Benchmark for NT

[autobot]

AMD Bulldozer or Intel Sandy Bridge

Hi.

Does anyone have any real world NT 7 performance comparison between these two CPUs?

It seems like some of the performance benchmarks are scoring higher due to Hyper Threading which may not be beneficial to NT. Any advise gleaned from experience is much appreciated.

[whittod]

What two CPU's are you referring to?

What did you get for a score?

[autobot]

Hi;

The two CPUs I'm referring to are in the subject line. AMD Bulldozer vs Intel Sandy Bridge. The benchmarks I'm referring to are on the web. I don't yet own either of these platforms but I'm trying to determine if 8 real cores are better than 4 with hyper threadding for NT backtest and optimization.

[whittod]

Here are some links for you review regarding CPU processing speed.

http://www.cpubenchmark.net/high_end_cpus.html

http://www.passmark.com/forum/showthread.php?t=3340

Intel will be releasing a new Ivy Bridge in Q2 of 2012 you might want to check out.

When it comes to back testing I assume the new AMD/Intel processors will get the job done to your satisfaction.

What processor are you currently running?

[autobot]

Thanks. I've seen the benchmarks and depending on which one you're using the fastest CPU will vary.

Are you suggesting that Passmark is the benchmark that equates to NT optimization performance?

[autobot]

Forgot to mention. I'm currently running a Q6600 which is a quad core intel at 2.4Ghz with 8GB RAM and Win7 x64

[whittod]

Looks like an average score of your Intel Q6600 @ 2.4Ghz is 2984

http://www.cpubenchmark.net/cpu.php?cpu=Intel+Core2+Quad+Q6600+%40+2.40GHz

I'm sure when you upgrade to the Sandy Bridge or Ivy Bridge CPU's your processing with be 5 time faster then what your experiencing now with your Q6600. You won't find a better bang for the buck then the current i7s 2600k Sandy Bridge CPU, you will not be disappointed.

I assume the if your back testing and optimizing you are in search of a better trading system or trying to develop a successful method. My experiences over the years suggest you first have to "get your mind right" about the trading business before you can even begin to develop a method or system. Also you'll spend a life time optimizing since the "market dynamic" is always changing so learn to trade from the markets perspective works every time.

Any other questions please let me know.

Thanks
TW (whittod)
www.EminiTradingSolutions.com

[autobot]

After another round of research, I ordered the SandyBridge 2600k and an "Intel" brand z68 chipset motherboard for cost and reliability given the performance. I'll also have double the RAM at 16GB.

Also, I agree with your comment about getting your mind right. Where I'm at right now is that I've realized there are unlimited "edges" it seems and the real holy grail is money management and compounding.

I want this extra hardware to help find and prove out more of these edges to use to create more systems so I can have a few combined together to smooth my equity curve.

Each system has some known psychological reason for the edge existing. I take that as a starting concept and build custom ways to exploit it from there.

My recent system is a mean reverting one that I've optimized on the ES but then backtested with positive expectancy on a few other markets without changing the parameters. It works as is on 6C, QM, YM and some others as well as ES of course.

[autobot]

Benchmark results of Intel Core 2 Quad Q6600 2.4Ghz

What two CPU's are you referring to?

What did you get for a score?

NT Benchmark v1.0a, build 11-08-2008
Elapsed time: 5 minutes, 41.6 seconds.

Intel Core 2 Quad Q6600 @ 2.4 Ghz

Windows 7 x64, 8GB DDR2

I noticed that the load on 2 cores was more than on the other 2 and the total CPU was at 25%. There was an occasional spike up to 50% or so (probably due to some other process)

With luck my next post will be made from my new Core i7 SandyBridge 2600K

[autobot]

New Sandybridge vs old Q6600 - smallish improvement on benchmark

I reran the benchmark and to my dismay there is not as much improvement as I hoped for considering the price of the upgrade. I was expecting closer to twice the performance on the benchmark.

I will say this. Outside of this benchmark my system rocks. One of the best things I did was go for the 120GB SSD. Windows install was about 9 minutes disk to desktop. NT install was on the order of 3 seconds. I was in disbelief. My drive is already getting on the full side. Go as big as you can afford and watch out for the tricks they play with the performance specs on the SSDs. Read some reviews.

Here are the results:
Both used NT Benchmark v1.0a, build 11-08-2008

Old Intel Core 2 Quad Q6600 @ 2.4 Ghz
vs
New Intel Core i7 SandyBridge 2600K stock clock speed 3.4Ghz

Old Elapsed time: 5 minutes, 41.6 seconds.
New Elapsed time: 4 minutes, 28.4 seconds.

Old Windows 7 x64, 8GB DDR2 PC2-8500
New Windows 7 x64, 16GB DDR3 1333 PC3-10666


On the new one I noticed that the load was on 4 cores only and the hyper threading (the other 4 "cores") did not appear to be used. It is enabled however. The total CPU was at 12% for 99.5% of the time with only two 1 second blips to 13% and 14% probably due to some other Windows process.

Perhaps the performance gains are directly related to the clock speed and not really benefiting from the more advanced features of the CPU?

I may try overclocking to around 4Ghz. If I do, I will post the results.

[Blackbull]

Guys, here is something you needed to consider: we do internal benchmarking always on the same (!) data set. I did not read whole thread, but so for I have not found where you would agree on a fixed (!) data set. May be you should consider tying some "free" data to your benchmark and make sure you all run the benchmark on the exact same time frame or even better: just iterate through OnBarUpdate only once, but execute the performance part say 100,000 times.

Also:
- you should shut down any other app (including e.g. Outlook) to eliminate any potential performance impact there.
- if I'm not mistaken the perf test for the most part is floating point math stuff. I would not think that this would be the typical perf footprint of a NT strategy...

Just my 2 cents

Dierk,
Generally speaking would these benchmark results relate to chart loading time performance/speed? ... Or do the benchmark speeds only relate to the actual backtest feature?
Thanks!

[whitmark]

Let me offer some comments that are published in the benchmark code that was first developed by cntlbrk, a.k.a. Big Mike:

Instructions: Set data series to Minute and interval to 1. Set time frame to one market day like start Mon end Mon. The general idea is the amount of data you feed Ninja doesn't matter, instead, the benchmark is based on how long it takes to calculate some math functions in the script.

While not a perfect benchmark for actual backtesting, I suspect the relative results and ranking will be useful to most.

I would be interested to see some updated times posted for PCs with 3rd generation i7 chip sets.

[paulg]

Let me offer some comments that are published in the benchmark code that was first developed by cntlbrk, a.k.a. Big Mike:



While not a perfect benchmark for actual backtesting, I suspect the relative results and ranking will be useful to most.

I would be interested to see some updated times posted for PCs with 3rd generation i7 chip sets.

Ya, I would like to see if does exist a PC that is able not to lag into a fast market with calculations active on a dozen of tick by tick charts w/ Rithmic feed.

[Global]

Just wondering, has anyone tried this with more recent NinjaTrader updates? Has the speed improved at all since you all first started trying this?