Not logged in
Viewing 15 posts - 1 through 15 (of 99 total)

Forums>StrategyQuant>General Discussion>PC setups, configurations, benchmarks and general recommendations

  • #250166 |
    ivan
    Participant
    143 Posts

    Because the interest is high among new users of SQX (i saw many topics with the same goal), i thought it would be a good idea to concentrate everything hardware related in one topic where new users but also more experienced ones, can easely find in one place what they are looking for. What to buy, what configuration, air or liquid cooled…etc.

    This topic is designed to help users in choosing the most apropriate hardware configuration according to their budget, a new computer or an upgrade. Also, posting benchmarks for processors, general raw performance benchmarks as a guide but also examples of settings in SQX and the resulting performance for others to have an idea what expectations they can have from a certain processor.

    I am sure the contribution of the more experienced users will be greatly appreciated by everyone.

    For starters, in the following days and weeks, i will post different results (pairs and timeframes) on my configuration, an AMD 3700X (3,6Ghz 8 cores) and 32GB RAM.

    To be relevant, we can agree on a minimum criteria, for example, each setting must be run at least 24 hours, and we can agree to count the accepted strategies/hour (in databank).

    Timisoara, Romania
    3700X 3.6 Ghz 8 cores, 64GB RAM DDR4 3000Mhz, Samsung 970 EVO Plus M.2 NVMe

    #250171
    Customer
    422 Posts

    many many benchmarks was already done – if you will not use the same data, same settings, results are not comparable…our “last” findings are in the table

    but i must say one thing: to be profitable doesnt mean to have strongest PC specs

    its more about users itself

    Attachments:
    You must be logged in to view attached files.

    You want to be a profitable algotrader? Sharing of final strategies with real traders just started. Fill in this FORM. 1500+ final SQ3 and SQX strategies for members running on demo accounts DEMO ACCS. We provide also strategies for indices - DAX and DOW JONES, because we have realtick data from brokers.

    #250176
    ivan
    Participant
    143 Posts

    @hankeys: interesting and useful chart. Question: what parameter is used in the chart?. The numbers from the chart, seem on par with CPU-Z benchmarks for raw performance

     

    i post a screenshot from the latest test, between 32 – 33 accepted strategies (databank)/hour on EURUSD H1. Around the same on GBPUSD H1. On other pairs and timeframes i got less. around 20.

    on the older SQ3 i never stoped below 5.000 in databank. That was the minimum. The initial goal was to reach 10.000 but it was impossible, i had a laptop.

    Attachments:
    You must be logged in to view attached files.

    Timisoara, Romania
    3700X 3.6 Ghz 8 cores, 64GB RAM DDR4 3000Mhz, Samsung 970 EVO Plus M.2 NVMe

    #250179
    Customer
    422 Posts

    no, its a number of strategies generated in 1 hour with custom project

    every user has the same data and same user, only by this way, we could compare something

    You want to be a profitable algotrader? Sharing of final strategies with real traders just started. Fill in this FORM. 1500+ final SQ3 and SQX strategies for members running on demo accounts DEMO ACCS. We provide also strategies for indices - DAX and DOW JONES, because we have realtick data from brokers.

    #250183
    ivan
    Participant
    143 Posts

    several conclusions can be drawn from those charts and this would be useful to beginers in SQ

    1. anything below a Core i7 8700K is too weak to make something useful with.

    2. sweet spot is anything modern, Core i7, Core i9 and Ryzen 2nd gen 7 and 9 with good prices and decent performance. Also their performance is close so it would depend on other factors in choosing.

    3. the 2nd gen Threadripper is the only family of processors with significant boost over the rest (double) and the best there is, but several times the price and power consumption.

    Regarding SQX benchmarks, they will be useful also in the future and i think we must continue updating the table, but in my personal view, taking the “accepted” strategies, in databank (the number is smaller and easier to remember) and a minimum of at least 12 hours, if not 24 hours would be more close to real life. With the same file of settings of course.

    Timisoara, Romania
    3700X 3.6 Ghz 8 cores, 64GB RAM DDR4 3000Mhz, Samsung 970 EVO Plus M.2 NVMe

    #250184
    Customer
    422 Posts

    its a number of generated strategies per hour, so without filtering with random generation – so i think, that this value represents the performance very good

    i agree, that below i7 its nonsense to try do anything

    for me are better workstations instead of laptops or desktops, because we need 100% full load for a long time

    You want to be a profitable algotrader? Sharing of final strategies with real traders just started. Fill in this FORM. 1500+ final SQ3 and SQX strategies for members running on demo accounts DEMO ACCS. We provide also strategies for indices - DAX and DOW JONES, because we have realtick data from brokers.

