what is a good way to "benchmark"?

[FoBoT]

i noticed this on the FAQ page

Q: Do you have any benchmarking data?
A: Not yet. This is one of the things we're currently working on, so check back later. If you've done any of your own benchmarking, we'd like to know about it.


can you tell me a good way to do benchmarks?

is there a way to get the client to write a text file of the timings, or do i need to just watch a clock or something?

[jjjjL]

only way to do it is just get your cEM/s number, right now. i'll make a benchmark feature in the next version that does some BS work just to test for proper results and give good speed measurements. thanks for reminding me. i've been meaning to do something like that.

i'll even make it report the result so it can be used to figure out relative speeds of procs w/o people emailing it all to me. sound good?


-L

[FoBoT]

Originally posted by jjjjL
i'll even make it report the result so it can be used to figure out relative speeds of procs w/o people emailing it all to me. sound good?


-L

that would be awesome! :D

[Chinasaur]

Louie,

Is the benchmark feature still in the feature set for new client?


:cheers:

[MAD-ness]

I just want to chime in and say that the proposed ideas for benchmarking are wonderful.

[FoBoT]

i need to benchmark to see how much worse celeron's are at SB

some DC projects, celeron's do ok, some, they are tremendously worse than similar Ghz of other cpu

[MAD-ness]

Q - Is there a built-in benchmarking feature in the Seventeen or Bust client?

A - No, there is not. There is a built-in progress tracker, which uses cEMs (corrected exponent mods) but this number is not accurate across a wide range of n values). Support for standardized benchmarking and other benchmarking related features is planned in the next version of the Seventeen or Bust client.

------

I know that 'sneaker-netting' is not supported and is difficult, but if I want to attempt it anyways, what should I do?

"people have been asking me about sneaker netting. it could be done in theory right now. linux and windows use slightly different caching methods.

w/ linux you can copy files directly. the "cache" file and the "zXXXXXXX" file can be moved from an online computer and put on another computer for processing. they will easily fit on a disk and all you have to do is copy the two files back to a directory with the sb binary in it on an online computer and run the file to have it upload results. if someone writes a script to help do this process, i'll post it on the website.

w/ windows you would have to copy the reg keys around with the "zXXXXXXX" file. in fact, you only really need to copy the reg keys to another computer to transfer it to the offline computer. when the block was done, it would need to be have the "zXXXXXXX" file copied back into the install directory of SB and then have the reg keys restored on the original PC
before running the client to submit the work."

(quote Louie)

------

I see two problems with benchmarking with the client the way it currently is. The first problem is having an identical exponent or number to test (k, n pair). The second is the way that the work unit is pretty hardwired into the client (registry keys, etc.).

The first *might* be overcome by copying a number/exponent from one of your computers (say a 1ghz amd cpu) to the Celeron.

Using the method suggested above (or other methods involving exporting registry keys, I have cut and pasted other methods out of this forum if you want me to post them here as well). In theory you could then have both computers offline and just time how long they take to complete the number/exponent. This would be best with a number with a smaller n value (less time to complete a test). I imagine you could write scripts to tell you when the client finishes a test (say by looking for cpu utilization to drop).

I haven't looked at the workings of the client enough or the files, to know if there is some progress tracker (other than cEMs).

Heck, what am I thinking?

If you have an identical k, n pair then the cEMs should be a sufficiently accurate benchmark measurement.

wtf, me stupid.

If you try any of this and it works, lemme know please.

I think I might try it now if i have time.

[MAD-ness]

Random statements I collected regarding moving exponents around:

The easiest way is to export the [HKEY_LOCAL_MACHINE\SOFTWARE\LhDn\sob\cache] key and copy the .reg file and the Zxxxx file to the other computer. Run the .reg file and start the SoB client

------------

How to exchange

I don't know how it is handled with the non-Windows clients (AFAIR it's easier there), so I can only give you the way you have to do it in Windows. When there's need for further instruction, I guess others can fill the gaps.

Here we go:
Write down the k and n values. Well, at least the k value. Then you open regedit and change the keys in HKLM/SOFTWARE/LhDn/sob/cache. Make sure you enter them in dec, not on hex.
Same with the other PC

---------

I did this a few times, by exporting the registry key and importing it on the other machine.


-------

Those are various accounts of users moving numbers around.

Being a noob, maybe someone can answer for me which reg keys need moved. I see the zzzz file in the sob client directory. In HKEY_LOCAL_MACHINE\SOFTWARE\LhDn\sob\cache I have 4 items.

Default Reg_SZ
Cache Reg_Dword
k Reg_Dword
n Reg_Dword

While it is pretty self-explanatory what each of the last 3 entries are for, I don't know what default reg_SZ is and I don't know for sure which would need to be exported/moved.

I have used regedit before of course, but typically only with explicit directions (I don't normally cruising around editing the registry for entertainment purposes).