AMD Ryzen 9 3950X Processor Review: Worth the Wait
Two Months After Launch, the 16-Core 3950X Still Amazes
What could possibly be left to say about AMD’s rise to the top of the desktop processor market? Just a few years ago the company’s future looked bleak, and it felt like they were light years behind Intel with CPU architecture. And then Zen happened. It was a monumental leap forward from where AMD had been with Bulldozer and its revisions, and it changed the desktop CPU landscape.
It has been nearly three years since we first reviewed a Ryzen CPU, and as disruptive as the first-gen Ryzen lineup was, and with the Zen+ Ryzen 2000 series offering even better performance in 2018, the 7nm Zen 2 architecture has finally given AMD a lead on desktop that still feels a bit surreal after months of covering these new CPUs.
Living Up to the Hype
Back when the Ryzen 9 3950X was announced at E3 (and I was sitting right there snapping photos like the one you see above) I struggled with the idea of a full 16-core bin (two fully-enabled 8-core chiplets) with boost clocks a full 100 MHz faster than a 12-core Ryzen 9 3900X. When it was delayed until November I assumed that yields were not going to be especially high. This was conjecture, of course, as was the thought that many of the 8-core chiplets needed for Ryzen 9 3950X might also find their way into AMD’s more profitable EPYC server processors.
When the Ryzen 9 3950X launched on November 25, 2019, supply was limited (to say the least). We weren’t able to get our hands on one with the initial round of review samples sent in early November, but as promised AMD did send us their flagship desktop CPU after all (along with a Threadripper 3960X – to be reviewed shortly). Sure we’re late to the party, but there was no way I was passing up the chance to go hands-on with this shiny new 16-core desktop monstrosity.
|AMD Ryzen 3000 Series Desktop Processors|
|Model||Cores / Threads||Base / Boost||Unlocked||GameCache||Memory||TDP||Price||Launch Date|
|Ryzen 9 3950X||16 / 32||3.5 Ghz / 4.7 GHz||Yes||72MB||DDR4 3200||105W||$749||11/25/19|
|Ryzen 9 3900X||12 / 24||3.8 Ghz / 4.6 GHz||Yes||70MB||DDR4 3200||105W||$499||7/7/2019|
|Ryzen 7 3800X||8 / 16||3.9 GHz / 4.5 GHz||Yes||36MB||DDR4 3200||105W||$399||7/7/2019|
|Ryzen 7 3700X||8 / 16||3.6 GHz / 4.4 GHz||Yes||36MB||DDR4 3200||65W||$329||7/7/2019|
|Ryzen 5 3600X||6 / 12||3.8 GHz / 4.4 GHz||Yes||36MB||DDR4 3200||95W||$249||7/7/2019|
|Ryzen 5 3600||6 / 12||3.6 GHz / 4.2 GHz||Yes||36MB||DDR4 3200||65W||$199||7/7/2019|
The results below won’t shock anyone, serving only to validate what we already know about the performance of AMD’s Zen 2 architecture. Clock speeds obviously play a role in the Ryzen 9 3950X position over the 3900X, as AMD was very aggressive in pursuing a higher max single-core frequency for the 16-core flagship.
From the charts to follow it’s clear that this higher 4.7 GHz boost clock produces better single-threaded numbers than the 3900X, and both of these processors were freshly benchmarked using AGESA 18.104.22.168 firmware.
|PC Perspective Test Platforms|
|Motherboard||AMD X570: MSI X570 GODLIKE
AMD X399: GIGABYTE AORUS X399 Gaming 7
Intel Z390: ASRock Z390 Phantom Gaming X
Intel X99: ASUS X99 Deluxe II
Intel X299: MSI Creator X299
|Memory||X570: Crucial Ballistix Elite 32GB (8GBx4) DDR4 @ 3200 MT/s
X399: Crucial Vengeance LPX DDR4 @ 2933 MT/s
Z390: 16GB (8GBx2) Crucial Ballistix Sport LT 32GB DDR4 @ 3200 MT/s
X99: 32GB (8GBx4) Crucial Ballistix Elite DDR4 @ 2133 MT/s (i7-5960X)
X299: 32GB (8GBx4) Crucial Ballistix Elite DDR4 @ 2933 MT/s (i9-10980XE), 2666 MT/s (i9-7980XE)
|GPU||NVIDIA GeForce RTX 2080 FE|
|Storage||Corsair Neutron Series XTi 480GB|
|Power Supply||CORSAIR RM1000x 1000W|
|Operating System||Windows 10 64-bit 1903|
|GPU Drivers||NVIDIA: 440.20|
This benchmark has been broken up into multiple charts to help illustrate the differences in the compression and decompression results, as the order of these CPUs changes considerably between the two workloads. I added an overall MIPS result as well, which is less useful but helps demonstrate some semblance of a “winner” here. The short version? CPU performance is so often workload specific.
