@Alan
Thanks for the breakdown on Geekbench's component benchmarks, i didn't know what they really tested or how they're used, beyond geekbench's short desctiption.
I wish this discussion on Dhrystone/Whetstone had happened a month ago, i randomly ended up reading about it, but i don't remember much now, but yeah it's an old benchmark. IIRC, for some reason, it's still deemed relevent, and i'm not sure but i think Sandra's implementaion is not the original algorithim, but slightly tweaked, i might be completely wrong here. And, the EMBC (folks who made Browser Bench, +1 on using that BTW, i was going to suggest it sooner or later to Chris or Andrew) have a modern implementation of Dhrystone, again if memory serves me correctly (They have CoreMark, MultiBench, and Energy Bench, how are those?).
Oh, and, what's this they're using? (You guys probably shouldn't bother running this, after all Anand's already run it):
We have one last web-based benchmark: WebXPRT by Principled Technologies (PT). WebXPRT measures performance in four HTML5/js workloads:
Photo Effects: Measures the time to apply effects to a set of six photos. The filters are Sharpen, Emboss, and Glow. WebXPRT applies each filter to two photos. This test uses HTML5 Canvas 2D and JavaScript.
Face Detect: Measures the average time to check for human faces in a photo. WebXPRT runs this test on five photos and uses the average time to calculate the final result. This test uses HTML5 Canvas 2D to get access to photo data. The detection algorithm is implemented in JavaScript.
Stocks Dashboard: Measures the time to calculate financial indicators of a stock based on historical data and display the result in a dashboard. The calculations are done in JavaScript, and the calculated stocks data is displayed using HTML tables and Canvas 2D.
Offline Notes: Measures the time to store notes securely in the browser's HTML5 local storage and display recent entries. This test uses using AES for security.
This is from the 2nd page of the Samsung Tab review.
Also on Vellamo...there's been a new version out, does it use single-threaded Linpack too? I've always seen Tegra 3 based devices score relatively higher on Vellamo (and AnTuTu)...
I've run it on a Xolo X900, and even with the separate Linpack download, the difference between one thread and two is minimal on Medfield, Anand found similar results back in May.
Also what about AnTuTu? They had a bug with x86 stuff (another example of why it might be a good idea to use more than one synthetic), and performance jumped quite a bit after they patched it.
Again, i'll bring up PassMark, as it's Windows/Android/iOS...it thinks my CPU (Core 2 Quad Q8400) does 4.4 MB/s per core in the encryption test, and my fourth generation ipod does 5.394 MB/s. the 200 MHz slower ZTE blade does 512 KB/s according to it. It just took one version update for the ipod's score to go from a few hundred KB/s to that insane figure. So yeah. Probably why Andrew doesn't trust this either. Of course, it may be single threaded, and the 16.6 MB/s it claims for my CPU might be for one thread, but in the config it specifically states it's using a multi-threaded benchmark.
Finally: Why not make your (as in, TH's) own benchmark?
P.S. Try and include the Xolo X900/Orange San Diego/Motorola Razar i as well, wouldn't hurt to compare Medfield to Clover Trail going forward. At least, wouldn't be a bad idea to keep it on the charts.