Following on from our previous graphics card for trading computers article we are back with a brand new for 2020 trading computer graphics card test.
This is a brand new test featuring four cards which have a wide range of different spec’s, capabilities, and pricing, from ultra-cheap to massively expensive.
But before we get into it all, let’s just have a brief recap on why these tests are important.
There’s a lot of different information out on the Internet about what you do or don’t need in your trading computer. One of the biggest areas of confusion are graphics cards.
The aim of these tests, and this website as a whole, is to show you the actual differences and impact of using one type of graphics card (or any other component) over another.
This is valuable information, dedicated graphics cards can cost as low as £30 right up to £1,200+, so picking the right card can save you a lot of money, just like going for the wrong one can be a costly mistake.
Let’s run through the cards and then a short section on the test system and how we have tested before diving straight into the results.
Graphics Cards on Test
For these tests we want to show the biggest range of cards possible without testing every card on the market. To achieve this we have gone with the lowest cost card we can, an nVidia GT 710 1GB which can be purchased for around £30.
On the other end of the scale we’ve gone straight to the nVidia RTX 2080 Ti 11GB, pretty much the most powerful gaming graphics card on the market at a cost of just under £1,200.
An nVidia GT 1030 2GB is also on test. This is a £70 card which we use a lot as it is essentially the most powerful graphics card that can be passively cooled (meaning it is silent). There are a few other passively cooled cards which pop up from time to time that may be faster, but the 1030 is widely available whereas others are perhaps not.
Finally, we have a professional class nVidia Quadro P620 2GB, this normally sells for around £200. Historically I’ve not been a fan of professional class cards however I am a fan of the P620 for one main reason, it can handle up to four 4K resolution monitors at full 60Hz, something many other cards simply can’t do. (No idea what a 4K resolution screen is? See our guide to monitors and resolutions for traders).
So that covers the cards, let’s now see how the tests are conducted.
Test System and Methodology
To make these tests as fair as possible we have use one computer and have then simply switched in and out the graphics cards with no other changes made.
The base system specification was an Intel i5 9600KF, with 16GB DDR4 2,666MHz RAM, a Gigabyte Z390 D motherboard, 240GB Kingston SSD, EVGA 600W power supply, Fractal Design Core 2300 case and an Arctic Freezer CPU cooler.
This represents a mid-priced computer setup, typical of what might be found in a decent trading computer system.
In terms of software we ran Windows 10 Home Edition build 1909, the latest version of Google Chrome, Handbrake v1.3.0, and then used Google’s older Octane browser test, WebXPRT v 2.93, and the latest JetStream 2 browser benchmarking tool.
All drivers for hardware components including graphics cards were the latest ones available from the manufacturers.
For our main computer testing software we used Passmark’s Performance Test suite, this runs individual tests on the major components in a PC and scores, with higher scores equalling better performance, very useful when comparing one component / setup to another.
All tests were run repeated three times and the scores given are the average of the three results.
2D and 3D Graphics Tests
As this is a graphics card test I think the best place to start would be with the 2D and 3D tests conducted by our Performance Test software.
2D graphics are basically lines, numbers, and fonts, standard interfaces for computers really. Crucially as well, pretty much all trading software uses 2D graphics to create their interfaces and charts. So that candlestick chart you watch religiously is basically a set of 2D shapes updated constantly with new market data.
3D graphics are mainly for games, to make them more lifelike, games use 3D shapes and models to draw things to your screen.
Imagine a simple green box, a 2D drawing of this would need four basic lines and then a block of colour.
A 3D representation would require considerably more information, the box now needs some depth which triples the amount of lines needed to be drawn, each face of the box now needs colouring as well which mean 6 times as much work to draw.
Throw in some textures, light sources, shadows, and reflections and the processing power required over our basic 2D box has jumped up substantially.
Going back to our trading software, no matter what your style of trading, it is important that data is conveyed to you as quickly as possible. For that reason alone it would be a bad idea for a trading platform to attempt to draw all of your charts in 3D, and for the main they don’t do this, they use quicker 2D graphics engines.
All that being said, lets look at charts for both the 2D and 3D performance of our four graphics cards:
So, two things instantly jump out to you here, the 2D performance difference across all the cards is relatively small, whereas for 3D the performance gap is massive.
