Single quad-core or dual Xeons for Blocks rig?

Did some prelim comparisons between two systems I have on hand:

SYSTEM 1:
AMD Athlon3700+ lightly OC'd to 2.8GHz (previously incorrectly referred to as 3400)
1GB RAM
XP Pro

SYSTEM 2:
Stock Lenovo ThinkPad X61s
Intel Mobile Core 2 Duo L7500 @ 1.6GHz
2GB RAM
XP Pro


I ran several controlled tests on both systems:
- time to start Blocks
- time to close Blocks
- time to run a custom scan on S&P100 watchlist on 2,000 5-minute candles per symbol
- time to run same custom scan on Nasdaq100 watchlist on 2,000 5-minute candles per symbol

Each test was repeated 5 times. The scans were run with CPU UTILIZATION set to HIGH.

Here are the results:
TIME TO START BLOCKS:
System1: 58 seconds avg
System2: 36 seconds avg

TIME TO CLOSE BLOCKS:
System1: 13 seconds avg
System2: 9 seconds avg

S&P100 x 2,000 bars scan:
System1: 4 minutes, 6 seconds avg
System2: 2 minutes, 11 seconds avg

Nasdaq100 x 2,000 bars scan:
System1: 4 minutes, 9 seconds avg
System2: 2 minutes, 12 seconds avg


Things noted:
novo_ndaq100x2000_01_edited949.png


^^ During a scan on System2, Blocks utilizes one core at 80-90%, while the other only at 10-20%. I wonder if this behavior is normal? Shouldn't it be evenly dispersing the load amongst the cores?




blocks_backscan_sp100x2000_01_edited807.png


^^ During a scan on System1, the single core was utilized in the 70-90% range. The HDD containing the data set saw regular reads with fewer writes.


Will post more results and tests when/if I have time.
_mad
 
It would be interesting to see how it does with the same amount of memory. Otherwise the system is really fighting a losing battle when one has 1 gig and the other 2 gig.


: o )
 
Quote from 1Reason:
It would be interesting to see how it does with the same amount of memory. Otherwise the system is really fighting a losing battle when one has 1 gig and the other 2 gig.
: o )
For this test, system 1 was only using 674 MB of memory.
 
UPDATE

The purpose of this post is to publish the results of a benchmark on 3 different systems against a constant layout and rule set which I have programmed. These tests will be performed using Worden's StockFinder (the new version of Blocks). The goal is to figure out "How to build a fast RealTime Scanning / BackTesting computer?" -- a question I have asked many times before, and also one that is of frequent discussion on this forum.

The layout used for these tests is modest: A 5 min chart and a Daily chart, a single watchlist, a debug window. The 5 min chart has a volume indicator added. To perform the tests, StockFinder's BackTester was used. Under normal market hours, I would not normally use the BackTester tool. The reason its being used is because it helps obtaining significantly more accurate benchmark results.

System 1 - Single Core
AMD Athlon 3700+ @ 2.81GHz (slightly OC'd)
2 GB DDR1 RAM
7200rpm HDD
WinXP Pro SP2

System 2 - Dual Core
Intel Core 2 Duo L7500 @ 1.6GHz
2 GB DDR2 RAM
7200rpm HDD
WinXP Pro SP2

System 3 - Quad Core
Intel Core i7 920 @ 2.67GHz
12 GB DDR3 RAM
Intel X25-E Extreme SSD
WinXP x64 SP1

Code:
[b]SCAN #1 - 250 bars - 5min candles - S&P 500 watchlist[/b]

[url=http://xs.to/xs.php?h=xs135&d=09051&f=s1_250666.jpg]S1: 2 min, 40 sec  [img]http://xs135.xs.to/xs135/09051/s1_250666.jpg.xs.jpg[/img][/url]
[url=http://xs.to/xs.php?h=xs135&d=09051&f=s2_250935.jpg]S2: 1 min, 45 sec  [img]http://xs135.xs.to/xs135/09051/s2_250935.jpg.xs.jpg[/img][/url]
[url=http://xs.to/xs.php?h=xs135&d=09051&f=s3_250573.jpg]S3: 57 sec  [img]http://xs135.xs.to/xs135/09051/s3_250573.jpg.xs.jpg[/img][/url]
Code:
[b]SCAN #2 - 250 bars - Daily candles - S&P 500 watchlist[/b]

S1: 1 min, 27 sec 
S2: 1 min, 4 sec
S3: 42 sec
Code:
[b]SCAN #3 - 500 bars - 5min candles - S&P 500 watchlist[/b]

S1: 3 min, 34 sec 
S2: 2 min, 12 sec
S3: 1 min, 24 sec
Code:
[b]SCAN #4 - 500 bars - Daily candles - S&P 500 watchlist[/b]

S1: 3 min, 26 sec 
S2: 1 min, 47 sec
[url=http://xs.to/xs.php?h=xs135&d=09051&f=s3_500123.jpg]S3: 1 min, 7 sec  [img]http://xs135.xs.to/xs135/09051/s3_500123.jpg.xs.jpg[/img][/url]
Code:
[b]SCAN #5 - 1000 bars - 5min candles - S&P 500 watchlist[/b]

S1: 7 min, 13 sec 
S2: 5 min, 3 sec
S3: 2 min, 17 sec
Code:
[b]SCAN #6 - 1000 bars - Daily candles - S&P 500 watchlist[/b]

S1: 4 min, 23 sec 
S2: 3 min, 4 sec
S3: 1 min, 54 sec
Code:
[b]SCAN #7 - 2000 bars - Daily candles - S&P 500 watchlist[/b]

S1: 10 min, 5 sec 
S2: 7 min, 12 sec
S3: 3 min, 23 sec

Note that Daily bars are taking less time to scan than Intraday bars.

Now what would we see if a 1,000 symbol watchlist was used instead (2X as many symbols than before)?

Code:
[b]SCAN #8 - 2000 bars - Daily candles - Russel 1,000 watchlist[/b]

S1: BSOD - computer crashed
S2: 5 min, 25 sec  [url=http://xs.to/xs.php?h=xs135&d=09051&f=s2_2kd01128.jpg][img]http://xs135.xs.to/xs135/09051/s2_2kd01128.jpg.xs.jpg[/img][/url]  < start  end >  [url=http://xs.to/xs.php?h=xs135&d=09051&f=s2_2000_daily02127.jpg][img]http://xs135.xs.to/xs135/09051/s2_2000_daily02127.jpg.xs.jpg[/img][/url]
S3: 3 min, 21 sec



And at this point it is getting late and I need sleep, to be continued...
_mad
 
Back
Top