Sidenote – The Post-Launch Evaluation of the Ryzen 9 3900X

A few months ago, I wrote a fun story about upgrading my main computer to the new Ryzen 9 3900X 12-core processor.

With the time that has passed, and the various stories of issues and concerns that have popped up, I think my story is worth telling – I discussed the immediate performance improvements in my prior post, but there was a lot of news that came out about performance not being met, boost clocks, voltages, and various other issues, many of which are fair criticisms but are also being exaggerated or overemphasized.

I’ll start this post by stating this: I am glad to have purchased the R9 3900X, it has been a great performance upgrade for me and I have been very happy with the behavior of my system and daily-use applications with it. I’ve also seen the issues many have talked about, as the 3900X amplifies them the most – high idle voltages often spiking or even sitting at 1.5V for prolonged periods, higher operating temperatures as a result, and an inability to hit the advertised boost clocks the CPU promises, with my highest single core boost settling at 4.5 GHz, 100 MHz short of the 4.6 GHz figure on the box. None of these issues have made it a measurably worse CPU for me, as even with these problems, it performs better in all regards than my R7 2700x, and does so while only slightly increasing average operating temperature under the same cooler as my 2700X.

However, I think it is safe to say that the 3rd-generation Ryzen chips suffer from a very AMD malady – AMD sucks at launching new products and shoots themselves in the foot frequently. This series of products had a world of hype behind it, some deserved, some produced by grifters and charlatans, and the whole tech world was buzzing over just how these parts would deliver once they were in the wild. The early benchmarks shown by AMD were promising, and further, honest – the scores shown were repeatable and the places where AMD is beat by Intel were clear and accurate to the real world scenarios. Intel is still king of pure gaming machines at the highest end of the mainstream, as the Core i9 9900k is still the best gaming CPU out there, although currently by single-digit percentages on average, which the new 9900ks should rectify by shoring up Intel’s multi-core performance.

In the launch window, however, AMD did some frankly foolish things that made their product look lesser than, and allowed Intel and others to question the validity of benchmarks and the reliability of the new Ryzen lineup. To recap:

The Precision Boost Overdrive and Auto-OC Video Was Dumb: AMD, before launch, pushed a YouTube video with information on the new iteration of Precision Boost Overdrive and the new Auto Overclock feature that ties into it. Basically, Precision Boost is AMD’s brand term for the CPU having a boost clock, which it handles differently than Intel’s CPUs by having a dynamically boosting and lowering clock speed. Intel CPUs with Turbo Boost basically have multiplier values for a given count of loaded threads, and the motherboards for them have a value that dictates how long the power delivery can maintain that value. On desktop boards, this is generally unlimited for Intel, so under a sustained load, an Intel CPU will lock in a boosted clockspeed and stay there unless it reaches a thermal danger zone, and either at thermal max or at completion of the workload, clock back down to the “stock” values. Intel’s boost is “dumb” in this way – much easier to understand and explain, but offers less headroom or flexibility. Ryzen, on the other hand, adjusts its clockspeeds on the fly in increments based on a mix of power delivery capacity, sustained power input, and temperature, and can nudge the clockspeed up or down in 25 MHz increments until reaching the maximum published clockspeed (first-gen Ryzen had a different way of implementing this, but it’s irrelevant for us today).

Base Precision Boost is the maximum rated clockspeed the CPU can reach, but not the maximum true clockspeed. AMD’s other boost feature, XFR, then kicks in, offering an additional increase past that point. In the first two generations of Ryzen, this was a feature that could extend clockspeeds further past the rated box speed, but was not considered an overclock for warranty purposes. What it meant in practice is that a CPU like the Ryzen 7 2700X, rated for maximum boost of 4.2 GHz, could then extend further to 4.35 GHz through XFR, provided that conditions were favorable. PBO then allowed the CPU to exceed the rated parameters programmed for the CPU on power delivery and maximum sustained current, which would allow these features to push harder if you had a good motherboard and cooling. On my R7 2700X, in practice, enabling PBO meant gaining around 200 MHz in bursts in a single-core setting (like WoW prior to 8.1.5), allowing my CPU to hit 4.55 GHz a tiny amount of the time, letting it burst through workloads and then quickly adjust downwards with the work complete.

