This year's Computex 2024 saw AMD making huge strides in the industry, with the company's CEO, Dr. Lisa Su, showcasing the next generation of computing "prowess."
AMD's Computex 2024 Keynote Showed That The Company Means Business, AI-Focused Reveals Being Center Of Attractions
AMD's progress into this year hasn't been on par compared to its past, especially since the firm has been overshadowed in the AI hype since its arch-rival NVIDIA has managed to attract the market spotlight. The Computex 2024 was seen as an opportunity for Team Red to come back, and it's safe to say that the firm did deliver our expectations, as not only did they showcase their 2nd-generation AM5 platform, but several AI-focused developments were discussed. Here's a quick rundown of AMD's Computex 2024 keynote highlights.
AMD's Desktop Ryzen 9000 Series Finally Made Official, Zen 5 Magic Coupled Up With Significant Upgrades
What's an AMD keynote without a new desktop CPU lineup? That's precisely what AMD thought before unveiling the new Ryzen 9000 "Granite Ridge" series. The highlight of the SKUs is the revamped Zen5 architecture, which not only features higher performance but the platform has seen a cumulative increase in capabilities. Here are some of its upgrades:
- Improved branch prediction accuracy and latency
- Higher throughput with wider pipelines and vectors
- Deeper window size across the design for more parallelism
The Zen 5 architecture has CCDs based on TSMC's 4nm process, while the IOD (Input/Output Die) utilizes TSMC's 6nm process. In terms of improvements, the Zen 5 platform gives a 2x increase in areas such as the Instruction Bandwidth for the front-end instructions, data bandwidth (L2 to L1 and L1 to FP), and AI performance (AI & AVX512 Throughput).
Combining such changes, the Zen 5 has brought in decent generational upgrades with an IPC uplift of 15% and, in some cases, reaches up to 35% as well, so overall, AMD has brought in solid gains.
Moving on to the Ryzen 9000 series SKUs, AMD has brought in a total of four different models, with the flagship one being the Ryzen 9 9950X, coming down to the Ryzen 5 9600X. We won't go into depth about the specifications since we already have a detailed post about the new models. Still, the maximum configuration in the new models is 16 cores, 32 threads, two Zen 5 CCDs, and a single IOD. The initial models are targeted for the higher segment for now, but as the Zen 5 architecture gets more mature with time, we'll see budget variants drop in.
Pricing details about the new models weren't disclosed, but AMD says they are expected to be available by this July. The Ryzen 9000 series models are going to put out some serious competition in the markets, and we can't wait to try them out once they hit retail. Apart from that, the firm debuted the new X870 motherboard chipset as well for the Granite Ridge family, which you can check out here.
AMD Can't Seem To Let Go Of The AM4 Platform, Adds New High-End Models With Sweet Upgrades
AMD's commitment to its AM4 socket is surprising yet exciting. The firm values its old customers, and the new CPU models will elevate the platform even further. At Computex 2024, Team Red decided to unveil two new AM4-based processors, the Ryzen 9 5900XT and the Ryzen 7 5800XT, which come with bumped-up core counts and clock speeds.
Coming into specifications, the Ryzen 9 5900XT has 16 cores and 32 threads based on the Zen 3-core architecture, an increased 72 MB cache, and clock speeds up to 4.8 GHz in a TDP design of 105W. The other new AM4 variant, the Ryzen 7 5800XT, has 8 cores, 16 threads, 36 MB of cache, and a nominal clock speed bump of +100 MHz to 4.8 GHz. This chip is also rated at a 105W TDP. So, when speaking about AMD's stance with the new processors, it's only to promote AM4's longevity since the performance gaps with counterparts aren't that huge.
2 of 9
Interestingly, AMD compared its new AM4 CPUs with Intel's 13th-generation SKUs, such as the Core i7-13700K and the Core i5-13600KF, and the overall performance difference was around 10%- 15%, favoring Team Red. AMD is still paying attention to the almost decade-old platform, and its commitment to AM4 is unmatched.
AMD's New "Strix" AI-Focused APUs Unveiled, Interestingly Naming Scheme & Combine Latest Zen 5, RDNA 3.5 & XDNA 2 IPs
AMD, with this year's Computex, decided to jump into the AI PC bandwagon like others, with their new Strix APU family targeted towards cutting-edge AI and compute performance on the mobile platform. Team Red has named the lineup the "Ryzen AI 300" series, and while it is a fresh approach, consumers will have to keep up with processor names these days. The naming scheme has evolved in a way that AMD decided to include the keyword "AI", either for industry attention or to maintain relevancy in the rapidly growing AI segment.
The new Strix APUs succeeded Phoenix, and Hawk Point series, and Team Red decided to significantly change at every level, incorporating new IPs such as the Zen 5, RDNA 3.5, and XDNA 2, all targeted at battling against competitors, which have up till now managed to "steal" the spotlight from AMD, especially the likes of Qualcomm's Snapdragon X series.
We have discussed Zen 5 above, but the RDNA 3.5 is new. As the name suggests, it's a mere refresh of the original RDNA 3 lineup, so there's nothing much in terms of generational upgrades, except for the fact that the graphics platform is more optimized and comes with a higher number of compute units, which add up to 16 for the flagship part.
