Upcoming MultiCharts Beta Dramatically Outperforms TS8.3

Questions about MultiCharts and user contributed studies.
User avatar
Andrew Kirillov
Posts: 1589
Joined: 28 Jul 2005
Has thanked: 2 times
Been thanked: 31 times
Contact:

Upcoming MultiCharts Beta Dramatically Outperforms TS8.3

Postby Andrew Kirillov » 23 Aug 2007

Please see our test results:
Image
Image
Disclaimer:
Please note that, while every effort has been made to ensure that information contained herein is accurate and up to date, TS Support, LLC accepts no responsibility for errors or omissions that may occur.
No person should place reliance on information contained herein where such reliance may result in loss, damage, or injury.
Attachments
benchmark.PNG
benchmark.PNG (22.88 KiB) Viewed 2659 times
comp.PNG
comp.PNG (24.67 KiB) Viewed 2670 times
Last edited by Andrew Kirillov on 19 Sep 2007, edited 5 times in total.

glam_100
Posts: 157
Joined: 14 Jun 2006
Has thanked: 2 times
Been thanked: 4 times

Postby glam_100 » 23 Aug 2007

Oh Andrew I love it I love it...

This is what I'm talking about...

Is there an ETA for the new multi chart?

User avatar
Andrew Kirillov
Posts: 1589
Joined: 28 Jul 2005
Has thanked: 2 times
Been thanked: 31 times
Contact:

Postby Andrew Kirillov » 23 Aug 2007

we believe the end of the next week is a realistic ETA.

User avatar
Andrew Kirillov
Posts: 1589
Joined: 28 Jul 2005
Has thanked: 2 times
Been thanked: 31 times
Contact:

Postby Andrew Kirillov » 23 Aug 2007

P.S.
Keep in mind MultiCharts results on quad-core will be almost 2 times better...

glam_100
Posts: 157
Joined: 14 Jun 2006
Has thanked: 2 times
Been thanked: 4 times

Postby glam_100 » 23 Aug 2007

This is really exciting. Looking forward to the new beta.

Guest

Postby Guest » 23 Aug 2007

that's good news.

but the color code on your table does not co-relate to the bar chart.

glam_100
Posts: 157
Joined: 14 Jun 2006
Has thanked: 2 times
Been thanked: 4 times

Postby glam_100 » 23 Aug 2007

BTW Andrew, how did you guys managed to outperform TS on a single thread basis? My understanding is that TS is highly optimized for code performance. Did you take advantage of the SSE floating point? If not would it be possible to exploit that and further leave TS in the dust? thx

Guest

Postby Guest » 23 Aug 2007

looks like you guys are good programmers !

Guest

Postby Guest » 23 Aug 2007

P.S.
Keep in mind MultiCharts results on quad-core will be almost 2 times better...


Just to confirm... are you saying that Multicharts is able to take advantage of Quad-core chips running on Vista ?

IF yes, does this only apply to backtesting or does it apply to Charting as well ?

thanks

User avatar
Andrew Kirillov
Posts: 1589
Joined: 28 Jul 2005
Has thanked: 2 times
Been thanked: 31 times
Contact:

Postby Andrew Kirillov » 24 Aug 2007

BTW Andrew, how did you guys managed to outperform TS on a single thread basis? My understanding is that TS is highly optimized for code performance. Did you take advantage of the SSE floating point? If not would it be possible to exploit that and further leave TS in the dust? thx


Actually we didn’t use any special technologies. We've revised our code and found some obvious bottlenecks. It is true that TS is highly optimized and it wasn’t simple task to outperform it.
It is possible to make it 30-50% faster, but it will involve significant efforts. So we leave it as is so far.

User avatar
Andrew Kirillov
Posts: 1589
Joined: 28 Jul 2005
Has thanked: 2 times
Been thanked: 31 times
Contact:

Postby Andrew Kirillov » 24 Aug 2007

P.S.
Keep in mind MultiCharts results on quad-core will be almost 2 times better...


Just to confirm... are you saying that Multicharts is able to take advantage of Quad-core chips running on Vista ?

IF yes, does this only apply to backtesting or does it apply to Charting as well ?

thanks


Yes, MultiCharts is able to use all available cores. If your PC has two quad-core CPU MC will use 8 cores. It is true for optimization only, because paralleling is possible for this process.
We are working on data loading optimization as well.

glam_100
Posts: 157
Joined: 14 Jun 2006
Has thanked: 2 times
Been thanked: 4 times

Postby glam_100 » 24 Aug 2007

BTW Andrew, how did you guys managed to outperform TS on a single thread basis? My understanding is that TS is highly optimized for code performance. Did you take advantage of the SSE floating point? If not would it be possible to exploit that and further leave TS in the dust? thx


Actually we didn’t use any special technologies. We've revised our code and found some obvious bottlenecks. It is true that TS is highly optimized and it wasn’t simple task to outperform it.
It is possible to make it 30-50% faster, but it will involve significant efforts. So we leave it as is so far.



