AMD Blitzes CES With Cache-Fueled Zen 4 CPUs, A Ryzen And Radeon 7000 Mobile Arsenal And AI
There’s a silicon slugfest going on right now between AMD (Advanced Micro Devices) and its chief rivals Intel and NVIDIA. If it were not for the continued execution machine that AMD has proven itself to be under the leadership of its CEO, Dr. Lisa Su, one might think this stiff competition is an extremely tall order. But AMD is apparently oblivious, because the company continues a relentless pace of innovation. In fact, on Wednesday evening during Dr. Su’s CES 2023 keynote, there was so much to unpack from the data center to the desktop and mobile client, that I’m kind of breaking a sweat at the heavy lifting required for even a high level synopsis. That’s OK though, I’m buckling my chinstrap and dropping a shoulder.
One aspect of Dr. Su’s stage presentation that impressed me were some of the partner endorsements, including Microsoft, which is an ecosystem partner that doesn’t casually appear on stage with another brand. However, seeing Panos Panay step in to share the limelight with AMD on the topic of AI, was just a feather in the company’s cap. But I digress; let’s limber-up and dive in.
Ryzen 7000 And Radeon RX 7000 Mobile Power A New Breed Of All-AMD Laptops With XDNA AI
A moment that laptop OEMs were apparently waiting for as well, was the announcement of AMD Ryzen 7000 and Radeon RX 7000 processors and graphics chips for laptops. Let’s start with the CPUs, and though its decoder ring is a bit of an eye chart here on Forbes, AMD unveiled a big family of chips based on its new Zen 4 and legacy Zen 3/3+ microarchitectures, some of which have the company’s new RDNA 3 graphics cores board, which based on the same tech as its new Radeon RX 7900 series discrete graphics cards.
As you can (hopefully) see above, AMD has a new naming convention that is a bit complex but transparent and detailed. The 7 designates the model year and family, and a 4, 3 or 2 in the third digit location designates the CPU core architecture as Zen 4, Zen 3 or even Zen 2 for the company’s entry level chips. So Ryzen 7030 and 7020 series CPUs are essentially scaled back and optimized legacy chips, while Ryzen 7040 series CPUs are where the new blood is.
Dragon Range Coupled With Discrete AMD Radeon RX 7000
AMD’s Ryzen 7045, aka Dragon Range, will be comprised of 16-core Zen 4 offerings with previous-gen AMD RDNA 2 (Navi 2X) graphics cores on board and support for DDR5 memory. Laptops powered by this platform will be larger enthusiast gaming or beastly content creation mobile workstation type machines. These machines will likely be equipped with discrete graphics chips as well, including AMD’s new Radeon RX 7000M series, that I invite you to check out at HotHardware for a full deep-dive.
AMD’s Radeon RX 7000 mobile family is comprised of 4 new mobile GPU offerings, known as the Radeon RX 7600M XT, RX 7600M, RX 7700S and RX 7600S, the latter of which are power-optimized for thinner and lighter gaming notebooks. The company is comparing its top-end Radeon RX 7600M XT mobile GPU to NVIDIA GeForce RTX 3060 desktop-class performance, which is definitely stronger than the competition’s mobile RTX 3060 offering. However, where it will line-up versus NVIDIA’s higher-end RTX 30 or just announced GeForce RTX 40 series mobile offering remains to be seen.
AMD Phoenix Infused With RDNA 3 And Xilinx DNA
The new AMD Ryzen 7040 Series, however, is designed for what AMD is calling the “Elite Ultrathin” laptop segment, and also sports Zen 4 CPU architecture with 8-core designs, along with up to 12-core on-chip AMD RNDA 3 graphics (Navi 33) and something new called AMD XDNA. This new AMD mobile CPU design is the most exciting to me, personally, both because of its strong on-board RDNA 3 GPU engine, as well as its Xilinx FPGA-based AI accelerator. This chip marks the first true AMD integration of Xilinx FPGA (Field Programmable Gate Array) technology with its CPU and GPU tech, on a single monolithic die, built on TSMC’s 4N process node.
So, in addition to strong GPU cores for content creation and some lighter-duty gaming, AMD’s Phoenix Ryzen 7040 series will have a more efficient, dedicated AI engine for accelerating workloads like machine vision and collaboration. Think background noise elimination on your Zoom calls, as well as head tracking and framing, studio background blur camera effects, etc. Intel has been driving the adoption of on-board AI accelerators for a while, but this is the first time a reconfigurable block of logic has come to a consumer laptop client device to my knowledge. The technology’s potential down the road, as new use cases and workloads evolve, will be very interesting to watch.
AMD’s Desktop Ryzen Zen 4 Onslaught Continues With 3D V-Cache And A Value Play For Gamers
Almost everybody in the industry saw this coming — and it was even hinted at in a recent interview I had with my colleagues and former AMD Director Of Technical Marketing, Robert Hallock on the initial launch of the company’s Zen 4-based Ryzen 7000 desktop chips — but yes, the day is finally here where AMD is unveiling new Ryzen 7000 series enthusiast processors strapped with sweet 3D V-Cache that helps these chips game so well. As with AMD’s previous gen Ryzen 7 5800X3D, the new Ryzen 7 7800X3D has been outfitted with 64MB 3D V-Cache bolted-on for a total of 96MB of shared L3 cache (104MB with L2 and L3 combined). What this affords is the ability for the CPU to go off-chip less often to system memory, with its cache complement maintaining a larger scratch-pad for frequently access data and metadata that game engines so frequently crave.
This additional cache memory pool required AMD to scale back clock speed by 400 MHz to mitigate power and thermals, but the company proclaims the Ryzen 7 7800X3D is the new “Ultimate Gaming Processor,” with significant gains of up to 30% over its previous gen Ryzen 7 5800X3D Zen 3 chip.
But that wasn’t the only 3D V-Cache infused CPU that AMD unveiled, the company also brought a 12-core Ryzen 9 7900X3D to the fold, as well as a beastly Ryzen 9 7950X3D 16-core chip, with 3D V-Cache strapped to one of its 8-core CCDs (Core Chiplet Die). This may sounds a touch unbalanced but AMD notes that it has worked with Microsoft on its scheduler functionality to accommodate for this. In short, apps like games that benefit from the added cache block will be able to take advantage of it, while those that don’t will run on the other CCD at higher clock speeds, or both at nominal clock speeds. As you can see in the slide above, the 12-core Ryzen 9 7900X3D and the 16-core Ryzen 9 7950X3D have top-end clocks of 5.6GHz and 5.7GHz, respectively.
In addition to this trio of high-end chips (pricing hasn’t been revealed yet but I wouldn’t expect these premium enthusiast CPUs to be cheap), AMD also revealed a trio of mainstream Zen 4-based non-X branded Ryzen 7000 chips, called Ryzen 9 7900, Ryzen 7 7700 and Ryzen 5 7600. These are completely unlocked (for overclocking), lower power 65 Watt CPUs that drop in at great price points, and will be available starting 1/10/23.
For The Data Center, Instinct MI300, Big And Strapped With Many Chiplets This Way Comes
There’s little question AI (Artificial Intelligence) or machine learning, is the new frontier for many industries, from big data analytics and supercomputing, to healthcare, science and finance. AMD of course has been hard at work for years developing AI acceleration engines that are built on its core CPU, GPU and FPGA technologies, and in Ms. Su’s keynote we were treated to the culmination of thousands of ours of engineering effort that have resulted in what the company has unveiled as its new Instinct MI300 AI Accelerator.
AMD’s new Instinct MI300 will be bound for data center servers tasked with some of the most intense machine learning workloads and is built on 24 Zen 4 AMD EPYC server CPU cores and an undisclosed number of its latest gen CDNA 3 GPU cores (similar to RDNA 3 for gaming but optimized for GPU-compute workloads). This massive slab of silicon is comprised of 146 billion transistors and is also equipped with 128GB of HBM3 (High Bandwidth Memory).
The company is claiming an 8X performance lift in AI workloads versus its previous generation Instinct MI250 accelerator, and a 5X lift in performance-per-watt, which if realized will be great for AMD’s AI accelerator TCO metrics as well as sustainability through better power efficiency.
Wrapping things up here isn’t so easy, and I still have a lot to digest personally with all of AMD’s disclosures during Dr. Lisa Su’s CES 2023 keynote address. However, there’s one thing for sure and its that this company isn’t taking its foot off the gas on any of its core technologies and market segments. In addition, the integration and rollout of its acquisition of Xilinx is progressing along seemingly very well, not only with a solid combined go-to-market strategy, but new silicon innovations on multiple fronts.