I Ditched Intel After Twenty Years

Yep, it happened to me too. 

I'm keeping this one short and sweet since there isn't a whole lot of new ground to cover here but, to the surprise of probably no one - myself included, my still fairly new 13th-generation Intel i5-13600KF started exhibiting what I could only imagine to be the onset of CPU instability. Two weekends ago I tried firing up a round of Assetto Corsa; which in my experience has been a very CPU intensive title if you're running with a sizable grid of cars, and my PC crashed straight to a black screen.

I initially thought this could be an issue with the game and I tried switching tracks and even reducing the grid size. Nothing worked, so knowing the current ongoing fiasco with the 13th and 14th gen chips, I decided to more thoroughly stress test my CPU to see if it was the potential culprit.

Normally I run my 13600KF with it overclocked and undervolted via XMP, and my chip has remained stable at 5.3 GHz P-core and 4.5 GHz E-core frequencies, with a -0.025 mV core voltage offset with temperatures under load hovering around the mid-to-high 70s in degrees C. Resetting these frequencies to the default 5.1 GHz P-core/3.9 GHz E-core fixed the crashing in-game, but I still wanted to do some more thorough testing to be completely sure.

Unfortunately there really isn't a definitive test out there that will say if your Intel processor is cooked or not. The most common suggestion is to use the official Intel Processor Diagnostic Tool to determine the condition of the CPU, while Intel also even suggested a dubious test involving the repeated installation of the latest Nvidia GPU drivers in an attempt to trigger an installation error. Neither of which is a guaranteed way to tell if the CPU is affected by instability or not.

So I decided to go about things in the most familiar way by using Cinebench R23's multi-core test along with Intel XTU. I ran my initial tests using the old microcode and all of the default CPU BIOS settings on my Gigabyte Aorus Z690 Ultra. Running the test on my original XTU settings crashed the PC to a black screen, while subsequent attempts at finding stability with or without an undervolt applied either caused Cinebench R23 to error out or the PC to blue screen.

My second go around was with the updated BIOS and subsequent microcode updates that Intel pushed out over the past month. Running the tests with the recommended "Intel Performance" setting enabled in BIOS actually stopped the crashing issues at my usual frequencies, however it enables undervolt protection by default, so I couldn't do anything about the 90 degrees C package temperature that I was getting hit with at that point. Turning off the "Intel Performance" setting allowed me to undervolt again, but once again caused it crash with the prior settings applied.

Regardless of whether or not I've come to the absolute right conclusion in this situation, I'm just not confident in the longevity of this chip generation anymore, nor am I keen on sticking with a company that has so far dragged their feet on providing answers and meaningful solutions beyond a two-year warranty extension and a microcode update that might at least prevent problems for people who have new chips and no interest in overclocking (which mind you, is the whole reason why "K" SKU chips even exist), but likely won't help people already affected by instability. 

Despite how widespread this problem seems to be, Intel still forces everyone to jump through existing RMA channels to resolve their issues. That includes OEMs for tray CPUs, who may have completely different standards on what they'd be willing to accept. The fact that Intel has refused to issue a recall or even just say what the affected dates ranges were in regards to the earlier oxidation issues makes me feel like they're trying to avoid owning up to the problem as much as possible. My trust in them is dead.

This of course, meant I needed to find a replacement, and I didn't want to have my system down for very long. I gave myself a loose budget of $500 as that was roughly what I spent on just the i5-13600KF and the Z690 board at the time. While I didn't expect to earn anywhere near that much back in return, it at least helped focus my search. And don't worry - I replaced the CPU before I sold it. That'd be a real dick move otherwise.

I at least wanted something on par with the 13600KF in both gaming and work applications and the least expensive chip that seemed to fit the bill was the $275 Ryzen 7 7700X. On the motherboard end, I'd be happy to take whatever I could get within that budget. A X670 with a decent VRM would be ideal so I could play with overclocking, but something like a good B650 probably would have held me over, and potentially opened up a little more headroom for a better CPU.

What I actually wound up with, was a little bit unexpected.

I managed to snag a bundle for $541 after tax that included an ASUS Prime X670-P motherboard with 32GB of G.Skill Ripjaws S5 DDR5 memory, and a Ryzen 9 7950X. Considering the CPU alone still retails at $513 as of writing, I'd say this was too good of a deal to pass up. I figured this would be about as good of an entry point into AMD as I could get for the money.

