Now that I’ve bolstered my confidence in being able to successfully install a modern CPU in a socket and robustly attach a heat sink to it, I’ve got more aggressive plans in mind going forward. In transitioning from “crawling” to “walking,” and in pondering what to do with the Intel Core i5-3570 I’d pulled out of the system, I realized that over the years I’d inadvertently collected nearly enough spare parts to put together not just one complete system based on the Core i5-3570 but also two others derived from the AMD A8-5600K Trinity 3.6GHz (3.9GHz Turbo) quad-core CPU I’d mentioned more than six years ago (but never got around to doing anything with) plus another AMD processor, an A10-7850K Kaveri 3.7 GHz (4 GHz Turbo) quad-core I’d subsequently obtained.
What do I plan to do with the systems after I build them up? Donate them to charity. Specifically, I’ve reached out to nearby Evergreen Christian Outreach, who (among other things) runs a highly-regarded homeless shelter program. More generally, their various clients (to clarify: not just Christians) are involved in, for example, therapy sessions and job searches-and-interviews, all of which have been complicated by the pandemic-induced cessation of face-to-face engagement. Computer interaction is an option, if computer access is available, of course, but also needed are webcams, microphones, and speakers for the systems so that they can be used as fuller-featured videoconferencing platforms. I’ve therefore also purchased these for each system, along with the stray DRAM module, enclosure, or other previously missing “puzzle piece,” and will be building the PCs in the weeks to come. I’m sure I’ll have plenty of experiences and results to share in upcoming blog posts.
I suspect, by the way, that at least some of the “Hackintosh” systems I’ve been writing about in recent months will end up as donations, too, once my experimentation with them is complete. And what do I plan to replace them with, for my own day-to-day use? Well, that’s where the “walking” to “running” transition part of my plan fits in. I actually plan to build up two modern PCs, one AMD-based and the other Intel-based, so that I can compare and contrast them and, more generally, come up to speed on both suppliers’ ecosystems.
First, the AMD system. Here I’ve selected the company’s Ryzen 7 3700X, an eight-core (sixteen-thread) 3.6 GHz (4.4 GHz max) CPU as its foundation.
AMD aficionados may be a bit surprised by the choice, since this processor is based on the third-generation Zen 2 spin of the Zen microarchitecture, versus the latest fourth-generation Zen 3, which delivers a not-insignificant improvement in average instructions per clock (IPC). My choice was driven by several key factors:
- It remains nearly impossible to find a Zen 3-based desktop processor at retail even months after their introduction, especially at 8-or-higher core counts. And assuming you can even get your hands on an 8-core Ryzen 7 5800X, for example, it will likely be priced at a significant markup to the $449 MSRP. Conversely, I bought the Ryzen 7 3700X on sale from Newegg for $319.99 (I tacked on an extra $35.99 for a three-year warranty).
- This CPU has only a 65W peak power consumption specification. I plan to shoehorn the Ryzen 7 3700X (plus a discrete graphics card, since this particular CPU doesn’t include an integrated graphics core) in a particularly diminutive mini-ITX case (Cooler Master’s 260mm wide × 208 mm high × 280 mm deep Elite 110) where fan count/size-limited airflow (specifically, heat removal) are always challenging, so the low power draw will be appreciated.
- Zen 2’s IPC is already quite good. More generally, the various Zen microarchitecture generations’ evolving capabilities, coupled with the advance processes they’re built on (and Intel’s contrasting manufacturing stumbles of recent years), have been key to AMD’s resurgence against Intel, specifically in enhancing the CPUs’ appeal to gamers (games remain notoriously single-thread dominant).
Source: Cooler Master
Now for the Intel system. Here again I’m not going with a processor (and accompanying chipset) based on a leading-edge architecture, although strictly speaking the 11th generation “Rocket Lake” products aren’t yet shipping as I write this in mid-February (they were officially announced at CES in January.) Instead I’ve selected the 10th-generation “Comet Lake”-based Core i9-10850K, a 10-core (20-thread) 3.6 GHz (5.2 GHz max) CPU.
