Raja Koduri: “AMD has no software ecosystem that’s meaningful without Intel”
During Intel's 2019 Investor Meeting, Intel has announced numerous updates to their product roadmap, but in the process of doing so, Intel's Raja Koduri had some pretty interesting things to say about the competition.
Intel Versus Competitors
Raja Koduri, Intel's Systems and Graphics Architect and Senior VP, analyzed what Intel's competition is doing for the future of computing. Koduri referred to both AMD and NVIDIA, though in a manner that was not direct, but rather a type of graphic with two red and green circles representing AMD and NVIDIA. According to Koduri, AMD has 'no ecosystem that's meaningful'.
This graphic essentially shows Intel, with NVIDIA and AMD as smaller competitors. Judging by this graphic, Intel is active in nine markets:
Intel's Competitors' Markets
Intel compares these nine markets to a few of those that AMD and NVIDIA are active in, listing the following:
- Cloud (AMD & NVIDIA)
- PC (AMD & NVIDIA)
- GPU (AMD & NVIDIA)
- CPU (AMD)
Intel also has competitors in the same markets, as well as alternate markets although lacks a robust ecosystem:
- Datacenter (AMD & NVIDIA)
- FPGA (Xilinx)
- AI (AMD & NVIDIA)
- Interconnect (AMD & NVIDIA)
- Memory (AMD)
- Semi-Custom Solutions (AMD)
- Embedded Solutions (AMD)
AMD has been gaining momentum and outsold Intel by two fold at a major German retailer - although this is a very small sample and not indicative of the wider markets.
Intel shows AI solutions to be one of their four core markets, yet Intel's only products in that field include Nervana and the Intel Neural Compute Stick, with Intel cancelling production of first-generation Neural Compute Sticks. Both of these products cannot compete with alternate FPGA solutions, let alone NVIDIA's Tesla V100 or AMD's Radeon Instinct MI60.
Current High-Speed Interconnect Solutions
Intel also has presence in the Interconnect market and faces competition from NVIDIA and AMD here as well. NVIDIA designed NVLink, a high-speed interconnect designed for GPU-to-GPU, GPU-to-CPU, and CPU-to-CPU communication, though GPU-to-CPU and CPU-to-CPU is currently supported only by IBM's POWER9. AMD founded the HyperTransport Consortium and has designed HyperTransport, a point-to-point interconnect since HyperTransport 1.0's release in 2001. HyperTransport has undergone multiple updates:
- HyperTransport 1.1 (2002)
- HyperTransport 2.0 (2004)
- HyperTransport 3.0 (2006)
- HyperTransport 3.1 (2008)
Despite HyperTransport 3.0's release in 2006, HyperTransport 3.0 manages to overtake Intel's fastest QuickPath Interconnect variant. (332.8Gb/s versus 307.2Gb/s)
AMD's Infinity Fabric is a superset of HyperTransport and is supported by Zen-based CPUs and Radeon GPUs with a maximum bandwidth of 4096Gb/s, making it the fastest interconnect available.
As far as memory development goes, AMD and Intel are the dominant players. AMD designed HBM, designed Athlon, the first CPU to integrate the memory controller onto the CPU, and was a major entity in the development of DDR3, therefore allowing Phenom II CPUs to be utilized in AM2/AM2+/AM3/AM3+ and utilize both DDR2 and DDR3 memory variants. Intel's most recent memory product comes from a collaboration with Micron and is known as 3D XPoint Memory, a type of non-volatile flash memory.
Semi-Custom & Embedded Markets
Two markets where AMD has an edge in are Semi-Custom Solutions and Embedded Solutions. AMD designed custom processors for the Wii U, Xbox One, PS4, and future consoles such as the Atari VCS, and Smach Z. AMD's lineup of embedded processors is:
- G Series
- R Series
AMD & NVIDIA In the Datacenter
NVIDIA and AMD will also present competition in GPU compute workloads, with most compute tasks being handled completely by GPUs and CPUs are used to execute commands to the GPUs. AMD has also established a presence with EPYC. EPYC has exceeded all expectations bringing to the table great value, massive amounts of PCIe lanes, high core counts, insane memory bandwidth and capacities, and great performance. Multiple entities have pledged support for EPYC. These entities include Microsoft, Cisco, Amazon, Google, and others. The US Department of Energy has contracted AMD to build the world's most powerful supercomputer.
Infinity Fabric - AMD's Interconnect Solution
Raja Koduri had previously made the following statement as reported by PCGamesN:
'Infinity Fabric allows us to join different engines together on a die much easier than before,” he once said. “As well it enables some really low latency and high-bandwidth interconnects. This is important to tie together our different IPs together efficiently and quickly. If forms the basis of all of our future ASIC designs.'
Koduri's most recent statement was as follows:
'[AMD] has no memory or interconnect strategy that I know of, and the size of the developer ecosystem is tiny'
Considering Raja was part of the team that worked on Infinity Fabric, these statements might seem contradictory at first but are actually talking about different things. Its worth noting here that Raja is talking about the developer ecosystem and strategy as opposed to whether that particular technology exists. Naturally having worked on the tech, he would know it does.
Developers Backing AMD
AMD's ecosystem currently has support from game-related entities such Microsoft, and The Khronos Group, both of which have developed low-level APIs, DirectX 12 and Vulkan, based off of AMD's Mantle. Crytek has already begun to implement DirectX 12 and Vulkan into CRYENGINE 5.7 and EA has partnered with The Khronos Group.
However, AMD's ecosystem in the video and professional market is very limited. To date, users in the video industry swear by NVIDIA graphics cards due to the presence of the CUDA framework and AMD has almost no presence there.
How Will AMD Compete with Intel?
One thing is clear: AMD is positioning itself to compete with Intel. AMD will have Ryzen 3000 CPUs and Navi GPUs on shelves in the next few months. 2018 was AMD's first profitable year in quite a few years, and revenues will only grow from this current point in time. AMD has already stated Zen 2 has a yield of 70%, twice of Intel's top server chips, which is quite impressive.