AMD’s pledge for third-gen Ryzen was that PBO and the new Auto OC feature would push 200 MHz, and was presented as without flaws – the only caveats were motherboard support for the power delivery required. The problem is that PBO kind of…didn’t work for launch, save for maybe the low-end R5 3600, much less the Auto OC. Both features were present for launch, but enabling PBO on my system made my CPU run hotter and higher voltage for no discernable gain. Most people who turned on Auto OC (I was not one of them) gained nothing from that setting, either – the CPUs were simply eating up more power and offering nothing in return. Naturally, attention turned to the video, and frustration mounted, while AMD attempted to explain away that thermal condition mattered for the maximum clockspeed. Which would be fine, if not for…

The Boost Fiasco – Where’s My Advertised Clock Speed, AMD?: Boost clocks on the boxes of the third-gen parts were…generous, to say the least. They are supposedly obtainable, but it has taken until September for the 1.0.0.3 ABBA revision AGESA BIOS code from AMD for most users to see it. My CPU did not hit the rated boost frequency until I installed the new BIOS. The word is that the earliest BIOSes could hit the advertised frequencies under extraordinary circumstances (GamersNexus did it with the 3900X under liquid nitrogen!), but not on all CPUs (the higher up the stack you got, the less likely it was) and there were fears that the power delivery behavior these BIOSes used was too severe for long-term stability. The response was the earlier 1.0.0.3 revisions, which attempted to curb voltage supply and reduce spikes to 1.5V, but the flipside to this is that it meant that most of the CPUs now couldn’t hit their maximum boost frequency, and if they could, they’d only sustain it for milliseconds, missed by most monitoring tools which only poll at select larger intervals. Then, the still-high default voltage behavior spooked a lot of techies, leading to a further fix in 1.0.0.3 ABB, which was where we were stuck until September 10th and the announcement of the ABBA revision. Most users with it have reported positive developments on both fronts – reduced idle and load voltages and CPUs that can hit their advertised boost clocks.

Destiny 2 Had a Weird Snafu: Destiny 2 would crash at launch on Ryzen 3rd gen systems, and it seemed completely baffling as to how that could happen. Lo and behold, somehow the RdRand instruction, a useful but not very common x86 instruction for random number generation, was used by Destiny 2 and would cause the problem due to it not being executed correctly. The ABB BIOS revisions fixed this, but if you bought a Ryzen 3rd gen part at launch and your favorite game was Destiny 2, this was a pretty bad experience!

Motherboard Issues: AMD has something great in their sockets, which are designed to work with a wide array of CPUs for a number of years. All of their sockets in the last decade have seen a 3+ year service life with multiple generations of products released for them, generally only switching to support new standards for DDR memory. AM4, launched in 2017 with the first Ryzen parts, has been long-lived already, with 3 Ryzen generations spanning 2 to 16 core CPUs on it, and a 4th generation of products if you count the A-series APUs that were brought out with first gen Ryzen as an integrated graphics solution ahead of the Ryzen G lineup and the newer Athlon parts. The problem this breeds, however, is that AMD relies heavily on their AGESA BIOS microcode to communicate with new CPUs, meaning that any number of different and weird problems can happen. MSI, for example, shipped most of their motherboard lineup in the past with 32 MB BIOS chips, which are now storage limited for the microcodes needed to run the full Ryzen lineup, and are having to drop support for some parts and some features to free up space, while they’ve also relaunched some of their more recent older-era parts as Max or V2 versions with larger BIOSes. Some motherboards support BIOS flashback, a handy feature that allows the motherboard to function without a CPU inserted, allowing the user to download a supported BIOS and run the flash utility to make the motherboard work – but this is far from universal and was often a premium product differentiation for older generation motherboards.