You guys are really doing a great job. My hat is off to TSSupport team.

User avatar
Andrew Kirillov
Posts: 1589
Joined: 28 Jul 2005
Has thanked: 2 times
Been thanked: 31 times
Contact:

Postby Andrew Kirillov » 24 Aug 2007

I would like to thank you for the recognition our work has received from all of you.

Guest

Postby Guest » 24 Aug 2007

Andrew,

Will the upcoming beta fix the bug that effectively limits the number of GA optimization parameters to 18.

User avatar
Andrew Kirillov
Posts: 1589
Joined: 28 Jul 2005
Has thanked: 2 times
Been thanked: 31 times
Contact:

Postby Andrew Kirillov » 24 Aug 2007

yes, it will.

widgetman
Posts: 81
Joined: 11 Oct 2005

Opentick FEED issue

Postby widgetman » 24 Aug 2007

I would really like to see the opentick 0 tick issue to be fixed in this beta release. This issue has been around since April or so. Good job on the speed improvement. I know writing this code has to be a challenge.

Thanks

User avatar
Andrew Kirillov
Posts: 1589
Joined: 28 Jul 2005
Has thanked: 2 times
Been thanked: 31 times
Contact:

Postby Andrew Kirillov » 24 Aug 2007

it was fixed a few days ago. So it will be released in the new beta.

Guest

Postby Guest » 24 Aug 2007

P.S.
Keep in mind MultiCharts results on quad-core will be almost 2 times better...


Just to confirm... are you saying that Multicharts is able to take advantage of Quad-core chips running on Vista ?

IF yes, does this only apply to backtesting or does it apply to Charting as well ?

thanks


Yes, MultiCharts is able to use all available cores. If your PC has two quad-core CPU MC will use 8 cores. It is true for optimization only, because paralleling is possible for this process.
We are working on data loading optimization as well.

oh, I see, there is no difference if you are doing charting only.

User avatar
Andrew Kirillov
Posts: 1589
Joined: 28 Jul 2005
Has thanked: 2 times
Been thanked: 31 times
Contact:

Postby Andrew Kirillov » 25 Aug 2007

It is not correct. If you run several charts and indicators MultiCharts will use all available cores to process indicators as quickly as possible.

Guest

Postby Guest » 25 Aug 2007

thanks for the update. This is good news because I run multiple instances of MC, with many chart windows, and many substantial indicators at the same time. Currently my CPU time is at 80% to 100%.

User avatar
Marina Pashkova
Posts: 2758
Joined: 27 Jul 2007

Postby Marina Pashkova » 29 Aug 2007

The situation with CPU running at 80-100% capacity is not normal. How many workspaces and charts did you use? Also how many indicators were applied to your charts? And one last thing: check if dumps are generated. To do that go to Start-> Programs -> Logs and check the size of the Logs folder.

jmi88
Posts: 29
Joined: 31 Jan 2007

Postby jmi88 » 30 Aug 2007

Is the release still going to be this week?

User avatar
Marina Pashkova
Posts: 2758
Joined: 27 Jul 2007

Postby Marina Pashkova » 04 Sep 2007

New 2.1.976.1109 beta version of MultiCharts is out and can be downloaded here : http://www.tssupport.com/support/downloads/

glam_100
Posts: 157
Joined: 14 Jun 2006
Has thanked: 2 times
Been thanked: 4 times

Postby glam_100 » 04 Sep 2007

I'm very happy with the new MC speed especially with quad core.

Any chance of improving the data loading speed? Loading 5 years of intraday data takes orders of magnitude longer than TS. It looks like the loading time grows exponentially in relation to number of bars loaded.

User avatar
Marina Pashkova
Posts: 2758
Joined: 27 Jul 2007

Postby Marina Pashkova » 04 Sep 2007

Hello,

We have been working on improving the loading time and we have aleady increased the downloading speed 2.5 times. Our goal is to make this process faster than in TS.

What MC version are you using? Have you started using the latest beta 2.1.976.1109 now available on our website?

Also, when you are saying that it takes a long time to download data, what data and it what resolution is it? And how long exactly does it take?

glam_100
Posts: 157
Joined: 14 Jun 2006
Has thanked: 2 times
Been thanked: 4 times

Postby glam_100 » 04 Sep 2007

Hello,

We have been working on improving the loading time and we have aleady increased the downloading speed 2.5 times. Our goal is to make this process faster than in TS.

What MC version are you using? Have you started using the latest beta 2.1.976.1109 now available on our website?

