2020 UPDATE – Whilst this test is still highly relevant, we have a brand new updated graphics card test now available: Trading PC Graphics Cards
When it comes to trading and multi-screen computers, one of the most misunderstood aspects is the graphics card setup.
I’ve lost count of the number of times people have ‘told’ me:
‘you need a really powerful graphics card to run a multi-screen trading computer’
‘my friend said I should get a <insert ridiculously expensive gaming graphics card here> as it will make my computer run faster’
Personally, I’ve always responded to this by advising that perhaps this might be flawed thinking, and that our own experience shows us that this is not actually true.
The problem is that when I say this I’m really only giving my opinion, and if you’re told one thing by one person and something different by me then who are you to believe?
Sure, my 10 years of running a company specialising in trading computers and multi-screen computing probably outweighs the anonymous guy in a forum who’s probably never built a computer in his life, but we still need a better way of answering this question.
So, my solution is to put these different types of graphics cards to the test to actually see the real differences between them.
Let’s actually see if your trading software will run faster on one card versus another.
Do you need that £500 gaming graphics card? Let’s find out.
Test System and Methodology
We put together a base computer using medium level PC components to produce results that someone purchasing a new computer for trading might expect to see. We then simply swapped the graphics cards for each series of tests.
Graphics drivers used were always the latest ones provided by the graphics card manufacturers, again this would be what a standard new computer owner would expect to be running.
We repeated each of our tests 3 times for each graphics card and then used the average result of these as the final benchmark result for the charts below. There is always a small difference between test results even on the same exact computers, taking an average of three results helps give a more realistic view of the actual performance level of the card.
The full system specification was: Intel Core i5 7600K CPU, Asus Z270-P Motherboard, 16GB 2,666Mhz Crucial RAM, 240GB Kingston Solid State Hard Drive, BeQuiet 600w Power Supply, Antec 302 Case, Arctic Freezer CPU Cooler.
Note: The tests for the Intel 630 UHD graphics chip and the nVidia GTX 1060 card were performed on our latest test system which uses the newer i5 8600K CPU and Asus Z370-P Motherboard combination. This did affect the results of some tests as the CPU change made a difference to them, to alleviate this we repeated the nVidia GT 1030 graphics card test on the original base system and on the newer testing machine and then reduced the results of the GTX 1060 and the Intel 630 test to factor out the CPU advantage. These two tests have been starred on the charts below.
Finally, the software used to perform the tests was: Windows 10 Home Edition, Google Chrome v59, Handbrake 1.07 (64-Bit)
Web Browser Tests
Our first series of tests are web browser tests, these are essentially websites that run in a browser and perform a variety of functions which are timed, they then output a score for your browser and PC, the higher the better.
We use these tests as a proxy for trading software performance. Many trading and charting platforms actually run in a web browser and they use the same (or very similar) functions that these tests benchmark, if a browser test performs well then your web based trading application is also going to perform well.
Non-browser based trading software, such as MetaTrader 4, often use a computer in a similar fashion to the browser based packages. Essentially they are receiving data feeds and then representing these on your screens in the form of charts and tables of numbers. Due to this we can say that the browser tests are also a good proxy for these types of trading packages.
Our first browser test is the JetStream tool, it performs a series of tests that a typical web browser application would use in general operation and gives us a score at the end.
Interestingly the results are all pretty much identical between the various graphics cards, there is just a 0.52% difference between the best and the worst result, that is well within the margin of error for this test (we typically see more than that between two tests run on the same hardware!).
To be honest when I first saw these results put together even I was surprised how close they were, it clearly shows that none of these graphics cards impact the performance of this web application what so ever.
This test is similar to the JetStream test in many ways, it does use some slightly different procedures though so it’s worth looking at to see if it shows anything different.
Again, what is clear again is that the graphics cards have pretty much no impact on performance levels in this test, the difference between the best and worst result is just 1.97%, greater than the JetStream deviance but still well within our margin of error.
After the JetStream result I’m not surprised about the above result, the two tests often produce similar results.
Our final browser test is the WebXPRT suite, it again uses a lot of calculations but also features a much more graphical test as a part of it. Part of the graphical tests are rendering charts and graphs, and manipulating tables of data, a very good proxy for many types of trading software applications.
Here we have a little bit more spread on the results. The GT 1030 and GTX 1060 cards come in with stronger results than the rest of the field whilst the GT 710 is a small way back, the other 4 cards are separated by just 2 points. This gives us a 7% difference between the best and worst results.
I’ve taken a closer look at the results for this test and it shows that the main difference was that the GTX 1060 and GT 1030 cards were able to render the 3D graphical parts of this test quicker than the other cards, they had a roughly 9% advantage. For actions such as dealing with tables of numbers and standard line charts the differences were a lot less, these came out around 2.5% between the best and worst results.
I guess this is to be expected, both the 1030 and 1060 cards are classed as gaming cards, and the advantage of them comes when dealing with 3D graphics.
Overall, what is clear across all three browser tests is that the graphics card has very little, if any, impact on how these tests perform. The only test to show even a marginal difference was only on the 3D graphical part of the test, not something your trading software is actually going to ever do.
