Thanks for the suggestion and pointer, apologies for subsequent hiatus, I needed to dig deep into the cryptographic underpinnings of folklore combiners such as x11.
I'm currently looking at a couple of sources of hard data.
1. The original
Classification of the SHA-3 candidates, including reported performance statistics in cpb (cycles per message block) for 256 and 512 output lengths (sometimes labelled "-32" or "-64") on 32 & 64-bit host architectures:
| Name_of_hash_function | 32-bit_arch | Class'n | 64-bit_arch | Class'n |
| ARIRANG-256 | 20 | A | 55.3 | E |
| ARIRANG-512 | 14.9 | AA | 11.2 | B |
| AURORA-256 | 24.3 | B | 15.4 | B |
| AURORA-512 | 46.9 | B | 27.4 | E |
| BLAKE-32 | 28.3 | B | 16.7 | B |
| BLAKE-64 | 61.7 | C | 12.3 | B |
| Blender-32 | 105.8 | E | 105.8 | E |
| Blender-64 | 122.4 | E | 164.2 | E |
| BMW-256 | 8.6 | AA | 7.85 | AA |
| BMW-512 | 13.37 | AA | 4.06 | AA |
| Cheetah-256 | 15.3 | A | 10.5 | A |
| Cheetah-512 | 83.8 | D | 15.6 | C |
| Chi-256 | 49 | C | 26 | D |
| Chi-512 | 78 | D | 16 | C |
| CRUNCH-256 | 29.9 | C | 16.9 | B |
| CRUNCH-512 | 86.4 | D | 46.9 | E |
| CubeHash8/1-256 | 14 | A | 11 | A |
| CubeHash16/1-512 | 14 | A | 11 | A |
| DynamicSHA-256 | 27.9 | B | 27.9 | D |
| DynamicSHA-512 | 47.2 | B | 47.2 | E |
| DynamicSHA2-256 | 21.9 | B | 21.9 | C |
| DynamicSHA2-512 | 67.3 | C | 67.1 | E |
| ECHO-256 | 38 | D | 32 | D |
| ECHO-256 | 83 | D | 66 | E |
| Edon-R-256 | 9.1 | AA | 5.9 | AA |
| Edon-R-512 | 13.7 | AA | 2.9 | AA |
| EnRUPT-256 | 8.3 | AA | 8.3 | A |
| EnRUPT-512 | 5.1 | AA | 5.1 | AA |
| Essence-256 | 149.8 | E | 19.5 | B |
| Essence-512 | 176.5 | E | 23.5 | D |
| Fugue-256 | 36.2 | C | 61 | E |
| Fugue-512 | 74.6 | D | 132.7 | E |
| Grøstl-256 | 22.9 | B | 22.4 | D |
| Grøstl-512 | 37.5 | A | 30.1 | E |
| Hamsi-256 | 22 | B | 25 | D |
| JH-256 | 21.3 | B | 16.8 | B |
| JH-512 | 21.3 | AA | 16.8 | D |
| Keccak-256 | 35.4 | C | 10.1 | A |
| Keccak-512 | 68.9 | C | 20.3 | D |
| LANE-256 | 40.4 | D | 25.6 | D |
| LANE-512 | 152.2 | E | 145.3 | E |
| Lesamnta-256 | 59.2 | E | 52.7 | E |
| Lesamnta-512 | 54.5 | B | 51.2 | E |
| Luffa-256 | 13.9 | AA | 13.4 | A |
| Luffa-512 | 25.5 | AA | 23.2 | D |
| Lux-256 | 16.7 | A | 28.2 | D |
| Lux-512 | 14.9 | AA | 12.5 | B |
| MD6-256 | 68 | E | 28 | D |
| MD6-512 | 106 | D | 44 | E |
| NaSHA-256 | 39 | D | 28.4 | D |
| NaSHA-512 | 38.9 | A | 29.3 | E |
| SANDstorm-256 | 62.5 | E | 36.5 | D |
| SANDstorm-512 | 296.8 | E | 95.3 | E |
| Sarmal-256 | 19.2 | A | 10 | A |
| Sarmal-512 | 23.3 | AA | 12.6 | B |
| SHA-256 | 29.3 | C | 20.1 | C |
| SHA-512 | 55.2 | C | 13.1 | C |
| Shabal-256 | 18.4 | A | 13.5 | A |
| Shabal-512 | 18.4 | AA | 13.5 | C |
| SHAvite-3-256 | 35.3 | C | 26.7 | C |
| SHAvite-3-512 | 55 | B | 38.2 | E |
| SIMD-256 | 12 | AA | 11 | A |
| SIMD-512 | 118 | E | 85 | E |
| Skein-256 | 21.6 | A | 7.6 | AA |
| Skein-512 | 20.1 | AA | 6.1 | AA |
| TIB3-256 | 12.9 | AA | 7.6 | A |
| TIB3-512 | 17.5 | AA | 6.3 | AA |
| Twister-256 | 35.8 | C | 15.8 | B |
| Twister-512 | 39.6 | A | 17.5 | D |
| Vortex-256 | 46.2 | D | 69.4 | E |
| Vortex-512 | 56 | C | 90 | E |
This is, of course, a preliminary classification performed in 2009.
2. Contemporary detail is less accessibly presented:
http://cryptography.gmu.edu/athenadb/asic_hash/table_viewhttp://bench.cr.yp.to/results-sha3.htmlLooking at those benchmarks, I can't immediately see anything that might explain why x11 should run cooler. You're probably closer to the x11 details than I am, (I'm still ploughing my way through the relevant literature on folklore and other combiners to see if there's any traction there), did Euan provide a technical appreciation for the choice and ordering of the chain of hash fns for x11?
Zooko Wilcox-O'Hearn has provided (IMO) the clearest
description of the trade-offs in a choice of basic folklore combiner.
Higher hashrates and lower power consumption
My rapid review of the x11/GPU issue (hands-off, I'm afraid; my altcoin experience is circumscribed by my entry-level Acer Sandybridge) suggests that there were contemporaneous changes to the coding of the mining app. This would be unfortunate if true; it introduces a confounding factor that undermines the credibility of claims of improved efficiency accruing from the introduction of the x11 combiner in that it's impossible to disentangle the relative contributions of performance gains from i) enhancements to the application code and/or ii) the change in combo.
But it's an ill wind that blows no good - I now have at least a plausible explanation of the difference in reception between c11 and x11 --- I've had a very hard time persuading myself that such a significant effect was due to a simple change in the batting order. Given what I now know, Occam's Razor tells me that it probably isn't and I should look elsewhere for the source of the difference in reception/perception. If there's any reliable, robust data which points to the opposite conclusion, I'd really appreciate a pointer.
Perhaps the most significant element to emerge from my consideration of a switch to x11 is that it would come at the cost of completely ruling out any possibility of merge-mining with our current altcoin kith and kin.
Cheers
Graham
Edit: extended header for more tabular space. enumerated