    #250185
    ivan
    Participant
    143 Posts

    are better workstations instead of laptops or desktops, because we need 100% full load for a long time

    i agree with you that servers have several advantages, one is the ECC memory. Others are the power supply and fans which have bearings designed for long use. Most ordinary desktop fans have a lifetime 2 – 3 years of continuous use. And last but not least, one must have experience to build to assemble a computer from ground up.

    However, for desktops, one major advantage in the last 2 years: price

    https://hothardware.com/news/amd-ryzen-7-2700x-deal-50-percent-off

    when AMD releases new generation of processors, it discounts the former. 160 EUR for 2700X, and you saw in the charts its very decent. Its a very powerful argument for desktops, especialy the AM4 motherboards are cheap. In extreme cases, you can buy 2 cheap mini tower cases and you build 2 computers and run 2 instances of SQX

     

    Timisoara, Romania
    3700X 3.6 Ghz 8 cores, 64GB RAM DDR4 3000Mhz, Samsung 970 EVO Plus M.2 NVMe

    #250364
    ivan
    Participant
    143 Posts

    i would like to post the configuration i put together last month. I think it might be useful for beginers. Modern motherboards dont have onboard video so a discrete graphic card is mandatory.

    Processor: AMD 3700X 3.6 Ghz 8 cores
    Motherboard: Gigabyte B450M DS3H
    Video card: Zotac GT730 2GB DDR3 Zone Edition (passive cooled)
    Memory: Corsair 32GB CMK32GX4M2D3000C16
    SSD M.2: Samsung 970 EVO Plus 500GB
    Power supply: Seasonic M12II EVO Bronze 520W
    Case: Coolermaster MasterBox Q300L
    Case fans: 3 x Arctic F12 PWM with PST 4 pins
    Cooler CPU: Arctic Freezer 33
    Wireless adapter: Archer T2U Plus

    Total cost around 760 EUR. All components in sealed box but some purchased from the grey market, only a few from a store. I assembled the system.

    Temperatures: 60 – 62C max on the CPU and 40 – 42C on the motherboard in ambient constant temperatures of around 24C. Very silent system and the fans rearely reach max rpm. The air that comes out is only slightly hotter than room temperatures.

    Power consumption: still in testing

    Timisoara, Romania
    3700X 3.6 Ghz 8 cores, 64GB RAM DDR4 3000Mhz, Samsung 970 EVO Plus M.2 NVMe

    #250410
    ivan
    Participant
    143 Posts

    With the system from above and a minimum 12 hours each (but most on 24 hours run) with general settings, the results would be as follows:

    major pairs, GBPUSD, EURUSD, GBPJPY: M15 – H4: from 20 to 35 accepted strategies/hour

    crosses like CADJPY, or like GBPCAD, same M15 – H4: from 1 to 8 accepted strategies/hour.

    Only one notable exception: EURUSD H4, aprox 300/hour. This is the single configuration when i reached 10.000 in databank in a reasonable timeframe..

    On SQ 3 i used a minimum of 5.000 in databank.

    By general settings i mean just reasonable, common sense building blocks, mandatory but loose SL and PT, minimum 300 trades and whole history.

     

    Timisoara, Romania
    3700X 3.6 Ghz 8 cores, 64GB RAM DDR4 3000Mhz, Samsung 970 EVO Plus M.2 NVMe

    #250429
    Customer
    422 Posts

    accepted strategies count be taken as some “benchmark”

    benchmark need to be done with the same amount of data, same settings, random generation without filtering – number of bruto strategies per hour

    You want to be a profitable algotrader? Sharing of final strategies with real traders just started. Fill in this FORM. 1500+ final SQ3 and SQX strategies for members running on demo accounts DEMO ACCS. We provide also strategies for indices - DAX and DOW JONES, because we have realtick data from brokers.

    #250456
    ivan
    Participant
    143 Posts

    Yes, of course, i understand your logic and i agree. I als owant to participate in running different settings files and whenever is needed, i will run and share the results.

     

    But, since the first builds of SQX i was interested in benchmarks and i saw the first results, those huge numbers, in their milions but those numbers dont answer to a very simple, basic and fundamental question: how much time (hours, days) do i need to obtain a decent stable EA on a specific pair and timeframe. And this is something every beginner is interested.

    That huge number doesnt say how many of those, reach the databank, especialy given the fact that the speed varies (decreases), so beginers should know that the speed of generation is not constant. So the number given is the average of the entire length of time. I reached 10.000 in databank starting from 300/hour in the first hours and ended up with 20/hour in the last hour before i stopped, being clear to me that the potential was exhausted or depleted.