2D and 3D Graphics Tests
Our testing tool, Passmark Performance Test, runs a number of system tests, two of these directly assess the 2D and 3D graphical performance of the overall system.
Seeing as we are assessing the graphical performance of the different graphics cards I think these would be relevant tests to look at next.
2D graphics make up the vast majority of the interfaces on your computer and software. Pretty much anything that is made up of lines, fonts and vector shapes are classed as 2D shapes and are processed by your 2D graphics engine in your PC.
For trading software anywhere you see columns of numbers or text you are seeing 2D graphics. All trading chart types such as line, bar, tick and candlestick charts are 2D items.
The results for our graphics cards show us that there is a bit of a spread here, the M2000 and GTX 1060 do have superior 2D graphics performance.
The worst performing card was the Intel 630 UHD, this is actually an ‘on-board’ graphics card, it is built into the motherboard. Historically on-board graphics tended to be really bad, here we can see that it still doesn’t quite match up to the dedicated graphics cards but it’s not a million miles away either.
The spread between the best and worst result is slightly over 13%, not a massive margin.
Interesting the most expensive card here, the NVS 510 is one of the worst performers. It only improves on the GT 710 by 1.8% yet costs around 12 times the amount!
The GT 1030 performs 3.4% better than the NVS 510 and costs 6 times less…
Looking back at the browser tests above, it shows that even the lowest performer here, the Intel 630 UHD, matches the performance of the top M2000 and GTX 1060 cards in both the Octane and JetStream Tests, and it also matches the M2000 in the WebXPRT test.
What can we learn from this?
I think, (and have been saying for years), that when it comes to 2D graphics, the level of performance required to display things like fonts, lines and vector shapes on a computer monitor was smashed through quite a few years ago.
If we imagine that on the chart above once a card scores more than say 600 or 700 then this is ‘fast enough’ to not be a bottleneck on your computers performance for 2D graphics, anything over this score would then offer no benefit to you when it comes to running your programs.
This line of thinking is backed up by the identical results in the browser tests between all the cards, essentially they are all ‘good enough’ to not impact your trading software performance levels.
Let’s now look at 3D test results.
The GTX 1060 is the clear winner here, it offers over 1400% improved performance over the GT 710, so if you are looking for strong 3D performance then you’d definitely want the 1060 card.
Looking back at the web tests though, what does this 1400% improvement give us when using our trading software packages? That’s right, it offers no improvement in the Octane and JetStream tests, and only 7% in the WebXPRT test (which drops to 2.5% when you factor out the 3D graphics tests in that particular test).
3D graphics are used in games and not too much else. Some high end photo, video and CAD editing / rendering packages will make use of 3D graphics hardware, but on the whole, outside of these use cases, nothing else will use your 3D graphics engine in any meaningful way.
Due to this, your graphics card’s 3D performance will not impact your trading software at all, imagine the 3D graphics engine as a tool your computer can use if it needs to, your trading software will never use this tool, so it doesn’t matter how effective the tool is if it’s never going to be used, does that make sense?
So, if we are looking at graphics card performance levels for a trading computer then why bother looking at 3D performance?
I think it kind of proves the point that it will not impact your trading, if that isn’t clear cut by looking at these charts then I don’t know what is.
As an interesting side note, the NVS 510 again scores really poorly here, it really is struggling to justify it’s high price in my opinion.
Even though our focus here is to test the impact of graphics performance on a trading PC, let’s now take a look at some CPU tests to see if the graphics card makes a difference to these.
The processor or CPU in a computer is essentially the heart of the system, it is often directly responsible for how fast your computer runs, as such many computer testing packages have lots of CPU benchmarks available.
Our Performance Test suite offers us an overall CPU test:
This benchmark is an overall score made up by a number of sub tests which look at things such as physics simulations and number sorting algorithms, all processor intensive tests.
The chart shows there is no benefit given to any of these tests by any of the graphics cards, they have zero impact on how the system performed this test.
The performance spread was 1.1% between the best and worst result.
A sub test of the main CPU test above, this measures the raw speed of the processor only and discounts other factors like its multi-tasking capability.
Again none of the graphics cards offer any benefit to this test, the performance difference was just 2.5%which is within our standard error margin for these tests.
This test uses a program called handbrake to change one video format into another. This type of action is heavily processor dependent and because Handbrake will scale up to use as many processing cores as a computer has, it is a good test to see how well a particular system performs in terms of high end multi-tasking and multi-threaded workloads.
Note: The score is the number of frames of video the program could process per second, obviously the higher the better.
Once again, as expected, this test is completely unaffected by the type of graphics card installed. The performance difference across all results was just 0.55%.
RAM & Hard Drive Tests
Our final two tests will look use the Passmark Performance Test suites RAM (memory) and hard drive benchmark tests to see if your graphics card affects either result.
RAM or memory is where your open and active files and programs are stored when in use, this test measures the speed at which the RAM can be accessed and written to on the system.
