He's going to be top for quite a bit for sure!
Last edited:
-
Save 15% on ALL SFF Network merch, until Dec 31st! Use code SFF2024 at checkout. Click here!
PPL - Performance Per Liter, do you SFF?
- Thread starter Duality92
- Start date
He's going to be top for quite a bit for sure!
I made the final design in Protocase Designer, so it's pretty simple. I started with a design that ReinPure from reddit gave me and modified using OnShape here. I'm going to make some more changes and then do a small run or two for some friends. It's still a bit too rough around the edges for me to feel comfortable with anyone other than myself using it as it is. It was my very first attempt at anything like this.
The first change is likely going to be adding an SFX PSU mount, but it'll make the case a little bigger. This will also give it more heatsink clearance. Right now it only fits the Raijintek Pallas with a 25mm fan and that too rather messily. (The heatsink juts out onto the video card for an extremely tight fit. Ideally I'd like to be able to fit the C1 and NH-L12 with both fans. The AIO as it is is also a sloppy fit. The tubing needs better routing
I tried a little too hard to chase small liter numbers, and even then ended up way higher than I planned. If I cave and let it go slightly over 8L, it can go from being a good case for exactly the hardware I have to being something quite versatile.
It seems like the H90 that I'm using is too old for the Corsair software, so I'm not sure. Looking at our CPU-Z scores, you're about 5C cooler than me. I tried to stay below 90C in CoreTemp. This CPU regularly runs in the 70's/80's in the rendering rig it calls home, though. I was more concerned about the PSU and VRM than the CPU.Awesome
What kind of water temps are you getting with that? I was getting quite uncomfortable with 50C+ water temps on my 7980XE at 4.2-4.4Ghz with a Corsair H80iv2.
And that was pulling "only" 350W on the CPU.
Thank you!
Definitely wouldn't say that. The new nVidia cards are just around the bend, not to mention Threadripper. There should be a wide variety of ways to beat this, either with ultra-small or very high-end setups, or something in between.
we're still using Haven bench though, so for the 2080ti performance might scale up towards the new CUDA processor count ._.
My opinion, current formula (CPUZMT*Heaven)/Volume is not fair in terms of SFF. In fact it is not performance per liter, it’s performance*performance per liter. It depend much more from performance, not from volume. As the result build based on STX MXM / Kaby Lake G / Threadripper don’t have any chance to win. 7980XE+2080Ti in the as small as possible case is two heads higher.
From my point of view (CPUZMT+Heaven)/Volume or similar will be much more fair.
For example:
n4ru - i7 8809G - RX Vega M GH - (2930*1426)/1.62 - Post #181. Will have 2688 points.
Gautam - i9 7960X - GTX 1080 Ti - (11537*4916)/7.9 - Post #232. Will have 2082 points, but with very high potential: 1080ti > 2080ti and 7.9L > 6.XL or 5.XL.
Confusis - i7 8700 - GTX 1060 6GB - (3879.7+2073)/3.5 - Post #121. Will have 1700 points, but with high potential too: moving to slim version of deskmini, 1060 > 1080, 8700k > 9900k etc.
All platforms will have the same opportunities.
From my point of view (CPUZMT+Heaven)/Volume or similar will be much more fair.
For example:
n4ru - i7 8809G - RX Vega M GH - (2930*1426)/1.62 - Post #181. Will have 2688 points.
Gautam - i9 7960X - GTX 1080 Ti - (11537*4916)/7.9 - Post #232. Will have 2082 points, but with very high potential: 1080ti > 2080ti and 7.9L > 6.XL or 5.XL.
Confusis - i7 8700 - GTX 1060 6GB - (3879.7+2073)/3.5 - Post #121. Will have 1700 points, but with high potential too: moving to slim version of deskmini, 1060 > 1080, 8700k > 9900k etc.
All platforms will have the same opportunities.
we could use the new blender benchmark that runs the same benchmark on cpu and gpu and scales well with multiple cpus and gpus, then it could be (cpu time + gpu time)*volume, a lower score is better.
But that's the way it is. There are two aspects of PC benchmarking ._.
Also "averaging" both CPU-based and GPU-based results will mean n4ru taking the top just for having the second smallest case (and Duality92 taking third just for having the tiniest).
(i.e. it heavily favours any beating heart in a matchbox)
Gautam deserves the top spot because not only it's small, but it's cleaning the floor with everyone's faces
(PS: we should find a public and good video exporting benchmark ._.)
Last edited:
Once we reach 25 entries, I will create a new thread, for PPL, round 2, different CPU and GPU benchmark, GPU benchmark will have SLI support as Testifier would've ranked higher with 1080 Ti SLI (4th place potentially instead of 11th).
we'll see how multi-GPU and geforce NVLink goes ._. as of now there's not much incentive to multi-GPU
I'm compiling all the data in a spreadsheet too and will do a recap with tons of fun statistics after the competition is over 
(I'm an Excel geek).
