The best CPU for trading?
A question we get asked at least once a week.
Our stock answer is that it depends on how you trade, which platforms do you use? how many different platforms do you run? how many charts do you open up?
Different traders have different workflows, some traders open up one chart and focus entirely on that, we have others that run 6 screens with up to 50 charts per screen. it’s a real mixed bag of requirements which leads to different solutions required on the PC / CPU side of things.
From experience of supporting traders and their various platforms over the last 13 years we have discovered some generalisations, the first being that all trading software performs better on a faster CPU.
I know what you are thinking: ‘duh!’
Yes that is a pretty obvious conclusion, but the nuance is how exactly do you measure the speed of a particular CPU? Is it the GHz rating? No. Is it the number of CPU cores? No. Is it the IPC (instructions per clock) No.
The only conclusive way to work out the speed of a CPU is to measure it directly with testing, to see how fast it can actually perform tasks.
Only then can we better understand the real differences between processors and then determine whether they are a good fit for your trading computer based on the type of trading software you are running.
What Are We Testing?
We have run and published processor tests previously, however due to a few factors we have not published any updated testing for over two years.
In that time we have had new processors from both Intel and AMD so this is a long overdue update.
Primarily we are going to look at the 10th generation Intel CPU’s versus the AMD Zen 3 processors, we have also re-run tests on the 9th gen Intel chips and a couple of AMD Zen 2 CPU’s to give a more complete picture of where everything stands, and to help you tie these results back to previous tests we have published.
A Quick Note On Test Results
We have used the latest versions of some of our testing tools, some of which give out a different range of numbers to previous tests we have published using the same tools.
For example previous JetStream browser test gave a score of around 257 for the Intel 9900K whereas the updated JetStream test now assigns a score of 156 for the same CPU, the chip has not gotten worse, the scoring method for this test has just been updated.
Due to this we have re-run all tests here on the latest version of the tools rather than relying on older tests, more work for us but better results for you.
Test System and Methodology
Where possible we have tried to keep all test systems identical, switching out just the CPU and motherboard if required.
All test systems feature 32GB DDR4 3,200MHz RAM, a 250GB Samsung 970 EVO NVMe M.2 SSD, an nVidia GT 1030 (2GB) Graphics card, 600w BeQuiet Power Supply, and BeQuiet CPU Air Cooling.
The 9th generation Intel CPU’s ran on a Gigabyte Z390 chipset motherboard, an AS Rock Z490 motherboard was used for the 10th gen Intel chips, and an MSI X570 motherboard handled both the AMD Zen 2 and 3 processors.
Windows 10 Home edition was the choice of operating system, with the latest version of Google Chrome running the Browser tests. PassMark’s Performance Test Suite was used for the CPU tests.
All tests were repeated 3 times and the average of the 3 results was reported as the final test score to avoid any randomly good or bad results which can sometimes occur in testing of this nature.
The CPU Test Results
I’m going to start off showing the test results for the CPU produced by the Passmark testing software.
This result is generated by running sub-tests on calculations involving things like prime numbers, physics calculations, sorting algorithms, integer math, encryption, compression, floating point math, and single thread performance.
Apart from the single thread test, all other tests are multi-threaded, i.e. the testing software will use as many threads as the computer has available to produce a result.
Due to this the result does favour processors with higher core and thread counts, does this correlate to how the CPUs perform in trading workloads though? Let’s find out.
Looking at these results it is pretty clear that the AMD chips dominate their Intel counterparts, the only Intel chip that manages to topple any AMD is the i9 10900KF.
The Zen 2 chips (3900X and 3950X) seem to be massively strong options, even when compared to the newer Zen 3 chips (5600X and 5800X), only losing out to the top end 5950X.
It would be easy to come to the conclusion that AMD is the best choice judged solely on this result.
Before you start hunting for a new AMD based trading PC though, let’s look at one of the sub test results in isolation, the single thread speed test.
Single Threaded Speed
The reason why we are now going to look at this metric is because the above test features a lot of multi-threaded sub tests which is great if you run multi-threaded software.
A lot of software (including trading software) is not massively multi-threaded though, sure some packages may be able to use 2, 4 or 6 threads but on the whole your software generally isn’t processing a big enough volume of data to really take advantage of multi-threaded working.
The big exception to this for traders would be back-testing which does require your CPU to shift through and process lots of data as quickly as it can, but let’s put that aside for the time being.
Let’s see if there are any differences between the above results and then the single thread speed.
Interesting. The overall chart looks a lot different now, the two AMD Zen 2 chips (3900X and 3950X) dropping from 2nd and 3rd place to the 2nd and 3rd places from the bottom.
The Intel i5 chip matches and slightly beats out both the 9th generation i7 and i9 CPU’s, that is a very strong result for the joint lowest cost CPU in the test.
The newer Zen 3 chips (5600X, 5800X, & 5950X) put in really strong results here as well proving what a leap forward AMD made moving from the Zen 2 to Zen 3 generation.
Summing up both tests it is clear the AMD Ryzen 5950X is the king of the hill in both single thread speed and overall multi-threaded performance. All the newer Zen 3 AMD CPUs perform brilliantly in single thread speeds and beat out the Intel chips in multi-threaded workloads as well.