Not surprisingly, graphics performance has no impact here, the spread between all results is just 2.6%, a negligible amount which is well within a standard error margin.
Your hard drive is permanent storage for your files and folders, when you open programs and files they are read from the hard drive and then transferred into your RAM. This test measures the read / write and seek speed of the hard drive.
These results show more differentiation, digging into the results the Passmark suite results shows us that multiple tests on the same exact system showed big differences between results. We have seen this on other system tests as well, the hard drive test seems to be pretty volatile and susceptible to what else is going happening on the test computer at the same time.
Even if we say that the graphics card is affecting theses tests, there is no logical pattern. The strongest card in the 2D and 3D tests is the GTX 1060 but it gives one of the worst results here. The strongest result here is from the GT 1030 which is not the best or the worst in any of the other benchmarks.
As such I would not put too much emphasis on these results, I honestly can’t see how the graphics card would impact these tests really.
We will also be publishing a hard drive test soon which will show the impact of different speed drives have on your trading computer, but as a quick heads up, the impact is negligible…
Overall Conclusions and Recommendations
Okay, so what have we found though these series of tests?
Let’s say that the majority of trading software packages either run through a web browser, or they take in a data feed and then display it on your screens using a combination of fonts, lines and vector based charts. I think we can agree that generally that is how they operate.
On that basis, the performance of your trading computer is practically unaffected by the graphics card you go for.
We have clearly seen that a relatively high end gaming graphics card, the GTX 1060 gives us no benefit in the Google Octane and JetStream browser tests, and it only offers a marginal improvement in the WebXPRT test which reduces significantly if we discount the 3D graphical aspects of that particular test.
We have also seen that an expensive NVS 510 card offers nothing in the browser tests that the much lower cost GT 710 could not match, and when we dig into the 2D and 3D tests it is one of the worst performers overall.
Finally, the CPU, RAM and Hard Drive tests show us that none of these are affected by your choice of graphics card. A more expensive, or ‘high end gaming’ card has zero impact on these aspects of your trading computers performance.
So, next time somebody tells you that you need a ‘high end’ graphics card for your trading computer then just point them here, they are wrong.
If you are in the market for a new trading computer then here are my recommendations in terms of graphics cards:
Scenario 1 – Cheapest Possible Trading Computer Running 1 or 2 Screens:
Use the onboard Intel 630 UHD graphics chip that probably comes on your motherboard, it will happily run your trading platforms and essentially costs you nothing on top of the cost of the motherboard (which you will need anyway).
Scenario 2 – Cheapest Possible Trading Computer Running 2 – 6 Screens:
Whilst the onboard Intel chipset is adequate it will only run 2 digital screens in most cases, going over this will require dedicated graphics cards.
In this case I recommend using the GT 710 cards, they are low cost, low power draw, can be obtained in passively cooled (silent) configurations and offer more than good enough 2D performance for your trading software needs.
Scenario 3 – Great Performing Trading Computer Running 2 – 6 Screens:
For this I’d recommend the GT 1030 card, it has great 2D performance, it also does pretty well on the 3D performance tests, the price is competitive and you could run it easily in a 3 card system to support up to 6 standard monitors. Finally, they can also be passively cooled for silent running.
Scenario 4 – Great Performing Trading Computer Running 8 – 12 Screens:
It’s best (and often required) that you use a maximum of three graphics cards in a PC, this means that to support 8 or more screens then you are going to need to use graphics cards that can support 4 screens each. This rules out the GT 710 and GT 1030 cards as they are dual monitor cards.
The AMD W4100, nVidia M2000 and the NVS 510 are all quad monitor capable graphics cards, i.e. they can each run 4 screens per card, so two of them would run 8 screens etc…
Looking at the test results and taking into account pricing, the AMD W4100 is the most cost effective card here. Definitely stay away from the NVS 510 (and anyone who recommends it as they obviously don’t know what they are talking about…), it’s an expensive and pointless card for our needs on a trading computer system. The M2000 card performed well but is more expensive for no real benefit.
There is one caveat here though. If you want to run four 4K screens at full 60Hz 4K resolution then the M2000 will do that, the W4100 nor the NVS 510 can’t support four screens at that resolution.
Scenario 5 – Great Performing Trading Computer That You Can Also Play Games On:
If you want to play games on your trading computer (I’m not judging!) then definitely go for a more higher end gaming card like the GTX 1060, or even look at a 1070 or 1090. They are unquestionably far superior at gaming than the rest of the tested cards and they will also match the top performance on all your trading software.
Also, just to be clear, when I say play games I’m talking about the latest 3D rendered games, if you simply want to play solitaire or Candy Crush then these GTX 1060 type cards are massively overkill, the GT 710 or 1030’s would easily run these types of lower graphics games.
Be aware that these higher end gaming cards are physically bigger cards, often require bigger power supplies, and have cooling fans which can add to your system noise level (sometimes substantially).
Hopefully that covers everything you’d ever want to know about how graphics cards affect your trading computers performance when it comes to running your trading software packages.
If you have further questions or just want to feedback on anything let me know in the comments below.