[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[amibroker] Re: Freakishly fast backtest using 64 cores



PureBytes Links

Trading Reference Links

> Hey Fred:  I am not sure that I understand your comment.

I started that one - based on an earlier comment from Paul I was 
speculating if CUDA might have other possible uses in AB besides 
optimization (that is how AT got dragged into it).

I am trying to think of other heavy computational loads that slow AB 
down.

brian_z



--- In amibroker@xxxxxxxxxxxxxxx, "dloyer123" <dloyer123@xxx> wrote:
>
> Hey Fred:  I am not sure that I understand your comment.
> 
> If you are talking about auto trading, not sure.  I use 
Tradestation 
> for that.  It is fast, cheap if you have any trading volume, and 
> solid.  It does not need much horse power to run.  Not a very good 
> optimization plaform.
> 
> For optimization, it seems to me, that the problem is inherent in 
the 
> problem domain, not just due to limitations or implementation 
details 
> of a specific tool.  
> 
> In Pardo's new book, he observes that the hedge funds build custom 
> solutions that are much faster and more powerful than anything 
> available to the retail invester.  It can be done with a room full 
of 
> unix servers, or the same power can now be acheived at best buy for 
> $150.  Got to love the power of the video game market.  
> 
> 
> 
> 
> --- In amibroker@xxxxxxxxxxxxxxx, Fred Tonetti <ftonetti@> wrote:
> >
> > It may well have a need for speed but unless I'm missing 
something 
> ( and I
> > may well be ) this only really has application with the currently 
> structured
> > AB in optimization .
> > 
> >  
> > 
> >   _____  
> > 
> > From: amibroker@xxxxxxxxxxxxxxx 
[mailto:amibroker@xxxxxxxxxxxxxxx] 
> On Behalf
> > Of brian_z111
> > Sent: Monday, August 11, 2008 11:52 PM
> > To: amibroker@xxxxxxxxxxxxxxx
> > Subject: [amibroker] Re: Freakishly fast backtest using 64 cores
> > 
> >  
> > 
> > It is generally the case, for AB, that the user base is 
relatively 
> > small, and there are competitors for that $pool, plus, Tomasz 
can, 
> > and is entitled to, include his version of a competing feature at 
> any 
> > time in the future and in the manner of his choosing.... so the 
rug 
> > can slip out from under any third party app that is aligned to 
core 
> > features.
> > 
> > Real Time AT seems one place that has a need for speed.
> > 
> > (I don't know that the majority of AB users are EOD - there are 
> over 
> > 1000 users in the AT forum).
> > 
> > If using the GPU was a solution for AT how many of those who are 
> > using AT right now would have the chutzpah to look after it 
> > themselves, at the code level.......not many I think.
> > 
> > Those with the chops can already make their own solutions based 
on 
> > your pathfinding efforts?
> > 
> > > I am not opposed to creating an open source project for this. 
It 
> > > might use parts of the code base I have now, such as the trade 
> > >match 
> > > code, but would need a very different calculation core. 
> > 
> > That is the litmus test.
> > 
> > How many will come forward now and sign their name on the dotted 
> > (forum) line?
> > 
> > Don't be disapointed if they don't - maybe they will take the 
view 
> > that Intel/MSFT will soon move this into the mainstream OR make 
> > quantum leaps on processing speed via other mainstream solutions?
> > 
> > brian_z
> > 
> > --- In amibroker@xxxxxxxxx <mailto:amibroker%40yahoogroups.com> 
> ps.com,
> > "dloyer123" <dloyer123@> wrote:
> > >
> > > TJ is right. The number of AmiBroker users that are hard core 
> > enough 
> > > to need this AND willing to pay for it is proably much too 
small 
> to 
> > > justify the time involved for a comercial project.
> > > 
> > > It is not really needed for EOD data. For that, a decent laptop 
> is 
> > > fast enough. I suspect that covers the majority of the 
amibroker 
> > > user base. Tradestation would be a better design target.
> > > 
> > > I am not opposed to creating an open source project for this. 
It 
> > > might use parts of the code base I have now, such as the trade 
> > match 
> > > code, but would need a very different calculation core. 
> > > 
> > > --- In amibroker@xxxxxxxxx <mailto:amibroker%40yahoogroups.com> 
> ps.com,
> > "Mike" <sfclimbers@> wrote:
> > > >
> > > > If the market proves insuficient to warrant the effort 
required 
> > to 
> > > > package and sell your work, or if you'd just rather spend 
your 
> > time 