As with 7-zip the results from the x264 (64-bit) benchmark have been broken up into multiple charts to help illustrate the differences between these CPUs and their first and second pass workloads. There is also a combined chart with total encode time at the end (which is the only important metric when you’re talking about a multi-pass encode anyhow).
Going into this review I wondered how the difference in both core count and clock speeds would affect gaming results between the Ryzen 9 processors. It’s not as simple a question as it might seem at first glance. Consider this: while the Ryzen 9 3950X offers four additional cores and a 100 MHz peak single-core clock speed advantage (4.70 GHz vs. 4.60 GHz), the Ryzen 9 3900X has a 300 MHz base clock speed advantage across its 12 cores (3.80 GHz vs. 3.50 GHz).
Within the context of a “desktop” computer chose to omit the HEDT CPUs from the gaming benchmarks, but afterwards I relented and added in Intel’s Core i9-10980XE. Why? This 18C/36T part is the only desktop CPU (high-end or otherwise) that Intel currently offers which can meet or exceed the 16-core mark set by the Ryzen 9 3950X, and I wanted to see if the higher core count made any difference in gaming benchmarks. (Spoiler alert: it didn’t.)
Far Cry 5 – 1920×1080
At 1080p Far Cry 5 shows virtually no difference between the two Ryzen 9 processors, with the Core i9-9900KS enjoying a large lead. The Core i9-10980XE finishes in last place among these CPUs, so it will be interesting to see if it continues to fall behind the less expensive processors in the remaining tests.
Far Cry 5 – 1280×720
Moving down to 720p there is a slight advantage for the Ryzen 9 3950X over the 3900X (1.6 FPS), with rest of the results in the same order as the 1080p test.
Metro Exodus – 1920×1080
Metro Exodus is a obviously a more GPU-bound test, with all results within a 7 FPS range. Once again the two Ryzen 9 CPUs offer virtually identical performance, while the i9-9900KS has a 5 FPS advantage. This time the Core i9-10980XE is much closer to the Ryzen 9 results.
Metro Exodus – 1280×720
At 720p Metro Exodus suddenly produces much more variance between the tested CPUs. The two Ryzen 9 CPUs still offer similar performance, though the 3900X’s advantage here might suggest an advantage for its base frequency edge vs. the 3950X. The Core i9-9900KS has a much larger lead at this lower resolution, while the Core i9-10980XE falls further behind with a result only a few FPS higher than we saw at 1080p.
World of Tanks enCore – 1920×1080
Even at 1080p “ultra” settings the World of Tanks enCore benchmark is no match for an RTX 2080 FE on these test platforms. While the Core i9-9900KS does offer a measurable advantage once again, the rest of the group finishes within 1.5 FPS of each other at this resolution.
World of Tanks enCore – 1280×720
At 720p – and down to the “medium” preset to really push these frame rates – the high average FPS numbers do highlight the potential differences between CPUs. Once again the Core i9-9900KS has a comfortable lead in this group. The Ryzen 9 3950X finishes ahead of the 3900X in this test, with the Core i9-10980XE moving the back of the pack once again.
Interpreting the Gaming Results
There are generally games that we consider more CPU-bound vs. GPU-bound, but after testing the 3900X against the 3950X I started to wonder if there might be a sub-category of this that would further differentiate CPU performance between these Ryzen 9 processors based on reliance on “bursty” single-core clocks or faster all-core speeds. Without going too far down this rabbit hole it’s interesting to consider the overall results, and how the 3900X occasionally had the higher average FPS (even after repeated testing to ensure this was not an outlier).
In the Intel side their fastest current desktop processor (as in, not an HEDT product), the Core i9-9900KS, provides the highest performance in the game benchmarks presented here. This CPU has the overall clock speed advantage with its 5.0 GHz Turbo Boost frequency on all cores, while the Ryzen 9 3950X tops out at 4.7 GHz – with all-core boost clocks lower than this, of course.
