Benchmark for NT

[ClayB1]

Holy Sh*t - are you sure you didn't take that out of a refrigerator!! What motherboard did you use in your build?

[ctrlbrk]

Here is all the info:

OS: Windows 7 x64
Motherboard: eVGA X58 SLI
CPU / Speed: Core i7 920 @ 3.8ghz
Memory / RAM: 12GB 7-7-7-20
Graphics Card(s): (2) GeForce 9500GT
Monitor Display(s): (3) Acer 22's
Sound Card: Onboard
Hard Drive(s): 3Ware 9690SA w/(4) Patriot 32GB SSD R0 + (2) WD Black 750GB R1
PSU: Corsair 750W
Case: CM Stacker 830
Cooling: CM V8
Internet Speed: Load-balanced / redundant DSL+Cable, plus 100mb seedbox
Antivirus: Avast







Mike

[ClayB1]

Very nice

[jdfagan]

I ran a few tests to see impacts of some variables:

Symbol: FDAX
Date: 5/4/2009
Background apps: Many including Firefox with Pandora playing
04m 24s: Intel Core 2 X6800 @ 2.93GHz / 8GB, Vista x64, NT 6.5.1000.9 (jdfagan)

Symbol: FDAX
Date: 5/4/2009
Background apps: Many without Firefox
04m 0s: Intel Core 2 X6800 @ 2.93GHz / 8GB, Vista x64, NT 6.5.1000.9 (jdfagan)

Symbol: FDAX
Date: 5/4/2009
Background apps: Many without Firefox, but this time FDAX is in premarket (> 11pm PST) and is actively receiving realtime data from ZenFire in 3 of my charts running in background to the Strategy Analyzer
04m 0s: Intel Core 2 X6800 @ 2.93GHz / 8GB, Vista x64, NT 6.5.1000.9 (jdfagan)

Symbol: ES
Date: 5/4/2009
Background apps: Many without Firefox, but this time FDAX is in premarket (> 11pm PST) and is actively receiving realtime data from ZenFire in 3 of my charts running in background to the Strategy Analyzer
04m 0s: Intel Core 2 X6800 @ 2.93GHz / 8GB, Vista x64, NT 6.5.1000.9 (jdfagan)

Symbol: ES
Date: 5/4/2009
Background apps: Many including Firefox with Pandora playing, but this time FDAX is in premarket (> 11pm PST) and is actively receiving realtime data from ZenFire in 3 of my charts running in background to the Strategy Analyzer
04m 15s: Intel Core 2 X6800 @ 2.93GHz / 8GB, Vista x64, NT 6.5.1000.9 (jdfagan)

I agree with Dierk's comments below. How about just taking a sample strategy (one that preferably makes use of OnOrderUpdate -> OnExecution -> OnPositionUpdate) to stress out the event system a bit in NT and then choosing a standard date period (preferably more than 1 day) and symbol from ZenFire demo feed for backtest benchmark (e.g., ES 06-09) - or maybe an optimization run would be even better? That way you may get a better idea of impact of changing processor (or other variables) for a real-world strategy.

[NinjaTrader_Dierk]

>> How about just taking a sample strategy (one that preferably makes use of OnOrderUpdate -> OnExecution -> OnPositionUpdate) to stress out the event system a bit in NT and then choosing a standard date period (preferably more than 1 day) and symbol from ZenFire demo feed for backtest benchmark (e.g., ES 06-09) - or maybe an optimization run would be even better? That way you may get a better idea of impact of changing processor (or other variables) for a real-world strategy.

I definitely would not want to take the lead on this thread, however I somewhat agree with these statements. Here are the setups I would run performance tests on (in fact we're doing pretty much exactly that on our internal performance tests):
- basic OnBarUpdate only strategy
* backtest on a defined bars series
* optimize on defined parameter
- enhanced OnBarUpdate opnly strategy(ies)
* backtest on a defined bars series
* optimize on defined parameter
- free lance strategies using OnExecution, OnOrder, OnPosition
* backtest on a defined bars series
* optimize on defined parameter

All strategies should create a "resonable" number of trades. Strategies not placing any trades unlikely would server as a NT benchmark (as the thread title suggests).

Please find attached a code snippet from our own internal perf test strategies.

Note: We will not provide any support regarding this code nor provide comments on if or not these samples would be any better than others nor will we provide any advice on how to compile etc.. It's just some "food for thought"...

[KrisS]

Excellent, an opportunity for distraction while this noob tries to figure out how to make money. I love a good drag race.

Using 6.5.1000.10, my systems achieved the following:

Q9450 (2.66) OC to 3.3GHz, 4GB, Vista x32, 3:19 27% CPU Max, using mostly 2 cores

i7 920 (2.66) OC to 3.4GHz 6GB, Vista x32, 2:18 13% CPU Max, with spikey use of all 8 (4 cores hyperthreaded)

The i7 sure will be fun once supported by NT. Now for a strategy...

[ctrlbrk]

Hi KrisS, welcome to the forum!

Updated!

Mike

[iwannatoscript]

>>

Note: We will not provide any support regarding this code nor provide comments on if or not these samples would be any better than others nor will we provide any advice on how to compile etc.. It's just some "food for thought"...

I think we should built standard example, like Dierk suggest us, and us the file in this post.

I need a suggestion from people that works on hardware or geeks.

i want to buy a pc to do only backtesting and optimizationing by ninja.

No keyboard, no monitor, i'd like access by remote control so i think that i don't need all other.

I am thinkin to intall oh this macine winxp64 bit
i think that is better than win7 in use of resources, what do you think?

i am not skill on hardware, i'd like buy:

good cpu
ram
VERY BIG BIG fan for case and very good cool system (heavy calculation to do),
network card
500 USD is my budget


Any suggestion?

