CUDA and openCL Benchmarks

Bikeman (Heinz-Bernd Eggenstein)
Bikeman (Heinz-...
Moderator
Joined: 28 Aug 06
Posts: 3522
Credit: 723477545
RAC: 1160962

Interesting.

Interesting.

Quote:
with a power draw of < 80 Watt.

I think the official TDP is 85 Watt, and I've seen reviews that claim a measured power draw in the 90s.

HBE

Beyond
Beyond
Joined: 28 Feb 05
Posts: 121
Credit: 2337466212
RAC: 5291515

RE: Interesting. RE: with

Quote:
Interesting.
Quote:
with a power draw of < 80 Watt.
I think the official TDP is 85 Watt, and I've seen reviews that claim a measured power draw in the 90s.


Einstein shouldn't push the GPU to the max so I'd like to see a Kill-A-Watt measurement on the 7700, 7790 & 7850 to see if there are indeed efficiency differences.

Beyond
Beyond
Joined: 28 Feb 05
Posts: 121
Credit: 2337466212
RAC: 5291515

RE: A User had the wish to

Quote:

A User had the wish to show credits per Day too. I looked into it and showing that now too from 1-3WUs in parallel.

[LINK]http://www.dskag.at/images/Research/EinsteinGPUperformancelist.pdf[/LINK]

Have fun :)


Wow thanks! Hate to ask for more work but could you add a column for 5 WUs (credits/day) as that seems to be optimal for some of the fastest GPUs (7970, and of course it's the only known value for Titan).

Sunny129
Sunny129
Joined: 5 Dec 05
Posts: 162
Credit: 160342159
RAC: 0

he might as well add PPD

he might as well add PPD columns for 4 simultaneous tasks and 6 simultaneous tasks too then...while those columns will probably remain empty for the most part, it would be a quick and easy task to add them to the spreadsheet...and its better to have them and not need them than it is to need them and not have them.

btw, i'm curious as to how people are coming up w/ their PPD values for the table. are you guys just going to your personal Einstein@Home web pages, viewing the computers on your account, and taking the numbers directly from the "avg. credit" column? and if so, would it not be wise to monitor that value for several days following any hardware changes just to make sure the PPD value settles down or converges on a specific value (ensuring that values taken prior to any hardware change(s) don't get averaged into the PPD value and skew it)?

Neil Newell
Neil Newell
Joined: 20 Nov 12
Posts: 176
Credit: 169699457
RAC: 0

I do it by reading the

I do it by reading the "job_log_einstein.....txt" file - this seems to keep a record of all work done, and how long it took. It's not hard to extract the relevant fields and plot them using GNUPlot (or Excel I imagine).

Also for precise comparision of GPUs, any BOINC application can be run standalone from the command line - so I have a BRP 'reference task' for comparing CPU/GPU/Motherboard combinations. This is a neat way to do it because it's precisely repeatable.

Neil Newell
Neil Newell
Joined: 20 Nov 12
Posts: 176
Credit: 169699457
RAC: 0

...and here's an example


...and here's an example graph generated for this host's 'job_log_einstein.phys.uwm.edu.txt':

The red line shows the raw credits per day; this host was brought up with low-performance GPU's to verify operation before putting in the current units, as can be seen over the first few days. The dip on Day 15777 is the effect of a power cut, and the slight rise in performance on Day 15790 is the result of switching the linux CPU governor from "on-demand" to "performance" (n.b. this only seems to make a difference with certain CPU's, here an AMD Athlon II X2 250).

The blue line is the output of an exponential averager, empirically tweaked to follow "Average Credit" (in the long term - I use a tweak to speed up the initial response).

Of course, graphing like this makes the assumption that all work units are accepted as valid; also the fine detail can be complicated if e.g. the points for a particular application change during use. Still, it's a quick way to see how well a host is working and to track the effect of changes (e.g. it would be hard to spot the effect of the CPU governor change from the "average credit" alone).

dskagcommunity
dskagcommunity
Joined: 16 Mar 11
Posts: 89
Credit: 1216734096
RAC: 188459

Added new values for 4 Cards

Added new values for 4 Cards and the wished 4-6 Simultan WU Credit Colums. It gets more and more difficult to get all on a single PDF Site, but i think the zoom Function works good on all PDF Readers ^^

[LINK]http://www.dskag.at/images/Research/EinsteinGPUperformancelist.pdf[/LINK]

So, now i will "celebrate" :( my Birth and get drunk, good evening ^^

DSKAG Austria Research Team: [LINK]http://www.research.dskag.at[/LINK]

Beyond
Beyond
Joined: 28 Feb 05
Posts: 121
Credit: 2337466212
RAC: 5291515

First of all, Happy

First of all, Happy Birthday!
Second, can't see the new columns...

Sunny129
Sunny129
Joined: 5 Dec 05
Posts: 162
Credit: 160342159
RAC: 0

yes, Happy Birthday

yes, Happy Birthday indeed!

...the revised spreadsheet is working fine for me...

Beyond
Beyond
Joined: 28 Feb 05
Posts: 121
Credit: 2337466212
RAC: 5291515

RE: yes, Happy Birthday

Quote:

yes, Happy Birthday indeed!

...the revised spreadsheet is working fine for me...


Now it's working here too :-)

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.