Admittedly, if I’d waited on “Rocket Lake” I’d have garnered some improvements:
- A more advanced integrated GPU
- Notably improved IPC
- Support for AVX-512 and “Deep Learning Boost” hardware-accelerated instruction sets
- Chipset support for PCI Express Gen. 4, and for four additional PCIE lanes (20 total)
- A 2× wider DMI CPU-to-chipset interconnect bus (8 total lanes)
- Hardware-accelerated support for more advanced video codec profiles
- And other nips and tucks
That said, as I previously stated, “Rocket Lake” won’t ship until (reportedly) around the time you read this; I bought my CPU on New Year’s Eve 2020.
That the CPU includes an integrated GPU in this particular setup is at-minimum convenient, in case a graphics add-in card goes wonky, and potentially essential, if power draw or heat dissipation limitations in the slightly larger but still tiny mini-ITX case (Cooler Master’s 240mm wide × 207.4 mm high × 401.4 mm deep Elite 120 Advanced) I plan to use in this case preclude the use of a discrete graphics card. And that it’s not the latest-and-greatest integrated graphics architecture isn’t a big deal, since I plan to use the system for content creation, not gaming.
Single-thread performance improvements are all well and good, but the “Rocket Lake” CPU family won’t come in a 10-core (20-thread) variant. Again, since I plan to mostly run multithread-friendly content creation apps on this system, I figured the tradeoff made sense in my particular case.
Lack of AVX-512 and Deep Learning Boost support was actually one of the tradeoffs that gave me greatest pause. But none of AMD’s current CPUs support these instruction sets either, I don’t plan to be doing much if any AI inference on the system, and as for deep learning model training, I’ll likely leverage the external GPU as my accelerator, anyway.
Currently, at least, PCI Express v4 support is primarily advantageous with advanced NVMe SSDs, and even here the benefits are dubious (especially considering the incremental prices of PCIe v4 SSDs vs their mainstream PCIe v3 counterparts). As I’ll describe in a future post, I have a straightforward upgrade plan.
Lack of hardware-accelerated support for advanced video codec profiles is a bummer, but unless throughput is paramount, many content creation folks rely on fine-tuned software-only encoding for highest quality at lowest bitrates, anyway.
Pragmatically, “Rocket Lake” feels like a stopgap, because it is. It only exists because of Intel’s earlier-mentioned struggles to get the latest 10 nm manufacturing process into high volume production; yield and speed problems forced the company to backport its latest “Sunny Cove” CPU architecture, found on mobile “Tiger Lake” CPUs, to 14 nm. Now that Intel’s got its 10 nm problems sorted out, its next CPU line, “Alder Lake” with a “big.LITTLE”-like mix of standard high-performance and Atom-derived low power cores, is reportedly coming out this fall, effectively leaving “Rocket Lake” with a six-month life. In contrast, “Comet Lake” processors have already been shipping for nearly a year.
And at the end of the day, I got a great price on the Core i9-10850K: on sale from Newegg for $389.99 (again, I tacked on an extra $43.99 for a three-year warranty). I’d suspected that since it was widely rumored Intel would be publicly launching its next-generation “Rocket Lake” processors less than two weeks later at CES, the company’s retail partners would be heavily discounting various “Comet Lake” CPUs in advance to move remaining inventory. And indeed, I haven’t seen the Core i9-10850K at or below this low price since I bought it nearly two months ago.
Source: Cooler Master
One other note on the Core i9-10850K (and other Intel CPUs, for that matter): It’s not enough to just select the base product; the presence or absence of various suffix letters in the full product name affect the features you’ll obtain (as well as the price you pay). The “K” in my CPU’s name, for example, indicates that it not only has higher “spec” base and turbo clocks (and power consumption to match, maybe), but can also be user-overclocked. And had there been an “F” suffix, it would mean that the CPU contained no integrated graphics (in reality, the GPU core is probably still sitting on the die, but untested and disabled). Here’s the full list.
So there’s where I am with my plans, at least on the CPU side of things. There’s much more to discuss, of course:
- DRAM generations, capacities, and speed bins
- SSD form factors, interfaces, and capacities
- Graphics cards
- Wired and wireless connectivity (network and other)
- Chipsets and motherboards
- Power supplies
I plan to touch on all of these topics in future blog posts, but for now I’ll press “pause” and await your comments!
—Brian Dipert is Editor-in-Chief of the Embedded Vision Alliance, and a Senior Analyst at BDTI and Editor-in-Chief of InsideDSP, the company’s online newsletter.