AMD Ryzen 3700X & 3900X: Zen 2 Brings More than Just Cores
It’s been two years since AMD’s Ryzen product line was released to consumers and so far the products offered seem to be well received by consumers. AMD brought an increase of cores and threads from the HEDT lineup down to the mainstream consumer. Consumers noticed – not only due to AMD’s offerings but the fact that Intel added cores to both the Core i7-8700K and Core i9-9900K. It made the case for upgrading beyond older AMD or Intel processors as well, encouraging consumers to embrace DDR4 memory and even faster NVMe SSDs that run on an m.2 socket without needing the top of the product line. In short – it’s been a great market for consumers over the past two years, and the roadmap showed more to come with Zen 2.
Both at CES and ahead of E3 we were told that a new generation of Ryzen was coming. And since the event in Los Angeles, there have been many questions about the new generation of Ryzen. What will more cores bring to consumers? Do people really benefit from those? How about overclocking? And what, if any, change will this bring to the market beyond the new generation of Ryzen? Some of those will be best answered when the Ryzen 9 3950X arrives in September, but we can start looking at how the 8-core, 16-thread 3700X compares to its two older siblings, and through the new 12-core, 24-thread Ryzen 9 3900X we can start to understand what the higher core counts may mean for consumers.
Note: The Ryzen 7 3700X, Ryzen 9 3900X and other processors/components used in this review have been provided by others for review/evaluation purposes. A full list of those items can be found in the Testing Setup section.
AMD Ryzen 9 3900X & Ryzen 7 3700X Unboxing
This year, I have a reason beyond packaging to be happy for AMD. The previous two generations were covered while I was in Okinawa – and so it was extremely frustrating for me to see a market where pricing was extremely attractive in the US but absolutely horrible in Japan. There are reasons why this can happen that are not within AMD’s control. Tariffs both on importing and exporting, exchange rates, shipping costs can all play a factor in the pricing of any product. So it makes me extremely happy to check 3700X pricing in Japan and find the situation much better in 2019 versus the pricing here in the US.
I inquired about this to a UK buyer of the 3600 and it turns out pricing on Amazon.co.uk was in line with what I saw in Japan. I checked against the Intel Core i9-9900K to see if everything had changed, but the discrepancy I have noted for previous years still remains. This is indeed very good news for buyers in both the UK and Japan. Hopefully, that has brought just as much good news to those interested buyers elsewhere.
I still find it hard to beat the packaging offered initially with Ryzen – the wooden box still holds a special place in my room and heart. As with many things the marketing team has brought more sophistication to the now-older Ryzen brand, and I’m very impressed with the style of this year’s packaging. We did receive some other items at the same time as the Ryzen 7 3700X and Ryzen 9 3900X, some of which will be in another review coming soon. Other components will be a part of our exploration on what the 500-series chipset-based motherboards offer to those who adopt it.
The packaging for the 3900X initially caught my eye because of the same message across the different languages. Now knowing the pricing situation has at least changed in two significant markets makes me happier that this was the direction chosen. I hope they continue this on other versions in the lineup. And whether intentional or not, the foam insert that the 3900X was seated in makes a phenomenal stand for Ryzen CPUs while still keeping them in the protective clear plastic case. I certainly wouldn’t mind having a few more of those around for photo or filming purposes.
Two X570 motherboards (AORUS and ASRock), a PCIe 4.0 SSD (AORUS) and a new DDR4-3600 memory kit (G.Skill) were received but not used in this initial review. We have added photos of some of them below, and you will see them in use soon. We will be explaining later in this review why we have not used them in our initial benchmarks.
It took some time to gather all of the results, but what was most surprising was learning more about our new test environment. This has led to some delays in finishing up this review, but in general, the tests already conducted have helped identify new points of concern that will streamline our tests in the future. Ambient room temperatures were severely skewed high, causing us to believe some issues were occurring due to overheating. This resulted in identifying a motherboard feature that was intended to be disabled was not, causing several crashes and failures where they did not previously exist. Our thoughts on how we do simple overclocking for reviews was revisited. Issues that we had in testing the 9900K last year – issues that were identified but not confirmed – not only proved to be the case but also validated a rare decision to not post a review on a product because we lacked a way to properly test it.