Could be a good idea that i change the cou evry 6 month?

Regards

[WhoKnows]

A question for the I7 920 owners:
how many parameters within a strategy can you typically run at the same time? And does the allocated memory to the optimizing make a difference?

On my Athlon 3500 with 3GB under XP the speed is apparently about 40% on your test, but what is more irritating for me is that if I try to optimize 3 parameters(40 steps each) at the same time the machine only just manages (1.5 hour runtime)... memory usage 0.4-0.5 GB.
4 parameters simultaneously sometimes works ... 8-10 hours .. but more often memory usage of NT will creep up to the 1 GB and the system locks up.
Doing the optimizing run with more variables would save me a lot of time and increase the urge to upgrade.

[ctrlbrk]

WhoKnows,

I have many strategies, but I would say on average when I am really running a big optimization job it is looking at at least 10 different parameters and each of those probably has 20 or more possibilities.

The memory usage will be determined mainly by the amount of historical data loaded. For instance, if you load 3 months of "5 minute" data, the memory footprint is relatively small. However, if you try to load 3 months of tick data, it will probably crash Ninja. Maybe it will work the first time, but due to memory issues, it will crash the second or third time. I usually re-load Ninja at least once an hour when I am developing (not trading), and I use it full time every day.

The Core i7 still brings performance improvements, but NT 6.5 is, let's just say, "not good" at using it. It is not like most other modern Windows applications that can utilize all cores efficiently. It will max out at about 25% usage on a Core i7, for instance. NT 7 is supposed to bring enhancements so I will reserve judgment until it is released. Hopefully it will be able to do what freeware Zip programs and video editing programs can do, fully utilize all cores and max out 100% my system.

Mike

[ctrlbrk]

iwannatoscript,

I suggest checking out www.newegg.com for reviews and good prices on computer hardware.

NT 6.5 does not support x64 officially, although I think every major OEM will be heavily pushing x64 once Seven rolls RTM in October. I have not used x86 for my primary desktop in years. I would recommend Seven x64 personally, it is more efficient than Vista by far, but I can't stand running XP for my primary desktop. However, there are many traders that would only consider XP and nothing else.

NT pretty much requires 2GB of memory but also cannot use any more than 2GB. So depending on what else you are doing on this machine, more memory is not beneficial to NT. This will not change with NT7 based on what has been posted. Personally, I have 12GB of memory but I usually have a lot of stuff going on. I am not a typical user.

CPU is the most important factor for building a NT machine. I wouldn't buy anything less than a high-end C2Q, or Core i7 920 if you can splurge. A basic low end video card for $50 bucks and a cheap 1TB hard drive will be fine. You can go as crazy as you want, but it won't really help NT.

Running headless is no problem. If you have more than one monitor, then Microsoft RDP will be an issue. But you can use other free alternatives like TeamViewer and be able to remote view multi-monitors (virtually, since it is headless, but presumably you still have more than one physical monitor from your main desktop to fit all the charts).

Mike

[iwannatoscript]

iwannatoscript,

I suggest checking out www.newegg.com for reviews and good prices on computer hardware.
NT pretty much requires 2GB of memory but also cannot use any more than 2GB. So depending on what else you are doing on this machine, more memory is not beneficial to NT. T
Core i7 920 if you can splurge.
Mike

I choose i7940 by cpu,

good cooler
Good Zalman 9900LED, if someone knows a better coller a little bit ceaper, no problem if it will do noiese but same cooling could advise please?

CASE MIDLE TOWER ATX 500W FULL BLACK By NILOX

Ram , mmm, i think to buy 8 giga, i use only ninja, so i waste 6 giga,
so could be better 4?



Motherboard i don't have any idea, any suggestions please?

[zoltran]

iwannatoscript ... Mike posted screen shots of his motherboard and cooler below.
Perhaps that will give you some ideas..

You now know that NT works better with a fast CPU and lots of RAM, and that the next release is reported to make much better use of multi-core CPU's and 64Bit OS.
So that should bias your picks to those type of systems.

Other than that, I don't think this Forum is the best place to be looking for more hardware info that that ...

There are many other hardware oriented sites where you can do your reasearch... Just google ' I7 Cooler' or 'i7 and motherboard' and look at sites such as Toms Hardware or ANDTech ...

[iwannatoscript]

iwannatoscript ... Mike posted screen shots of his motherboard and cooler below.
Perhaps that will give you some ideas..

You now know that NT works better with a fast CPU and lots of RAM, and that the next release is reported to make much better use of multi-core CPU's and 64Bit OS.
So that should bias your picks to those type of systems.

There are many other hardware oriented sites where you can do your reasearch... Just google ' I7 Cooler' or 'i7 and motherboard' and look at sites such as Toms Hardware or ANDTech ...

Thanks for the suggestions

[WhoKnows]

WhoKnows,

I have many strategies, but I would say on average when I am really running a big optimization job it is looking at at least 10 different parameters and each of those probably has 20 or more possibilities.

The memory usage will be determined mainly by the amount of historical data loaded. For instance, if you load 3 months of "5 minute" data, the memory footprint is relatively small.
Mike

Thanks for the info
10 parameters times 20 settings sounds like a dream to me.
As mentioned I get maxed out at 4 parameters with 40 settings.

The data were typically one year of 15 or 20 minute data .. don't think that should be too much memory footprint. BTW what also amazes me is the millions of pagefaults I get with NT. As I understand that happens when the computer can't find the data in its direct memory.

Anyway, probably time to upgrade. As NT6.5 cannot use a multicore I first thought to wait until NT7 to see what would be the best option forward. But the info about VMware here and on your blog makes me think that is the way to go. At least I then can run optimizations parallel.
Just hope the learning curve on VMware is not too steep.