Before we get into the testing, I want to address the elephant in the room – who is this CPU launch for? AMD themselves have confirmed Zen3 is still coming later this year so why would anyone buy this soon to be old architecture? I believe AMD has released this chip for the many early adopters who are still running Zen1 CPUs with older motherboards like 300 series. This will be the last chip that is officially supported by those motherboards and providing you already have a full system, it may make more sense to get this CPU over its predecessor.
In the tests we will cover the standard synthetic benchmarking, throw in some creative workloads with rendering times and also gaming as this is where the boost clocks should shine the most. But first let me talk you through our test bench.
ASUS TUF Gaming B550m Plus motherboard KLEVV CRAS X RGB DDR4 RAM 2 sticks of 8GB @3600MHz ASUS ROG Strix 2080 Super KLEVV CRAS C700 RGB NVME SSD 480GB Fractal Design Celsius + S24 Dynamic AIO cooler
We’ve also done a few tests with Wraith Prism RGB cooler which is packaged with 3900X by default. This is to show how 3900XT will behave, but let me tell you something – there is a reason why it was gathering dust in the cupboard.
Heading into the settings – for all tests we have turned off performance enhancement, enabled DOCP memory profile and set fan and pump speed to full. Ambient temperature recorded at 27.5 °C. Purpose here is to see variance between the two CPU’s while maintaining consistency in our testing.
For overclocking, we have managed to get both CPUs at stable 4.3GHz but we had to use different voltages. The older 3900x was only stable at 1.3V while the new 3900XT managed to keep up with 1.25V.
In the first few tests you will see a lot of similar results but bear with me as there are a few very significant differences. Starting with the Cinebench R15 test – we see almost no performance change between these CPUs.
On the other hand, the Cinebench R20 single core test is much longer and we do see a single core score increase but it is barely noticeable.
Looking at the multicore test, we see again basically the same results.
Let’s follow up on the comment I made earlier about temperatures between AIO and stock coolers – AIO tops out at 77 °C while Stock cooler hits 92 °C in addition to being considerably louder and this is just at stock settings. We will remove this set-up from the tests from here onwards.
Talking about temps, let’s dive a little bit deeper to see what is happening in the background of these results. In this graph we are looking at both 3900XT and 3900X using AIO Cooler at stock running R20 single core test, we see 3900XT actually hitting higher temperatures throughout the whole test (on average 4.3 °C higher). Based on lack of improvement in score, this seems more like a heat tax, doesn’t it?
Here I have isolated the fastest cores on each of the chips and zoomed in – you can see that 3900XT does constantly go faster than the 3900X but also has more speed drops. I wonder if this can be improved in future optimisation.
When it comes to multi core tests that difference in temperature decreases to just 3.4 °C, but 3900XT is still running hotter.
Moving on to 7Zip benchmark, we yet again are greeted with uninspiring results. There is a slight variance between the scores, especially in decompression at stock settings as 3900X is lagging behind by 3% but it is not that significant.
Next up Handbrake test and 3900XT seems to be in the lead at first glance, but it is actually being matched by the older chip.
Then we ran a few different Blender tests starting with a short BMW render and followed by a longer Classroom render and saw yet again the similar results. When it is run on GPU using 2080 Super – both CPUs perform identical.
When we check the temperatures while running stock it shows that 3900XT is yet again running on average 5 °C hotter.
When we check the onboard package power it is basically identical as it is hitting limits on both CPUs.
At this point we do the same test and check temperatures in the overclocked state and find something quite unexpected. Here temperatures should be lower as our voltage is 1.25v instead of 1.3v but we still have slightly higher temps across the test.
Then we look into the reported power consumption and here it starts to make more sense. The 3900XT is running on lower voltage and thus using up less power but still maintaining the same performance as 3900X – it does that at a cost of running hotter though.
To verify this, we ran Prime95 for 10minutes on all of the same settings and found at stock XT variant boosts slightly better with higher temps and when overclocked – it stays a few degrees cooler while maintaining the same 4.3GHz overclock.
Lastly we jump into gaming to see what are the differences here. Let me quickly cover our testing methodology. We are using lower graphical settings than normal. This is to remove the GPU bottleneck in games – overwise results would be skewed. This is simply the best way to keep all the variables stable and concentrate specifically on the CPU benchmarking.
Now that is out of the way let’s jump into first game which is Total War Three Kingdoms and here if we focus on average FPS the overclocked 3900X is actually taking a 3% lead.
When benchmarking CS:GO the 3900XT wins by just over 3% on average score, and just under 2% in 1 percentiles. This is pretty expectant as CS:GO scales really well with higher performance.
Next we have Formula 1 and here we have overclocked 3900XT leading the average FPS by 2% but when we flip it to 1 percentiles we see 3900X stock keeping those framerates about 1.6% higher and surprisingly 3900XT stock is way down at the bottom.
Shadow of the Tomb Raider comes in next and we see 3900XT taking top 2 spots again, with only a few percent distribution from best to worst.
Lastly we have Doom Eternal. This test gives us a tie in average FPS performance between overclocked 3900X and 3900XT.
But when it comes to stock, the new CPU is 11% lower on average FPS and 7% lower on 1 percentiles. When re-running the test, I had the same results and I could not figure out why it shows such a big difference here. We will investigate this further in a future.
With all the game tests complete we find basically the same results as we did in the synthetic tests previously, so does the same apply for the temperatures and power efficiency?
Upon further review we found that while at stock settings the 3900XT is between 4 and 5 degrees higher than the 3900x while keeping all the other values basically on par. In the overclocked setup 3900XT temperature is only 2-3 °C higher but it is consuming on average 5 to 8W less power.
So what is 3900XT? Ultimately this chip is the better binned 3900X but that’s exactly what AMD said it is – simply put, the improvement in their manufacturing process has produced a slightly more efficient chip and AMD has changed the boost limit from 4.6GHz to 4.7GHz and in some cases it does deliver a slight improvement. Also it certainly has a better overclocking potential. Be aware that it is only a sample of one on each of the chips tested – it is also possible that other CPUs will have a bigger performance delta.
My ultimate advice here – if you are building a new system and you are not in a rush, then try to wait it out until Zen3 comes out and decide then if you need the latest CPU or snag a great deal as shops will likely be trying to sell off their old stock.
If on the other hand you are upgrading or building a new system and cannot wait, then for ultimate performance 3900XT is the better CPU.
However if AMD does not drop the price straight away – then a few percent improvement is not worth the 90 dollar premium. Get 3900X and put the money towards upgrading your other components.
Affiliate disclosure: as an Amazon Associate, we may earn commissions from qualifying purchases from Amazon.
Comments