In practice, what this means is that if you are building a new PC, you can buy the affordable B450 motherboard of your choice, but it may not work with a 3rd-gen CPU, and if your board doesn’t have BIOS flashback, you can’t boot. You can request a bootkit from AMD, which uses an old CPU that the microcode should support, allowing you to socket the loaner CPU, update BIOS, and then swap for your real part, but this is a poor option given shipping times and the delay it creates. Intel’s frequent socket changes and forced obsolescence are annoying, but their design idea is a good one – the motherboard and BIOS are effectively dumb hardware that allows the CPU to complete a circuit and therefore even with newer CPUs, you often don’t see the level of difficulty with BIOS on Intel systems that you do on AMD, although BIOS updates are often required for proper compatibility and feature support. As an enthusiast who upgraded, I didn’t much mind – it was easy to update my BIOS with my 2700X to prepare, and even if I didn’t have that, I could have just as easily used BIOS flashback on my board. However, when you’re working with the layman (like my guild leader, who just got a 3rd-gen system with a B450 motherboard that did not have flashback), even just explaining all of this is too much to expect an average user to put up with.

That is without getting into various possible problems like iffy power delivery on older motherboards which can handicap the higher-end third-gen parts, the X570 motherboards using chipset fans that can sometimes be suffocated by graphics cards, issues with on-again/off-again PCI-Express 4.0 support on older motherboards, and the new X570 motherboards not supporting the first generation Ryzen CPUs officially (a very minor point). On my Asus Crosshair VII Hero Wi-fi, new AGESA updates that AMD released often took weeks to be available in a complete BIOS for my board, meaning that as AMD rapid-fired new BIOS revisions, I was waiting on Asus to deliver new BIOS revisions which could help my performance and CPU lifespan.

Overall, these issues all point at something that tends to happen a lot with AMD. I like their products quite generally, and I made a point of buying into the Ryzen ecosystem when I did my new system build in summer 2018 as a measure of support. However, AMD often releases new products in a sort of slap-dash way, with things not fully baked or ready and various issues – usually small ones, but given that AMD is often the number 2 in their respective markets, people often are quick to judge based on those flaws.

Ryzen third-generation CPUs are fantastic products with great pricing that deliver excellent performance even absent the pricing (ignore the 3600X and 3800X and the rest of the stack is phenomenal), but the issues with the CPUs at launch make it easy for those wanting to write off the company to do so. In the price bracket my system is in, I could have built either an Intel Core i9 9900K system or the one I have with the Ryzen 9 3900X. Sure, in my case I would have also needed a new motherboard for the Intel option, but in pricing terms, they’re pretty close, and the Intel system still wins in gaming overall by around 3% due to the higher single-core clock speeds. The boost clock fix is a good start that can go a long way to offering better, more reliable performance in that comparison, but next month Intel is set to launch the 9900KS which is going to offer 5GHz clockspeed on all 8 cores, which will likely mean that the CPU will beat the 3900X. By how much and in what tasks remains to be seen, but it will retake a substantial performance lead rather than the simple 3% the current 9900K has.

This is without diving into issues with the Radeon 5700XT video card AMD launched the same day (a terrible blower-style cooler that was noise-throttled and allows the memory to reach near danger territory temperatures for GDDR6, a cooler design that can be improved substantially with the addition of plastic washers to the main mounting screws, software and driver issues, noise, and the weird management of the card’s pricing).

AMD has a history of doing this with new products, and while they get a bit of leeway when the products are really good (as the Ryzen third-gen parts are), it gets them an earned reputation as being bad at launches, or the “fine wine” meme around their GPU drivers improving performance of their graphics cards so much over time that they often can advance past competitors that are beating them at launch.

While someone like me can excuse the minor hiccups I experienced (none of mine were materially affecting at all other than having a warmer office), those that had bigger issues like the Destiny 2 problem would be more angry, and rightly so.

I hope for an AMD that someday has a flawless new product launch when delivering something as excellent as the Ryzen third-generation parts.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.