The difference between the lowest rated 2D card, our £30 GT 710 and the top result on the £200 P620 is around 13.8%, nothing too big but a definite difference.
The £70 GT 1030 and the £1,200 RXT 2080 are within 0.1% of each other, and the Quadro P620 achieves a 6% lead over them both.
If we now turn our attention to the 3D results we can definitely see which of the cards is best for a gaming type workload. The RTX 2080 result is almost 2150% better than what the GT 710 can muster, it also comes in 414% faster than the GT 1030 and 245% faster than the Quadro P620.
I think it is also interesting to see the differences between the other cards, for a extra £40 the GT 1030 gives a 337% increase on the GT 710 performance, this is a significant jump for a relatively small extra cost.
The Quadro P620 scores 551% better than the GT 710 and over 48% better than the GT 1030, surprising as technically the GT 1030 is a gaming card and the Quadro is a professional class (non-gaming card).
These two charts are interesting and a good way to assess relative performance differences however what does this mean in the real world and how does this impact your trading software?
Web Browser Tests
To asses trading software performance levels we use web browser benchmark tests.
Why do we do this?
When you get down to the core of what trading software actually is, you end up with some client that pulls in a data feed from the Internet and then converts it into numbers and 2D graphics.
This is very similar to how a web browser works, when you access a website it pulls the data that makes up this site from the Internet and converts it into fonts, lines and graphics for you to see it.
A lot of trading platforms actually run in a web browser without a dedicated program, so browser tests like the ones we run are a highly reliable way to asses real world performance for trading type workloads.
If your trading software has a downloadable client then this will also act essentially in a similar fashion, so whilst it might not use exactly the same programming tools as a web browser to perform its tasks, the underlying processes will be similar.
We run three different browser tests to assess performance levels. The first is an older Google test which performs a series of tests similar to what web based applications would carry out.
The second test is a newer test called Jetstream 2, very similar to the Google test, it simply uses a wider range of mini-tests using new types of programming frameworks and methods.
The third test is a little different, it actually measures how fast a computer / browser can perform a series of more graphical tasks like resizing photos, encrypting data, and it even shows some stock charts which is useful for us.
This is Google’s original browser benchmark test detailed above.
As you can see there is less than 1% difference in the results across all four cards.
What does this mean? Let’s look at the Jetstream 2 tests before we jump to conclusions:
Here we see a very similar pattern, the spread between best and worst is just 1.3%
These two tests are showing us that there is hardly any difference to these kind of workloads no matter whether you run a £30 graphics card or a £1,200 one, and this makes sense really.
Whilst we are seeing the results of these tests displayed in the web browser, it is your computers CPU which is doing the work to produce the test results. The graphics side of the equation is just displaying numbers and fonts between some 2D shapes.
Our 2D graphics test up above already highlighted that there wasn’t too much difference between the performance levels for 2D work and these two browser tests are real world examples of what this actually means for your day to day trading use.
There’s basically no impact at all.
Our third browser test is the more graphical one, let’s see what happens here:
We are now seeing roughly a 10% difference between our £30 GT 710 card and the P620 / RTX 2080 cards, so finally there is an impact.
The GT 1030 scores better than the 710 but falls behind the two leaders.
If you remember from above, I mentioned that one of the sub-tests of the WebXPRT test is solely about displaying stock charts, would it be helpful to break this section out to examine it in more detail? I think it would, so here goes:
Here a lower score is better as this is the time in milliseconds to complete this task.
The gap narrows considerably, the difference between first and last is just 3.6% or 5 milliseconds…
So that extra £1,170 spend on the GTX 2080 Ti will buy you 5 milliseconds better performance when looking at your stock charts.
By many online accounts, anything less than 20 milliseconds is classed as imperceptible to humans, so I personally don’t think that £1,200 graphics card is worth the cost for traders.
Summary of Graphics Card Trading Workload Tests
Overall what have we learned?
We have seen that graphics cards can cost a lot of money, and this extra cost can lead to massive performance gains in a cards area of expertise. The RTX 2080 Ti costs 40 times more than a GT 710 and offers 2150% more performance for 3D graphics rendering.
But when we look at 2D performance, that extra spend becomes far harder to justify.
