March2011_OTA_ReferenceROM_v1.zip ( 99,606,802 Bytes; MD5: 770786915545a16f564a42e344ce7bb6 )
( 02/15/2012 - Re-hosted download link - thanks to XDA user doogald for providing a home for this ROM )
This is what you would get if you took the HTC 2.1 Leak-V3 PB00IMG.zip (same thing as the "Official 2.1 RUU") ROM, and then upgraded it twice - first with the July 2010 OTA, and then after that with the March 2011 OTA. (Using this ROM flash basically saves you that trouble.)
The ROM itself is completely stock, but unlike the "Flashback21" series of ROMs, this flash does not restore a stock recovery, nor flash any radio firmware - you can use it just like any other ROM.
(It is not meant to used for preparing a phone for resale or return - use the FlashBack21 ROM for that).
AGAIN, please note that this ROM is completely stock - it is not even rooted.
Feel free to use it as the basis for other 2.1 Sense ROMs.
If you want to have a "pure HTC stock ROM", but with the benefits of overclocking and rooting, I suggest you flash the "engtools3.zip" and Conap's "CFSv9" kernel immediately after flashing this. This would get you "stock+root+overclocking" - kind of like the original "PlainJane" ROM, but patched up to the current (3/2011) HTC software release.
One other note. Out of curiosity, I decided to run some benchmarks on a GB ROM vs. this HTC ROM, but using the same kernel (CFSv9), and clocked at a fixed rate (setCPU min = max = 748Mhz). So, I performed the kernel replacement and then ran a couple benchmarks.
This allows a crude comparison of what the improvements in the underlying dalvik engine have been between the Eclair release and the Gingerbread release - it is sort of "apples to apples" because the same kernel, scheduler, applications, and clock rates are used.
The numbers here are not the absolute benchmark values, but rather their RATIO - GB:Eclair. (In the cases where the benchmark value is "lower = better", or elapsed time, the value is inverted so that in all cases, a number greater than 1.0 indicates an improvement in Gingerbread). I used JIT = on, and compcache off for the GB ROM.
Ratio Benchmark 1.57 Quadrant Full 1.16 Elapsed Time (wall clock time) for Quadrant Full 1.10 setCPU "Long Bench" 0.96 setCPU "Native Bench" ** -.-- Linpack for Android - Skipped (not meaningful - too sensitive to JIT)
In "real world" terms, Quadrant's raw score probably is not too meaningful - but notice that in elapsed time for the full benchmark, the GB ROM was consistently about 16% faster to completion. That measurement - wall clock time - is a "real world" metric. So, in addition to all the other goodness of Froyo/GB, you do seem to get a decent improvement - 10 to 15% - by switching to GB.