable-Tie Ninja
Jul 17, 2017
228
165
F

For power bricks, this seems kinda silly, but do we need to include the volume of the cable and the DC-in jack? That might be kinda a pain...

Only if you include the volume of the AC cable on brickless builds too! ;D
 
  • Like
Reactions: ZMan

Duality92

Airflow Optimizer
Original poster
Apr 12, 2018
307
330
EDITS: I tried to make the images display more cleanly. I failed miserably.

When I saw @confusis's article about this thread, I knew I had to throw my hat into the ring. I made an account and everything this afternoon and began the benchmarking process (with some hiccups).

System Components and Dimensions (in cubic inches)

  1. Apple 13" retina MacBook Pro (i5-4278u, 16GB 1600Mhz DDR3): 0.71 • 12.35 • 8.62
  2. Apple 85 watt MagSafe 2 Power Brick: 1.125 • 3.125 • 3.125
  3. Gigabyte Aorus GTX 1070 Gaming Box: 8.34 • 3.77 • 6.37

System volume
  • Laptop: 75.584 cu. in.
  • eGPU: 200.284 cu. in.
  • Brick: 10.986 cu. in.
  • Total: 286.854 cu. in. = 4.701 L

Testing methods
  • Do note that because CPU-Z and GPU-Z are not available for macOS, I had to substitute in alternative utilities, namely Intel's MacCPUID, CUDA-Z, and GeekBench 3 for the CPU benchmark. In the actual performance per liter calculation, I used the median CPU-Z score for my MacBook's processor from CPU-Z Validator's database, since it was lower than the average score.
  • The eGPU is connected to the MacBook using a Thunderbolt 3 to Thunderbolt 2 adapter and a Thunderbolt 2 cable, which is equivalent to a 2-lane PCIe connection. As you might have guessed, this bottlenecks the graphics card's performance.
Results
  • Setup verification photo:
  • Benchmark Scores Screenshot (see testing methods for clarification):
  • (1860 • 2703) ÷ 4.701L = 1069470.325

Conclusion
I don't know if the community will accept this score or not, but I certainly hope so. If anybody has recommendations, questions, or comments for me, please shoot them my way!
This may be an extremely unorthodox setup compared to what folks here are used to, but the volume of the system compares extremely favorably to desktop-class SFF builds and has the performance to match, which i think is pretty cool.

Updated everything, but I can't accept this methodology because if I compare my i5 7300U, which is an ES sample, so technically the way it's programmed acts like an i7 7500U, but your CPU-Z score is almost 800 points (80% more) than my CPU. The CPU-Z database has the first, initial benchmark, which was very Intel sided, so they changed it in like 1.83, since then, the benchmark scores have changed considerably. A stock i7 4790K scores 2300 for example, you're not 500 points off that CPU. I'd estimate your CPU score somewhere between 700 and 800.

I hope you understand and I think if you'd like to participate, you'd need to find yourself a an old SSD/HDD that you could do these benches.
 
  • Like
Reactions: GermanDrifter97