Both NVidia and Intel have not suffered from AMD’s competition for a long time.
AMD has both GPU and CPU chips in portfolio.
5 years ago, Lisa Su became CEO of AMD.
The goal of each CPU producer is the lucrative server market segment.
But AMD has promised a decade ago about much in terms of server CPU and then disillusioned.
Until about two years ago, Intel thus had 99% market share in the server market.
AMD was 5 years ago so dying after death.
Lisa Su has put AMD back on the map, and that was a huge feat.
It has made enormous improvements not only on the technical side but also on the marketing side.
Technically, a roadmap has been introduced that describes the different architectures that first had to bring AMD back as a reliable CPU producer with “value for money”.
Via Chiplets (eg the IO ring is still 14nm technology of GF and the chiplets themselves are 7nm TSMC chiplets) and via focus on architecture she has put the right people in the right place that have implemented this architecture.
At the same time, the marketing communication was magijet.
The first Ryzen chips were good enough to compete with the lower Intel processors.
AMD came back to attention, via desktop and laptop.
Partly thanks to production problems at Intel, which first delivered erver chips, then mobile and then desktop.
This allowed AMD to show that they were back and pick up Intel’s market share.
That was the foot in the door to get into the server market EPYC (the server variant of Ryzen which is roughly the same in architecture but has more cores), by letting the procedure of certification go (CPU and motherboard need about a year to be tested and Be certfied for server customers they can roll out).
Su promised what was possible and never made the mistake of the past.
Ryzen 2000 series was better and could already compete mid-range with Intel.
Ryzen 3000 is in 7nm technology and is therefore better in performance per watt.
Also regarding the number of transistors per unit surface, this technology is better than what Intel now has (namely 14nm + +).
In the EPYC series, Naples (32 core) was already rolled out and Rome is almost available (64-core), taking over AMD’s leadership in Technology node (7NM) and up to maximum cores per CPU (64 versus 56 planned cores for Intel).
The pipeline that started 5 years ago has now reached full speed at AMD.
Intel pipeline is struggling with problems.
To the extent that Intel sets record figures thanks to past products.
But the products that have to bring in over 2 years the large amounts, there one can put large question marks.
Desktops and laptops have long been 2, 4, or 8 cores with 4, 8, or 16 threads.
The evolution is lean in terms of parallel processing.
For KI, we need a lot of parallel calculations that go slow on a CPU but quickly on a GPU.
Graphics chips need to handle a lot of pixels in parallel and are therefore extremely suitable for machine learning and AI.
NVidia is still top here.
With Turing, they claim the first real-time ray tracing.
It is actually still a hybrid solution, but that leads us too far.
Ray tracing is possible on older video cards but slower (fewer frames per second).
Now the RTC (ray tracing cores) from NVidia do not mean as much for AI and ML.
AMD has made mainly Gpus, such as the CPU strategy, which was re-mapped via an architecture roadmap in AMD steps.
First low end, now also closer to high-end.
But AMD has always taken a different market than NVidia.
There is overlap but AMD was strong in game consoles.
Both the PlayStation 5 and the Xbox (code name Scarlett) will get the new NAVI GPU.
NVidia has previously been targeted to PC gamers and video/image processing companies.
Here too, the server market is the market with high profit margins.
And the cloud needs GPU and FPGA hardware acceleration for AI and ML.
AMD has a CPU and a GPU and can both work seamlessly together at the highest speed.
Intel has no discrete GPU (they’re way too late “awake” shot, their Intel Xe would come out in 2020, way too late) but Cpus and FPGA (former Altera).
NVidia has no CPU but does have powerful GPU accelerators.
They also bought Mellanoz, because the communication of the entire server system is also a very important factor.
Fast, low latency and high bandwidth communication between all components of the system is crucial.
That’s what AMD has not pronounced again (as Mellanox has it).
- Hardware, IE chips, are the basis of everything.
Software can only be efficient and fast thanks to custom hardware.
Thanks to the boom in software, a very wide software world has come, from low end (web related) to high end. And there’s definitely a race to the bottom busy for low end software work (Fiverr, Odesk,…). This has also had a devastating effect on hardware design. Quantity on quality. Hardware must work first time, there is no Tuesday patch day.
Under Lisa Su, AMD has launched a good technical pipeline with experts who know what they are doing.
The marketing piece has also been given a makeover under Lisa Su.
She is someone who can have the right people in the right place to do the right things (I do not do her with Jobs, that is out of category, but he also mastered the same strategy that has proven to be hugely successful).
Compare that to Intel and your pants are dropped off.
You see the train arriving, first from very far, time enough to react.
Is a little closer, another little glass to learn it off?
Very close, hmm, time for what caviar I think.
The train walst over you, well that hurts, but look at our profits.
Train is over and there is still some damage, maybe we will release a discrete GPU chip next year, what do you think?
A classic of an institute that lived for years on the lack of innovation and slowly calended.
Now there is still little movement in it.
But if the profit starts to sag, the market share drops to 85 a 80% in the server market, then the man comes with the axe and drastic measures will have to be taken.
NVidia is in a less precarious situation, but they also laughed AMD away and continued to praise their top chips far too high.
But they still do have things in the pipeline and still have a market outside the common GPu competition with AMD, namely image processing and high-end gamers.
But they too can no longer be more competitive.
It is exciting again (market) and the innovation is dripping back off.
Something that has been gone for a long time.
I’ve always said that any startup with a tenth of the budget of a large semiconductor giant but with a few hardware experts, can make a much better chip that costs so much less (NRE) and at the same time can be on the market much faster (TTM).
Incompetence and inefficiency is everywhere.
Lisa Su proved that it can also be different.
By placing good products on the market with associated marketing.