I've also added a bit of color to the OP to show visually part brands
(I'm an Excel geek).I've also added a bit of color to the OP to show visually part brands
No challengers with a 2990WX? I am still waiting and hoping from the sidelines out here in the audience that someone will be that one who does it. Besides, the 2990WX easily scores in the 16000 point territory in CPU Z. A 11-12L InWin case or a modded NCASE M1 might just do the trick to hit all the magic tabulations.
Last edited:
I'm waiting for this too, a 32 core mATX monster with a Titan V or RTX 2080 Ti.
The arithmetic mean is generally bad practice to use when averaging benchmark scores. The geometric mean and harmonic mean are more commonly used.
Using an arithmetic mean doesn't account for difference in scale, and just skews things to whatever the most extreme number is. It just so happens that Heaven and CPU-Z happen to be at the same order of magnitude, but if we did something like use Heaven FPS, in the 10's to 100's, instead of the final score, it would become nearly pointless next to the CPU-Z score, which is in the thousands. And here, the volume in the denominator ends up dominating.
The PPL formula here is basically a geometric mean squared; strictly speaking it would be more correct to square root the benchmark scores product, or square the liter measure, but that also depends on how strongly you want to weigh the volume in the calculation. The geometric mean is a solid measure when averaging measurements of different scales, which applies here.
The industry standard SPEC benchmarks use the geometric mean across a benchmark suite to calculate a final score.
However, both the CPU-Z and Heaven benchmark effectively measure rates; instructions per second for CPU-Z and frames per second for Heaven, so the harmonic mean can also be appropriate.
It's incorrect mean to average rates. If you travel 10 miles at 10 MPH and another 10 miles at 50 MPH, your average speed is not 30 MPH, (10+50)/2, but is actually 8.33 MPH, the harmonic mean, 1/(10/10+10/50).
I have some qualms about the type of hardware it takes, but it's also objectively how it works.
Going from a Ryzen 2700X to a 7980XE doesn't exactly require double the volume, but really does more than double the number of cores. The latter takes a lot more cooling, but even if it takes double, it only adds to the total rather than doubling it. The radiator I'm using adds about 1.5L. An L9a could count as zero volume on top of the motherboard.
Maybe brickless builds and STX builds need to be pushed a bit harder, too. Anything less than a 500W+ system is effectively wasting volume with an SFX PSU, potentially a lot. In some ways using anything less than a Core i9 on a mini ITX board is technically also wasting volume. Full-size GPU's also pretty much unconditionally waste volume when it comes to raw scores. Most of the time, though, for normal, usable builds, these are sensible options, but they're not going to peak in raw performance per liter.
@Gautam I just expected that the volume would be more important on the SFF Forum than performance. Now we have almost the same as hwbot.org (world overclockers community) has, 7980XE + 1080Ti easily win in any cases. I expected to see in top something like STX MXM, Kaby G, Udoo, Ryzen Embedded, 3.5” mobos, short GPUs, thin Mini ITX and other progressive SFF platforms. Instead of that we see standart money-dependent rating.
@chinevo even if we more heavily weigh volume money will win out. A 7980XE + T315 with a 4k rpm delta 92mm fan will be very stable, easily passing any benchmarks while still fitting in a case with a mere 52mm of cooler height. Pair this with a half height GPU in a custom case (sub 4L is feasible) and you have a build that demolishes in a volume heavy competition.
oops I just realized that doesnt work, if the render time for each device is added you get a worse score for having a gpu and every additional gpu. I feel like this calculation should be simple but i cant seem to figure it out.
There's alot of suggestions for formula changes which might be valid but I kinda like the status quo.
However, cost is a huge factor. It is also going to change over time which will throw off the results. I can see grouping the current scores into cost brackets or dividing the current scores by the current cost of items... performance per litte per $ (PPL$). The cost should be something like launch price that doesn't change.
It would be an interesting trend to graph... the date of each build vs PPL$. You could see if future prices (and efforts) will net higher scores in smaller packages.
However, cost is a huge factor. It is also going to change over time which will throw off the results. I can see grouping the current scores into cost brackets or dividing the current scores by the current cost of items... performance per litte per $ (PPL$). The cost should be something like launch price that doesn't change.
It would be an interesting trend to graph... the date of each build vs PPL$. You could see if future prices (and efforts) will net higher scores in smaller packages.
I like this. One of the major barriers to entry for builders isn't always the size but the price itself. When I was a part of my university's eSports club for a brief stint, one reason some did not purchase SFF was namely because of the price. They figured they might as well get the bigger, cheaper cases when the smaller, more pricy ones had less airflow, expandability, and features.
We can't include costs because it varies per country, per season, used and new, etc... I get my hardware sponsored so since I'm at $0, does it mean I win? It won't work. benchmarking competitions are pay to win and we're doing it for brags, it will stay as is for now.
Similar threads