Also, when you are saying that it takes a long time to download data, what data and it what resolution is it? And how long exactly does it take?


Hi,

I'm using 1109 build. Just forked over $499 and purchased a license.

It's not the downloading time. It's the loading time. I'm using a ASCII data source and it's so slow and I changed to a Globalserver data source, and it's about the same speed. It's the time MC takes to load the data into price bar that is slow. It has nothing to do with the downloading because all my data is locally stored in my computer.

glam_100
Posts: 157
Joined: 14 Jun 2006
Has thanked: 2 times
Been thanked: 4 times

Postby glam_100 » 04 Sep 2007

Just now I'm making a change to the data compression and it takes at least 3 minutes to load the data. I'm using 3 data series and it takes 9 minutes in total just to change the data compression!

User avatar
Marina Pashkova
Posts: 2758
Joined: 27 Jul 2007

Postby Marina Pashkova » 04 Sep 2007

What data compression and over what period of time are you loading?

glam_100
Posts: 157
Joined: 14 Jun 2006
Has thanked: 2 times
Been thanked: 4 times

Postby glam_100 » 04 Sep 2007

What data compression and over what period of time are you loading?


5 years of 15 min data. Just the regular session, not even the night data or else it would take 3 times longer.

User avatar
Marina Pashkova
Posts: 2758
Joined: 27 Jul 2007

Postby Marina Pashkova » 05 Sep 2007

We reproduced the situation you are describing. You are right. It does take a long time to load a chart.

We are currently working on speeding up this process. It will be at least 3 times faster in the 2.2 release.

glam_100
Posts: 157
Joined: 14 Jun 2006
Has thanked: 2 times
Been thanked: 4 times

Postby glam_100 » 05 Sep 2007

We reproduced the situation you are describing. You are right. It does take a long time to load a chart.

We are currently working on speeding up this process. It will be at least 3 times faster in the 2.2 release.


I look forward for the speed upgrade.

For comparison purpose, the same data that took MC 3 minutes to load only takes 5 seconds in TS. This is a really dramatic difference. Please see what TSSupport can do. Thx.

User avatar
Marina Pashkova
Posts: 2758
Joined: 27 Jul 2007

Postby Marina Pashkova » 06 Sep 2007

3 minutes vs. 5 seconds is indeed a dramatic difference.

However, the tests we have run here show different results. According to our tests, the previous beta is about 3-4 times slower than TS, the latest beta 2.1.976.1109 is 1.5 times slower than TS. None of the tests in these two versions yielded the difference you are talking about.

Could you run tests comparing the loading time again and give us the exact results?

Thank you.

glam_100
Posts: 157
Joined: 14 Jun 2006
Has thanked: 2 times
Been thanked: 4 times

Postby glam_100 » 06 Sep 2007

3 minutes vs. 5 seconds is indeed a dramatic difference.

However, the tests we have run here show different results. According to our tests, the previous beta is about 3-4 times slower than TS, the latest beta 2.1.976.1109 is 1.5 times slower than TS. None of the tests in these two versions yielded the difference you are talking about.

Could you run tests comparing the loading time again and give us the exact results?

Thank you.



Attached is a csv data file for ES daily data from 9/5/2001 - 9/6/2007 (6 years).

The exact load time is:

Compression 15 min
Date: 9/5/2001 - 9/6/2007 (total 6 years)

TS time: 3 seconds
MC time: 55 seconds


So it wasn't as bad as 3 min that I mentioned above. But the difference still exceeds 18x.
Last edited by glam_100 on 06 Sep 2007, edited 1 time in total.

User avatar
Marina Pashkova
Posts: 2758
Joined: 27 Jul 2007

Postby Marina Pashkova » 06 Sep 2007

Could you attach the file please ?

glam_100
Posts: 157
Joined: 14 Jun 2006
Has thanked: 2 times
Been thanked: 4 times

Postby glam_100 » 06 Sep 2007

I tried to attach the data file twice but somehow the attachment doesn't show. Can I email the file to you? It's about 3 Mb zipped.

User avatar
Marina Pashkova
Posts: 2758
Joined: 27 Jul 2007

Postby Marina Pashkova » 06 Sep 2007

Yes, you can e-mail it to mpashkova@tssupport.com

glam_100
Posts: 157
Joined: 14 Jun 2006
Has thanked: 2 times
Been thanked: 4 times

Postby glam_100 » 06 Sep 2007

Sent. Please check mail.

User avatar
Marina Pashkova
Posts: 2758
Joined: 27 Jul 2007

Postby Marina Pashkova » 06 Sep 2007

One more question before we run the tests: did you use ASCI import or ASCI mapping for the file?


Return to “MultiCharts”