For workloads where the output is very simple numbers and 2D lines then there is practically no difference between any of the cards we tested.
If we narrow our focus on just looking at how fast stock charts can be drawn on to your screen then there is basically an imperceivably small difference.
Should everyone just go with the cheapest option then?
No, whilst this test highlights the difference between the various cards for trading workloads, we have to take into account that people sometimes do other things on their computers other than trading.
Watching a Bloomberg TV feed will require more work than drawing a stock chart for example. Likewise if you want to run high resolution screens then the GT 710 and GT 1030 cards can’t do this.
The point of these tests are to show that you should get a card that meets your needs.
We do have many customers that want a fast trading PC which will be used for nothing other than charting and execution. For this use case then it is far more beneficial to put any extra spend into the CPU rather than the graphics cards.
If you want to run 4K or QHD monitors then get a card which can do that, like the Quadro P620, but don’t worry about going for anything more powerful or expensive, there is no point and again, you’d end up with a faster computer putting any extra spend into the CPU.
(See our latest CPU benchmark tests)
Finally, do not make the mistake of going for a high end gaming graphics card unless you actually want to play games.
I get ‘told’ regularly how a computer with a GT 1080, or 1060, or any other gaming card you can think of, must be a better trading computer than one with a GT 1030 in it. This just is plain wrong.
Virtually any graphics card review you will read will focus on gaming performance, that’s a far bigger market than anything else in terms of graphics cards, but for our purposes it’s meaningless to look at 3D performance.
To finalise this review, and for completeness, I’ve included charts below for the four test graphics cards showing their impact on how other elements in the computer perform.
Spoiler Alert! – The graphics card has no impact on the rest of the PC’s performance. It impacts only 2D and 3D performance and any ‘combined’ performance differences are highlighted in the web browser tests above.
I hope this up to date trading graphics card test is useful.
As ever, any questions just leave a comment.
All the best!
Anand Chandavarkar says
This is the most informative and authentic technical support on trading pcs I have come across on net so far.
Thanks a lot for helping me save pretty big amount without loosing performance.
No problem at all Anand, thanks for the comment!
George Humphreys says
I’ve been thinking about either getting a new graphics card or an entirely new computer. My current build is: I7 6700 CPU , PNY 810 card (2 cores 4GB), 32BG ram. I find when running thinkorswim it is extremely slow in processing the data. At times there is a 1-3 sec delay between when a sale is registered to when it is displayed on the chart. I have interactive brokers too which is a bit better …… their data comes in 2-3 seconds faster than tos and it gets displayed faster on their charts but its still slow. According to your tests there is no value in upgrading my card since it only renders in 2D I don’t do any gaming. BUT the think or swim tech guys say that my card is the problem. They recommend 2 GB per screen. Looking at other peoples charts their card processes the sale data onto the chart instantly like it should but I don’t know their exact computer build but I’m sure they have a better card. It would be interesting to compare the performance of different cards using actual trading software.
Hi George, Thanks for the comment.
I’ve looked into this for you and it does seem linke Think or Swim is a platform which requires more graphical resources to run. This is something we will look into a bit further in a future post as it appears TOS behaves differently to virtually all other trading platforms in this regard. I would also say your NVS 810 card is not particularly powerful really so this does correlate with what the TOS tech support are saying. Do you need to actually drive 8 screens? If so that is going to affect what graphics setup you can move to.
George Humphreys says
Thanks for the reply Darren. I recently purchased a Geforce RTX 2070 super card to test it with tos using my existing pc build. It made a big difference in real time display performance. It’s still not truly real time and sometimes I have to close and open tos in order for it to make any difference. Since this worked better I just purchased another 2070 to run more than 4 monitors. Now I need a whole new computer to run those 2 big cards.
Drew Gravin says
Hey George and Darren,
I have struggled with TOS as well and am now looking to build out a PC that might help. George, when you said the Geoforce helped with “real time” performance, did you mean the order lag between when you press buy and when the order appears/executes in the software? This lag is my biggest issue with TOS, that and the hotkey “order confirmation” bug which I can only hope the new card has solved!
And Darren, these articles are awesome and just what I have been looking for! Would you agree that TOS and others would be best served with the highest single thread performance? I have now seen conflicting views that TOS is indeed still single thread. Would you agree that most trading software –even those like Lightspeed which take advantage of multi-thread CPU– are still best served with the highest single-thread CPU?