The R9 7950X is the flagship CPU of AMD's non-X3D Zen 4 lineup and it's mainly positioned as a competitor to Intel's i9 series CPUs, specifically the 12900K, 13900K, and 14900K SKUs. It packs 16 cores and 32 threads, and while it isn't quite as power hungry as it's rivals from Intel, it's not far off. A CPU like this is generally better suited for workstation applications that can really benefit from the higher core and thread count, gaming less so.

That's not to say you certainly can't game on it like a fiend too. It's just kind of a waste of money when less expensive AM5 CPUs within it's own generation like the 7700X, and especially the 7800X3D with it's increased 3D V-Cache can match or even blow the 7950X out of the water in those scenarios - especially in efficiency. More cores doesn't always equal better, but at least it won't prematurely degrade, right?

One cavernous VRM heatsink.

I'm not super thrilled about the motherboard though. This ASUS X690-P is an overall downgrade from my Gigabyte Z690 Ultra in almost every conceivable way. It's equipped with a 12 + 2 phase VRM versus the 16 + 1 + 2 phase design in the Z690, far less effective cooling solutions for the VRM and the M.2 slots, and it only has 9 USB ports versus 12 - two of which are still USB 2.0. It even still has a fucking PS/2 port on it for crying out loud. Who is this thing for?

But as a holdover board, I think I can live with it. I'm probably never going to attempt a genuine overclock on it, seeing how the 7950X is a pretty sizable step up from the 13600KF anyways. The G.Skill RAM DIMMs were actually slight upgrade rom my existing Corsair Vengeance-5600 DIMMs at 6000 MHz, though. I can't imagine there's going to be much of a noticeable performance uplift with the new RAM, but I'm not going to say no to more speed.

Performance-wise, the 7950X is an odd bird. AMD designed this CPU to boost to TJMAX under multi-core loads - a whopping 95 degrees C, and that's regardless of how beefy of a cooler you strap to it. It crushes multi-core workloads (I pulled a 36822 on my first Cinebench R23 run with it), but it's runs like a goddamn nuclear space heater. I instantly looked into what my options were for undervolting the beast and was introduced to Ryzen Master; effectively AMD's equivalent to Intel's Extreme Tuning Utility, and PBO. 

Essentially, undervolting a Zen 4 CPU in this manner requires you to set a negative curve offset in BIOS, either all core for simplicity, or per core for more control and potentially better results. I opted for the the latter option and like any undervolting setup, you'll want to test this with some trial and error. Being that I had sixteen cores to work my way though however, I used Ryzen Master's auto calculate feature to quickly determine what initial values to set, and then I'd adjust each accordingly after testing.

My Cinebench R23 results from a -11 all-core curve offset; which was ultimately unstable (top), and 105W Eco Mode with no curve offset (bottom).

Unfortunately, this still turned out to be a lot more of a struggle than I initially thought. While I saw some nice gains under load, I realized instability crashes would occur randomly under idle which is something I can't easily test for. Finding stable offsets ended up being such a pain in the ass that I ultimately just settled for using the 105W Eco Mode preset, which dropped the load temperatures down to a more manageable ~75 C with minimal performance loss. I might revisit the curve offset when I have more time to sink into testing it, but right now it isn't worth the effort for me.

Overall, I don't notice a dramatic improvement in gaming performance over the 13600KF but in CPU heavy workloads, the 7950X absolutely smokes. It's certainly still a step backwards in efficiency and overall thermals for my build, even against an already toasty Raptor Lake CPU, but as an introduction to Ryzen I'm pretty damn happy with it. As long as it just continues to work for the foreseeable future, I don't think I'll have many complaints.

As for Intel, they need to win back a lot of trust, and that's going to take some time. Arrow Lake doesn't appear to be much more than an efficiency focused generation, which is fine considering the focus there on Intel's front has been severely lacking in recent years, but I'd personally be weary of touching any of their offerings until long term reliability verdicts are available. I definitely want to see them rebound though, because competition keeps the market moving in a positive direction - but hopefully that's sooner rather than later.