> > > > on other things, another alternative might be to publish what 
> > > you've 
> > > > got, "as is" with the agreement that you offer no support 
> beyond 
> > > what 
> > > > is in the code itself.
> > > > 
> > > > You could let the community run with it, posting back their 
> > > > modifications for the benifet of all. It wouldn't necessarily 
> > have 
> > > to 
> > > > be you, or at least not only you, that put the effort in to 
> > > > generalize the approach, or at least render it useful to 
others 
> > > (even 
> > > > if only as an example of how to write the custom code).
> > > > 
> > > > I suspect that there are enough developers in this forum that 
> at 
> > > > least a couple would be curious enough to have a look at it.
> > > > 
> > > > Unless, of course, what you currently have somehow exposes 
your 
> > > > strategy? In which case, perhaps a simple MA crossover 
example 
> > > could 
> > > > be published.
> > > > 
> > > > Mike
> > > > 
> > > > --- In amibroker@xxxxxxxxx <mailto:amibroker%
40yahoogroups.com> 
> ps.com,
> > "dloyer123" <dloyer123@> wrote:
> > > > >
> > > > > I am currently getting 133 portfolio backtests per second, 
> > > > including 
> > > > > trade matching and fitness function evaluation on the host 
> > > system. 
> > > > > These are on 1 year of 5 minute bars, plus higher time 
scale 
> > > data, 
> > > > for 
> > > > > > 850 symbols. 
> > > > > 
> > > > > The card I am running on costs < $200 retail, less if you 
> shop 
> > > > online. 
> > > > > I would get the new Nvida card with 240 cores, but there is 
> > > really 
> > > > not 
> > > > > much point.
> > > > > 
> > > > > Walkforward tests run in no time at all. 
> > > > > 
> > > > > As it stands, it just takes a lot of time and code to do 
> this. 
> > > > But, 
> > > > > there is no other way that I know of to get this level of 
> > > > performance.
> > > > > 
> > > > > I am very tempted to write a micro kernel that could 
execute 
> a 
> > > set 
> > > > of 
> > > > > functions on command from afl code. That way system design 
> > could 
> > > > be 
> > > > > done in afl, where it belongs, but execute on the GPU. 
> > > > > 
> > > > > If enough people where willing to pay for it, I would do it.
> > > > > 
> > > > > 
> > > > > --- In amibroker@xxxxxxxxx <mailto:amibroker%
> 40yahoogroups.com>
> > ps.com, "ozzyapeman" <zoopfree@> 
> > wrote:
> > > > > >
> > > > > > Man, somebody design a plug-in or something to make this 
> > > useable 
> > > > by 
> > > > > the
> > > > > > rest of us! I would love to have this capacity. I'm sure 
> > people 
> > > > would 
> > > > > be
> > > > > > willing to shell out $100-200 or so for a plug-in like 
this 
> > > that 
> > > > > allows
> > > > > > us to use our graphics cards to boost backtest speed.
> > > > > > 
> > > > > > There are some multivariable optimizations I would like 
to 
> > run, 
> > > > but at
> > > > > > my current computer capacity it would take over a year. 
But 
> > > that 
> > > > could
> > > > > > be shrunk down to a day or so using this graphics card 
> > > > acceleration
> > > > > > method.
> > > > > > 
> > > > > > I want it!!!
> > > > > > 
> > > > > > :-) :-)
> > > > > >
> > > > >
> > > >
> > >
> > 
> >  
> > 
> > 
> >   _____  
> > 
> > I am using the free version of SPAMfighter for private users.
> > It has removed 523 spam emails to date.
> > Paying users do not have this message in their emails.
> > Try SPAMfighter <http://www.spamfighter.com/len>  for free now!
> >
>



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

Please note that this group is for discussion between users only.

To get support from AmiBroker please send an e-mail directly to 
SUPPORT {at} amibroker.com

For NEW RELEASE ANNOUNCEMENTS and other news always check DEVLOG:
http://www.amibroker.com/devlog/

For other support material please check also:
http://www.amibroker.com/support.html
Yahoo! Groups Links

<*> To visit your group on the web, go to:
    http://groups.yahoo.com/group/amibroker/

<*> Your email settings:
    Individual Email | Traditional

<*> To change settings online go to:
    http://groups.yahoo.com/group/amibroker/join
    (Yahoo! ID required)

<*> To change settings via email:
    mailto:amibroker-digest@xxxxxxxxxxxxxxx 
    mailto:amibroker-fullfeatured@xxxxxxxxxxxxxxx

<*> To unsubscribe from this group, send an email to:
    amibroker-unsubscribe@xxxxxxxxxxxxxxx

<*> Your use of Yahoo! Groups is subject to:
    http://docs.yahoo.com/info/terms/