The XDNA 2 NPU, however, deserves a separate paragraph. Dedicated AI engines in processors have become the next norm in the markets, and what started with Intel's Meteor Lake has evolved into another race in the industry. AMD's XDNA 2 NPU is reported to offer up to 50 TOPS of AI power, which is massive and higher than Qualcomm's Snapdragon X Elite and also Intel's Lunar Lake.
Team Red might get a breakthrough here with XDNA 2, and with the integration of the upcoming Ryzen AI 300 series APUs into Microsoft's Copilot+ AI PCs, the adoption rates can massively soar.
Quickly going through the new APUs released, AMD announced two different models, the Ryzen AI 9 HX 370 and the Ryzen AI 9 365, both coming in with tremendous capabilities. Yet again, we won't dive much into specifications since you can view them in our detailed post here, but as far as the new variants are concerned, they do bring in serious computing power. However, the interesting fact is that the new Ryzen AI 300 models stack up against market offerings, and here's how the firm has compared the APUs with competitors. Overall, AMD looks to take the lead here.
In terms of availability, AMD states that its first Ryzen AI 300 "Strix" Copilot+ PCs will start rolling out in July 2024, and there will be over 100 platforms to choose from by leading manufacturers such as ACER, ASUS, HP, Lenovo, and MSI.
AMD's Workstation Arsenal Sees Inclusion Of New 5th Gen EPYC “Turin” CPUs & Radeon PRO W7900 Dual Slot GPU
At this year's Computex, AMD decided to integrate the Zen 5 magic into its latest workstation processors, the EPYC 9005 "Turin" series. These processors feature various chips, including compute, cloud, telecommunication, and edge-optimized variants. Team Red hasn't yet provided us with the lineup specifics, but the firm has showcased how the new Turin lineup will stack up against existing Intel offerings. For that, AMD compared a "128-core" variant of its EPYC 9005 lineup of processors against an Intel 5th Gen Xeon Emerald Rapids CPU, the Xeon Platinum 8592+.
Official slides show that Team Red managed to gain an impressive 2.5x to 5.4x gains in performance across various workloads. Since AMD decided to put the curtains on its EPYC 9005 models, we are unaware of the specific details, such as core configurations, for now, but we do have a list of preliminary specifications, and we have discussed them in our dedicated post here, so make sure to check them out.
Apart from the EPYC Turin, Team Red decided to introduce a new workstation GPU variant under its Radeon Pro series, a new offering that complies with the demands of the AI workstation segment. AMD unveiled the Radeon PRO W7900 workstation model, which comes with a 2-slot compact design and the same Navi 31 GPU with 96 Compute Units and 6144 cores based on the RDNA 3 GPU architecture.
The firm claims that their Radeon PRO W7900 offers up to 2.01x better perf per dollar in workstation-oriented workloads against NVIDIA's Ada 6000 GPUs. The firm also unveiled its newest ROCm 6.1.2 suite, which included more AI frameworks, support for Windows, and data parallelism.
The AMD EPYC Turin "5th Gen" family of servers is expected to debut this year after the formal launch of the Zen 5 and Zen 5C architectures for desktop and client PC platforms such as Granite Ridge and Strix Point. Conversely, the Radeon PRO W7900 will be available by 19th June with a SEP of $3,499 US.
AMD & The AI Industry, The New Instinct AI Accelerators Yet Another Attempt To Gain Market Interest
Team Red's entrance into the AI markets was undoubtedly delayed, and the firm hasn't received hype similar to that of competitors like NVIDIA. The company made strides with options such as the Instinct MI350X, which brought tremendous value onboard, but unfortunately, it couldn't attract much market interest.
However, AMD looks committed, as the firm has disclosed additions to its AI Instinct accelerator roadmap and has revealed that it will offer a new AI accelerator, either a refresh or a completely new product, each year, similar to NVIDIA's strategy.
What was the best announcement from AMD's Computex keynote?
For Computex, AMD first announced the "refresh" Instinct MI325X AI accelerator, which will use the same CDNA 3 architecture, 288 GB of HBM3E memory, 6 TB/s memory bandwidth, 1.3 PFLOPs of FP16, and 2.6 PFLOPs of FP8 compute performance. The firm also revealed a new "MI350" lineup of AI accelerators, which would be based on a 3nm process node, HBM3E memory,and next-gen CDNA 4 architecture.
Apart from that, the firm revealed an update to its Instinct AI roadmap, now including the cutting-edge Instinct MI400 AI accelerators, which will be based upon a future yet unannounced CDNA architecture. In the end, AMD provided an update on the recently reported UALink platform, a new high-performance, open, and scalable AI interconnect infrastructure developed by some of the big names in the industry.
Well, it looks like AMD has big plans for the future of the AI markets and we have only seen a glismpe of it in this year's Computex. The firm has shown tremendous commitment to rapidly evolving its Instinct lineup of accelerators to ensure market relevancy and competitiveness.
That's it from AMD for this year's Computex. The announcements were indeed long and huge, but that's where Team Red is the best. It revolutionizes every segment of the market through its unique and capable offerings. We look forward to how the above-mentioned products contribute to the betterment of the industry.