The older AMD’s whilst strong in multi-threaded workloads drop off significantly in single thread tests which might mean they are not great trading CPUs.
For Intel their 10th gen chips offer solid single thread speed with decent progression from their previous generation chips, but there is no doubt they are currently now behind AMD in terms of overall CPU performance.
Let’s now see how this all relates to ‘real-world’ performance.
Web Browser Tests
A web browser test is a benchmarking test that is run in a web browser, we use Google Chrome for these tests.
Browser testing is a great proxy for many trading software packages as they operate in a very similar way, i.e. browsers are a locally installed (on your computer) package, which then accepts a data feed (via the Internet), and translates the incoming data into media you can interact with (charts, lists, data entry boxes, etc…).
A lot of trading platforms offer entry to their data via a web browser (Trading View, CMC, IG Index) which reinforces this assertion, and downloaded platforms like MT4, NinjaTrader, and TradeStation will use similar coding and programming languages as a modern web browser.
Our first browser test is the JetStream test which is made up of lots of sub tests which in turn execute different code segments measuring the time taken to complete, a total score is then given once all the tests have completed, the higher the score the better.
This correlates almost perfectly with our single thread CPU test above.
The Intel chips get stronger and better results as you move up through each level and generation of chip, exactly what the single thread test shows.
The AMD Zen 2 chips again drop right back, whereas the newer Zen 3 chips take the crown.
The one strange result is that the 5600X and the 5800X actually seem to beat out the 5950X with really strong results.
Now let’s turn our attention to another browser test, the Google Octane test.
This is similar to Jetstream but uses some different libraries and code segments, an older test but still definitely worth a look.
This mirrors the Jetstream browser test results, which you would kind of expect, but it also serves to back up the results we are seeing throughout this investigation.
Interestingly the 5950X claws back against the 5600X but still loses out to the 5800X showing that the Jetstream test was not a freak result.
Our final browser test is for the WebXprt suite, this test includes more graphical tasks such as photo and text manipulation as well as producing charts and graphs which is of particular interest to us attempting to simulate trader style workloads.
Despite the different test routines a similar order is established to the other browser tests.
The one glaring difference is the drop off of the Intel i5 10600KF, this is difficult to explain as the result was repeated 3 times (like all of our tests) and it consistently performed worse. This is a result I may revisit in the future.
I mentioned above that part of this test looks at the speed of drawing charts to the screen, as this is a sub-test which strongly matches what a trading computer would be expected to do well, I’ve gone through and pulled out these results for just this sub test into its own chart.
(Note that a lower result is better as this is a time in MS to render the results).
You can instantly see again the i5 10600KF had a bad result here, if we exclude that then the two next worst results were the other 10th generation Intels, the joint best results were the 9th gen Intel chips with the 5600X AMD tying.
Taking a step back though the difference between the best and worst result (excluding the strange i5 10600KF CPU) was just 14 milliseconds, which is apparently around a tenth of the time it takes to blink, so not a massive margin really…
That sums up the results of our testing but what can we interpret from the data?
If we assume that most traditional trading software operates in a way described in our browser tests, i.e. accessed through a browser or uses a locally installed client to process a data feed and display the results, then we can see that the single thread speed of a CPU is clearly the best way to assess how well a processor will handle these kinds of tasks.
Despite some strong multi-threaded test results, the AMD Zen 2 chips (3900X and 3950X) show us that having lots of CPU cores does not mean top performance for trading workloads.
For a trader running just one or two platforms or charting packages then the Intel i5 10600KF is a really strong option, especially when you take the price into account, it is also a chip we love over at Multiple Monitors for this very reason.
For someone looking to stick with Intel then the 10th gen i7 and i9 chips do push performance further.
Realistically though, for the price of an i9 you could bring in an AMD 5600X and almost get yourself a 5800X for not a lot more, both of which offer ever increasing performance jumps.
The AMD 5950X is obviously a beast in any type of workload, so for serious multi-taskers, back-testing workloads, or any trading wanting the absolute best that money can buy then this is where to look.
It is a £700 + CPU though which is a lot, especially when you take into account the required extra cooling systems needed to handle it, but then again, compare this with some of the Intel Xeon server class CPUs (which some people strangely still think are better options) that can cost well over £2.5k per chip then they are a pretty good deal!
Does anyone need a 5950X? We do have some customers over at Multiple Monitors that run a lot of intensive platforms, things like Bloomberg, and Eikon, alongside some customised packages which churn through a lot of data, so a 5950X is a great option for them.
If you’re running NinjaTrader, or TradeStation, then the trusty i5 10600KF or the i7 10700Kf would handle them pretty well really, if you are more of a power user though, i.e. lots of charts open, lots of indicators per chart then you may be better off looking towards the i9 or the AMD Zen 3 options.
Personally I can’t see any reason to go with the older AMD Zen 2 chips. I know they are still offered by some websites which make claims to their top performance levels but as you now know, there are much better options available at both lower and higher price points, so don’t waste your money.
As ever, I hope this helps?
Any questions just fire them into the comments below.