Advanced Micro Devices (AMD, Financial) is rallying as the company unveiled the world’s first 7nm graphics processing unit (GPU) alongside the new 32 core Ryzen Threadripper processor at Computex Tapai 2018 through a live stream yesterday.
The next generation Vega GPU products will be based on GlobalFoundries’ 7nm technology and are expected to be launched during the second half of 2018. The new 32 core Threadripper CPU, however, will be based on a 12nm process technology and is set to debut during the third quarter of 2018. Furthermore, AMD continues to make strides on the server side as it will start sampling the 7nm EPYC – processors targeting data centers and servers – during the second half of 2018.
“At Computex 2018 we demonstrated how the strongest CPU and GPU product portfolio in the industry gets even stronger in the coming months,” said AMD President and CEO Dr. Lisa Su in a press release.
Investors are cheering on the news as the stock added 5% to the gains on Wednesday to reach a decade-high of $15.
Thread-Ripper takes the lead
Just when Intel (INTC, Financial) thought it finally has the lead in the core wars, AMD produced a 32 core processors from up the sleeve. Just hours before AMD’s announcement at Computex 2018, Intel demoed its 28-core processor running at 5 Gigahertz. To take the lead, AMD followed with a 32 core processor demonstration.
Advanced Micro Devices is also set to beat Intel to the market as Threadripper will be available during the third quarter of 2018 while Intel is eyeing a launch towards the end of the year. A report from AnandTech mentioned an August launch for the new 32 core processing behemoth from AMD.
However, it has yet to be seen how both the processors fair against each other. AMD refrained from showing benchmarks for its Threadripper during the demonstration while Intel showed an impressive Cinebench score – a software-based score that evaluates computing performance capabilities – of around 7000 for its 28 core processor.
The hesitation from AMD, given its history of relying heavily on benchmarks, indicates that the company might not have been able to beat Intel in synthetic benchmarks. Moreover, AMD didn’t disclose the clock speed of its new leading edge processor, which makes the comparison with Intel’s 28 core processor quite difficult.
Nonetheless, Advanced Micro Devices has the core-lead with a process technology that’s at par with Intel’s 28 core processor. Note that Intel’s core intensive processor in based on an advanced iteration of its 14nm process. To review, AMD will give Intel a tough time in enthusiasts’ market amid higher core count, improved process density and quick time-to-market of its new 32 core processor.
AMD is set to join the machine learning party
The high-performance computing market is expected to grow at 7% per year during the next five years or so, according to a research report from Research&Markets. Moreover, Nvidia’s (NVDA, Financial) data center revenue reached around $2 billion during the trailing 12 months, up more than 100% on a year-over-year basis. This indicates the growing potential of the data center GPU market.
The new 7nm Vega GPU is optimized for machine learning and is a part of Radeon Instinct – AMD’s line of GPU accelerators for artificial intelligence, high performance computing and deep learning applications. AMD has already stated sampling the product to initial customers. 7nm-based GPU from AMD will first appear in datacenter and server farms followed by professional applications, according to the company’s press release.
It is worth mentioning that 7nm Vega for machine learning applications will hit the market earlier than expected. Originally, AMD was planning to sample the new Vega by the end of 2018, with mass production slated for next year. Now, the chips are already in sampling phase, with production expected by the end of 2018.
Quick execution of AMD’s roadmap doesn’t bode well for counterparts like Nvidia who have yet to move to smaller manufacturing processes. Nvidia is expected to continue to produce its GPU at 12nm process, which is not as dense as the 7nm. Note that AMD claimed that its 7nm node will result in twice the density improvement leading to 2x power efficiency. Tirias Research, a provider of custom research and advisory on technologies, noted, “32GB of HBM2 will bring Vega to parity with NVIDIA’s high-end Tesla V100 datacenter GPUs.”
In plain English, memory expansion in the new Vega makes AMD a worthy competitor against Nvidia’s machine learning silicon. Given machine learning’s potential growth, AMD is set to benefit as it reaches parity with Nvidia’s machine learning products. The lead in process technology alongside memory bump will certainly improve performance and power dynamics for AMD on the GPU side, especially in machine learning.
What about servers?
The server success story is taking shape now as AMD showcased new developments. Cisco is using AMD’s EPYC server processors in one of its high density servers. Tencent is also offering EPYC for customers through its cloud deployments. The new one socket servers from Hewlett-Packard (HPQ) Enterprise for virtual workloads offer 27% lower cost per virtual machine compared to Intel’s offerings. The most interesting development is the sampling of 7nm EPYC, code named Rome, during the second half of 2018. AMD is set to launch 7nm server processors during 2019.
To review, the server success story seems intact. As data center players and cloud players are finally warning up to the idea of EPYC, AMD will certainly post revenue growth on the server side. Moreover, the launch of 7nm based EPYC will consolidate AMD’s position in servers going forward.
Takeaways
- AMD’s only weak link – high performance computing – is finally disappearing. The company has not been up to par with Nvidia in data center computing for the past couple of years. This allowed Nvidia to gain material revenue in the space. This, however, is set to change amid new Vega chips that are optimized for machine learning. Thanks to 7nm process technology and high memory bandwidth, AMD will probably steal some market share from Nvidia during the next year.
- Enthusiasts’ market will also reward AMD due to early availability of its 32 core processors, which will probably outperform Intel in multi-threaded performance.
- Server processors are turning out to be AMD’s strongest suit as the 1-Socket (1-S) strategy starts to play out. New 7nm EPYC server processors, that are slated to launch next year, will further boost AMD’s revenue as Intel has yet to release its 10nm processors.
- From an investor's perspective, as Advanced Micro Devices gains ground in computing, graphics and servers going forward, market share will be the primary catalyst for share price growth.
Disclosure: I have no positions in any stocks mentioned and have no plans to initiate any positions within the next 72 hours.