With all of these lessons in mind, we made a few decisions early on about how testing would be conducted. All AMD AM4 samples were tested not on a 500-series motherboard, but instead on the previous generation. This allowed us to limit the variables that changing motherboards could introduce during the testing. After the initial data was gathered we performed a build test on one of the new motherboards – which showed no identifiable performance difference at stock.
As we have done with previous reviews, we will identify the component and how it was acquired. We will be listing the components by type this time, due to the addition of multiple components. Motherboard BIOS versions are also provided.
Test Bench/Case (All Self-Purchased)
- Lian Li PC-O11 Dynamic (TR1950X testing)
- Lian Li PC-T60 Test Bench (Black)
- Lian Li PC-T70 Test Bench (Black)
Power Supply (All Self-Purchased)
- GIGABYTE GA-Z170X-Gaming 7 – BIOS F22m – provided by GIGABYTE
- Z370 AORUS Ultra Gaming – BIOS F14 – provided by GIGABYTE
- ASUS ROG STRIX X299-E GAMING – BIOS 1704 – Self-purchased
- MSI X470 GAMING M7 AC – BIOS 1.94/1.9O – provided by AMD
Note: 1.9O was required for 3700X/3900X testing and only used for these. 1800X tested but failed on 1.1, 1.94 resolved the issue.
- MSI X399 GAMING PRO CARBON AC – BILS 1.B0 – provided by AMD
Processor (All provided by Intel/AMD)
- Intel i7-7700K
- Intel i7-8700K
- Intel i9-9900K
- Intel i9-7900X
- AMD Ryzen 7 1800X
- AMD Ryzen 7 2700X
- AMD Ryzen 7 3700X
- AMD Ryzen 9 3900X
- AMD Ryzen Threadripper 1950X
- Corsair Vengeance 2x8GB – 3200MHz, CAS 16 – Provided by AMD
- Apacer Blade – 4x4GB – 3000 MHz, CAS16 – Provided by Cybermedia on behalf of Apacer
- G.Skill Flare X 2x8GB – 3200 MHz, CAS14 – Provided by AMD
GPU (All Self-Purchased)
- Sapphire RX580 8GB
- EVGA GeForce GTX 1060 6GB
- HP Geforce RTX 2080 (believed to be from PNY, blower style)
M.2 NVMe Storage (All Self-Purchased)
- Samsung MZ-VLW512A (2 identical parts)
- ID-Cooling Chromaflow 240mm – Fans from same kit – Provided by ID-Cooling
- Deepcool Captain 240 EX – Fans from same kit – Provided by Deepcool,
- Enermax TR4 AIO Cooler (Fans not used) – Self-Purchased
- 9 x ID-Cooling 120mm RGB fans (used with Enermax) – Provided by ID-Cooling
Additional Components (Self-Purchased)
- MSI AC905C Wireless Network Card (Used with Z170/Z370 motherboards)
As we have in previous testing, our tests are conducted using a publicly available document. We set up and tested this on the first AMD processor, then tried to clone it for Intel testing. This did not provide reliable results so we instead wiped and re-created using the same process. A file is available in Google Drive to view more of the notes as well as a comparison of the 3 generations of AMD Ryzen 8-core, 16-thread processors.
- Operating System: Ubuntu 18.04 LTS
- NVIDIA Drivers – latest nvidia-### available in standard PPAs
- AMD Drivers – AMDGPU (open source version)
Overclock testing was not performed on the Threadripper 1950X due to inconsistent results, even when just setting on Core Performance Boost only. All other overclocks were done by multiplier only to all cores. Our determination of a stable overclock was only when all tests passed without a single failure or crash.
Benchmark Notes: Phoronix Test Suite’s CPU suite offers a plethora of tests and not all are included in this review. The full list of tests and results are available here, with the exception of our LineageOS build times. Those will be included later in the article. Color scheme for benchmarks continues to follow XDA’s traditional color scheme.
This is pretty similar to our previous findings except for the 2700X and 9900K when overclocked. It’s the only one that performed better at stock speeds than at overclock, and that’s odd given the results from this test increase based on the clock speed. The 9900K could have reached a thermal threshold, unlike the 2700X that usually has issues first in power consumption instead of heat.
