Hello all, today we are going to be discussing everyone’s favorite meme stock,
Tesla AMD! Before we get into the meat of the arguments for AMD and Intel, this is intended to be a follow up of another pretty crappy post I made a few months back, as such we will go into much further depth than before. Also I will not be providing price targets due to macro environmental circumstances.
In this post I will first have to cover some terms needed to talk about both companies’ products, and some history. Then I will talk about how both are positioned in the present and the near term (next 6 months to a year), followed by long term speculation which I’ll release later (next week maybe). Please note that I will briefly touch on NVIDIA and their products in the future speculation, but I will not go into depth due to there being a lot of misinformation circulating. I will most likely make another post doing a deep dive into the relation between NVIDIA, Intel, and AMD in late October/December.
Also note that I do not hold any shares or derivatives in AMD, Intel or NVIDIA.
An introduction into the semiconductor industry
The semiconductor industry is a massive industry currently valued at a massive 433b1, it is what powers all electronics ranging from the weakest smartwatch all the way up to the largest supercomputers. Semiconductors are made out of silicon oxide with transistors on them which perform logic functions and the likes. The semiconductor market has two large segments, that being the CPU market and the GPU market, with CPUs being used to compute a wide ranging set of data, whereas a GPU is used to perform lots of very repetitive tasks such as rendering. There are four primary companies in this space, that being Intel, AMD, NVIDIA, and ARM (note that Qualcomm and Samsung license their core architectures from ARM).
Some important terms that are needed to understand basic things that will affect performance/competitiveness.
- Architecture: Refers to the design of a product think about it like how a house is built from a frame.
- Clock speed: How many cycles a CPU/GPU can perform, each cycle is the effectuation of a certain number of calculations.
- IPC: Instructions per clock refers to how many calculations a CPU/GPU can perform per clock cycle.
- Core: The “processor” in a CPU/GPU that performs the calculation
- Thread: present only in CPUs, allow multiple code paths to access a singular core at once thereby ensuring that the core is always fed with information.
- Monolithic design: A CPU/GPU using only 1 piece of silicon.
- MCM: Multi-chip module, a CPU/GPU that is composed of a multitude of pieces of silicon.
- Wafer: A disc of varying size in which the transistors are inlaid on, this will later be cut up into chips
- Chips: Smaller pieces of silicon which are typically a final product
- Node: Refers to how small a transistor is, measured in nano-meters (NM). Note however, it is more of a marketing term and the real size of the transistors are typically much larger, furthermore different nodes are intended to compete with each other, such as Intel’s 10nm node being intended to be on parity with TSMC’s 7nm node.
- Reticle limit: The maximum size a chip can be made on a certain node.
- EUV: Extreme ultraviolet, is the cutting edge of semiconductor manufacturing, uses high energy ultraviolet light to imprint the transistors on the wafer, allows for increased efficiency and better yields.
- Yields: Defect density of a wafer, if there is a defect on wafer the corresponding chip may be compromised and have to be thrown out or binned lower.
- Binning: The practice of sorting chips by how many defects they have, a chip with a lot of defects will have lower performance then one that has none.
- Cache: Small, low latency storage that allows for cores to quickly access data. Further divided into L1, L2, and L3 cache, with L1 cache having the lowest latency and lowest storage and L3 having the highest latency and the highest capacity.
- Memory/RAM: Fast storage that a CPU can directly access, note that it can not store as much data as traditional solutions, but latency and transfer speed are much higher. There are currently four generations of it, that being DDR1, DDR2, DDR3, and DDR4, with DDR4 being the newest. Memory also has clock speeds which indicates indirectly how fast the data transfer is.
- Fab/Foundry: Where wafers are made, the three largest ones are TSMC, Intel and Samsung.
A brief history of the relation between Intel and AMD
Intel was founded in 1968 by a few people you may know about, Gordon Moore, Robert Noyce, and Arthur Rock. Gordon Moore and Robert Noyce had left Fairchild Semiconductors with Arthur Rock providing the initial capital. Together the three made the first commercial CPU, the 4004.
AMD was founded a year later in 1969 and initially produced memory, but in 1981 they entered in a 10 year partnership with Intel as a secondary supplier as requested by IBM. At the time, most OEMs required a secondary supplier to ensure that they would not become too reliant on one company and to ensure that they would have enough supply should one company have issues. This meant that Intel licensed their technology to AMD so they could produce them at their own fabs.
This partnership did not last long when in 1984 Intel broke the partnership, and violated the terms which in and after numerous appeals and cases of discovery abuse on the side of Intel resulted in the court’s ruling in AMD’s favor in 19902. Because of this AMD was forced to set out and create their own chips, initially copying Intel’s 80C287, which would have been shared under the 10 year partnership. This resulted in Intel suing AMD for patent infringement, however the courts ruled in AMD’s favour citing the previously mentioned case.
As the decade progressed, and the 2000s rolled around AMD came out with their own products, which handily beat Intel’s own offerings, but the damage was already done. Intel was able to make so much money that they essentially had a monopoly on the CPU market.
But despite having the performance crown for about 7 years, their market share never came close to that of Intel, and the reason behind this would be revealed in 2009.
In 2009, the European commission fined Intel $USD 1.4b3. The reason behind this was Intel had bribed OEMs with cash incentives to not use AMD processors. This was to the point that Intel paid Dell 1b per year to not use AMD4. These bribes were so significant that when in 2006, in order not to be laughed out of business, they started to use AMD products, resulting in a massive 50% decrease in profit in 20065. That being said the damage was already done once again, and AMD was relegated to being a secondary competitor to Intel as they had simply run out of money. AMD released the Phenom in 2008, which was thoroughly crushed by the Intel competitors and after three years of refreshing the product with minor improvements, they released the horrible Bulldozer in 2011 which was legendary for its horrible performance, and insane power consumption. In 2016, they were near insolvency, with a share price bottoming out at 2$ per share.
