The best competition in CPU History is NOW

Why NOW is the best time for CPU competition

How did we get here? 

As anybody reading this probably already knows that we're in the middle of a CPU revolution. With the rise of AMD's zen architecture in 2017 we have been faced with unparalleled innovation in the CPU space with rapidly rising core and thread counts, increasing clock speeds, architectural and node improvements, and falling prices. This is no more evident than with the release of Intel's 10th generation desktop CPUs. 

I can hear many of you already shouting at your screen, and don't tear me apart. It's no secret that even now, Intel has been stuck on minor iterations of a process node introduced in 2014, and an architecture introduced in 2015.  Intel has not only abandoned the tick/tock cadence they had for years, but even their process/architecture/optimization cadence has been utterly abandoned with the 10th generation offering none of the above. This seems a far cry from the claims of innovation I made before. Despite this, they still have yet to totally fall behind, and with 10th gen, they're truthfully more competitive now than they have been since AMD launched the Zen architecture in 2017. 

Note: As always, I'm including section headers, so if you understand the section I'm talking about, feel free to skip it; this is so I can write to a larger spectrum of knowledge, and hopefully by the end, everyone's at the same page.

What is the significance of process node? (Understanding nm)

There are several notable things about process node. The process node also referred to as lithography describes the process in which processors are created, by literally laser-printing them onto a large circular wafer that may include dozens or hundreds of processors, depending on the size of the wafer and the size of the processor die (the term to describe the rectangular chunk of silicon that composes the logical processor itself, not the package which you may see and handle directly, on which you will notice pins, a PCB board, and an Integrated Heat Spreader IHS)

There are 3 main things that a better process node can help with. Clock speeddensity, and efficiency. Clock speed (measured in Hz, for example 4.0GHz) mainly determines how stable the processors are at a given clock speed, with some nodes allowing notably higher clock speeds. Higher density allows you to fit more transistors into the same area, allowing you to leverage those transistors for higher performance. Transistor density is the primary subject Moore's Law. Efficiency affects how much power (measured in W) is required to run at a certain level of performance. 

You will hear people refer to different process nodes in terms of nanometers nm, which typically describes the size and thus density of the transistors. However, the nm number of the process node has become somewhat arbitrary, and often much different between manufacturing companies like Intel and TSMC. It's important to note that Intel's 7nm and TSMC's 7nm are not the same. In reality, Intel's 10nm is closer to TSMC's 7nm and Intel's 7nm is closer to TSMC's 5nm and that's okay. That's neither good or bad for either company, as long as we understand the difference. When I say 'closer to' what I really mean is 'of similar density'. In other metrics, such as performance (clock speed) and efficiency, they may differ. 

What is the significance of Architecture? (understanding IPC)

A processor's architecture plays a large role in the performance of the product. While it plays a role in clock speed and efficiency, its main measuring stick is IPC, or Instructions Per Clock. IPC can be viewed as how much performance you will get from a CPU at a set clock speed. You will see some reviews compare the performance of two different CPUs that are locked to the same clock speed, in order to compare IPC. It is unfortunately not a single figure that can be attributed to a CPU architecture. It would be very convenient if we could just multiply IPC by clock speed to determine general performance, but a processor will have a different IPC in different tasks depending on how well-suited that processor's architecture is to that given task. Generally speaking, IPC is compared relatively to other CPUs; the generational improvement in IPC from the Zen+ architecture to the Zen2 architecture was often stated to be around 15%. 

There will be more discussion later on why Intel wins in some tasks and loses in others. Notably, Intel is currently top dog in gaming, and I'll give you a hint now that it's not just because Intel's clock speeds are higher, which is a commonly-repeated misconception (though as you should understand now, it does play some role)

Note: Intel is one of the few companies that both designs and manufactures their own CPUs, something that presents both benefits and risks. TSMC is a manufacturer, meaning they do not design chips, they just manufacture them. They're notably the current primary manufacturer for chips designed by AMD, Radeon, Nvidia, and Apple among of course many others.

