Folding Forum

Hi,

I was wondering if there is some sort of comparison available for the FaH Android App and the Desktop Version. In the Desktop Version, I can see very clearly, that my GTX 1080 scores about ~600K/day and an old Windows-Tablet scores less than 1k/day. Therefore (by my view), it is not really worth it, to run FaH on the Tablet as the contribution is negligile... If I run my GTX for one more Work Unit, it is "worth" several months of the Tablet (especially with regards to electrical consumption).

I was wondering, how much of a contribution an average mobile device with the Andoid App is making... Is it "worth" using Is there some form of benchmark or translation table available or something?

Thanks

Since we don't have any comparison for FAH maybe the BOINC data will help: I run WCG on a Kitkat Tablet (Nexus 5 or 7; forgot)

SCC1_ 0001215_ Bct-E_ 27718_ 1-- android_34b7a68 3.63 cpu time in minutes

Compare to Ryzen1700 (Ubuntu)
SCC1_ 0001274_ Bct-E_ 24245_ 0-- Ryzen1700 0.72 CPU time in minutes

Actually I'm surprised how well the little ARM is doing; but also not 100% about what math they use for the type of work units. I have seen other examples where the ARM was 10x slower.
Is it worth ? For me yes ... every cycle counts ...

ChristianVirtual wrote:...For me yes ... every cycle counts ...

That's certainy true.

It's not easy to compare the ARM processing power with the desktop CPU.

It should be noted that the early x86 hardware was extraordinarily different than today's hardware with AVX support. I've lost track of how many "generations" the experts call the radical improvements, but it's many, even ignoring the other factors which contribute to CPU productivity. The SoC processor is proceeding along a path containing number of "generational" jumps in productivity.

In the same fashion, the processing capabilities in early ARM processors were a lot more limited than the hardware in the latest portable phone. The current FAH software was aimed at a relatively low level of hardware capability. FAH's software hasn't yet been rewritten to take advantage of all of the changes to instruction sets. In other words, we would need to compare what is currently being done with what COULD be done with a new generation of software as well as figure out how the inherent processing power of the ARM compares with X86.

None of us has any solid numbers on which to make that comparison, so CV's comparison is probably as accurate as you're going to find ... and making a direct comparison of the scientific value of a CPU to a SoC probably hasn't been done. I think it's fair to GUESS than the Android App is not going to favor that App but it'e anybody's guess how radical that comparison would turn out.

Back to the quote above. The Android App is processing a few specific projects that are not being worked on my the CPU.(The same is true for FAH's support for {wide variations in SMP count/AVX vs. non-AVX/GPU vs. CPU/etc.} within the desktop platform) One way or another, that scientific research needs to be done. We can HOPE that projects chosen for each platform are ones that are best suited for that platform's capabilities. That, too, would make a direct comparison all the more difficult.

Did I hear that the latest iPhone hardware is incorporating AI/Deep_Learning capabilities? Those buzz-words are also used to describe features that are primarily available in GPUs. Who knows where the FAH App will be in a few years? (We can dream, can't we?) A lot depends on finding and harnessing the developmental resources of ( Sony ) or Stanford or somebody else.

Keep on contributing to FAH with whatever equipment you have.

I’m quite interested in the ML capabilities of the new iOS ; if just had more time ...

I guess the next big thing would be FAH using the smartphone GPUs.

Actually from first reading that is part of the iOS ML concept, compute on GPU. But who knows yet how flexible it will be ...