The AMD 3rd Gen Ryzen Deep Dive Review: 3700X and 3900X Raising The Bar
by Andrei Frumusanu & Gavin Bonshor on July 7, 2019 9:00 AM ESTSection by Gavin Bonshor
X570 Motherboards: PCIe 4.0 For Everybody
One of the biggest additions to AMD's AM4 socket is the introduction of the PCIe 4.0 interface. The new generation of X570 motherboards marks the first consumer motherboard chipset to feature PCIe 4.0 natively, which looks to offer users looking for even faster storage, and potentially better bandwidth for next-generation graphics cards over previous iterations of the current GPU architecture. We know that the Zen 2 processors have implemented the new TSMC 7nm manufacturing process with double the L3 cache compared with Zen 1. This new centrally focused IO chiplet is there regardless of the core count and uses the Infinity Fabric interconnect; the AMD X570 chipset uses four PCIe 4.0 lanes to uplink and downlink to the CPU IO die.
Looking at a direct comparison between AMD's AM4 X series chipsets, the X570 chipset adds PCIe 4.0 lanes over the previous X470 and X370's reliance on PCIe 3.0. A big plus point to the new X570 chipset is more support for USB 3.1 Gen2 with AMD allowing motherboard manufacturers to play with 12 flexible PCIe 4.0 lanes and implement features how they wish. This includes 8 x PCIe 4.0 lanes, with two blocks of PCIe 4.0 x4 to play with which vendors can add SATA, PCIe 4.0 x1 slots, and even support for 3 x PCIe 4.0 NVMe M.2 slots.
| AMD X570, X470 and X370 Chipset Comparison | |||
| Feature | X570 | X470 | X370 |
| PCIe Interface (to peripherals) | 4.0 | 2.0 | 2.0 |
| Max PCH PCIe Lanes | 24 | 24 | 24 |
| USB 3.1 Gen2 | 8 | 2 | 2 |
| Max USB 3.1 (Gen2/Gen1) | 8/4 | 2/6 | 2/6 |
| DDR4 Support | 3200 | 2933 | 2667 |
| Max SATA Ports | 8 | 8 | 8 |
| PCIe GPU Config | x16 x8/x8 x8/x8/x8* |
x16 x8/x8 x8/x8/x4 |
x16 x8/x8 x8/x8/x4 |
| Memory Channels (Dual) | 2/2 | 2/2 | 2/2 |
| Integrated 802.11ac WiFi MAC | N | N | N |
| Chipset TDP | 11W | 4.8W | 6.8W |
| Overclocking Support | Y | Y | Y |
| XFR2/PB2 Support | Y | Y | N |
One of the biggest changes in the chipset is within its architecture. The X570 chipset is the first Ryzen chipset to be manufactured and designed in-house by AMD, with some helping ASMedia IP blocks, whereas previously with the X470 and X370 chipsets, ASMedia directly developed and produced it using a 55nm process. While going from X370 at 6.8 W TDP at maximum load, X470 was improved upon in terms of power consumption to a lower TDP of 4.8 W. For X570, this has increased massively to an 11 W TDP which causes most vendors to now require small active cooling of the new chip.
Another major change due to the increased power consumption of the X570 chipset when compared to X470 and X370 is the cooling required. All but one of the launched product stack features an actively cooled chipset heatsink which is needed due to the increased power draw when using PCIe 4.0 due to the more complex implementation requirements over PCIe 3.0. While it is expected AMD will work on improving the TDP on future generations when using PCIe 4.0, it's forced manufacturers to implement more premium and more effective ways of keeping componentry on X570 cooler.
This also stretches to the power delivery, as AMD announced that a 16-core desktop Ryzen 3950X processor is set to launch later on in the year, meaning motherboard manufacturers needed to implement the new power deliveries on the new X570 boards with requirements of the high-end chip in mind, with better heatsinks capable of keeping the 105 W TDP processors efficient.
Memory support has also been improved with a seemingly better IMC on the Ryzen 3000 line-up when compared against the Ryzen 2000 and 1000 series of processors. Some motherboard vendors are advertising speeds of up to DDR4-4400 which until X570, was unheard of. X570 also marks a jump up to DDR4-3200 up from DDR4-2933 on X470, and DDR4-2667 on X370. As we investigated in our Ryzen 7 Memory Scaling piece back in 2017, we found out that the Infinity Fabric Interconnect scales well with frequency, and it is something that we will be analyzing once we get the launch of X570 out of the way, and potentially allow motherboard vendors to work on their infant firmware for AMD's new 7nm silicon.
