I have this algo --- 13.2 MH/s

It looks not even bad, thanks.

poclbm 27.06.2011 comes with phatk instead of the old poclbm kernel.

Latest poclbm exe (20110627) also had the phatk patched.

All leaked passwords are either MD5 hashes or FreeBSD MD5 crypt() hashes. The attacked must have changed all hashes to a known one in order to be able to log into each other users' account. It's not impossible to execute a simple UPDATE query given the fact that the entire passwords database was dumped. Then it would be impossible for you to log into your MtGox account unless the hacker has restored the original hashes.

$1$salt$hash is the standard FreeBSD MD5-based crypt() format. It was first developed for FreeBSD back in the days when export of DES code outside USA was forbidden. Then all major Unix variants switched to using it as it is much more secure than the original Unix DES-based crypt() and allows passwords longer than 8 symbols. It employs fixed number of salted MD5 rounds and is considered fairly secure given that lots of special symbols and combination of upper and lowercase letters are used. There is another Blowfish-based variant from OpenBSD that is clearly recognisable by the $2$ sentinel. It is much stronger as it takes a lot more CPU/GPU power to compute it compared to the MD5-based one.

I'm running 32-bit Windows 7. What you can probably do is to modify your card's BIOS. I've read stories of people flashing their cards' bioses with reference ones from ATI and/or editing existing power states. There are number of tools to do that, for Windows only I presume. Personally I have never done such a thing as my mining rig is a dedicated one and I had access to a genuine Windows 7 license.

I'm getting 413 Mhash/s on both my 5870s clocked at 935 MHz core and 319 MHz memory. The point in downclocking the memory is to conserve power and to reduce the heat output, thus lowering card temperature and displacing the inevitable failure further in time. GDDR5 uses lots of power even when idle and lowering it's clock speed two- or threefold reduces the memory power usage accordingly.

Finding the optimal clock speeds is a tricky optimization task. CMOS logic power consumption is proportional to the clock speed and to the square of the voltage. Hashing still uses the card's memory for some storage and time misalignment between GPU's and memory's clocks can lead to wait cycles when incommensurable frequencies are set.

AMD clock tool is for Windows only. Try the AMDOverdriveCtrl tool on Linux.

I have 2x Sapphire Vapor-X 5870 OC running at stock voltage (1.16 V if I remember correctly), 935 MHz core and 319 MHz mem clocks. No CrossFireX bridges. Windows 7 32-bit, Catalyst 11.5 w/ bundled SDK 2.4. Running Phoenix 1.48 + phatk. Options are: VECTORS BFI_INT WORKSIZE=256 FASTLOOP=false AGGRESSION=13. Getting 413 Mhash/s from each card. The rig is located in a conditioned room alongside other IT equipment. With both fans at 100% and with case's side covers removed core temperatures are 45C and 75C (the second card sits too close to the first because of bad PCI-E slots separation).

Well, Sandy Bridge packs up to 12 EUs (equivalent to ATI's SPs) and probably that OpenCL SDK would give programmers access to them. There is nothing wrong with your CPU contributing some Mhash/s more

There is a noticeable difference in hashing speeds between 319 MHz and 300 MHz. What miner are you using? What options? How about running Phoenix with phatk kernel that is specifically optimised for 5870 and SDK 2.4? My options are:
VECTORS BFI_INT FASTLOOP=false AGGRESSION=13 WORKSIZE=256
WORKSIZE=128 gives a little lower result with SDK 2.4. With SDK 2.1 worksize of 128 gives better hashing speed but still lower than worksize of 256 with SDK 2.4. Beware, that's not a general rule but specific to the phatk kernel!

I'm getting 413 MH/s on each of 2x5870 at 935 MHz core and 319 MHz memory clock, fans at 100%. The lower card runs at 45C and the upper - at 75C (cards are too close - bad mobo).

First, try asking in the "Mining" forum where all the hardware geeks lurk.
Second, is your miner running while you tweak the memory clock rate? My system crashes if I tweak card settings while running Phoenix.
Also did you try with some other tweaking tools? I personally use ATI Tray Tools since they work best with my two Sapphire Vapor-X 5870. I've tried Sapphire Trixx, MSI Afterburner and AMD GPU Clock Tool, but only ATT works the way it is supposed to work.

Tianhe-1A is a mixed architecture: 14366 CPUs + 7168 GPUs. A rough estimate of hashing power is 14366 * Xeon X5670 (8 Mhps) + 7168 * nVidia M2050 (100 Mhps) = 831 Gh/s. Xeon X5670 peak performance is rated as 2.93 GHz * 4 DP * 6 cores = 70.32 DP Gflops (DP == double-precision floating point as used to measure performance in Top500). nVidia states that M2050 achieves 515.2 DP Gflops. Thus for the CPU the Gflops/Mhps ratio is 8.8 and for the GPU it is 5.2. The combined CPU+GPU ratio is 5.7. For a single HD5870 GPU this ratio is 1.5 (544 Gflops and 375 Mhps).

There are few mixed CPU/GPU computers in Top500. Amongst the top 10 are only Tinahe-1A, Nebulae and TSUBAME. Roadrunner is using Cell processors which are similar to CPUs in performance. There are only 10 systems using nVidia technology at the Top500 list from November 2010. All other supercomputers are only CPU based thus the combined Gflops/Mhps ratio for the whole Top500 group would be around 8. At 43.6 Tflops combined this leads to an estimate of about 5.5 Thps. We are currently at 3.5 Thps.

As most miners are running GPU and not CPU miners, I think the estimated Tflops rate of the Bitcoin network is a very large overestimate: 42895 Tflops / 3377 Ghps = 12.7 is a ridiculous ratio. It's most likely that the ratio should be around 2 or 3, depending on the portion of remaining CPU miners.

FASTLOOP is enabled by default in recent Phoenix versions and one must specify FASTLOOP=false in order to disable it.

phatk works best with WORKSIZE=256. Getting 413 Mhash/s on both 5870 935/319 in a dual-card mining rig (w/o CrossFireX bridges).

The funny thing is that FLOPS means floating point operations (per second) and SHA256 computations (the core Bitcoin hashing algorithm) take exactly 0 (zero) floating point ops as they consist of integer and bitwise ops only. Besides floating point operations are usually slower than integer and logical ones.

Bottom line is - measuring Bitcoin performance in FLOPS is wrong!

My dual 5870 (w/o CF bridges) maxes out at WORKSIZE=128.

What happened to luke.dashjr.org? It's down and no JSON stats are available though the RPC server seems to work fine.