GZip is a common compression method and so it makes sense to check out the performance here. We continue to see the gap decrease between AMD and Intel single thread performance, and AMD’s willing to cede some to Intel. Additional cores and the improved overclocks help. The 2700X results at stock are a bit head-scratching and suspected of being an outlier.
SciMark 2 (Java)
The SciMark 2 benchmark utilizes Java for arithmetic operations and then provides scoring based on those results. In three generations AMD has closed the gap in respect to the difference in performance and can close that even further when overclocked. The 9900K appears to hit another thermal limit, which is unfortunate given the boosts that the previous 2 generations provide when overclocked.
John The Ripper
On the cryptography front, John The Ripper offers similar results as before. More cores and higher clock speeds have definitely done well for both AMD and Intel, but AMD seems to have far more headroom to improve performance. Given the 3900X results, it will be very interesting to see how its bigger brother, the 3950X, will fare in these tests.
C-Ray demonstrates a similar result as in previous years. I have to wonder if there may be some optimizations in place to explain the performance increase over Intel. The increases seem especially noteworthy between the first and second generations of AMD Ryzen, whereas in other situations that is noticed more often between second and third generation processors. We do have more processors that will be tested and added to the lineup, so we hope that may help shed some further light on the jumps.
Benchmarks: Build Performance
Build Test: LLVM
We did not get results for ImageMagick build times on all CPUs. Instead, we will look at LLVM build times, that should offer XDA readers some relevant information. And this tells a very interesting story given the increase in performance comparing the 2nd and 3rd generations of Ryzen. It has closed the gap as far as build times with its Intel counterpart on stock speeds and takes the lead as clock speeds increase. Similar results were observed in other PTS tests where compile time was measured. In some of them, AMD took the top slot, in others, Intel did – but AMD is not trying to win them all.
Build Test: LineageOS lineage-16.0 marlin
LineageOS tests were conducted using lineageOS 16. Initial build attempts were performed using the Pixel 3, but all build attempts failed. We have gone back to the Pixel 2 XL (marlin) since these were building without issue.
Just as we saw with the LLVM build times, AMD has not only closed the gap but outperformed its Intel counterpart. But there is another data point that may prove to be extremely valuable to those building Android from source. Two years ago we looked at the high-end desktop (HEDT) lineup to see just how well more cores and threads improved build times. At the end of that we noticed that adding more cores did not always result in dramatic
The third generation makes a significant impact even on ccache performance, allowing it to keep within or below the comparative Intel processor. There is a staggering jump between the second and third generation of Ryzen, one that can be attributed to the processor itself given that it was the only variable between each AMD processor tested. It also does not consider PCIe 4.0 performance, which may decrease the times even further.
This is just the opening salvo of the third generation of AMD Ryzen. There will be more CPUs to test and evaluate, culminating in the release of the first mainstream processor with 16 cores and 32 threads. Intel normally also releases a new lineup in the Fall – we expect to have these to test as well. With that in mind, we will be holding on some of the “big picture” analysis until those arrive on the scene and are able to be included in our consideration.
As it stands, AMD is doing exactly what they said their strategy was from even the initial release of Ryzen. The goal is not to outperform Intel processors all of the time – but to offer a product that not only stays competitive to Intel and does so at a better price. For the third year in a row, they have done that. They’ve also challenged the status quo by increasing the cores and threads available to mainstream platforms – which has a significant difference in cost compared to HEDT systems. This is the first time we’ve seen Intel match that count and there is no guarantee that they could do it again.
Normally, it’s not good to repeat a message for three years or more. AMD’s case and message shows the exception to the rule. They have remained competitive and help bring better offerings to consumers at price levels that just about any general consumer would be interested in. That means that the good days for consumers are here to stay for the time being. I continue to laud this, and I do even more this time because we see that going beyond the borders of the USA and into other nations. For that larger group of consumers, it’s been long overdue – and so we welcome to the party of great processor options available to just about every price point. This was the norm for quite some time. It’s nice to see this being a norm once again.