Impact of Intel Microcode 0x12B on Content Creation Performance

Introduction

By now, most people in the broader PC world have heard of the alarming failure rates among Intel’s 13th- and 14th-generation desktop processors. This instability, which has come to be identified as the “Vmin Shift Instability,” was caused by elevated temperatures and voltages in a portion of a clock tree circuit in the IA cores. Those elevated voltages and temperatures were, in turn, caused by flawed or buggy microcode resulting in improperly elevated SVID (Voltage) requests from the processor, particularly when under light load or when idle.

Intel has now released the 0x12B microcode, which they claim solves the root cause of the instability and should be the last microcode update necessary to prevent further CPU degradation. Although this microcode (and the other prior microcode and BIOS tweaks Intel released to help mitigate the issue) affects voltages and boosting algorithms, Intel claims that the performance impact should mostly be “within run-to-run variation,” elsewhere stated to be about 3%. Since we will be rolling out these updates to our customers and have an upcoming review of the new Intel Core Ultra 200 series processors, we wanted to examine how much this update affects performance in the content creation applications and workflows we regularly test.

Background

Over the last 6 months or so, Intel has taken steps to identify and address the underlying issues leading to elevated failure rates for 13th- and 14th-generation Intel Core desktop processors. Although some end-users reported failure rates of up to 50%, many others—including ourselves—were seeing much lower failure rates, although still above the typical average and continuing to trend upwards. This is because the underlying cause of the instability was multifaceted but, interestingly, appeared to primarily affect processors that were using power settings that kept the processor in a high state of readiness while running relatively light workloads.

In June, Intel released its first microcode update that attempted to address the issue: 0x125. This update fixed a bug in Intel’s eTVB algorithm, which was causing processors to continue boosting at temperatures above which those processors should be boosting using TVB. Around this same time, Intel also released guidance to end-users and motherboard manufacturers advising following Intel’s power settings for their processors, including such things as limiting maximum power draw, current, and voltage, as well as enabling CPU protection algorithms.

Two months later, in August, Intel released the 0x129 microcode update, which limited VID requests by the CPU to 1.55 V. Finally, in October, Intel announced that it had discovered the root cause of the majority of CPU failures and released microcode 0x12B. This microcode update prevents the processor from improperly requesting elevated voltages during idle and light-load states. Importantly, although these microcode updates should prevent future CPU degradation, they cannot “fix” processors that are currently experiencing the Vmin Shift instability/degradation, and those processors should be replaced or refunded under warranty.

Going forward, Intel does not believe that any of its upcoming products will be affected by this issue. In particular, they claimed that the just-announced Arrow Lake (Core Ultra 200 series) processors are unaffected by the Vmin Shift Instability. Other unaffected parts include 12th Gen Intel Core, mobile processors (including HX processors), and Xeon processors.

Test Setup

In order to test the performance impacts of the Intel microcode updates, we preserved our i9 testbed from its configuration in our Ryzen 9000 series review, including all software, OS, and driver versions. We then updated our motherboard BIOS to the most recent version and enabled the “Intel Performance Profile” in the BIOS. Although how exactly it is enabled may vary, this option should be available on all new motherboard BIOSes for Intel 13th- and 14th-gen processors. After testing with that, we adjusted the settings to match what we use for our production systems. While these settings align with Intel guidance, they are somewhat more conservative than the default Intel Performance Profile as we typically focus on stability and reliability first and foremost.

Below, we have listed the primary differences between some of the profiles for the 14900K and 13900K. Note that lower-tier SKUs will have slightly different PL1, PL2, and ICCMax. For more info, see Intel’s blog post.

BIOS Profiles	PL1	PL2	Tau	ICCMax	CEP*
Intel Baseline Profile	125 W	188 W	56 S	249 A	Enabled
Intel Performance Profile	253† W	253 W	56 S	307 A	Enabled
Intel Extreme Profile	253 W	253 W	56 S	400 A	Enabled
Puget Profile	125 W	253 W	56 S	307 A	Enabled
Old Motherboard Defaults	4096 W	4096 W	56 S	Unlimited	Disabled

As you can see, the primary difference between the Puget profile and Intel’s Performance Profile is that we use the optional, lower PL1 of 125 W. Historically, we have found small performance differences with these settings in most applications, and for applications where they provide significant performance improvements, we have recommended other CPUs (typically, AMD’s Ryzen desktop processors) due to their superior performance in those workloads. Because of this, we have chosen to prefer more conservative power settings which we believe have helped increase the stability of our systems prior to the release of the new microcodes, and which results in cooler and quieter systems. However, this is an area we are continuing to explore with the newly announced Intel Core Ultra Desktop Processors (Series 2).