    With this type of bechmark, you get a more clear picture of how much time you need, depending on how many strategies you want, for example a minimum of 1.000 or 3.000 or 10.000.

    Regarding the exact strategy settings, changing just slightly the settings like a SL, PT or number of trades, history, doesnt alter much the time needed. I mean if i only change from a minimum of 300 trades to 350, the final time needed will not differ significantly, or the SL from 1.000 pips to 1.500, same. I wont jump from 20/hour to 100/hour. I made tests. Only with dramatic, big changes, i will impact the final time lenght.

    Beginners must understand that history of the pair is the biggest factor, not the software or the processor.

    Timisoara, Romania
    3700X 3.6 Ghz 8 cores, 64GB RAM DDR4 3000Mhz, Samsung 970 EVO Plus M.2 NVMe

    #250468
    Customer
    422 Posts

    nobody knows what is ACTEPTED for you – its a final strategy after whole worklow, after WFM or what?

    what are your ranking criterias, genetics settings, number of conditions, building blocks, data settings, your workflow?

    You want to be a profitable algotrader? Sharing of final strategies with real traders just started. Fill in this FORM. 1500+ final SQ3 and SQX strategies for members running on demo accounts DEMO ACCS. We provide also strategies for indices - DAX and DOW JONES, because we have realtick data from brokers.

    #250483
    ivan
    Participant
    143 Posts

    for the tests i made, it was much simpler, i just loaded the default configuration file offered by SQ, made just a few adjustments and hit start (those adjustments are too small to have a visible impact on the speed)

    its obvious the test doesn’t include retest processes like Walk Forward or other retest types, just generation of strategies. The process is stopped manualy if it closes very fast to almost 0. In one generation, i stoped when after 3 – 4 hours, the spped was 0/hour.

    Regarding the relationship between spped and settings, its a constant myth i come across many times. For example i generated on USDCAD M15 or USDCHF M15 and changed the settings to a certain point. Doesn’t imact at all, still 2 – 4/hour. Only if i make dramatic changes like disabling the SL or the minimum trades i get a visible difference but even in that case, its not dramatic, maybe double, from 6 to 12/hour bot not 50 or 100/hour.

    Only on major pairs like GBPUSD or EURUSD, GBPJPY i get decent, reasonable speed.

    Timisoara, Romania
    3700X 3.6 Ghz 8 cores, 64GB RAM DDR4 3000Mhz, Samsung 970 EVO Plus M.2 NVMe

    #250543
    ivan
    Participant
    143 Posts

    on more restrictive settings, the speed drops sharply, a normal behavior i would say, so the more complex the strategy, the lower generation speed is

    i decreased the maximum SL or disabled the entry/exit symmetry and/or other parameters and the speed dropped from lets say 25 accepted/hour to 4 or 6/hour

    the final criteria is how many strategies in databank one needs but from my tests, below 1.000 is not enough, at least 3.000 if not 5.000 but with 4 or 6/hour, you need 1 month or 2 months

    Timisoara, Romania
    3700X 3.6 Ghz 8 cores, 64GB RAM DDR4 3000Mhz, Samsung 970 EVO Plus M.2 NVMe

    #250634
    ivan
    Participant
    143 Posts

    a few new potential useful information:

    -on XAUUSD, I’ve hit probably an absolute all time record, on H1 I’ve had 1.400 added strategies in databank/hour so I’ve hit the max 10.000 in aprox 8 hours or a bit less. From SQ support I’ve learned that once the max capacity of 10.000 is hit, the generation process continues to generate new strategies and replacing them in the databank if they are better, and eliminate the weaker ones. This is very important.

    -on such a fast generation process, the memory usage skyrocketed to 99% ( of 32GB DDR4) (measured in Windows Task Manager) compared to other pairs and timeframes when the usage was around 30 – 40%. The conclusion is that unfortunately for our budget, 32 GB is not enough on such computers and 64GB is more desirable.

    -monthly power consumption on my system (3700X) is aprox 50 – 60 kw/month, same as a small air conditioning unit in the summer. In my country, Romania, this translates to aprox 10 EUR/month.

    -case fans: only fluid dynamic bearings and dual ball bearings are suited for such systems. Sleeve bearing fans are inappropriate due to the short lifespan of about 3 years of continuous use.

    Timisoara, Romania
    3700X 3.6 Ghz 8 cores, 64GB RAM DDR4 3000Mhz, Samsung 970 EVO Plus M.2 NVMe

Viewing 15 posts - 1 through 15 (of 99 total)

You must be logged in to reply to this topic.