All the while, Intel made small iterative performance upgrades while making all the money they could ever want. This is why your computer from 2010 is still very serviceable in 2020, 10 years later.
In 2017, AMD released their new Zen architecture with a massive 40%+ leap in IPC bringing it back into competition with Intel. Furthermore, Zen was an MCM design meaning that you could pair a few small chips to act as one large chip via Infinity Fabric, an interconnect that acted as the highways permitting multiple chips, or chiplets to communicate with each other.
Intel VS. AMD in the modern age
A lot has happened in the last 5 years. Currently, Intel’s technology portfolio is a complete and utter disaster, rife with stagnation. For example, Intel’s primary product, Intel’s current offerings are a mere 7% better in applications that only use 1-2 cores, than those of AMD’s while using 50% more energy and having vastly inferior in the increasingly important applications which leverage more than four cores. Intel’s highest end desktop product, the w3175x, coming in at 3k only has 28 cores, whilst the highest end AMD cpu the 3990x, has over double at 64 cores. The performance disparity is clear, AMD is the only viable option for productivity. The story continues in servers where Intel is outclassed in a similar manner to that of the desktop with performance on the AMD side being well over 200% higher. The slaughter continues into laptops where AMD’s 4900HS, completely embarrasses the highest end Intel competition while consuming ⅓ the power. The only performance crown that Intel can claim is gaming, and even then it is a mere 7% higher, due to extreme clock speeds with their newest architecture, when compared to AMD’s design, which is over a year old.
There are multiple reasons why Intel is so behind in performance, the primary being nodes. Intel’s 10 nm node was slated to launch four years ago. Four years is a very long time for any company, especially for the semiconductor space, and the sole reason why Intel was not severely punished for these stagnations was that they were miles ahead of any competitor. Earlier on I mentioned that AMD had fabs and was a secondary supplier, however as they were approaching insolvency, they sold off their fabs and partnered with Global Foundries, and later TSMC. However TSMC ramped the 7 nm, which was designed to compete with 10nm, extremely fast, handedly taking Intel’s previous position as owning the most advanced fabs. Looking at Intel’s road map, volume 10 nm will most likely hit the market by 2021, if there are no further delays, but TSMC is not sitting still, they are also innovating with the 5 nm, which was designed to compete with Intel’s 7nm hitting markets in 2020. If Intel does not get 10 nm functioning by 2021, it will spell the end of Intel’s fabs. TSMC will be not one, but two nodes ahead of Intel.
This is also compounded, though to a lesser extent by there being little to no IPC improvements in 5 years, meaning that AMD was free to catch up and in many cases surpass them.
The final nail in the coffin when it comes to Intel in the near term is their security vulnerabilities. Before 2018, all security vulnerabilities were mainly from software attacks, but that changed with the discovery of Meltdown and Specter. To show how important these issues were, let me illustrate an example. Say you are a CEO of a major investment firm (lol), and you are using Intel servers to do whatever you need to do, you go about your day, the server is 100% secure from a software standpoint. Night comes, and you decide to look at how profitable your company was this quarter (again lol), but on the front page you see “Major leak at X company all confidential data was leaked,” you panic and call up your IT guy and he is in a panic, the head lines are correct all company data stored on any server that you own has been leaked and there is nothing you can do about it. That is the danger of these vulnerabilities. There is absolutely nothing you or your staff could have done, and worse yet, there are over a dozen similarly dangerous vulnerabilities, and over 600 more minor ones. The only way to try and prevent these leaks are to apply patches from Intel, but this results from anywhere between a 10% to a 60% loss in performance, which needs to be made up by buying new servers as you are already at 90-100% capacity. Then you hear that a certain company, AMD, was immune to this attack and their products are vastly superior. This is the major issue plaguing Intel at the moment. For the time being you will see a major rise in server sales as people scramble to buy up servers to remain operational, but in time they will switch to AMD, with AMD forecast to take 10% in Q2 20206, up from near 0% in 2016.
Intel VS. AMD in the near future
At the moment AMD is uniquely positioned, their competition has yet to compete truely with their 1 year old tech, and they are on track to release brand new CPUs this year in late Q3/Q4. As a warning, we are entering grounds of speculation, but do note that this speculation is based on reliable leakers like AdoredTV7 (correctly leaked entire AMD CPU product stack from last year with only minor errors), Moore’s Law is Dead8 (correctly leaked entire CPU product stack form Intel that was released this year with near 0 errors thus far).
AMD: Intel(New product titled Zen 3, releasing this year) (New product released this year)
16 cores with 32 threads N/A
12 cores with 24 threads 10 cores with 20 threads
8 cores with 16 threads 8 cores with 16 threads
6 cores with 12 threads 6 cores with 12 threads
Things to remember
– Zen 3 ~13-15% IPC bump7, with maybe 100ish MHz clock speed bump7, which will allow AMD to claim the final performance crown of gaming from Intel, this coupled with Intel being nowhere in sight will allow AMD to reign supreme in all faculties for a while.
AMD is in an excellent position to capitalize on Intel’s stagnation. The security flaws could have not come at a better time, accelerating the world’s adoption of AMD CPUs and their new products will beat Intel in gaming thanks to the IPC and clock speed increase improving their gaming performance by at least 15%. Overall, Intel is a precarious spot, that being said they have vast sums of cash and they had a few key developments in the past few years which could allow them to catch up in 2022 or 2023. I will discuss that possibility in a later post along with an analysis of how AMD will perform in the GPU market with consoles on their side.
Disclaimer: This information is only for educational purposes. Do not make any investment decisions based on the information in this article. Do you own due diligence.