The Work Is Never Finished
The tests you just saw represent a very small group of CPUs, just one GPU, only three games (at two resolutions each, but still), and yet the results encompass nearly a full month of testing. At one in the process I made the decision to throw out all of the completed benchmark results and re-test after our X570 motherboard displayed some instability, and a different motherboard/BIOS combination with the same components proved this. It doubled the workload, but I have to trust the results before I’m willing to publish a review.
Re-testing has become the norm, as AGESA revisions alone have prompted us to totally re-benchmark all CPUs and GPUs more than once since last July (that happened again between the Core i9-10980XE review and the first of our Ryzen 9 3950X and new 3900X tests). Add security mitigations to the mix and it seems that this need to re-test will never end.
Here is where the binning process that produced a 16-core processor with higher single-core clocks than the 12-core bin reveals another advantage: power consumption. Yes, as others have discovered in their Ryzen 9 3950X testing, along with higher overall performance the 3950X also offers lower power draw.
Just in case I haven’t emphasized this enough already: this processor has an additional four cores and still offers lower power consumption. I tested maximum power draw from the wall (using an 80 Plus Gold rated PSU) with loads generated using the all-core test in Cinebench R20. Not a lightweight workload. And still the system with the Ryzen 9 3950X is pulling a total of 201W from the wall – a 13W drop from our Ryzen 9 3900X sample. Impressive.
Until just recently the availability of the Ryzen 9 3950X was so poor and pricing so high (even Newegg was charging $899 during brief in-stock moments) that I was recommending the Ryzen 9 3900X to anyone considering this new flagship. Technically the 3900X is still the better deal in a pure price/performance argument, though I’ve become even more impressed by the relatively small premium for the 3950X considering its higher clocks and lower power draw.
Consider the per-core price of all tested CPUs from this review. We could pontificate endlessly on pricing if we consider variable like post-launch discounts and the availability of used parts, but for this review I wanted to provide a clear picture of how these CPUs stacked up at their original launch prices – and with the exception of the older Intel HEDT parts this is still pretty accurate:
With a premium of just over $5 per core, the Ryzen 9 3950X is a far better value that I would have originally considered when it was announced, and it’s actually remarkable how much more AMD is giving customers with this 16-core part. Just as the Core i9-9900KS is a binned part, providing a “pre-overclocked” experience out of the box, the Ryzen 9 3950X offers a hyper-binned experience with not only slightly higher single-core clocks than the 12-core 3900X, but consistently lower power draw as well (!).
It still feels to me like AMD could have charged more for the Ryzen 9 processors this generation, but for the enthusiast market I’m glad they didn’t. The performance capabilities of the Ryzen 9 3950X were so great at even $749 that it’s far to say that Intel had no choice but to slash the price of their new Extreme Edition CPU to stay competitive. Would we have ever seen Intel go from $1979 with the i9-9980XE to $999 with the i9-10980XE without this part? No way. And AMD’s Ryzen Threadripper parts continue to push things on the high-end desktop side, with no immediate answer from Intel for the new 3rd generation Threadripper 3960X and 3970X.
Bottom line, for $5.23 more per core than the already impressive Ryzen 9 3900X you can buy a 16-core desktop processor with significantly higher multi-threaded performance and lower overall power draw. It’s a testament to AMD’s chiplet strategy – and the rapidly maturing 7nm process. Now that prices have seemingly stabilized the Ryzen 9 3950X is a tempting proposition for an ultra-high end desktop build, though the Ryzen 9 3900X almost feels like a steal in comparison at “only” $499.
From a pure CPU performance standpoint nothing on the desktop side can touch the Ryzen 9 3950X, which is a tremendous value even at its $749 MSRP considering how much compute power you’re getting in a single 16C/32T package. Even moving up to Intel’s flagship HEDT product with the 18C/36T Core i9-10980XE isn’t an assurance of higher performance in many CPU workloads, and it was the weaker choice in our gaming tests as well.
Intel’s gaming edge is alive and well, with the Core i9-9900KS the leader of every test we ran, and if you were targeting gaming alone – with no other workloads to consider – Intel is still the way to go. That has become a hard sell, however, as a slight reduction in gaming performance garners so much more total compute performance on the Ryzen side – particularly with the 12-core Ryzen 9 3900X priced within a few dollars of Intel’s i9-9900KS when they are both found at MSRP.
Intel may still have the edge in gaming, but if you’re going AMD – especially now that Threadripper starts at $1399 – the Ryzen 9 3900X and 3950X processors are unbeatable right now for workstation-class performance on a budget. It really is staggering just how good the Ryzen 9 3950X is for the money, even at $749.