Facebook
Twitter
RSS
447 Comments
View All Comments
Yaldabaoth - Sunday, July 7, 2019 - link
Thank you so much!What more needs to be said about the 3900X relative to its peers? So glad I didn't spring for the i9-9900K a few months back!
The 3700K is in a more interesting position at its price, but it looks like there are still some conditional performance advantages for the i7-9700K (read: single threaded workloads), but many times not. In any case, it's a great value and product. Kudos to AMD.
What I _really_ want to see is the 3600X (and the 3600) vs the i5-9600K. Historically, AMD has been seen as a budget play, but I suspect the single-threaded performance of the Intel CPU will shore it up vs the higher threads of these AMD chips in many user-relevant workloads vs. benchmarks. (Heck, it probably is just as good as the 3900X in gaming due to GPUs not catching up to increases in effects and resolution, right?) There could be very interesting values at these price points. Could you imagine 3 years ago telling someone, "For a good budget gaming PC, the Intel chips are alright, but if you want something REALLY nice once your GPU budget is used, put another $80 toward an AMD CPU!" Amazing.
BloodyBunnySlippers - Sunday, July 7, 2019 - link
I too am interested in the 3600x. I don't know if my math is relevant but the TDP per core count is higher for the 3600x than for the 3700x AND 3800x. Isn't it possible that it would then have a higher OC and better gaming performance? I mostly game, but moving from 4 threads on my current system to 12 threads isn't going to hurt either.allenb - Sunday, July 7, 2019 - link
This is a great result for AMD. They’ve won some benchmarks, conceded some others to Intel, and embarrassed themselves nowhere. A few thoughts:1. Was worried about the memory latency graphs at the start, but it was a relief to see how effectively those regressions have been mitigated in actual workloads.
2. In the 3D particle movement with AVX test, the 7920X jumps out to a *huge* lead. While everything else roughly doubles, it shows 9x speed up. Granted it is the only AVX512-capable cpu but it shouldn’t be *that* good unless the revised code is able to brilliantly utilize some of the new -512 features beyond just doubling vector length.
3. I work in an industry that’s heavily focused on performance and relatively cost-insensitive. That’s traditionally meant Intel but suddenly we are looking very hard at AMD on the server side.
4. Can’t wait to see what the 3950X can do!
Kevin G - Sunday, July 7, 2019 - link
1) Yes, AMD seemingly knew what they were doing and addressed that concern in the end product. I'd be curious what impact reducing the L3 capacity would have on benchmarks to valid some of the theories (or better yet, AMD release a larger L3 cache chiplet).2) There are a couple of other instructions with AVX-512 that can make vector code more efficient beyond just doubling the vector width for a linear increase in performance. Also worth noting that leveraging AVX-512 comes at a clock speed hit so the gains per clock is even higher.
3) While this is consumer side, the 12 core results are the most important to look like for Rome performance projection. That is the one that shares the IO die. (OTOH, a single CPU chiplet Rome that is allowed to clock to the moon with a 512 bit memory interface and 130 PCIe 4.0 lanes is an interesting single threaded proposal.)
4) Agreed but I would predict a very small down tick in IPC. The 3900 has all the cache enabled with fewer cores competing for contention. Everything else looks to be upward.
just4U - Sunday, July 7, 2019 - link
It's been a long time since I actually was chomping at the bit to see a review.. A Big day for Amd.guachi - Sunday, July 7, 2019 - link
That power efficiency!!!! Oh, my!As to gaming: Taking only the 1080 results the 3700X and 3900X are 98% as fast as the 9700K/9900K.
I'll take 98% as fast for all the other benefits. The multi-thread performance is spectacular.
RSAUser - Monday, July 8, 2019 - link
That's 2% lead without exploit patches, this is why I am annoyed at Anandtech not having applied them.If it was a 10% difference, sure, then you would definitely know which is better, but at 2% the patch could mean the AMD CPU would win in their benchmark.
A5 - Monday, July 8, 2019 - link
2% is MOE either way. It's a tie.RSAUser - Thursday, July 11, 2019 - link
It's a tie now without the patch, zombieload patches seem to show cases of up to 40% I/O hits, so could influence a lot of benchmarks/games.Meteor2 - Monday, July 15, 2019 - link
Yes, testing without the May security update is inexplicable.