Static overclocking of CPUs has been a thing of the past, inefficient and literally uncool, and not just since yesterday. Instead, dynamic OCs as an optimization of the boost behavior ex works are often the method of choice nowadays. Not least because AMD had completely deactivated static overclocking in the past with X3D CPUs especially popular for gaming. With the 9800X3D, however, the age-old all-core OC was released again, not least because thermal limitations were solved by shifting cores and cache silicon layers.
Actually, today was supposed to be about a 9800X3D tuning guide, including RAM OC, Infinity Fabric and co (CO). But while testing for it, I unfortunately came across a bug that may not be a bug at all, but should definitely be examined more closely – at least in my opinion. It’s about a considerable loss of performance in demanding computing tasks with high power consumption and AVX512, such as y-cruncher. For most use cases of a 9800X3D this is probably of little relevance, but with the upcoming 12- and 16-core 9000X3D chips and workstation use cases, things could soon look very different.
Test setup and methodology
As a benchmark, I use y-cruncher with preset 1b, version 0.8.5 from Benchmate 12.1.0. You can optimize various parameters here to speed up the calculation, but for most tests I don’t bother here, because the aim is to compare the different CPU overclocks. For workload, “Intel Cannon Lake (AVX512)” is selected, which is optimized for AVX512 and will become interesting later, and the priority is set to “Realtime” for the most constant results possible. Three runs are carried out for each configuration and the best result is used. Incidentally, HWinfo runs in the background during all tests, which affects performance. If you want to reproduce the results, please bear this in mind.
The main motherboard used is the Asus Crosshair X670E Gene, but the behavior can also be reproduced with other motherboards. More on this later. For the RAM, I use DDR5-6000 with manually fixed timings (also subtimings) for the sake of simplicity, see screenshots. The FCLK runs at 2100 MHz. Every board and every CPU can do this and so we control these variables.
Ryzen 9 7950X as reference
Before we put the Ryzen 7 9800X3D through its paces, I would like to establish a baseline first. A non-X3D chip is used for this, in this case a Ryzen 9 7950X. We simply run this CPU through the y-cruncher with standard clock settings and note the result. We then configure the same all-core clock statically and compare the results. The RAM and FCLK settings are slightly different here, but that doesn’t matter because we are only interested in the difference in performance between the dynamic and static clock.
Out of the box, the cores of the 7950X clock to approx. 5.4 GHz on the first CCD and 5.2 GHz on the second CCD, i.e. around 5.3 GHz on average. In y-cruncher 1b, the 16-core Raphael CPU then needs approx. 14.1 seconds.
If we statically overclock the CPU to 5.3 GHz on all cores (with 1.35 V at LLC5), the CPU performs roughly the same at just under 14 seconds. The difference is probably due to the slightly different clock rate compared to stock. Same clock – same performance. This is not rocket science and is actually completely self-evident, but I wanted to prove this in advance before we descend into the strange clock abysses of the 9800X3D.
- 1 - Test setup and Ryzen 9 7950X as baseline
- 2 - BIOS 2602 – Asus Crosshair X670E Gene
- 3 - Clock Streching or AVX throttling and workarounds (LN2 mode)
- 4 - The "Fix" BIOS 0029 – Asus Crosshair X670E Gene
- 5 - ASRock B650I Lightning WiFi for comparison
- 6 - BIOSes 2604 and 2506 – Asus Crosshair X670E Gene
- 7 - X3D-Cache confuses y-cruncher (developer)
- 8 - Cinebench R23 with "AVX-Light" as reference
- 9 - Summary and many unanswered questions
19 Antworten
Kommentar
Lade neue Kommentare
Urgestein
Urgestein
Urgestein
Mitglied
Veteran
Veteran
Urgestein
Urgestein
Mitglied
Urgestein
Veteran
Urgestein
Urgestein
Veteran
Urgestein
Urgestein
Veteran
Mitglied
Veteran
Alle Kommentare lesen unter igor´sLAB Community →