And one last question, I am looking at the Intel Core i5-10600K @ 4.10GHz (integrated Intel UHD Graphics 630) and the Intel Core i7-10700KF @ 3.80GHz (pre-order-no integrated). They both make the top 10 in highest single thread performance.
I am looking to drive 3 or 4 QHD monitors 24″ to 27″, as I don’t think 4K will display well at this size. Which card would make the most sense, the UHD 360 or maybe the Intel Core i7-10700KF with the Quadro P620? Could I get the Intel Core i5-10600K and then remove the integrated 630 and put in the Quadro?
…I would really like to stay in the top 10 highest single thread performance CPU.
Both TOS & IB’s TWS platform are java based programs which allow one to increase the amount of RAM they are allowed to use. Have you tried increasing the amount of RAM allowed for each program? You have more than enough RAM so feel free to allow IB up to 2GB (2048mb) and a similar number for TOS. There’s plenty of guides you can google to find out which setting to change in the program itself. They also benefit from being restarted after a few hours of use so restarting them a few times each day will help too (Java isn’t the best at memory management).
Thank you for putting all of these tests together. I don’t have access to multiple different CPUs & GPUs like you guys do, but I’ve also come to similar conclusions based on my own testing with other people’s computers. Get the best CPU single thread performance you can afford, make sure you have a decent CPU cooler setup to ensure it doesn’t throttle due to high temperatures, and if you aren’t going to game at all then any budget GPU that allows you to run as many displays as you need (clearly the “professional” GPUs do better in this regard too).
George Humphreys says
Yes I did try allocating the maximum ram to tos but it didn’t make any difference. Their tech people said that y bumping up the ram allocation it simply secures more of it for only tos. Ram was not bottlenecking the display performance so that’s why there was no difference in performance.
And yes you are correct closing and relaunching tos does help. Especially now after I got a new card.
So say you are running Bloomberg tv or some other tv stream, still ok with the basic integrated graphics, or lowest graphics for screen needs, or need to move up?
Streaming a live TV feed generally won’t cause a problem even with the most basic graphics cards. The only time you might want to think about this is if you have a larger number of feeds streaming simultaneously, we have some sports betting customers who have this specific requirement.
I am a total beginner in this tech world and I wish to warmly thank you for your clarity, simplicity and accuracy of reasoning. Very very helpful 🙂 Karine
I am day trader that using multiple charts and technical analysis. I plan on having 3 QHD 27inch monitor setup. Would the NVIDIA Rtx 2060 6gb card be sufficient enough? Or can I go lower? Thanks!
tobor yenom says
wow, great article thanks alot ! i got my graphics card ! now ;
Rilker Frota Cavalcante says
Hi, you did just a almost (Kudos) perfect description and tests for a trading system. I tested 3 different setups, and with ToS… No way. Doesn’t matter what you do, new releases always (and Java itself) change something in the final setup behavior, always decreased overall performance (I tested AMD cards, 2060/2070, 1080, 2K monitors). My solution was get 2 PCs, one for Watchlists, alarms and scanners, and the for order entry with DAS pro (now they have API for TD Ameritrade). And using TradingView for charting. I hope I had read your post before. Find a way to test the charting and order entry software “encapsulated” to gather some performance numbers woldbe great!
I know this is an older study but based on it I built a “budget” power house pc for $1480 taxes and shipping. Thank you for the write up.
Intel i9 10850k comet lake 10 core
Msi mpg z490 gaming edge wifi motherboard
Samsung 970 evo m2 2280 1tb ssd
Crucial ballistix 3600mhz ddr4 32gb ram
Noctua nh-d15s dual tower cpu cooler
Msi geforce gt1030 lp oc ddr5 pci express
Some fans all in a fractal design meshify case
what about 2021 quadro t400 : ddr6, Memory Clock 1250 MHz10 Gbps effective and better 2d ratings than p620
Thanks for the comment, the T400 and T600 cards are very new however you are correct in that they do look to be a stronger option than the P400 and P620 cards. We have some coming into the workshop shortly, when they arrive we will get them tested and will likely update these results with the additional cards.