To say it in advance: The Ryzen 7 5800X3D is AMD’s almost condescending gesture of nonchalance to make use of an Epyc chiplet (“Milan”) with 3D-V cache as a “secondary purpose”, pack it into a normal consumer CPU and “purely by chance” send Intel’s waffle toast in the form of the completely overpowered Core i9-12900KS gallantly into the desert. Let’s call it planned efficiency versus ecological obsolescence or simply the downfall of a CPU with strongly anachronistic features.
Yes, the huge V-cache is obviously not a panacea for all circumstances and it will not be the general solution for all scenarios, but even in games (and applications) that do not benefit from the V-cache, the CPU still performs quite excellently even within its own fraternity. So it is not a miracle CPU, but it also shows why there is only a Ryzen 7 5800X3D and no Ryzen 9 5900X3D or even 5950X3D. AMD surely thinks it would be enough for a decent scent mark in Intel’s direction. Well, let’s see.
Besides that, the chiplets can be sold much too well as Epyc and above all more expensively than to have to unravel and fritter away its resources before the imminent launch of the next AMD generation. The cooling is hardly the reason, because there are up to 8 chiplets in an Epyc and they don’t burn up. No, AMD is probably just self-confident enough here to challenge the Intel Core i9-12900KS with only an 8-core. Will that at least be enough for gaming in the end? That’s exactly what we want to find out today. And for those who didn’t read it the other day, here’s the theory on today’s practice in pictures again:
And what actually happens when you cross-test and compare the fastest AMD and Intel CPUs with the respective fastest graphics cards from AMD and NVIDIA? Driver overhead of the GeForce RTX 3090 Ti in some games below 720p then versus its brute 1080p violence? Or would you prefer the more frugal Radeon RX 6900XT, which really comes into its own in 720p? And if you don’t know which of the graphics cards you should take in the end, it’s better to take both. So that was exactly the plan.
There are games that are better suited to the Radeon as a graphics card, but still prefer an Intel CPU, or vice versa. In contrast to games like “Shadow of The Tomb Raider” or “World War Z”, where the huge cache of the Ryzen 7 5800X3D virtually pushes everything else into the dust, the case is completely different in other games when the software generally does not like Ryzen CPUs. It is almost impossible to achieve a balanced mixture. That’s why I’ll dedicate my conclusion to the whole package later and not just the pure FPS. There are plenty of such tests and they sometimes end up in the direction of Intel or AMD, depending on the game selection and hardware.
Since many readers complained last time about having to run the CPUs with uniform RAM settings, I (logically) opted for a setup with DDR4 RAM. So that the Intel CPUs don’t get any memory hiccups and we can still stay fair, I opted for DDR4 3600. Firstly, this is the clock where even the Ryzens can still officially be operated with FCLK 1800 directly via memory setting and secondly, no instability is yet felt with Intel and Gear 1. My current Core i7-12700K is apparently a real potato chip that doesn’t run DDR4-3733 stably in Gear 1. So all CPUs run with DDR4 3600 CL16-20-40. That fits perfectly.
Test system and evaluation software
The measurement of the detailed power consumption and other, more in-depth things is done here in the lab (where the thermographic infrared recordings are also created with a high-resolution industrial camera in the air-conditioned room at the end) on two tracks by means of high-resolution oscilloscope technology (there are also various follow-ups!) and the self-created, MCU-based measurement setup for motherboards and graphics cards (pictures below) or NVIDIA’s PCAT.
The audio measurements are done outside in my Chamber (room within a room). But everything in its own time, because today it’s all about gaming (for now).
I have also summarized the individual components of the test system in a table: