tl;dr: See "Facts" section towards the bottom.
For almost 4 years, I had an AMD FX-8350 CPU on a GIGABYTE 970A-UD3 motherboard as part of a general-purpose computer with no compute performance requirements.
In 2017, I repurposed this computer so that I now require at least modest single-threaded performance, with decent multi-core performance being a welcome optional feature.
The AMD FX-8350 should still have been adequate, but to my surprise, trying to use the full potential of the chip resulted in about one-third of the expected performance!
Symptoms (AMD FX-8350 & GIGABYTE 970A-UD3)
Despite Windows and Linux showing 100% CPU usage, any CPU benchmark I tried showed 10% to 33% of the score value I expected based on other people's benchmarks of the same CPU.
For example, on Cinebench R15, other people get a multi-core score of around 650 cb. My score was 103 cb at 100% usage across all cores clocked at 4400MHz, drawing about 40W at 36°C.
I tried adjusting various BIOS settings, using stock settings, applying different voltages, overclocking, underclocking, replacing power cables and reapplying thermal grease.
The best I got was almost correct CPU performance at 576 cb just out of the blue before performance tanked again without warning. I recorded this screenshot when I was trying to decompress a .zip file and noticed that it was running at a dismal 5KB/s:
Finally, I took a shot in the dark and guessed that the motherboard was faulty. (After all, I did have other issues with it before.) I went out and bought an ASUS M5A97 LE R2.0 motherboard.
This new motherboard seemed to be adequate. The CPU was barely able to keep up with the desired compute tasks, so I thought things were all right.
Symptoms (AMD FX-8350 & ASUS M5A97 LE R2.0)
Here was where I felt that the issue was with the CPU rather than with the motherboard.
Running multi-core stress tests, I found that the CPU ran at 100% for about 17 seconds before dropping to 34% for about 50 seconds. Then it would jump back to 100% and repeat the cycle. The cycles did not run for consistent amounts of time.
It was as if the motherboard were trying to cope with a deficiency in the CPU.
So I thought to try replacing the CPU. Why not upgrade as well, so that I'm not borderline keeping up with the workload?
I picked up an AMD Ryzen 5 1500X CPU. Since it's a new architecture, I also needed a new motherboard, so I got a MSI X370 SLI PLUS motherboard. To resolve the other incompatibilities, I had to buy new cooling components and RAM.
This is practically a new computer now! Surely my AMD Piledriver problems couldn't follow me into this new AMD Zen build, right?
Well…
Symptoms (AMD Ryzen 5 1500X & MSI X370 SLI PLUS)
Things were going quite well until a friend advised me that the RAM I had was suboptimal. Supposedly, this new microarchitecture depends on faster RAM to perform at its intended potential.
Okay, no problem. I'll just get faster RAM with shorter timings.
After replacing the RAM and configuring the BIOS to use the advertised frequency and timings, I found that the motherboard beeped some fault code and wouldn't boot the system.
More troubleshooting ensued, and I was able to apply BIOS version 7A33v34 (factory version: 3.10), which had a changelog reading
Improved memory compatibility.
Sounds like just what I needed.
Even though the BIOS update exposed more features, the RAM was still not stable at its advertised speeds, so I tried running it at a more failsafe speed.
This was when the CPU performance issues came back. Running a multi-core stress test, the CPU frequency was capped at 1.35GHz (ignoring the BIOS configured 3.90GHz), and Windows refused to use more than 35% of each logical core:
I figure Windows is probably measuring core usage based on the highest frequency allowed by BIOS, since this is what Core Temp shows at max load:
UserBenchmark indicates that the CPU is performing in the 1st percentile, which is by far among the lowest among all other users' tests. Cinebench R15 shows the multi-core score is 258 cb and the single-core score is 55 cb, but this same machine was capable of 828 cb multi-core and 153 cb single-core just a day ago.
Giving up on this new RAM, I went back to the original RAM, but the same symptoms persisted.
Toggling AMD Cool'n'Quiet, ramping up fan speeds, changing voltages, resetting BIOS to default settings, and trying other CPU power settings haven't helped.
I will continue to try to troubleshoot this, including doing a BIOS downgrade (can't do at the time of writing because the MSI support website is down) and using Linux tools for more insight into CPU performance.
Until then, I am here, completely baffled by this CPU performance degradation affecting what are practically two different AMD systems. I'm beginning to wonder if there is a defect in AMD CPUs that I just happened to encounter twice in a row.
Facts
First problematic build:
- CPU: AMD FX-8350
- CPU installation date: 2013-07-20
- Motherboard: GIGABYTE 970A-UD3 (later: ASUS M5A97 LE R2.0)
- Memory: Kingston HyperX Savage DDR3-1600 4×8GB
- Cooling: Corsair Hydro Series H75 Liquid CPU Cooler
- Power Supply: EVGA SuperNOVA 750 G2
Second problematic build:
- CPU: AMD Ryzen 5 1500X
- CPU installation date: 2017-06-28
- Motherboard: MSI X370 SLI PLUS
- Memory: Crucial Micron Ballistix DDR4-2400 2×8GB (or: Corsair Vengeance LPX DDR4-3000 2×8GB)
- Cooling: Corsair Hydro Series H75 Liquid CPU Cooler
- Power Supply: EVGA SuperNOVA 750 G2
General symptoms:
- Choppy mouse movements in Windows and Linux
- User input is laggy, even in this Super User question box I'm typing in right now
- Compute performance ⅓ what it should be, compared to crowdsourced benchmarks
- CPU frequency sits well below configured frequency in BIOS
- Temperatures sit around 40°C as measured by Core Temp
Why are both AMD computer builds seemingly affected by the same issue?
How can I restore full CPU performance?
