AMD X870E Motherboard Mash-Up: Three Socket AM5 Mainboards Reviewed
Of course, it wouldn't be a HotHardware review without some hard benchmark data. Unfortunately, we're not extreme overclockers with easy access to liquid nitrogen, so trying for benchmark records is out of the question.
For all of our testing with these boards, we had Precision Boost Overdrive enabled with a 10X scalar, but no clock boost configured, so the CPU had power-limit headroom to boost but was still limited to 5.65 GHz. For comparison's sake, we also ran an older and cheaper X670 motherboard with the same CPU and memory at fully-stock settings through our benchmark suite; you'll see it listed as "AMD X670" in our charts.
Given that we were essentially thermally limited (and that wasn't going to change between motherboards), we instead focused on memory overclocking. We had two fast 48GB DDR5 memory kits available. One was a Patriot Viper Xtreme 5 kit spec'd for 7600 MT/s operation, while the other was a brand-new Kingston Fury kit of CUDIMMs spec'd for 8400 MT/s. However, using CUDIMMs on AMD platforms isn't necessarily a good idea.
The first time you boot either the MSI Carbon or ASUS Hero boards with CUDIMMs installed, you'll get this lovely message popping up. It was the exact same message on either board, so we suspect this originates in the AMD AGESA firmware. Selecting "Yes" reboots the system with the memory in "Bypass Mode", effectively operating the CUDIMMs like regular old DDR5 UDIMMs. However, on that first boot, you'll be stuck at 3200 MT/s—not 3200 MHz, which wouldn't be a 6400 MT/s Double Data Rate, but in fact 1600 MHz DDR.
Even in Bypass Mode, we were able to hit 8200 MT/s with that kit on the Carbon motherboard, and 8400 MT/s on the ASUS machine. However, neither of these configurations proved to be optimal. Using hand-tuned timings at 6400 MT/s ultimately ended up being the best option on each board. The reason for this is because at 6400 MT/s, we can run the memory controller at the same clock rate, 3200 MHz, as the memory itself. This results in significantly reduced memory latency, and it seems like the XMP timings on the Kingston Fury kit we have aren't tight enough to overcome this penalty.
The reason you don't get this message on the ASRock Nova board is because, for whatever reason, it will not complete POST with CUDIMMs installed. We tried several times, reseating the RAM and CPU multiple times, and it just wouldn't work. Because of this our results from the Nova board are using the Patriot kit instead, but as you'll see in a moment, it didn't exactly hold that board back, and in fact was able to run the same tight 6400 MT/s timings as the other kit.
X870E Motherboard Cinebench 2024 Benchmarks
Our first benchmark is the ever-popular Cinebench, in its latest 2024 variation. Cinebench is not a synthetic benchmark, contrary to popular belief; it's a test of a system's suitability for use with Maxon Cinema4D, a professional 3D rendering and animation package. We like it because it's easy to run and provides relatively consistent results—maybe a bit too consistent.
Cinebench is mostly concerned about CPU architecture and clock rate, and anything else is basically secondary. While the configurations with tighter memory latencies do see slightly improved results, the real story here is the JEDEC configuration, which drops off hard in multi-core performance thanks to the lack of PBO, meaning the relatively strict 120W power limit is enforced for the 12-core Zen 5 CPU. If you do any kind of serious multi-core work on your Ryzen PC, enabling PBO is a quick way to dramatically boost your performance with just a few clicks in your UEFI setup or Ryzen Master.
X870E Motherboard Geekbench 6 Benchmarks
Next up we have the similarly-popular Geekbench 6. Geekbench is mostly popular because it is both very easy to run and also cross-platform, so you can compare results from a desktop PC, to a smartphone, and then to a server system, and in theory, the results should be broadly comparable. In practice, well, some people have their doubts, but in any case, let's see what happened.
Geekbench likes low memory latency, as the majority of its tests take place in short bursts. However, the multi-core tests are also thirsty for memory bandwidth, particularly on a wide chip like our 12-core Ryzen 9 9900X. So it goes that the mid-range between our latency-tuned 6400 setups and our high-bandwidth XMP setups scores the best in multi-core—but the best single-core score once again goes to the ASRock PG Nova board.
X870E Motherboard Y-Cruncher Benchmarks
Arguably the most "purely synthetic" benchmark in our suite, y-cruncher is an application that uses tightly-optimized code to calculate many digits of pi as fast as possible. On modern hardware, that turns out to be very, very fast.