Intel 10th gen summary

From this point, while there are no public reviews of Intel's 10th gen, we can make some obvious conclusions about the performance as it's bringing absolutely no surprises, so you will see me make claims of how they perform. I'll update this post if I'm wrong, but it's a pretty safe bet. It's virtually identical in process node and architecture to 9th gen, changing only in core and thread counts, as follows (with 7th and 8th gens included for reference)

Mainstream Product tier
7th gen core/thread
8th gen core/thread
9th gen core/thread
10th gen core/thread
k-series price
i9
N/A
N/A
8/16
10/20
$500
i7
4/8
6/12
8/8
8/16
$350-375
i5
4/4
6/6
6/6
6/12
$240-260
i3
2/4
4/4
4/4
4/8
$185

Here, you can clearly see the effects of AMD's competition as Intel has increased core and thread counts in some way every generation, 10th gen being no different. Their highest-end model has increased by two hyper-threaded cores each generation, and our new i3 will actually outperform 2017's i7 due to higher clock speeds on a more optimized (higher performance; higher clock speeds) version of the 14nm process even though the architecture in all of these CPUs is fundamentally the same with only minor adjustments for security updates and other minor tasks, as seen below: 

Generation
Process Node
Max Clock Speed (stock)
IPC generational improvement
6th gen
Intel 14nm
4.2GHz
7th gen
Intel 14nm+
4.5GHz
~0%
8th gen
Intel 14nm++
4.7GHz
~0%
9th gen
Intel 14nm++
5.0GHz
~0%
10th gen
Intel 14nm++
5.2GHz
~0%
* I'm ignoring velocity boost because it realistically is unlikely to come into play in a significant way as it requires temperatures below 70C. 

AMD 3rd gen summary

AMD on the other hand, has has a nearly opposite path up to this point as compared to Intel. They have not drastically reduced the price per thread, instead offering better performance per thread. 

Mainstream Product tier
1st gen core/thread
2nd gen core/thread
3rd gen core/thread
k-series MSRP
Ryzen 9
N/A
N/A
16/32
$750
Ryzen 7/9
8/16
N/A
12/24
$500
Ryzen 7
8/16
8/16
8/16
$330-400
Ryzen 5
6/12
6/12
6/12
$220-250
Ryzen 3
4/4
4/4
4/8
$100-120

Since the chart maps their core and thread counts to price only, it looks as though there is very little happening generation on generation, but in reality, we see the a different story: 

Generation
Process Node
Max Clock Speed (stock)
IPC generational improvement
1st gen
Global Foundries 14nm
4.0GHz
~52%*
2nd gen
G.F. 12nm
4.3GHz
~5%
3rd gen
TSMC 7nm
4.4-4.7GHz**
~15%
* represents a major architectural redesign from the largely failed bulldozer-era designs
** the 3950X claims 4.7GHz and 3900X claims 4.6GHz, but accounts suggest that those processors rarely if ever achieve those speeds. The more common 3700X will hit 4.4GHz. 

The evolution of competition

2017

When AMD launched Ryzen 1st generation, the very first iteration of the Zen architecture, they had literally double the number of cores and threads for the same price as Intel had on the 7th gen i7 7700k, but even though they had comparable IPC in many productivity tasks, Intel's high clock speed gave them an obvious per-core advantage, which means in any single or lightly-threaded workloads, Intel maintained the clear advantage, and AMD Ryzen dominated the value department whenever all cores could be leveraged. Competition had finally returned, even though there was often a clear better-performer depending on your type of workload. Intel was clearly faster in gaming, since games at the time did not do much to leverage more than 4 multi-threaded cores, despite arguments (that would turn out to be largely accurate) that games would soon begin to demand the greater resources available to the AMD CPUs with more cores. 

2019

With the release of Zen2 to face off against Intel's more expensive i9 9900k, AMD came into the strongest position it had ever had. AMD not only continued to offer unilaterally more cores and threads than Intel, but had finally reached core-performance parity. Even though Intel's clock speed is higher, AMD holds an IPC advantage that allows them to edge out Intel's CPUs in many per-core workloads, the notable holdout being Gaming.

At this time, AMD is closing in on, but unable to overcome Intel's lead in gaming, with Intel typically leading in gaming performance by single-digit percentages. However, there's more to the story. At the midrange and low-end, Intel gamers are starting to find that the lack of CPU resources in their i3 and i5 systems are affecting stability of performance. While they have a higher average performance, Intel's failures tend to be more severe, with some games exhibiting stuttering and visible frame drops during moments of more intensive gameplay, which are not present when running on AMD's systems with more threads. These issues have lead influencers to recommend Intel only at the high-end of gaming-only systems, and AMD Ryzen otherwise. 