However, you can also see the very large differences between the Puget profile and the old motherboard defaults you would find on most consumer motherboards; the Puget profile has remained largely the same since the launch of 13th Gen, with major changes being enabling CEP when we were advised to do so, and increasing ICCMax from 280 A. The previous defaults would run CPUs with no power limits, effectively no current limits, and disable most of the important protection mechanisms built into the processor. It is perhaps understandable why it took Intel some time to realize that there was an underlying microcode bug rather than merely motherboard vendors running roughshod over the processors. Hopefully, going forward, Intel will take ensure that motherboard vendors comply with its power delivery profiles and guidance.

Results

As discussed above, we tested 3 different configurations of power and microcode for this article: our own power profile on the pre-0x125 microcode, our own power profile on the 0x12B microcode, and Intel’s Performance Profile on the 0x12B microcode. In the first graph of raw performance numbers, we provided 5% error bars—our typical margin of error. Note that the chart is logarithmic in order to fit the results all on one chart despite the two orders of magnitude of disparity. The second chart reframes the same data to look at the performance change from the old microcode with Puget’s settings as the 0% baseline.

We’ll start by examining the performance differences between the new and old microcode with the Puget Systems BIOS settings. Looking at the charts, the absolute score differences tend to be very small, to the point where, without labels, it would be difficult to see any difference in line length. Admittedly, this is due in part to the logarithmic scaling, but the error bars do help put the change into perspective—anywhere the bars overlap, we can’t conclusively draw a conclusion as to whether the performance is different. That being said, especially when examining chart #2, it appears as if there is a trend across all of the tests. Namely that, on average, the new microcode is about 2% slower. We definitely see that this is workflow-dependent, with the more heavily multithreaded (and CPU-based) tests displaying a larger difference of up to about 5%. Nonetheless, we feel confident in saying that the overall impact is negligible.

Although not a direct look at the microcode update itself, given that we have once again tweaked our BIOS settings away from the defaults, we wanted to see what the performance impact of doing so was. In the past, we have looked at the performance implications of running at PL1 = 125 W rather than PL1 = 253 W, so we have a good idea of what the results should be.

Digging into it, we find that in anything that is lightly threaded, the performance difference between our power profile and the Intel Performance Profile is essentially non-existent: less than one per cent. However, in heavily threaded applications like Cinebench multi-core or V-Ray, we see a performance loss of 11-20% from the lower PL1. This lines up with our previous testing and, although significant, does result in much higher temperatures. It is not merely a case of “free performance.” Further, we don’t configure or recommend these processors for heavy all-core CPU tasks as we typically recommend AMD Ryzen (Desktop and Threadripper) for our customers in those situations.

Conclusion

Overall, the recent microcode updates for Intel 13th and 14th Gen CPUs have little impact on performance in content creation applications. The largest decreases were seen for heavily multi-threaded applications, with Light Baking in Unreal Engine at -5% and Blender at -4% when using our internal settings. Both of these are within our typical margin of error, and most other results are smaller, approximately -1%—small enough that we cannot conclusively say they are present. Due to this, we are confident that there is, at most, a negligible impact on content creation performance when updating to the newest microcode in order to protect CPUs from the Vmin Shift instability.

When looking at the Intel Performance Profile, much like the last time we examined the 14th Gen power draw, increasing PL1 to 253 W has a relatively small impact on performance in most applications where you would want to use an Intel Core 14th Gen CPU. However, in heavily multi-threaded applications like Unreal Engine and Cinebench, we can see performance improvements of up to 22% as compared to PL1 = 125 W. Although, at this point in time, we continue to recommend that systems be configured with PL1 = 125 W, out of an abundance of caution, we expect to match PL1 and PL2 starting with Intel Arrow Lake.

We strongly encourage everyone to update their BIOS to a version that has microcode version 0x12B or newer in order to protect their processors from the degradation associated with the Vmin Shift instability. Most motherboard vendors should have at least a Beta version BIOS update that contains the fix, typically downloadable from the particular motherboard’s support/download section. If you are one of our customers, expect us to be reaching out to walk you through the update process.

Remember that the microcode updates cannot undo any damage that has already been done to affected CPUs. Intel has committed to extending the warranties of any retail CPUs, and most system integrators have followed suit (including us!), so reach out to your processor or system seller if you believe you are experiencing instability related to the Vmin Shift Instability. For our customers, our support team is here to help!

---

Finding the perfect workstation doesn’t have to be complicated. Explore our solutions page for a curated selection of recommended systems for a multitude of applications and workflows, or visit our custom configuration page if you already know the ideal hardware for your needs. If you need assistance with tailoring a system to a unique workflow or have any other questions, we encourage you to reach out to our dedicated technology consultants.