For our benchmark, we're calculating one billion digits of pi, and our best result finishes in a hair over twelve and a half seconds. All three boards with UCLK=MCLK configurations are neck-and-neck here, with the DDR5-7600 config not far behind. However, both CUDIMM XMP configurations fall away, with the X870E Hero's DDR5-8400 setup showing its first hints of some real problems as it actually turns in an even worse result than the 5600 MT/s JEDEC configuration.
X870E Motherboard FFmpeg VP9 Benchmarks
Video encoding is a demanding workload that many people still do on CPUs because software encoding allows you to get the best possible compression ratio, cramming the highest video quality possible into your chosen bit rate. We encoded a 30-second 4K-at-60-FPS video to the VP9 codec using the libvpx-vp9 encoder and row-based multi-threading.
Encoding VP9 is not fast, and it's primarily a CPU compute task, so our memory overclocking only had a minor effect. Once again, we see the tuned machines come out ahead, while the XMP configurations lag behind somewhat. Let this be a lesson, kids -- your XMP profile may not be the best settings for your memory—but they're probably better than stock.
X870E Motherboard 7-Zip Benchmarks
Instead of lossy video compression, how about some lossless file compression? We used the built-in 7-Zip benchmark to test each of the motherboards in their two memory configurations and got some pretty interesting results. These numbers are the final results that 7-Zip spits out, combining compression and decompression performance into one result each for single- and multi-core.
The 7-Zip benchmark is pretty variable, but that's why the utility performs ten runs of the benchmark and averages them for you. Overall, we see a pretty clear preference for low latency here, and the MSI Carbon takes home its first win in the multi-core score, although the ASRock board once again wins the single-threaded crown. Of course, all of these benchmarks are within a few percentage points, save, again, for the JEDEC setup.
X870E Motherboard Final Fantasy XIV Dawntrail Benchmarks
Naturally, we had to do some gaming benchmarks, too. Surprisingly, only the ASRock motherboard has "Gaming" in the title (as part of "Phantom Gaming"), but that's arguably the greatest concern of a majority of DIY builders. We tested Final Fantasy XIV in 1080p on the "Maximum" preset, as we wanted to avoid a GPU bottleneck but make sure to load our CPU with visual effects.
There are basically two stories here. The first is that FFXIV really likes low memory latency; while the average FPS doesn't change that much, the minimum sure does. The second is that there is something wrong with the 8400 MT/s configuration on the ROG X870E Hero board. It was stable, and able to past TestMem5, but it suffered from strange stuttering in both this game and Cyberpunk 2077.
X870E Motherboard Cyberpunk 2077 Benchmarks
When you think of "CPU-heavy games," Cyberpunk 2077 is an easy pick. This game will readily use available CPU cores to perform texture streaming, NPC AI processing, and other background tasks. It just so happens that it's pretty convenient to benchmark, too. The Cyberpunk 2077 benchmark isn't completely deterministic, but that's why we we do multiple runs of each config and average them.
Outside of the relatively poor 8400 MT/s result, all of these benchmarks differ only by tiny fractions that you will never notice. There's nothing wrong with our X870E Hero board, though; it actually gave us the highest minimum frame rate in this test when configured for low-latency operation. The JEDEC result is remarkably good, pointing to a GPU bottleneck in this particular test, even with frame generation enabled.
X870E Motherboard RPCS3 Emulator Benchmarks
Finally, we wanted to test the RPCS3 PlayStation 3 emulator because it is known to be very sensitive to memory latency. We went down to the local GameXChange and picked up a cheap copy of the original PS3 release, as the GOTY edition is known to run slightly worse on PS3. Using the settings listed on the game's RPCS3 wiki page, we checked performance in the starting area, just as John gets off the train in Armadillo.
No real surprises here, if you've been following along so far. RPCS3 wants tight memory timings, and it turns out that running your memory at 6400 1:1 is better for gaming than trying to brute-force a higher clock rate, at least on Ryzen. We're sure if we put some time into hand-tuning those DDR5-8200 results we could achieve something superior, but for now, it's gotta be 6400 MT/s.
X870E Motherboard VRM Temperature Benchmarks
This last one isn't really a benchmark, per se, but we wanted to see what VRM temperatures were like on these boards when the CPU was under a brutal load. With the power limit unlocked as it was, our CPU was easily pulling over 250W in Prime95, and we were bumping up against thermal limits in this "power virus" stress test. What about the CPU VRMs, though?
As you can see, they were cool as a cucumber. The ASRock board stands out from the others because it actually includes a small virtually-silent fan behind the I/O shield. Active cooling helps keep the VRM barely above room temperature, but the other boards are not even halfway to their thermal limits. VRM hardware typically tops out around 125°C, warmer than CPUs and GPUs can handle safely. At 59.2°C in the worst case, VRM cooling is not a concern on these boards.