Overall 2019 was a rough year for Intel, as the performance wasn't the only thing they lost. With their increasing core counts and clock speeds on a now-aged process node, Intel's CPUs are getting not only increasingly difficult to cool, but more difficult to supply stable power to, requiring more expensive motherboards with more robust VRMs. AMD also took the lead in platform, offering not only more PCIe lanes, but lanes that are twice as fast, with the new PCIe 4.0. Furthermore, AMD platforms tended to be more appealing as Intel retires motherboard lines every two years, making upgrading more difficult, where AMD's platforms promise a much longer path for future upgrades, not only for future generations of chips, but the possibility of buying a cheap used 16 core CPU a few years down the road.

Technical Note: It turns out that Intel has two main architectural advantages when it comes to Gaming. The first is Intel's core ring bus that allows on average, faster core-to-core communication, and the second is Intel's superior memory latency. Both of these advantages tend to get overwhelmed in most other workloads by AMD's otherwise more robust and modern designs. It's very common for people to say that Intel's faster at gaming because it's got a faster single core performance, which is wrong; it's that its IPC times the clock speed for that workload is faster. 

2020 - The best competition in CPU history

While in 2017 we saw AMD push the envelope with core counts but come up short in single-core performance, and in 2019 AMD nearly entirely eclipsed Intel in performance and features, at a lower price point. Now, in 2020, we're going to see the closest battle across the board we've ever had the pleasure of seeing.

With Intel now reaching segment-core parity with AMD, who now both offer 4c8t in the i3/ryzen 3 range, 6c12t in the i5/ryzen5 range, and 8c16t in the i7/ryzen7 range. We know their performance per core is similar, with AMD edging out Intel in most productivity, and Intel edging out AMD in gaming. Neither has a dominant categorical advantage.  Regardless of whether someone wants to game or has productivity work, they both offer similar performance at a similar price point. Intel is a bit more expensive at each price point, but offers a bit better gaming, and AMD is a bit cheaper while offering a bit better productivity. 

Intel series
Core/thread
Intel Price
AMD Price
Core/thread
AMD series
i3
4/8
$185
$120
4/8
Ryzen 3
i5
6/12
$260
$205
6/12
Ryzen 5
i7
8/16
$375
$340
8/16
Ryzen 7
i9
10/20
$500
$435
12/24
Ryzen 9



$740
16/32
Ryzen 9

The most important battle will be fought in the i7/Ryzen 7 range. Below it, AMD's prices clearly offer better value, and above it, Intel's own offering is cannibalized by the i7 which won't be able to offer any meaningful improvement in gaming on top of the fact that it's where Intel loses grip on core-parity with AMD. 

But of course this will all change with the release of AMD's Zen3 architecture in 4th gen Ryzen. So Intel gets a few months of the best competition they can muster. 

What's next? 

When I started writing this article, I started with the intention of supporting and leading to my conclusions for what's to come, as I'm very excited by it. Competition breeds innovation after all. However, regarding the length of this article I thought here was a good time to call it, with the caveat that it's sort of part 1 of 2, as we explore more factors in the landscape that support my conclusions for what's about to happen. I'll leave you with a bit of a teaser though: I believe AMD Ryzen 4th gen will see the largest increase in core counts to its product lines seen yet (though still top out at 16 cores)

Thank you for reading! 

If you enjoyed this, feel free to visit my Discord and my Patreon to tell me why I'm wrong discuss tech with me.

Discord: https://discord.gg/CHfha8V
Patreon: https://www.patreon.com/MeyerTechRants

Comments

  1. Replies
    1. sorry for the spam mobile browser glitched out and there were 4 comments instead of one.

      Delete
  2. This comment has been removed by the author.

    ReplyDelete
  3. This comment has been removed by the author.

    ReplyDelete
  4. This comment has been removed by the author.

    ReplyDelete

Post a Comment

Popular posts from this blog

Hardware-accelerated IO in consoles coming to PC

PS5 vs XSX vs PC featuring storage architectrures